Two cDNA libraries were generated with mRNA from control (22°C) or 24 hours chilling-treated (2°C) plants of C. bungeana and sequenced by Illumina deep-sequencing. 41,499,576 and 40,009,694 clean reads of 90 bp were generated from control and chilling-treated cDNA libraries, respectively (Table 1). De novo assembly was carried out by Trinity method 42 and final unigenes were obtained by TGICL clustering 43 . Overviews of the assembly results were shown in Table 2. The sequence reads were finally assembled into 54,870 non-redundant unigenes, spanning a total of 48.7 Mb of sequence. All unigenes were longer than 200 bp. Mean length of final unigenes was 888 bp and N50 was 1401 bp. With the Trinity de novo assembly method, no N remained in the final unigenes. We also tried de novo assembly with SOAPdenovo program 44 . However, the assembly quality was worse than that of the Trinity method, with a mean length of 596 bp and N50 of 809 bp, and 13.9% of the final unigenes had at least one N remained (Table 3). The results were similar to the transcriptome assembly report of Aegilops variabilis 45 , in which the assembly qualities of the Trinity method were superior to that of the SOAPdenovo method. Therefore, the assembly results from the Trinity method were used for all the following analysis.

Functions of the unigenes were annotated based on sequence similarities to sequences in the three public databases (NR, Swissprot and KEGG). Among the 54,870 non-redundant unigenes, 43,524 (79.4%) had at least one hit in BLASTX search with E-value < =1e-5 (Additional file 1). Functional classifications of GO terms of all unigenes were shown in Figure 1. In the category of biological process, the largest groups were “cellular process”, “metabolic process” and “response to stimulus”. In the category of molecular function, unigenes with “binding” and “catalytic” activities were the largest groups.

The expressions of unigenes were analyzed with DEGseq R package. Firstly, we tried to identify DEGs by applying screening thresholds of 2 fold changes and Benjamini q value <0.001. We got 12,808 DEG candidates out of 52,753 expressed unigenes (Additional file 2). However, when we verified the expressions of the top 10 up-regulated and down-regulated unigenes by RT-PCR and qPCR, only 3 of them were amplified and none of them showed up or down-regulated trends in chilling-treated seedlings (data not shown). In addition, we found that 80% and 90% of the top 200 up and down-regulated unigenes presented only in one sample’s RNA-seq data, respectively. PCR amplification failures of the selected sequences suggested that such genes were most likely to be the artifacts of de novo assembly.

To identify DEGs accurately, we dropped off all unigenes with RPKM < 1 in both sequencing libraries before DEGseq analysis. By this method, 8,055 DEGs (25.7%; 3,484 up-regulated, 4,571 down-regulated) out of 31,295 unigenes with minimal 1.0 RPKM in both cDNA samples were identified (Additional file 3). The top 50 most up- or down-regulated unigenes were listed in Table 4 and Table 5, respectively. A number of genes involved in cold or other stresses showed up in the top 50 up-regulated list, such as putative orthologous genes (POGs) of COR15A, ABR1, pectin methylesterase inhibitor gene, MAPKKK13, heat shock transcription factor A1E and LTI65 genes. A putative ortholog of Arabidopsis COR15A, which encodes a cryoprotective protein located to the chloroplast stroma 46 , was identified as the most up-regulated unigene in C. bungeana.

The top 20 up-regulated DEGs were selected to verify the expressions of the indentified DEGs by qPCR analysis. To get more reliable quantification results, we performed an experiment in advance to screen reference genes for qPCR (see Methods for details), and the relative expression levels of unigenes were normalized to 3 stable expressed reference genes. The results showed that 18 of the top 20 up-regulated DEGs (90%) were verified to be up-regulated by qPCR analysis, although their fold changes differed from that of RNA-seq (Figure 2). Except for CBT7920 and CBT22908, the expressions of all other tested unigenes showed at least 3-fold increases after 24-hour chilling treatment. The most up-regulated unigene were POGs encoded a plant invertase/pectin methylesterase inhibitor superfamily protein (CBT4773, 552 folds). COR15A (CBT13817, 318 folds) was also induced remarkably by chilling.

