Introduction

Hazelnut (Corylus avellana L.) is a diploid (2n = 22), monoecious and allogamous tree species belonging to the Betulaceae family. Among 25 Corylus species, C. avellana (European hazelnut) is the most important crop with a major impact on industry and global agricultural sector (Rowley et al. 2014). In ancient times, approximately 2500 years ago, hazelnut was cultivated in the Anatolian region and about 400 clonal cultivars were distributed across Europe at that time (Boccacci et al. 2013). The plant is highly rich in valuable nutrients such as oleic acid, phytosterols (especially β-sitosterol), vitamin E, and squalene. Additionally, the nut possesses several antioxidants such as tocopherols (mainly α-tocopherol) and phenolics (especially tannins, found in the brown skin) (Delgado et al. 2010).

One of the most important ecological factor in deciduous plant survival is the control of bud burst depending on a temperature-related mechanism. In perennial plants, on-time leaf bud burst allows them to start growing early enough to take advantage of the spring conditions, at the same time to grow late enough to escape from late frost to reduce the risk of tissue damage (Azad 2012). Hazelnut and other perennial plants sense environmental signals like photoperiodicity and temperature to adjust timing of actively growing and dormancy stages (Shim et al. 2014). Leaf bud burst time is associated with bud dormancy release which depends on suppression of growth inhibitors, and accumulation of growth promoters. Bud burst of perennial plants generally rely on exposure to a particular duration of non-freezing temperatures to release dormancy, followed by a warm temperature to allow growth in the spring (Faust et al. 1997). There are three types of bud dormancy; (1) para-dormancy (inhibition of meristem growth in buds due to unfavorable conditions), (2) endo-dormancy (inhibition of meristematic growth by itself, chilling requirement is required to overcome) and (3) eco-dormancy (a state of dormancy in the presence of harsh conditions for cell division, insufficient temperature increment for bud break) (Lang 1987). Leaf bud of hazelnut requires certain amount of chilling for the transition from endo-dormancy to eco-dormancy stage and from eco-dormancy to bud and sprouting (Lang GA 1987; Howe et al. 2015). The chilling requirement varies among genotypes and plant’s organs and is an important consideration in choosing cultivars adapted to local climatic conditions. For example, leaf buds of hazelnut have a chilling requirement of 365–1690 h below 7 °C, but the duration of chilling is variable between cultivars (Mehlenbacher 1991; Ma et al. 2016).

Turkey is the largest hazelnut producer in the world. Roughly, 70% of the world’s hazelnut supply is grown in the north side of country. In 2012, a period of severe frost in Turkey reduced the output from 660,000 tonnes (728,000 tons) to 549,000 tonnes. Another instance of cold temperatures in the following year caused the yield to tumble to 381,000 tonnes in 2013. These types of disasters have been also reported in other hazelnut producer countries. Only way to overcome these types of disasters is the development of new cultivars that have more cold hardiness and late bud burst. However, selection and improvement of these types of hazelnut cultivars strongly depends on identification of bud burst related genes and their regulation networks that could be used as molecular markers in breeding studies.

The current study was aimed to conduct a transcriptome analysis for the identification of genes and related regulatory networks responsible for leaf bud burst time in C. avellana genotypes that have different bud burst time. The results of expression profiling generated through transcriptome sequencing will increase the knowledge about molecular mechanism of leaf bud burst and will contribute to identifying important genes involved in bud burst for the ultimate improvement of hazelnut.

Materials and methods

Plant material

Transcriptome sequencing and de novo assembly was carried out with Çakıldak (late vegetative bud burst) and Palaz (early vegetative bud burst) for comparative transcriptome analysis. Eco-dormant leaf buds of these hazelnut genotypes have been previously reported to differ for 15–20 days in terms of their bud-burst times (Köksal 2002). All leaf buds were collected in same developmental stage before the start of swelling. Eco-dormant buds containing two main branches for each genotype were collected at the same time from 22-year-old C. avellana L. trees grown on a field site in Giresun, Turkey and stored in liquid nitrogen until RNA isolation. Ten buds were isolated from each branch using sterile scalpel blades and pooled for RNA analysis. By this way, two biological replicates for each genotype were used for transcriptome sequencing.

RNA isolation and sequencing and assembly

Total RNA was extracted from approximately 0.3 g of frozen sample using a modified method of Chang et al. (1993). A modified protocol including CTAB, Qiagen RNeasy kit and Phase Lock Gel (5 Prime) was preferred because of high polyphenol content in hazelnut buds. Kits and columns were used according to the manufacturer’s protocols. Genomic DNA contamination in extracted total RNA was eliminated using RNase-free DNase I (Thermo Scientific). The genomic DNA contamination and integrity of RNA samples were evaluated by an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). High quality samples with RNA integrity number ≥ 6.5, 28S:18S > 1, OD260/280 ≥ 2 and OD260/230 ≥ 2 were selected to construct cDNA library.

