Improved hybrid de novo genome assembly of domesticated apple (Malus x domestica)

Background

Domesticated apple (Malus × domestica Borkh) is a popular temperate fruit with high nutrient levels and diverse flavors. In 2012, global apple production accounted for at least one tenth of all harvested fruits. A high-quality apple genome assembly is crucial for the selection and breeding of new cultivars. Currently, a single reference genome is available for apple, assembled from 16.9 × genome coverage short reads via Sanger and 454 sequencing technologies. Although a useful resource, this assembly covers only ~89 % of the non-repetitive portion of the genome, and has a relatively short (16.7 kb) contig N50 length. These downsides make it difficult to apply this reference in transcriptive or whole-genome re-sequencing analyses.

Findings

Here we present an improved hybrid de novo genomic assembly of apple (Golden Delicious), which was obtained from 76 Gb (~102 × genome coverage) Illumina HiSeq data and 21.7 Gb (~29 × genome coverage) PacBio data. The final draft genome is approximately 632.4 Mb, representing ~ 90 % of the estimated genome. The contig N50 size is 111,619 bp, representing a 7 fold improvement. Further annotation analyses predicted 53,922 protein-coding genes and 2,765 non-coding RNA genes.

Conclusions

The new apple genome assembly will serve as a valuable resource for investigating complex apple traits at the genomic level. It is not only suitable for genome editing and gene cloning, but also for RNA-seq and whole-genome re-sequencing studies.

Whole-genome shotgun sequencing of ‘Golden Delicious’ apple on the Illumina platform

Genomic DNA was extracted from leaf tissues of a single ‘Golden Delicious’ apple tree with the GenElute™ Plant Genomic DNA Miniprep Kit (Sigma-Aldrich; St. Louis, USA). Paired-end libraries with insert sizes ranging from 350–500 bp were constructed with Next UltraTM DNA Library Prep Kit for Illumina (NEB; USA) according to the manufacturer’s instructions. These libraries were sequenced on an Illumina HiSeq 4000 platform (Illumina; CA, USA) using the PE-150 module [1], and yielded about 86 Gb of raw data. These data were then subjected to filtering to remove: (1) reads in which more than 5 % of bases were N or poly-A; (2) reads in which more than 30 bases were of low quality; (3) reads with adapter contamination; (4) reads shorter than 30 bp; and (5) PCR duplicates. These steps yielded a clean sequence of ~76 GB, representing about 102 × genome coverage (Additional file 1: Table S1). De novo assembly was performed with with SOAPec_v2.01 [2] using default parameters.

Single-molecule long reads from the PacBio RS II platform (Pacific Biosciences, USA) were used to assist the subsequent de novo genome assembly [3]. In brief, 15 μg of sheared DNA was used to construct five SMRT Bell libraries with an insert size of 17 kb. The libraries were then sequenced in 20 single-molecule real-time DNA sequencing cells using the P6 polymerase/C4 chemistry combination, and a data collection time of 240 min per cell. The sequencing produced about 21.7 Gb data, consisting of 2,759,937 reads with an average read length of 7,863 bp (Additional file 1: Figure S1). The polymerase read N50 length after single passing was around 16.6 kb, and the polymerase read quality was greater than 82.4 % (Additional file 1: Table S1).

Quality-filtered reads from the Illumina platform were subjected to 23-mer frequency distribution analysis with Jellyfish [4]. Analysis parameters were set at -k 23, and the final result was plotted as a frequency graph (Additional file 1: Figure S2). Two distinctive modes were observed from the distribution curve: the higher peak at a depth of 88 reflected the high heterozygosity of the apple genome; the lower peak provided a peak depth of 179 for the estimation of its genome size. Based on the total number of k-mers (125,428,662,216), the apple genome size was calculated to be approximately 701 Mb, using the following formula: genome size = k-mer_Number/Peak_Depth.

A hybrid genome assembly pipeline was used to overcome challenges posed by heterozygous apple genome (Additional file 1: Figure S3). An Illumina-based de novo genome assembly was first generated using Platanus [2], yielding a total length of 1.05 Gb, with a contig N50 length of 534 bp. Then, all PacBio RS reads were used in the hybrid assembly process via the DBG2OLC [5] pipeline with the following parameters: LD10, MinLen 200, KmerCovTh 2, MinOverlap 10, AdaptiveTh 0.001, and RemoveChimera 1. This led to a preliminary apple genome assembly of 632.4 Mb with a contig N50 size of 111,619 bp, representing ~90 % of the estimated apple genome (701 Mb). The contig N50 size represents a ~6.9 fold improvement in length from the previously reported 16.1 kb [6]. These improvements were made possible by introducing the long-read sequencing strategy (Additional file 1: Figure S4), which increased the sequencing precision of repeats.

