De novo transcriptome assembly and characterization of nine tissues of Lonicera japonica to identify potential candidate genes involved in chlorogenic acid, luteolosides, and secoiridoid biosynthesis pathways

In order to achieve comprehensive representation and characterization of L. japonica transcriptome, we performed RNA-seq-based deep sequencing for nine tissues, namely, shoot apex, stem, leaf-1 (youngest leaf), leaf-2 (second leaf), leaf-3 (mature leaf), green floral bud, white floral bud, white flower, and yellow flower (Fig. 1). Total RNA was extracted from all nine tissues, and quality was assessed using a bioanalyzer. RNA samples with RIN (RNA integrity number) over 8 were selected for mRNA preparation, fragmentation, cDNA synthesis, and library preparation for an RNA-seq experiment. Each library, thus prepared, was sequenced using the Illumina HiSeq™ 2000 platform, yielding a total of 120 M paired-end reads with 101 bps as average sequence length.

Raw reads generated from Illumina HiSeq™ were pre-processed using the Trimmomatic program [42] for the removal of adaptor sequences, reads with a sequence length <500, low quality and ambiguous reads, yielding over 110 M paired-end clean reads, or 22 Gbps (base pairs) of reads in total (Table S1). Mean phred score, which serves as a bench mark for assessing the quality of the sequenced reads, were >36 across all nine tissues of Lonicera japonica; <1 % of raw reads were dropped by the Trimmomatic program based on low-quality reads or being adaptor sequences. This indicates that our RNA-seq data was of high quality, and adequate for de novo transcriptome assembly. The study overview is shown in supplementary Fig. S1.

Success for any transcriptome-based study, especially when complete genome sequences are not available, depends upon the completeness and quality of an assembled transcriptome, which in turn depends upon the assembler program, and assembly parameters, particularly kmer size [43, 44]. Although, most de novo transcriptome assemblers rely on partitioning the sequence data into many individual de Bruijin graphs based on certain kmer size, where kmers are shorter than the reads length, output from each assembler is very different. Comparison of several popular transcriptome assemblers using different datasets revealed that none of them consistently performed to generate best assemblies [45]. Recent studies, therefore, have proposed the use of multiple assemblers in order to maximize diversity of the de novo assembled transcripts [46‐48]. Therefore, in order to generate a complete de novo transcriptome assembly for L. japonica, we used three popular assemblers, namely, SOAPdenovo-Trans [49], Trinity [50], and CLC Genomics Workbench 8.0.3 (https://​www.​qiagenbioinforma​tics.​com/​).

For SOAPdenovo-Trans, we performed de novo transcriptome assemblies using different kmer values, namely, 31, 41, 51, 63, 71, and 91, resulting in six different transcriptome assemblies. Analyzing assembly stats for six de novo transcriptome assemblies generated using SOAPdenovo-Trans revealed kmer 31-based transcriptome assembly as the best, with 95,718,128 bps reads incorporated into 120,798 unigenes, with an N50 value of 1420 and the number of unigenes with sequence length >500 bps as 52,789 (43.7 %) (Table 1). The Trinity program-based de novo transcriptome assembly using default kmer (kmer as 25) resulted in the incorporation of 309,874,152 bps into 351,356 unigenes, with an N50 value of 1480 bps and the number of unigenes with sequence length >500 bps as 175,121 (49.8 %) (Table 1). For the CLC Genomics Workbench, we used default kmer size to perform de novo transcriptome assembly, resulting in assembly of 88,253,035 bps into 132,053 unigenes, with an N50 value of 975 and the number of unigenes with sequence length >500 bps as 49,831 (37.7 %) (Table 1).

The transcriptome assemblies, thus obtained from SOAPdenovo-Trans (kmer31), Trinity, and CLC Genomics Workbench were combined, and sequence redundancies were removed using the CD-HIT-EST program [51, 52], resulting in a final de novo transcriptome assembly for L. japonica by incorporating 220,651,304 bps into 243,185 unigenes, with an N50 value of 1561 and the number of unigenes with sequence length >500 bps as 122,493 (50.4 %) (Table 1). Comparison of transcriptome assemblies resulting from individual assemblers, or one resulting by combining the output of three assemblers showed an advantage for this approach of combining multiple assemblers, as we observed a significant gain in N50 values, average unigene length, mean unigene length, and percentage of sequences with a length >500 bps for the combined assembly. The guanine-cytosine (GC) % and length distribution for resultant transcriptome assembly for L. japonica is shown in Fig. S2a, b, respectively, with average GC % for all unigenes being 40.82 %, while 10,374 unigenes had a sequence length >3000 bps. The resultant de novo transcriptome assembly for L. japonica was improved compared to earlier published transcriptome assemblies, with a significant increase in N50 value, and overall contig length distribution.