The poly(A)-containing mRNA was isolated from total RNA to carry out fragmentation. The first-strand cDNA synthesis was achieved with random hexamer primers by using approximately 170 bp fragments as templates. DNA polymerase I, dNTPs, and RNase H were used to synthesize the second-strand cDNA. To collect detailed information about the whole expression profiles at mRNA level, RNA-Seq was conducted in four sequencing libraries (2 genotypes X 2 biological replicates). Transcriptome sequencing was performed at Beijing Genomics Institute (http://www.genomics.cn) with Illumina Hi-Seq 4000 platform using a 100 bp paired-end sequencing approach. The standard sequencing protocols recommended by Illumina was followed to construct sequencing library and sequencing. After removal of adaptors, unknown nucleotides larger than 5%, reads that were shorter than 50 base pairs (bps) and low-quality reads with more than 20% low-quality bases (base quality ≤ 10), reads were obtained in FASTQ format and submitted to NCBI Sequence Read Archive (SRX1665375- SRX1665376- SRX1665377- SRX1665378). Assembly of all cleaned reads were carried out on Trinity (Grabherr et al. 2011), a software package optimized for de novo assembly of short-read data with default K-mers = 25. Assembled transcript file was deposited into NCBI (TSA) database under the SUB4195016 accession number. The resulting sequences of Trinity was called as transcripts (raw sequences). Transcriptome completeness were evaluated using BUSCO v3.0 using the plantae linage with default parameters (Simao et al. 2015). The assembled reads were further processed for redundancy removal and clustered with TGICL (Pertea et al. 2003) in order to obtain final non-redundant unigenes (consensus sequences) as long as possible. The non-redundant unigenes were separated into two class which were singletons and clusters. The prefix CL were assigned to the latter group called as clusters. In one cluster, there were several Unigenes with more than 70% similarity among them. The other one was singletons, for which the prefix was unigene.

Functional annotation

The final unigenes were aligned through BLASTx (E value < 10−5) to the publicly available protein databases, including NR, Swiss-Prot, Interpro, the Kyoto Encyclopedia of Genes and Genomes (KEGG), and Clusters of Orthologous Groups (COG). Gene names were assigned based on the best BLAST hit (Altschul et al. 1997). The Open reading frames (ORFs) finder software called as “GetORF” was used for prediction of ORF (http://www.bioinformatics.nl/cgi-bin/emboss/getorf). The longest ORF obtained from each unigene was used to determine coding sequence (CDS). Amino acid sequences of the CDSs were predicted by using the standard codon table. To determine coding sequences of unigenes with no hits in these databases, ESTScan software (Christian et al. 1999) was used. GO annotations of the unigenes was carried out by using Blast2Go program with an E value ≤ 10 − 5 on the basis of molecular function, biological process, and cellular component categories (Conesa et al. 2005), and InterProScan software.

Determination of gene expression levels

To estimate the expression levels of genes, we mapped clean reads to unigenes with Bowtie 2.0 (Langmead et al. 2009). Expression profile of each unigenes was calculated using RSEM software (Li and Dewey 2011). Reads that could be uniquely mapped to a gene by using SAMtools (http://samtools.sourceforge.net/) were used to calculate expression levels. Gene expression levels were measured according to the number of uniquely mapped fragment per kilobase of exon region per million mapped reads (FPKM). Since FKPM eliminates the effect of gene length on expression pattern, FPKM values was used directly for comparing differences in gene expression among samples.

Screening and pathway enrichment of DEGs

To detect the differentially expressed genes (DEGs) between Çakıldak (control) and Palaz, DeSeq 2 R software package (which use a negative binomial distribution to determine differential expression in digital gene expression data) was used as described by Love et al. (2014). More than two-fold change of expression levels and the expression with significant difference (P < 0.05) were considered differentially expressed between samples. Then, DEGs across samples were further annotated by GO enrichment analysis and KEGG pathway enrichment analysis. GO terms and KEGG pathways fulfilling the criterion of a Bonferroni-corrected P value ≤ 0.05 were defined as significantly enriched in DEGs. Based on the NR annotation, the Blast2GO program was used (Conesa et al. 2005) to get the GO annotation of all consensus sequences.

Quantitative reverse transcription (qRT)-PCR validation

Sixteen selected unigenes differentially expressed in phenylpropanoid and plant hormone synthesis pathways were chosen for validation using real time qPCR. Total RNA from leaf buds belonging to Çakıldak and Palaz was extracted with optimized protocol. Reverse transcription to cDNA was performed on 200 ng total RNA using RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) by following the manufacturer’s instructions. qRT-PCR was performed with cDNA template, 1 μl; 5 × HOT FIREPol® EvaGreen® qPCR Supermix, 4 μl; forward primer (10 μM), 0.4 μl; reverse primer (10 μM), 0.4 μl and dH2O, 14.2 μl in a BioRad CFX96 instrument. The qRT-PCR reactions were started with an initial denaturation step (12 min at 95 °C) and proceeded with 40 cycles of amplification (15 s at 95 °C and 20 s at 60 °C). Melting curve analyses were performed to check for specificity of the amplifications. 18S rRNA gene was chosen as reference gene in relative quantification of gene expression. All reactions were performed with triple biological replicates and repeated two times. Transcript fold-changes describing the change in expression of the target gene in samples from Çakıldak (control) to that of the Palaz transcript were calculated using the 2−ΔΔCt method described by Schmittgen and Livak (2008). The primers used in qRT-PCR analysis are listed in Table S1.

