Construction of Pará rubber tree genome and multi-transcriptome database accelerates rubber researches

Background Natural rubber is an economically important material. Currently the Pará rubber tree, Hevea brasiliensis is the main commercial source. Little is known about rubber biosynthesis at the molecular level. Next-generation sequencing (NGS) technologies brought draft genomes of three rubber cultivars and a variety of RNA sequencing (RNA-seq) data. However, no current genome or transcriptome databases (DB) are organized by gene. Results A gene-oriented database is a valuable support for rubber research. Based on our original draft genome sequence of H. brasiliensis RRIM600, we constructed a rubber tree genome and transcriptome DB. Our DB provides genome information including gene functional annotations and multi-transcriptome data of RNA-seq, full-length cDNAs including PacBio Isoform sequencing (Iso-Seq), ESTs and genome wide transcription start sites (TSSs) derived from CAGE technology. Using our original and publically available RNA-seq data, we calculated co-expressed genes for identifying functionally related gene sets and/or genes regulated by the same transcription factor (TF). Users can access multi-transcriptome data through both a gene-oriented web page and a genome browser. For the gene searching system, we provide keyword search, sequence homology search and gene expression search; users can also select their expression threshold easily. Conclusion The rubber genome and transcriptome DB provides rubber tree genome sequence and multi-transcriptomics data. This DB is useful for comprehensive understanding of the rubber transcriptome. This will assist both industrial and academic researchers for rubber and economically important close relatives such as R. communis, M. esculenta and J. curcas. The Rubber Transcriptome DB release 2017.03 is accessible at http://matsui-lab.riken.jp/rubber/. Electronic supplementary material The online version of this article (10.1186/s12864-017-4333-y) contains supplementary material, which is available to authorized users.


Background
Natural rubber is an indispensable material for many industrial applications such as in tires and medical devices [1]. Although more than 2500 plants produce latex, currently the Pará rubber tree (Hevea brasiliensis Muell. Arg.) is the only main commercial source for rubber production [2]. Even compared with petro-chemically synthesized rubber, natural rubber has advantages in adhesion, elasticity and durability.
Natural rubber is produced from specialized differentiated cells called laticifer cells in the outer layer of bark. Rubber latex is composed of rubber serum and rubber particles where cis-1,4-polyisoprene biosynthesis occurs. The molecular mechanism of rubber production is not well understood. Additionally, disease-resistance is an important trait to identify for rubber research and breeding. Rubber trees are susceptible to several fungal infections including South American leaf blight (Microcyclus ulei) and different cultivars show different sensitivity [3].
To accelerate molecular biological research in H. brasiliensis, we first determined its draft genome sequence and annotated 84,443 protein-coding genes [4]. After the rubber tree genome was determined, transcriptome data is important to identify gene expression level and precise gene structure, such as transcription start sites (TSS) and isoforms. Currently, RNA-seq is the most widely used transcriptome technology, providing gene expression levels in many conditions and tissues. For the rubber tree, latex where cis-1,4-polyisoprene biosynthesis occurs is the key tissue to understand the mechanism of rubber production. Many researchers determined gene expression of latex in different cultivars or conditions [5][6][7][8][9]. Although RNA-seq is a powerful tool to know gene expression, it is difficult to predict full-length splice isoforms and TSSs accurately. For better understanding of transcription in rubber tree, we constructed fulllength cDNA libraries and determined their sequence with Sanger and Illumina [10]. Pootakham et al. released Pacific Biosciences (PacBio) Isoform sequencing (Iso-Seq), single-molecule real-time long-read isoform sequencing in BPM24 cultivar [11].
When we predict 5′-end with EST and/or RNA-seq, we tend to predict the longest TSSs instead of major expressed TSSs [12]. To know major expressed TSS in different tissues, we applied cap analysis gene Fig. 1 The structure of rubber transcriptome database. Gene-oriented DB has functional annotation and multi-transcriptome data. Based on 11 RNA-seq data, we calculated co-expressed genes and their networks. All transcriptome data are available from JBrowse, a fast visualization genome browser. Three types of search systems are also available. See Fig. 2 for the details of the gene expression search  We use the JBrowse fast and embeddable genome browser. Sequence information is also available from JBrowse expression (CAGE) method that captures the 5′ end of the transcribed and capped mRNAs. CAGE provided us with a genome wide single base-pair resolution map of TSSs [13]. One of the main difficulties for gene annotation in non-model plants is the number of genes with unknown functions. In our case, 22,991 genes are rubber specific and functionally unknown and 33,213 genes have homologous sequences but are still functionally unknown. To overcome the problem, we carried out co-expressed analysis to predict functionally related gene groups. We can expect genes regulated by the same transcription factor (TF) or genes involved in the same biological pathway to show a similar expression pattern. There are many co-expression databases in plant [12,14]. In this DB, we show the top 20 similarly expressed genes in rubber.