The unigene sequences derived from L. japonica transcriptome assembly were subjected to further characterization. A Blastx program-based [53, 54] homology search was performed for L. japonica unigenes against the NCBI non-redundant (nr) protein database (http://​www.​ncbi.​nlm.​nih.​gov; formatted on Oct 2015) using an E value cut-off of <10⁻⁵, and the maximum number of allowed hits for each query was limited to 20. The top hit for each query sequence was used for the transcriptome annotation, and subsequent analysis and characterization. A Blastx-based homology search for L. japonica resulted in the annotation of 99,938 (41.1 %) unigenes (Table S2), while 143,251 unigenes had no significant sequence homology against the NCBI-nr database (Fig. S3a). Blastx results were used to extract associated gene ontology (GO) terms, to assign an enzyme commission (EC) number, and associated Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. A total of 91,745 unigenes were assigned to at least one GO term or KEGG pathway information (Fig. S3a). An E-value distribution plot for unigenes with a blast hit showed >89 % of aligned sequences having significant sequence homology against the NCBI-nr database (Fig. S3b). Sequence similarity distribution analysis for sequences with a blast hit showed 65,213 (65.23 %) unigenes having sequence similarity >75 % (Fig. S3c). These results, therefore, suggest that the annotation unigenes from Blastx results can be used reliably for further functional characterization.

Top-hit species distribution analysis for unigenes showed >85 % of all annotated transcripts having high sequence similarity against six plant species, namely, Vitis vinifera, Populus trichocarpa, Ricinus communis, Glycine max, Medicago truncatula, and Arabidopsis thaliana (Fig. S3d). Within these six plant species, L. japonica transcriptome assembly showed highest similarity to Vitis vinifera, with 50 % of annotated top-hit unigenes being derived from it. Top-hit species distribution results were similar to previous reports, where >50 % of annotated transcripts were assigned to Vitis vinifera [6].

GO-based functional classification for L. japonica transcriptome assembly was performed using the Blast2GO program v 3.0 [55], resulting in a total of 178 GO categories being assigned to the annotated unigenes in three broad categories, namely, biological process (89), molecular function (53), and cellular component (36) (Table S3). GO distribution analysis by level 3, and GO level distribution for three broad categories for L. japonica transcriptome assembly are shown in Fig. 2a, b. Several metabolic processes, such as organic substance metabolic process, primary metabolic process, cellular metabolic process, biosynthetic process, and nitrogen compound metabolic process were the top five GO terms being enriched within L. japonica transcriptome. For the molecular function category, GO terms corresponding to heterocyclic compound binding, organic cyclic compound binding, transferase activity, small molecule binding, and hydrolase activity were the top five processes being enriched. For the cellular component category, GO categories corresponding to intracellular, intracellular part, intracellular organelle, membrane-bounded organelle, and cell periphery were the first five GO terms assigned to the majority of unigenes.

The KEGG pathway database serves as a catalog for different cellular components and their interactions within various metabolic pathways, and allows functional annotation of the unigenes by assigning their potential role in metabolic pathways [56]. Pathway-based transcriptome classification provides an overview of active metabolic processes within an organism, which when coupled with transcriptome expression analysis, gives an important insight of different metabolic processes across different tissues or conditions. Blastx results for L. japonica were used to assign associated KEGG pathways to all the unigenes, resulting in 25,268 unigenes grouped into 142 pathways (Table S4). The first fifty KEGG pathways based on the assigned number of unigenes are shown in Fig. 3. Starch and sucrose metabolism, purine metabolism, phenylpropanoid biosynthesis, pyrimidine metabolism, and glycolysis/gluconeogenesis were the top five KEGG pathways based on the number of assigned unigenes. Key metabolic pathways, such as phenylpropanoid metabolism, flavonoid biosynthesis, and terpenoid backbone biosynthesis, which synthesizes precursors for the biosynthesis of CGA and secoiridoids, were assigned to 231, 137, and 124 unigenes, respectively (Table S4).