Results and discussion

Illumina sequencing and de novo sequence assembly

Dormancy and leaf bud burst is a highly complex physiological phenomenon used by perennial plants to survive in unfavorable environmental conditions. There are many woody plants that require a period of non-freezing temperature to break the dormancy of their buds. The physiological and molecular aspects of bud dormancy release in plants have been previously examined in several studies (Besford et al. 1996; Fuchigami and Wisniewski 1997; Hakkinen et al. 1998; Pletsers et al. 2015; Bhandawat et al. 2017; Singh et al. 2017). This is the first transcriptome study carried out on hazelnut genotypes to identify transcripts and gene regulation networks behind early and late bud burst.

To gather information about genes expressed at the time of bud burst period, RNA-Seq was performed on leaf buds of C. avellana L. cv Çakıldak and Palaz hazelnut genotypes having altered bud burst time. A total of 192,340,000 raw reads were generated after Illumina Hiseq sequencing from four sequencing libraries. A total of 179,830,250 clean reads with about 17.99 Gb bases was obtained with a Q30 percentage of 96.32% (Table 1).

Table 1 Summary of sequencing reads after filtering
Full size table

De novo assembly of cleaned reads were carried out via Trinity software with default K-mers = 25. Redundant sequences in assembled sequences were removed and final sequences were clustered by using the TIGR Gene Indices clustering tools (TGICL). The result of Trinity assembly revealed 74,342, 75,000, 81,667 and 74,234 raw sequence reads (transcripts) from four RNA-Seq libraries (Table S2). The most abundant raw sequences were 300 bp (a total of 106,642 from four sequencing libraries) and the least abundant were 3000 bp (1885); sequences over 3000 bp were grouped together (Figure S1).

Based on gene family clustering analysis, consensus sequences (non-redundant unigenes) were separated into two classes which were singletons and clusters. The prefix CL were assigned to the latter group called as clusters. In one cluster, there were several Unigenes having more than 70% similarity. The other one was singletons, for which the prefix was unigene. Altogether, considering both repetitions, 86,542 unigenes, including 40,387 clusters and 46,155 singletons, with a mean length of 1189 nt and an N50-value of 1916 bps (Table 2) was obtained within the current study. We further quantified the completeness of our assembled transcriptome by comparing their sequences to a core set of plantae genes using BUSCO. The result revealed that 95.2% of BUSCO genes were “complete”, 2.5% were “fragmented”, and the remaining 2.3% were “missing”. The unigene size distribution is summarized in Fig. 1, which shows that 78.97% of the total unigenes are shorter than 2000 nucleotides. Differences in the length of assembled unigenes ranged from 300 to 12,227 bps, with 47,556 unigenes being shorter than 1000 bps, and 5702 unigenes sequence lengths being more than 3000 bps (Fig. 1).

Table 2 Quality metrics of Unigenes
Full size table
Fig. 1
figure 1

Unigene length distribution in assembly. X axis represents the length of unigenes. Y axis represents the number of unigenes

Full size image

Consensus sequences were aligned with BLASTdb using BLASTx (e value < 0.00001). Sequence orientations were determined according to the best hit in the database. ESTScan was used to predict the orientation of CDS for the unigenes that were unannotated with BLAST search. Prediction summary is shown in Table S3. A total of 60,453 CDS was identified with BLAST search and ESTScan. Small amount of CDS (3858) was predicted by using ESTScan software.

Annotation of predicted proteins

To identify the maximum number of similar genes, the unigene sequences were searched in different databases including NR, NT, KEGG, COG, SwissProt, Interpro and GO databases with a cutoff E value < 0.00001. According to the functional annotation result, 56,748; 55,919; 43,571; 26,394; 39,424; 44,965 and 17,379 unigenes were annotated, respectively (Table S4). Top-hit species distribution analysis based on BLASTx results indicated that a total of 10,080 sequences (18%) had top matches to sequences from Vitis vinifera (Fig. 2). The other closest matches with hazelnuts are Prunus persica (18%), Theobroma cacao (13%), Morus notabilis (7%), Populus trichocarpa (6%) and Ricinus communis (5%). Maximum transcripts showed similarity with three species of tree implying that the transcripts were assembled adequately. The KEGG pathway database is a collection of databases which shows networks of molecular interactions in cells for particular organisms. Analysis based on pathway gives additional information about the biological roles of target genes. KEGG pathway annotations were carried out for all annotated sequences. To determine active biological pathways responsible for differential bud break time of C. avellena genotypes, sequences were mapped to the reference canonical pathways in the Kyoto Encyclopedia of Genes and Genomes (KEGG). In total, 43,571 sequences were assigned to 137 KEGG pathways. The most enriched pathways were “metabolic pathways” (9880 unigenes), “biosynthesis of secondary metabolites” (5669 unigenes), “plant-pathogen interaction” (1761 unigenes), ribosome (1605 unigenes) and “RNA transport” (1584 unigenes) in two different bud burst stages (Table S5). Our results were in consensus with previous work on grape, in which comparison of transcriptomic analysis of bud dormancy release period showed the most enriched pathways were plant-pathogen interaction and biosynthesis of secondary metabolites (Khalil-Ur-Rehman et al. 2017).