Construction, content and utility
General structure and content of rubber Transcriptome DB Based on our original draft genome sequence of H. barasiliensis RRIM 600, we predicted 84,443 proteincoding genes [4]. Users can access functional annotation and three kinds of transcriptome data through web pages organized by gene. The basic structure of our DB is summarized in Fig. 1. Users can access functional annotation and links to original databases; NCBI protein, KEGG, UniProt (both Swiss-Prot and TrEMBLE) and Gene Ontology [15][16][17][18]. We also provide multitranscriptome data of our original full length cDNA (FL-cDNA), RNA-seq and CAGE, and publicly available ESTs, RNA-seq and Iso-Seq (Table 1) [4,10,11,[19][20][21][22]. Since FL-cDNA sequences help to improve predicted gene structures, we constructed two FL-cDNA libraries and obtained c.a. 20,000 clones [10]. Our DB also includes public Iso-Seq and ESTs to capture various isoforms [11,[19][20][21]. Using RNA-seq technology, we aimed to reveal expression features of natural rubber biosynthetic genes. We previously obtained latex and non-latex (leaf, bark and petiole) in RRIM 600 genome cultivar and latex of other cultivars (RRIM 900 and PB350). Additionally, we downloaded latex and non-latex (bark and leaf ) in RRIM 928 and latex under normal and stressed condition in RRII 105 cultivars. All data were re-analyzed with the same protocol [10] and descriptions of all transcriptomic data were summarized in additional file 1 (Table S1). All our original data can be downloaded from our web site.

Data search system
We provide three types of searching systems: keyword search, blast homology search and gene expression search. Since functional annotation of rubber genes is still limited, keyword search and homology search are not enough. We prepared gene expression search so that users can easily change gene expression threshold for samples and retrieve gene sets such as latex specific, for example. In our system, users can select gene expression value of FPKM (Fragments Per Kilobase Million) or select fold-change and obtain gene sets according users' specific demands (Fig. 2).

Genome browser
We visualized our multi-transcriptome data in a genome browser (Fig. 3). We made links for the genome browser on each gene web page. Users can visualize eleven RNAseq, two FL-cDNA and three CAGE data and compare tissue specificity easily.

Co-expressed genes and their network view
Genes that are regulated by the same transcription factors or that work on the same biological pathways often show similar expression patterns. To suggest candidate genes that may have similar biological function with a user's genes of interest, we calculated the top 20 similar expressed genes using our original and publically available 11 samples (RRIM600 cultivar: latex, bark, leaf and petiole, RRIM901 cultivar: latex, PB350 cultivar: latex [10], RRIM928 cultivar: latex, bark and leaf (SRP022257), RRIM105 latex: control latex and stressed latex (SRP017288)). To visualize the co-expressed gene network (Fig. 4), we used Gviz and highlighted transcription factors (Fig. 5).

Rubber biosynthetic pathway search
The natural rubber biosynthetic pathway is not fully understood, especially regarding the polymerization process with isopentenyl diphosphates (IPPs). To clear known factors and their paralogous genes, we prepared a natural rubber biosynthetic pathway (Fig. 6). Users can easily understand the whole picture of the pathway and access gene information.  [23]. Gene expression search was built with the Shiny package in R (https://shiny.rstudio.com/). We configured Shiny server to run multiple Shiny processes. Gene description pages are generated as static web pages.

Discussion
Recently, RNA-seq is a powerful tool and the most widely spread transcriptome technology. However, RNAseq is not suitable for predicting TSSs and full-length splice isoforms accurately. To obtain an entire picture of the rubber transcriptome, it is important to integrate multi-transcriptome data. In this DB, we integrated three types of transcriptome data: RNA-seq, FL-cDNA and CAGE. FL-cDNA provide us precise gene structures including findings of novel genes and novel alternative isoforms. To enrich the quality and quantity of splicing information it is necessary to know the correct protein sequence and estimate the protein function. Iso-Seq, PacBio single molecule long-read, is a powerful technology for this purpose. To know the genome-wide TSSs, we carried out CAGE and obtained precise TSSs in latex, leaf and bark. With CAGE data, we can observe tissue-dependent alternative TSSs, multiple TSSs in a gene and variations of TSS, such as strictly determined TSS in a base or others. In latex biosynthetic pathway, 1-deoxy-D-xylulose-5-phosphate synthase gene and pyruvate dehydrogenase gene showed the different TSS patterns between samples. Precise TSSs assist to find consensus sequences or motif sequences for transcription factors and their regulations.
As a next target, we plan to expand cis-elements that are predicted after co-expressed gene set.

Conclusions
We have developed the rubber tree genome and transcriptome database, a comprehensive and searchable database of economically important plant, H. brasiliensis. To assist researchers and breeders in using the H. brasiliensis genome, we prepared a simple and user-friendly interface. We believe this database assists both industrial and academic researchers for rubber and important industrial close relatives such as M. esculenta, R. communis and J. curcas.