CEGMA was used to evaluate the quality of the final assembly with a set of 248 ultra-conserved core eukaryotic genes [7]. Comparison analysis showed that 231 of 248 genes could be fully annotated (93.15 % completeness, see Table 1), and 243 of 248 genes met the criteria for partial annotation (97.98 % completeness). Using the same evaluation parameters, the completeness of the ‘Golden Delicious’ apple genome assembly v1.0 by Velasco et al. [6] was also evaluated, and a completeness of 88.71 % was obtained (220 of 248 genes could be fully annotated, see Additional file 1: Table S3). This benchmark further demonstrates the improved quality of the genome assembly reported herein.

Tandem Repeat Finder [8] was used to identify tandem repeats in the ‘Golden Delicious’ apple genome. RepeatMasker and RepeatProteinMasker [9] were used against Repbase [10] to identify known transposable element repeats. In addition, RepeatModeler [11] and LTR FINDER [12] were used to identify de novo evolved repeats. The combined results show that the total length of repeated sequences is about 382 Mb, accounting for ~60 % of the ‘Golden Delicious’ apple genome assembly (Additional file 1: Table S4).

Genes for the ‘Golden Delicious’ genome were annotated using multiple methods, including transcriptome-based predictions, de novo predictions, and homology-based predictions. For de novo predictions, Augustus [13], GenScan [14], glimmerHMM [15] and SNAP [16] analysis were performed on the repeat-masked genome, with parameters trained from Arabidopsis thaliana. Partial sequences and genes with fewer than 150 bp of coding sequence length were removed. Predicted protein sequences from B. oleracea, G. max, O. sativa, P. mume, P. trichocarpa, P. persica, P. communis, V. vinifera, and Z. mays were used (Phytozome v10.3 [17]) for homology-based predictions. First, query sequences were subjected to TBLASTN analysis with an Expect (E)-value cutoff of 1 e^-5. BLAST hits corresponding to reference proteins were concatenated by Solar software (The Beijing Genomics Institute (BGI) development), and low-quality records were removed. The genomic sequence of each reference protein was extended upstream and downstream by 2,000 bp to represent a protein-coding region. GeneWise software [18] was used to predict gene structure contained in each protein region. For transcriptome-based predictions, RNA from three structures (leaves, flowers, and stems) was isolated and RNA-seq data (NCBI SRP067376) were used for gene annotation, processed by Tophat and Cufflinks [19]. The homology, de novo and transcriptomic gene sets were merged to form a comprehensive and non-redundant reference gene set using EVidenceModeler [20] software. Our analysis indicates that the ‘Golden Delicious’ apple genome contains 53,922 protein-coding genes (Table 2). This is slightly fewer than the previous prediction of 57,386 genes [6]. Approximately 60 % of predicted genes were represented in our transcriptome data.

tRNAscan-SE (version 1.31) [21] software with default parameters for eukaryotes was used for tRNA annotation. rRNA annotation was based on homology with rRNAs from several diverse higher plant species (not shown), using BLASTN with ‘E-value = 1e^-5’. miRNA and snRNA genes were predicted by INFERNAL software [22] using the Rfam database (release 11.0) [23]. The final results included 321 miRNAs, 274 tRNAs, 605 rRNAs, and 480 snRNAs (Additional file 1: Table S5).

Sequencing reads of each sequencing library and RNA-seq data have been deposited at NCBI with the project ID SRP067376. Supporting data are also available in the GigaScience database, GigaDB [24]. All supplementary figures and tables are provided in Additional file 1.

CDS, coding DNA sequence; NCBI, National Center for Biotechnology Information

ᅟ

Funding

This work was support by the National Science Foundation of China (31572106) and QG is supported by funding from the Thousand Talents Plan.

Authors’ contributions

FM, QG, YD and WW designed the study. JZ assembled the genome. XL, YX and NW extracted DNA, LK constructed libraries, LW and YY analyzed the data. QG, JZ XL, and SN wrote the manuscript. All authors read and approved the final manuscript.

Competing interests

JZ is a shareholder of Agri-biotech Lab Company (Kunming, Yunnan).

Affiliations

1. State Key Laboratory of Crop Stress Biology for Arid Areas, College of Horticulture, Northwest A&F University, Yangling, 712100, China
   • Xuewei Li
   • , Yinpeng Xie
   • , Liping Wang
   • , Yan Yan
   • , Na Wang
   • , Jidi Xu
   • , Cuiying Li
   • , Fengwang Ma
   •  & Qingmei Guan
2. State Key Laboratory of Genetic Resources and Evolution, Kunming Institute of Zoology, Chinese Academy of Sciences, Kunming, 650223, China
   • Ling Kui
   •  & Wen Wang
3. Agri-biotech-lab Company, Kunming, 650220, China
   • Jing Zhang
4. Department of Horticulture, Michigan State University, East Lansing, MI, 48824, USA
   • Steve van Nocker
5. College of Biological big data, Yunnan Agriculture University, Kunming, 650504, China
   • Yang Dong
6. Faculty of Life Science and Technology, Kunming University of Science and Technology, Kunming, 650500, China
   • Yang Dong

Corresponding authors

Correspondence to Yang Dong or Fengwang Ma or Qingmei Guan.

Reprints and Permissions