Fig. 2
figure 2

Distribution of annotated species. This figure shows the species distribution of unigene BLASTX results against the NR protein database with a cutoff E value < 10−5 and also shows the proportions of each species. Different colors represent different species. Species with proportions of less than 5% are not shown (colour figure online)

Full size image

Another functional annotation study was completed by using COG database, where the proteins have been classified on the basis of their orthology concept with the help of completely sequenced genomes (Tatusov et al. 1997). The COG database is built on genes from sequenced genomes and also show evolutionary relationships between organisms including bacteria, algae and eukaryotes. To predict possible roles of all unigenes of C. avellena, sequences were searched in the COG database. On the basis of NCBI COG classification, the 26,394 unigenes were divided into 25 clusters. The groups with the highest representation were found in the clusters R “general function prediction only (7555)”, K “transcription (3799)”, L “replication, recombination and repair (3399)” and “MJ translation, ribosomal structure and biogenesis (3004) (Fig. 3).

Fig. 3
figure 3

Functional distribution of COG annotation. X axis represents the number of Unigenes. Y axis represents the COG functional category

Full size image

Based on GO analysis of all unigenes obtained by using Blast2GO, 17,379 C. avellena sequences were grouped into 54 GO categories; molecular function (15), biological process (22), and cellular component (17). Unigenes with GO annotation accounted for 20.08% of all the unigenes. Results indicated that a big portion of genes was from the categories of “cellular process (8395)”, “metabolic process (10,020)”, “single-organism process (6497), “cell (6121)”, “cell part (6121)”, “membrane (3565)” “organelle (4316)”, “catalytic activity (9963)” and “binding activity (8389)” (Fig. 4). Enriched GO terms recognized in our study, response to stimulus, response to stress, oxidation–reduction process and transmembrane transport, were also in agreement with previous reports (Horvath et al. 2008; Khalil-Ur-Rehman et al. 2017). A total of 39,424 unigenes out of 86,542 has been successfully annotated in Swissprot database by using BLASTx (E value ≤ 1.0E−5) search. Using a five-way Venn diagram, we studied the overlap of the annotated unigene sequences within each of the databases (NR, Interpro, Swissprot, KEGG ve COG) (Fig. 5). Interestingly, a total of 20,010 unigenes had hits in all five databases, whereas there were overlaps of 2221, 1136, 3287, 111 and 155 unigenes between the NR and KEGG, Swissprot and NR, NR and Interpro, Nr and COG and COG and Interpro databases, respectively. We assume that unigenes having no hits in all databases were small transcripts lacking recognizable protein domains. The valuable information related with specific processes, structures, functions, and pathways obtained from these annotations will facilitate the research on C. avellena.

Fig. 4
figure 4

Gene Ontology (GO) categories of unigenes. Unigenes were annotated in three categories: cellular components, molecular functions, and biological processes

Full size image
Fig. 5
figure 5

Venn diagram between NR, COG, KEGG, Swissprot and Interpro. Venn diagram showing the BLASTX results of the C. avellena transcriptome against five protein databases. Using BLASTX search, de novo reconstructed unigene sequences were queried against the following public databases: NCBI-NR, Swiss-Prot, Interpro, KEGG and COG. The number of transcripts that have significant hits (E value ≤ 1e−5) against the five databases is shown in each intersection of the Venn diagram