SSRs are the tandem iterations of short oligonucleotides ubiquitously distributed within the genome. It consists of simple motifs of 1–6 nucleotides repeated from two to a few dozen times at a locus, and are considered as indel mutational hotspots within the genome [57, 58]. SSRs serve as an important marker for determining genetic variations, including paternity determination, population genetics studies, genetic diversity assessment, and for the development of genetic maps [57, 59]. Several reports have also implicated SSRs in affecting gene expression and, therefore, polymorphism of SSR tracts is considered important in the evolution of gene regulation [60]. In order to identify SSRs for L. japonica, we searched transcriptome assembly for mono- to hexa-nucleotide motifs with a minimum of ten repetitions using MISA software [61]. Overall, we identified 17,992 SSRs spread across 14,702 unigenes, with 2615 unigenes having more than one SSR (Table 2). Within the identified SSRs, mono-nucleotide represented the largest fraction (67.61 %) of all SSRs, followed by tri-nucleotides (17.95 %), and di-nucleotides (11.64 %). The number of SSRs identified as tetra-, penta-, and hexa-nucleotide repeat classes were relatively small but significant. All detected SSRs for L. japonica transcriptome assembly are shown in supplementary Table S5. Identified SSRs of L. japonica in this study may provide potential genetic markers, which will be important for population genetics and comparative genomic studies across different species or eco-types.

Transcriptome expression profiling provides a key insight into the different ongoing cellular processes under various conditions. To estimate expression abundance for unigenes across all nine tissues of L. japonica, clean paired-end reads were aligned to the de novo transcriptome assembly using the Bowtie 2 program [62], and transcript expression as direct count and the FPKM (fragments per kilobase of exon per million mapped fragments) values were determined by the RSEM program [63]. Among nine tissues of L. japonica, green and white floral bud tissues with 169,203 (69.57 %) and 160,454 (66 %) unigenes, respectively, showed the highest number of transcriptionally active unigenes (FPKM >0), while yellow flower and leaf-1 with 138,138 (56.8 %) and 141,099 (58 %), respectively, represented tissues with the lowest number of transcriptionally active unigenes (Table S6). Transcriptome expression analysis showed greater overlap in terms of transcriptionally active unigenes, with 15,551 (6.39 %) unigenes being expressed in only one of the nine tissues of L. japonica. Overall, FPKM value distribution for unigenes across all nine tissues was uniform except for leaf-1 and leaf-3, which showed the majority of its unigenes having lower FPKM values but the number of unigenes with an expression value >500 FPKM were highest when compared to the rest of the tissues.

In order to understand the relationship between the transcriptome dataset from all nine tissues, we used count read data to perform unsupervised principal component analysis (PCA) using the DESeq2 program [64]. The PCA plot showed all nine tissues of L. japonica being clustered in four major groups (Fig. 4a). Along the PC1 axis, we observed two clusters, including floral buds (green and white) and flower tissues (white and yellow) within the first group, while the second group included all leaf tissues (leaf-1,2,3), shoot apex and stem. Along the PC2 axis, these two groups were further separated to form two new groups, with flowers (white and yellow), and floral buds (green and white) being separated into two groups, while stem, shoot apex, and leaf-2 formed a separate group from leaf-1 and leaf-3. Correlation analysis based on Pearson’s distance matrix using the entire transcriptome dataset for all nine tissues was formed, and the results are shown in Fig. 4b. Similar to the PCA plot, we observed two major groups, one that included flower-related tissues while the other included all leaves and stem. Within each of these groups, we observed a higher correlation between developmentally related tissues. For example, transcripts from yellow and white flower tissues, and those from green and white floral buds were highly correlated. Green floral buds showed a high correlation with white floral buds, but a low correlation with white and yellow flowers. On the other hand, white floral buds shared a high correlation with green floral buds, and white and yellow flowers. Green floral buds represent the early development stage for flowers, which then turn into white floral buds, followed by conversion to white flowers and yellow flowers. Our PCA and correlation analysis for all nine tissues of L. japonica suggests the presence of signature unigenes associated with and specific to each tissue, and overlap of transcript expression across tissues was related to their association at different stages of development.