Full size image

Unigene TF prediction

Transcription factors (TFs) play important roles in various biological processes by regulating the gene expression. Determination of conserved domains located in deduced polypeptide can provide valuable information about the function, regulation and localization of the predicted protein. In order to increase our knowledge about control and regulation of gene expression in C. avellena, all transcription factors were predicted according to the family assignment rules illustrated in PlantTFDB (Riano-Pachon et al. 2007). A total of 2163 unigenes distributed into 59 transcription factor families were predicted to be involved in the regulation of transcription (Figure S1). Among these TF gene families, MYB, MYB-related and bHLH were the most abundant TF families. MYB proteins constitute a diverse class of DNA-binding proteins, which are important to the control of proliferation and differentiation in a number of cell types and regulation of secondary metabolism (Jin and Martin 1999). In hazelnut transcriptome, we identified 105 unigenes encoding MYB TFs. The expression profiles of these unigenes were compared between Palaz and Çakıldak genotypes to elucidate the roles of MYB TFs in dormancy release period. It was found that a total of nine unigenes have down-regulated expression pattern and 18 unigenes up-regulated (Table S6). Similar results about possible function of MYB TFs in bud burst have been reported by several researchers (Xu et al. 2016; Bhandawat et al. 2017; Hao et al. 2017). It was shown that the expression of genes encoding MYB TFs was increased between endo-dormancy, eco-dormancy and bud flush in bamboo (Bhandawat et al. 2017). bHLH proteins responsible for the controlling of a diversity of processes including cell proliferation constitute the second largest TF family in plants (Heim et al. 2003). By this context, it was reported that bHLH TFs may be involved in dormancy release by regulating CBF transcription expression (Ren et al. 2016). In this study, we identified 62 bHLH TF encoding unigenes in hazelnut transcriptome. Another TF family highly related to dormancy release called as Dormancy Associated MADS-box (DAM), has been described in several other perennial species, including poplar (Ruttink et al. 2007; Yıldırım and Kaya 2017), leafy spurge (Horvath et al. 2008) and peach (Jiminez et al. 2010). It was reported that MADS-box TFs have crucial roles in signaling and dormancy transition in several species (Mazzitelli et al. 2007; Hedley et al. 2010). They show high homology to two Arabidopsis genes, SHORT VEGETATIVE PHASE (SVP) and AGAMOUSLIKE 24 (AGL24), responsible for the regulation of flowering. By this context, mutation in DAM genes cause incomplete dormancy in peach (Prunus persica) (Horvath et al. 2008). In contrast to previous studies, there are just three differentially expressed MADS-box unigenes between Palaz and Çakıldak genotypes. Among them, the expression of unigene32100 and CL3418 were up-regulated and CL4239 were down-regulated (Table S6). In addition to these TFs, a total of 53 SBP/SPL transcription factors related unigenes have been identified in hazelnut genome. A unigene (CL2147) annotated to squamosa promoter-binding like TF was differentially expressed among palaz and çakıldak hazelnut. According to the comparative transcriptome analysis between two genotypes, it was found that expression of this unigene was increased 11.62-fold between Çakıldak and Palaz genotypes. Since squamosa promoter-binding-like (SBP/SPL) transcription factor family members have role in diverse biological process including development, our findings showing increased expression of SBPs during leaf bud dormancy release supports previous studies and suggests that they could be involved in the formation or development of new leaf primordia (Schwarz et al. 2008; Howe et al. 2015). Unigenes encoding AP2 transcription factors were also up-regulated during leaf bud burst. Our study revealed that there were 76 AP2 unigenes and one of the AP2 member (CL5542) showed 11.85-fold increased expression between Çakıldak and Palaz. This result support general information about the role of AP2 transcription factor in dormancy release process and re-initiation of development by inhibiting the synthesis of abscisic acid (ABA) (Pandey et al. 2005).

Variations in gene expression between Çakıldak and Palaz

In order to determine the variations in gene expression between leaf buds of Çakıldak and Palaz, two Çakıldak (Cakildak1 + Cakildak2) and two Palaz (Palaz1 + Palaz2) libraries were compared. More than two-fold change of expression levels and the expression with significant difference (P < 0.05) were considered differentially expressed between samples. Results obtained from comparison of two samples showed that there were 8125 significantly differentially expressed unigenes. Among these unigenes, 4348 were upregulated and 3777 were downregulated. The significantly differentially expressed unigenes were further analyzed to determine their biological and molecular functions. Their functional categories were determined according to information in GO and KEGG databases.

In present study, we also focused on DEGs that showed 10-fold and more increased expression between two genotypes. One of the highly expressed unigene identified in hazelnut transcriptome was responsible from the production of serpin-ZX-like protein. Serpins have a highly-conserved core structure that is critical for their functioning as serine protease inhibitors (Huntington 2011). Hazelnut CL976 having homology with Citrus sinensis serpin-ZX-like (LOC102629272) gene showed 11.8-fold increased expression pattern in Palaz genotype. Another highly expressed and annotated as actin related protein like-4 unigene was CL5247 (11.98-fold change). It was reported that plant actin genes have roles in stomatal closure, cell division and differentiation by regulating actin filaments in plant actin cytoskeleton (Kandasamy et al. 2004; Li et al. 2004). Although, there was no study revealing the functions of plant actin genes in dormancy release process, homolog of AT1G18450 gene (CL5247) was required to study in more detail because of its highly-increased expression level. According to comparative transcriptome analysis of two genotypes, it was determined that Unigene12494 encoding an ankyrin repeat-containing protein was differentially up-regulated (10.79-fold increase) in Palaz genotype. In previous studies, it was revealed that this group of proteins carried out crucial functions in cell cycle, differentiation, lateral root development, leaf formation, plant microbe interaction, pollen germination and pollen tube development by controlling the protein–protein interactions (Mou et al. 2013).

Functional classification of DEGs

GO and KEGG assignments were used to classify the functions of the DEGs identified in the pairwise comparisons of the Çakıldak and Palaz cDNA libraries. Then, we performed Gene Ontology (GO) classification and functional enrichment analysis. Based on GO analysis of DEGs between Çakıldak and Palaz, 8125 C. avellena sequences were categorized into 47 functional groups, belonging to three main GO ontologies: molecular function (13), cellular component (14) and biological process (20). Results indicated that a big portion of genes was from the categories of “cellular process (1174)”, “metabolic process (1422)”, “single-organism process (968), “cell (755)”, “cell part (755)”, “membrane (466)” “organelle (521)”, “catalytic activity (1429)” and “binding activity (1166)” with only a few genes related to “developmental process (94)”, “growth (27)”, “nucleoid (2)”, “electron carrier activity (37)”, “enzyme regulator activity (19)” (Fig. 6). “Metabolic Process” and “Cellular Process are the two most abundant GO terms among Biological Process groups.

Fig. 6
figure 6

GO classification of DEGs. X axis represents number of DEG. Y axis represents GO term

Full size image

In fact, both metabolic and cellular processes containing genes involved in the most basic of life processes and metabolic changes have been widely reported in dormancy releasing process (Lesur et al. 2015; Qi et al. 2015). The increased energy and biosynthetic needs of dormancy releasing process and bud burst are generally provided by metabolic regulation (Lesur et al. 2015). The heading “Metabolic Process” includes nitrogen compound metabolic process, catabolic process, single organism metabolic process, and biosynthetic process. “Cellular Process” includes cellular aromatic compound metabolic process, organic cyclic compound metabolic process, and organonitrogen compound metabolic process. “Biological Regulation” includes the regulation of apoptosis, metabolism, cell cycle, translation, catalytic activity, and homeostasis. For the molecular function categories, several GO categories including binding, catalytic and transporter activity were enriched. Our GO annotation of C. avellena transcriptome, therefore, suggests presence of several transcripts associated with active metabolism and dormancy release process. GO functional enrichment analysis were carried out for three GO categories separately. l-Phenylalanine metabolic process (11 unigenes), erythrose 4-phosphate/phosphoenolpyruvate family amino acid metabolic process (11 unigenes), phenylpropanoid metabolic process (17 unigenes), l-phenylalanine catabolic process (6 unigenes) and cinnamic acid biosynthetic process (6 unigenes) are the most enriched biological processes. Based on analysis of cellular component terms, MCM complex, microtubule, Golgi apparatus part, microtubule cytoskeleton, cytoskeletal part and microtubule associated complex are identified as significantly enriched in Palaz libraries (Figure S2). By influencing the deposition direction of cellulose microfibrils, the plant microtubule cytoskeleton have a crucial role in growth and development of plants (Gardiner et al. 2012).

Pathway analysis of DEGs

KEGG pathway classification and functional enrichment for DEGs were carried out to determine the important pathways having roles in dormancy release and development. Based on gene set enrichment analysis of DEGs, biosynthesis of secondary metabolites, starch and sucrose metabolism and phenylpropanoid biosynthesis pathways were the most represented pathways in de novo assembled transcriptome of Palaz (Fig. 7). The starch and sucrose metabolism and phenylpropanoid biosynthesis pathways, important pathways for the plant development and growth, were assigned to 182 and 46 of differentially expressed unigenes, respectively (Table S7).

Fig. 7
figure 7

Pathway functional enrichment of DEGs. X axis represents enrichment factor. Y axis represents pathway name. Coloring indicate q value (high: white, low: blue), the lower q value indicate the more significant enriched. Point size indicate DEG number (more: big, less: small) (colour figure online)

Full size image

Pathways of DEGs was classified into six categories. These are cellular processes, environmental information processing, genetic information processing, metabolism and organismal systems. The largest proportion of the DEGs belonged to metabolism such as global and overview maps (1468 unigenes), carbohydrate metabolism (513 unigenes) and amino acid metabolism (296 unigenes) (Fig. 8). A total of 133 DEGs was assigned to glycan biosynthesis and metabolism, which have an important function in plant cell wall biosynthesis (von Schaewen et al. 2008). These findings were consistent with GO analysis since in plant cells, glycans attached to asparagine (N) residues of proteins undergo various modifications in the endoplasmic reticulum and the Golgi apparatus. One of the most enriched enzyme was pectin methyl esterase (PME) (EC 3.1.1.11) a member of starch and sucrose metabolism pathway. PME catalyzes the hydrolysis of methylester groups of cell wall pectins. Ren and Kermode (2000) reported that PME activity was positively correlated to the degree of dormancy breakage of yellow cedar seeds. In our study, a total 22 unigenes encoding PME showed differentially upregulated expression pattern between two samples (Table 3).

Fig. 8
figure 8

Pathway classification of DEGs. X axis represents number of DEG. Y axis represents pathway name

Full size image
Table 3 Expression values and annotation of bud burst related genes
Full size table

As a consequence of low temperature, there will be an increase of reactive oxygen species (ROS) formation due to the damage of membrane lipids and proteins (Prasad 1996). Likewise, the bud dormancy release period coincides with up-regulation of the antioxidant system (Faust and Wang 1993). There are many studies reporting that bud dormancy release is associated with activated antioxidant system including catalase (CAT), ascorbate peroxidase (APX), superoxide dismutase (SOD), glutathione-S-transferase (GS) and glutathione reductase (GR) (Bartolini et al. 2006; Ben Mohamed et al. 2012; Viti et al. 2012). It was found that the expression of several genes whose products are related with antioxidant mechanism, including APX (CL2105), CAT (Unigene21988, Unigene22264), GS (Unigene8147), SOD (Unigene19084) was increased during bud burst. One of the most prominent antioxidant enzyme was APX (EC:1.11.1.11) in ascorbate and aldorate metabolism. The APX enzyme-encoding gene (CL2105) was differentially up-regulated with 5.2-fold change between Çakıldak and Palaz (Table 3). The other antioxidant enzyme encoding genes also showed similar expression patterns during leaf bud burst period.