CGA and luteolosides, both derived from phenylalanine, are key metabolites from L. japonica that contribute to its medicinal properties. CGA biosynthesis is proposed to occur through three alternative routes as shown in Fig. 5a. p-Coumaroyl-CoA serves as an important branch point for CGA and luteoloside biosynthesis, which leads to its biosynthesis via being converted into either p-coumaroyl quinate, p-coumaroyl shikimate, or caffeoyl-CoA, while it may also lead to luteoloside biosynthesis via naringenin chalcone (Fig. 5a). To identify potential candidate unigenes from CGA and luteoloside biosynthetic pathways, we screened the annotated transcriptome assembly of L. japonica, and identified homologs for all known enzymes involved in these pathways. A total of 61 unigenes were identified as being associated with CGA biosynthesis pathways, while 27 unigenes were identified as being associated with luteoloside biosynthesis (Table S7). In order to narrow down the most potential candidate unigenes associated with these key metabolic pathways, we selected unigenes with lengths >500 bps and sequence similarity >90 % with a positive alignment of at least 500 bps to its corresponding blast hit. This approach resulted in the identification of 31 unigenes associated with CGA and luteoloside biosynthesis pathways (Table S8). Expression values for unigenes thus obtained were used to perform correlation analysis using Pearson’s distance matrix across all nine tissues of L. japonica. A correlation plot (supplementary Fig. S4) showed the formation of two clusters, with all unigenes associated with CGA except unigene 071319 (HQT) being grouped into cluster 1, and were highly correlated, while unigenes associated with luteolosides were grouped into cluster 2. Expression levels of unigenes associated with CGA biosynthesis were highly upregulated in the stem, followed by leaf-2, shoot apex, and leaf-1, while transcripts associated with luteoloside biosynthesis were highly expressed in the stem, followed by yellow flowers, white flowers, and green floral buds (Fig. 5b). The stem emerged as the tissue with the highest expression of unigenes associated with both CGA and luteoloside biosynthesis. CGA biosynthesis-associated unigenes were highly expressed in the leaf, while unigenes associated with luteolosides were highly expressed in the flowers, suggesting tissue-based preferential expression of unigenes across L. japonica. Previous reports on CGA accumulation in L. macranthoides, a closely related medicinal plant to L. japonica, showed a high accumulation of CGA in the young leaf, followed by the young stem, while low levels were reported in mature flowers [38].

Metabolites derived from secoiridoid metabolic pathways (Fig. 6a) constitute an important part of the metabolite pool of L. japonica, and contribute to its overall medicinal properties. Several studies have suggested anti-tumor, anti-inflammatory, and anti-oxidant activities, and hepatoprotective effects of the metabolites derived from secoiridoid pathways [1, 33‐35, 37]. In order to identify unigenes associated with secoiridoid metabolic pathways, we screened annotated L. japonica transcriptome assembly for unigenes with sequence lengths >500 bps, and 85 % or more sequence similarity with at least 500 bps positive alignment to its closest homolog. This resulted in the identification of 24 unigenes representing all known enzymes from secoiridoid metabolic pathways (Table S9) [65]. Correlation analysis for these unigenes based on their expression values across all nine tissues of L. japonica using Pearson’s distance matrix showed formation of two highly correlated clusters of unigenes (Fig. S5). Among these, cluster 1, with 14 unigenes, included homologs for all enzymes associated with secoiridoid biosynthesis pathways. All these 14 unigenes were highly correlated and, therefore, were considered as potential candidate unigenes involved in secoiridoid metabolic pathways. Transcript expression analysis for unigenes associated with secoiridoid biosynthesis pathways across all nine tissues of L. japonica showed highest expression in leaf-1, followed by shoot apex and leaf-3 (Fig. 6b). Lower expression of unigenes associated with secoiridoid biosynthesis pathways was observed for the stem, leaf-2, floral buds (green and white), and flowers (yellow and white). While we observed high expression of CGA and luteoloside biosynthesis-associated unigenes in the stem, the same was not true for secoiridoid biosynthesis pathways. Comparing all nine tissues, leaf-1 and shoot apex emerged as key tissues which showed high expression of unigenes associated with secoiridoid, CGA, and luteoloside biosynthesis.

Cytochrome P450 (CYP) represents a large superfamily that plays an important role in oxidation and hydroxylation reactions, and is involved in key secondary metabolic pathways. UDP-glucosyl transferases (GTs) represent another super family which participates in conjugation of sugar moieties to secondary metabolites, and is responsible for huge metabolic diversity in plants. Biosynthesis of CGA, luteolin, and secoiridoids involves several CYP enzymes, while GTs play an important role in bringing metabolic diversity and regulating pools of bioactive metabolites. In this study, a total of 285 and 470 unigenes, with sequence lengths >500 bps were annotated CYPs and GTs, respectively, and are listed in Supplementary Tables 10 and 11, respectively. These unigenes are regarded as important enzyme coding genes in many secondary metabolic processes, and will serve as an important resource for future functional characterization attempts.