Transcripts related to phenylpropanoid biosynthesis and flavonoid biosynthesis pathway

A variety of structural and defense compounds in plants are synthesized through the phenylpropanoid pathway (Ververidis et al. 2007). Each part of this pathway is responsible for the synthesis of the different phenolic secondary metabolites, such as flavonoids, coumarins, and lignin. These compounds have different roles in metabolism including protecting against oxidative stress, participating in developmental and stress signaling, mediating plant–microbe interactions, and defense against pathogens and pests (Barber and Mitchell 1997; Weisshaar and Jenkins 1998; Vogt 2010; Jones et al. 2012; Fan et al. 2015). Based on the KEGG annotated sequences, enzymes involved in phenylpropanoid biosynthesis pathway that are considered to be involved in development and dormancy release were visualized and colored in the pathway-maps obtained from KEGG database (Figure S3). There are 140 DEGs that translate to 12 enzymes in the phenylpropanoid biosynthesis pathway. The three most important enzymes in the phenylpropanoid biosynthesis are phenylalanine ammonia lyase (PAL, EC:4.3.1.24), cinnamoyl-CoA reductase (CCR, EC:1.2.1.44), 4-coumarate-CoA ligase (4CL, EC:6.2.1.12), which were represented by 3, 8, and 4 DEGs respectively. All PAL encoding genes were differentially up-regulated between Çakıldak and Palaz. For instance, the expression of CL2676 was increased significantly between Çakıldak and Palaz transcriptome. Several copies of PAL-genes are found in all plant species, comprising four genes in Arabidopsis, five genes in willow and five in poplar (Cochrane et al. 2004; Tsai et al. 2006; de Jong et al. 2015). In contrast to PAL encoding genes, there are both up and down-regulated expression pattern in 4CL and CCR pathway related genes. Additionally, we identified that the unigene27109 annotated to 4CL enzyme was highly expressed between Çakıldak and Palaz genotypes (Table 3).

Another group of genes found in phenylpropanoid pathways and differentially expressed during dormancy release period are associated with secondary metabolism, specifically those leading to flavonoid biosynthesis. This groups included the genes encoding chalcone flavone isomerase, chalcone synthase, chalcone isomerase and flavones. They have been reported to be significantly regulated during dormancy release in variety of plants including raspberry (Mazzitelli et al. 2007) and leafy spurge (Horvath et al. 2006). There are 5 different genes encoding these enzymes in hazelnut transcriptome. CL7570 and CL658 annotated to chalcone synthase were identified as differentially upregulated with 2 and 1.2-fold change expression value between Çakıldak and Palaz transcriptome, respectively. There were 2 homolog genes encoding for chalcone isomerase, CL8598 and unigene13134. Among them just CL8598 was differentially down-regulated in Palaz (Table 3).

Transcripts related to phytohormone biosynthesis and signaling

Phytohormones play important roles in plant growth and development. Bud burst process involves an interaction between plant growth inhibiting substances such as, abscisic acid (ABA) and other plant hormones including auxin, gibberellins (GA) and ethylene (Rodriguez and Sancheztames 1986). Gibberellin type of phytohormones have many roles in plant life cycle. It has been reported that they are required in germination of seed, triggering of vegetative growth, and development of floral and seed (Michalczuk 2005). GID1 and Della encoding genes were enriched in diterpenoid biosynthesis pathway when compared two hazelnut transcriptomes (Figure S4). GID1 is a soluble protein that interacts directly with GA, triggering the gibberellin response (García-Martinez and Gil 2001). There were 4 DEGs encoding GID1 in hazelnut transcriptome. CL1560, CL3178, CL7872 and Unigene21953 have been upregulated between two hazelnut sequencing libraries. Similar results have been obtained with different plants including grape and leafy spurge (Mathiason et al. 2009; Dogramaci et al. 2010). Additionally, it was reported that GID1 encoding gene was overexpressed during eco-dormancy period in oak leaf bud (Ueno et al. 2013). The Della proteins, which have role in repressing of GA-dependent, are the most studied components of the GA signaling pathway (Silverstone et al. 1997; Mutasa-Göttgens and Hedden 2009; Li et al. 2016). There are four differentially expressed genes encoding Della proteins in hazelnut. Among them, unigene6641, unigene440 and CL1163 were differentially upregulated with 3.2, 1.5 and 1.0-fold change value between Çakıldak and Palaz transcriptome, respectively (Table 3). However, CL1210 was down regulated (-2-fold change). Another important dormancy related phytohormone is auxin that is generally up-regulated during bud burst period. Auxin have role in regulation of plant growth and development. It regulates the expression of related genes via the degradation of a group of repressor proteins known as Aux/IAA proteins. Auxin induce three important gene families. They are SAUR, Aux/IAA, and GH3 (Guilfoyle et al. 1998). There were 11 DEGS unigenes encoding auxin receptor (TIR1), auxin/indole 3-acetic acid (AUX/IAA) and auxin response factor (ARF) proteins involved in plant hormone signal transduction pathway. TIR1/AFBs, members of F-box proteins, is responsible for the perception of auxin. These proteins have been classified into four groups such as TIR1, AFB2, AFB4, and AFB6 in flowering plants. Members of TIR1 and AFB2 groups have been reported as regulators of auxin signaling by promoting the degradation of the Aux/IAA transcriptional repressors (Prigge et al. 2016). In our study, genes that encode TIR1 were both down-regulated (CL7563) and up-regulated (CL9080) during bud dormancy release process. It was reported that there was another important auxin related element having roles in dormancy release process. It was called as monopteros homologue or auxin response factor and their expressions were increased during bud burst (Derory et al. 2006). There were two differentially expressed monopteros homologue gene in hazelnut transcriptome. In transcriptome of Palaz genotype, CL93 showed six-fold higher expression pattern when compared to Çakıldak genotype. Likewise, it was observed that expression of CL283 increased 3 times from Çakıldak to Palaz (Table 3). In addition to these genes, there were also other differentially expressed unigenes encoding auxin induced protein 5NG4. In previous study conducted with Cunninghamia lanceolate, the expression of this gene was increased in induced leaf bud during dormancy release period (Xu et al. 2016). We identified three homolog 5NG4 genes in C. avellana transcriptome. These were CL4416, CL3309 and CL124 and their expression increased 2.74, 5 and 5.7-fold between two genotypes, respectively. The other DEGs in this pathway was auxin repressed (ARP) protein like gene. It was reported that some of the homologs of this gene have roles in dormancy and inhibition of vegetative growth. For instance, there was negative correlation between shoot growth and expression of this gene in Robinia pseudoacacia (Park and Han 2003). In our study, the homolog of ARP (CL4303) was down-regulated (-1.3-fold) in Palaz early leaf developed genotype.