All nine tissues for L. japonica, namely, shoot apex, stem, leaf-1 (youngest leaf near shoot apex), leaf-2 (second leaf), leaf-3 (mature leaf), green floral bud, white floral bud, white flower, and yellow flower were harvested in June 2014. L. japonica plants were cultivated in the natural environment of Chiba University pharmaceutical garden, Chiba (located at 35°36′17.7″N; 140°08′06.9″E). All tissues from L. japonica were harvested on ice, cut into small pieces, and were snap-freezed by liquid N₂ before storing at −80 °C prior to RNA extraction.

mRNA for each sample was isolated from the total RNA by using beads with oligo (dT), and were added with fragmentation buffer to shear mRNA into short fragments, which were then used as a template for the synthesis of first-strand cDNA using random hexamer primers. cDNA library for Illumina sequencing was prepared using SureSelect Strand specific RNA library kit (Agilent Technology, USA) according to the manufacturer’s instructions.

Raw reads thus obtained through Illumina sequencing were pre-processed using the Trimmomatic program [42] for the removal of adaptor sequences, empty reads, reads with ambiguous ‘N’ base >5 %, low-quality raw reads (Phred score <20), and raw reads with an average length <50 bps. The clean reads thus obtained were in the form of paired reads, or unpaired clean reads (forward and reverse), and were all used to perform de novo transcriptome assembly.

De novo transcriptome assembly for L. japonica was obtained by merging three popular assemblers, namely, SOAPdenovo-Trans, Trinity v 2.0.6, and CLC Genomics workbench v8.0.3 (https://​www.​qiagenbioinforma​tics.​com/​) (Qiagen, USA). For SOAPdenovo-Trans, we performed six independent de novo transcriptome assemblies using kmer sizes as 31, 41, 51, 63, 71, and 91, and resultant assemblies were analyzed using perl script from assemblathon_2 to obtain N50 values and other assembly-related stats [66]. De novo transcriptome assembly using Trinity and CLC Genomics Workbench were performed using default kmer size and default parameters. Resultant transcriptome assemblies from SOAPdenovo-Trans using kmer size as 31 emerged as the best assembly on the basis of different assembly parameters, which were then pooled together with assemblies from Trinity [50] and CLC Genomics Workbench into one merged assembly, and were processed by CD-HIT-EST v 4.6 (built on Mar 5, 2015) [51, 52] with parameters used as ‘−c 0.95 −n 8’ to remove sequence redundancy. Sequences with a length <200 bps were dropped, and the resulting de novo transcriptome assembly was used for further characterization. For transcriptome expression analysis, clean paired reads for each tissue were used for alignment over L. japonica transcriptome assembly using the Bowtie 2.0 program [62], and the RSEM program [63] was used for abundance estimation. To calculate unigene expression, we used the FPKM method. Unsupervised principal component analysis for all nine tissues was performed by the DESeq2 program [64] using count data for unigenes obtained from the RSEM program. GC content and basic statistic values for unigenes were calculated as described previously [67].

We performed a homology search based on the Blastx program using L. japonica transcriptome assembly as a query against the NCBI-non redundant (nr) protein database (http://​www.​ncbi.​nlm.​nih.​gov; formatted on Oct, 2015) using a cut-off E value of <10⁻⁵ with a maximum number of allowed hits of 20. The top hit for each unigene was used to annotate the transcriptome. For further characterization of L. japonica transcriptome assembly, we used the Blast2GO v 3.0 program [55] to assign GO terms, EC number, and KEGG pathway information to the unigenes using associated Blastx results. GO level distribution, and visualization of the top 20 GO terms from three broad categories (biological process, molecular function, and cellular component) at level 3 for L. japonica transcriptome assembly were performed using Blast2GO.

The transcriptome assembly for L. japonica was searched to identify the composition, frequency, and distribution of SSRs using the microsatellite identification tool (MISA) (http://​pgrc.​ipk-gatersleben.​de/​misa/​) [61]. The search parameters for maximum motif length group were set to recognize hexamers with each SSR length-based category to have at least ten repeats.