One of the most important player in dormancy process is ABA. This phytohormone have function starting of dormancy by interaction with other phytohormones. In this context, it was thought that enzymes responsible for degrading of ABA have roles in dormancy release process of leaf buds (Zheng et al. 2015). The enzyme abscisic-aldehyde oxidase (AAO3) have critical role in this process, since it is found in the last step of biosynthesis of ABA (Kushiro et al. 2004). According to the comparative expression analysis of Palaz and Çakıldak, it was found that CL1901 a homolog of Arabidopsis AAO3 has a reduced activity in Palaz (Table 3).

Confirmation of candidate DEGs by qRT-PCR analysis

To experimentally confirm the Illumina RNA-Seq results, a subset of 16 genes in phenylpropanoid (PP) and plant hormone (PH) synthesis pathways, which were found to be differentially up-regulated in Palaz libraries, was chosen for quantitative real-time PCR (qRT PCR) analysis. Although different optimization conditions were performed, amplification of CL6063 could not be achieved. According to the results, the expressions of CL1469, CL1560, CL3668, CL8248, unigene28358, unigene4384, unigene4817 and unigene6641 were upregulated in Palaz variety when compared to that in Çakıldak variety (Fig. 9). Furthermore, unigene 28358 expression levels exhibited the highest increase in Palaz variety compared to Çakıldak with a maximum of 16.2-fold more expression value. On the other hand, CL2140, CL5737, CL483, CL7825, unigene15907, and unigene3 gene expressions were shown to be downregulated in Palaz variety as compared to that in the Cakıldak. The most drastic expression decrease was recorded in CL5737 with a 4.96-fold reduced expression level in Palaz. Four of six down regulated genes in Palaz variety were found to be members of phenylpropanoid pathway. Among them, CL7825 is responsible for formation of xyloglucosyl transferase. However, five of eight up-regulated genes in Palaz variety belong to plant hormone synthesis pathway. Of these, Unigene6641 encodes for DELLA, unigene4383 encodes for SAUR, CL8248 encodes for JAZ (jasmonate zym domain) and CL1560 encodes for gibberellins. The qRT-PCR results that show upregulation of CL1469, CL1560, CL3668, CL8248, unigene4384 and unigene6641 gene expression in Palaz variety confirm the data obtained from transcriptome results.

Fig. 9
figure 9

Relative expression levels of unigenes from phenylpropanoid (PP) and plant hormone (PH) synthesis pathways in Çakıldak and Palaz genotypes

Full size image

Conclusion

Transcriptome sequencing analysis was carried to elucidate the major molecular mechanism in bud burst of hazelnut plant. The expression profiling analysis was done using buds in the ecodormancy and bud flush stages. Generally, phenylpropanoid and phytohormone biosynthesis pathway genes and gene sets that are significantly enriched have been indicated to exhibit crucial roles in regulation mechanisms of bud burst. By using transcription factor identification studies, the key transcription factors involved in bud burst process were highlighted, including MADS-box related, MYB and bHLH gene family. We further reported that bud burst process involves an interaction between plant growth inhibiting substances such as, abscisic acid (ABA) and other plant hormones including auxin, gibberellins (GA) and ethylene. These results will increase the knowledge about the molecular mechanism of bud burst process in hazelnut plant.