High throughput deep-sequencing is a powerful tool for DEGs screening, especially for species without available genomic information 45 47 48 . However, since Illumina sequencing is highly sensitive to templates presented in DNA samples, some traced transcripts or contaminants can be sequenced in one sample but not in other samples. This will have huge effects on the results of de novo assembly and increase false positive rate in DEGs identification. One strategy to reduce the false positive results is to set up biological repeats for sequencing and increase sequencing depth, but it will greatly increase the experimental costs. In this study, by simply applying an additional threshold (RPKM > =1) for DEGs screening without increasing costs, we got a high quality (confirmed by qPCR) list of chilling regulated DEGs.

Since both C. bungeana and Arabidopsis are Cruciferae species, it is more reliable to use the well-established GO and KEGG annotation systems of Arabidopsis to analyze the functions of C. bungeana DEGs. GO term and KEGG pathway enrichment analysis of DEGs were conducted with BiNGO 49 , a Cytoscape plugin assessing overrepresentation of ontologies in biological networks, using the list of all unigenes with a minimal RPKM of 1 in both sequencing libraries as a reference set. To compare the chilling responding network of C. bungeana with Arabidopsis, the networks of chilling-regulated DEGs of Arabidopsis were constructed using previously published RNA-seq and microarray data (referred to ATH-SR and ATH-MA, respectively; see Methods for details).

In chilling up-regulated DEGs of C. bungeana and Arabidopsis, two similar clusters in the networks of GO biological process, “regulation processes” and “stimulus responses”, were found among all three networks/datasets (Figure 3). In BiNGO constructed networks, most biological information can be inferred from end nodes and their relations with their source nodes such as gene numbers (node sizes) and p values (node colors) 49 . In “regulation processes” cluster of all three networks, genes involved in “regulation of transcription, DNA-dependent” accounted for the enrichments of all other nodes in this network branch since the end node was almost the same size and color as its source nodes, suggesting that transcriptional regulations might have common contributions in plants responding to chilling stress. In the cluster of “stimulus responses”, the network patterns showed that cellular responses to a wide range of stresses were aroused by chilling stress in both C. bungeana and Arabidopsis, which were probably due to the cross-tolerance mechanisms of plants. The cluster of “metabolism processes” comprised much more over-representative terms in the network of C. bungeana than that of Arabidopsis. “Flavonoid biosynthetic process” was the only over-representative term of this cluster presented in both C. bungeana and Arabidopsis (ATH-SR).

Twelve biological processes (end nodes in the networks) were found to be common in both C. bungeana and Arabidopsis (ATH-SR or ATH-MA), and ten of them were related to stimulus responses (Table 6). Genes “response to cold” were over-representative in all three networks, suggesting that our chilling stress treatments were efficient. However, the genes involved in “cold acclimation” did not over-represent in C. bungeana as did in Arabidopsis (Figure 3), indicating that cold acclimation mechanisms were not activated by chilling in C. bungeana. The results imply that C. bungeana may not have a cold acclimated mechanism or may have cold acclimated mechanisms different from that of Arabidopsis. For plants from temperate regions, cold acclimation is critical for them to tolerate freezing temperatures 8 . However, since cold acclimation requires a relatively long period of time to get freezing tolerance, such mechanisms may not be suitable for plants like C. bungeana in harsh environments. More rapid and efficient mechanisms are needed for such plants.

* Only GO terms of the end nodes in the network were presented.

Besides abscisic acid 50 and chitin responses 51 , which were known to be involved in cold tolerance of plants, the biological process “response to karrikin” was found to be a common response to chilling stress in both C. bungeana and Arabidopsis. To our knowledge, no previous study reported the involvement of karrikins in cold tolerance of plants. Karrikins are a new group of plant growth regulators discovered in smoke that can stimulate seed germination 52 . The biological and molecular functions of karrikins are largely unknown at present. Our results suggested that karrikins might play important roles in chilling tolerance of C. bungeana and Arabidopsis.

Nineteen biological processes were over-represented in chilling-treated C. bungeana but not in Arabidopsis. Nonetheless, it did not mean that such processes were specific to chilling responses of C. bungeana since most of them, such as salicylic acid 53 54 , jasmonic acid 54 , and immune response 55 , were reported to be involved in chilling response of Arabidopsis or other plants. However, two processes, “protein phosphorylation” and “protein autoubiquitination”, should be emphasized. Post-translational modifications of pre-existing proteins are believed to be a rapid pathway to get tolerance in plant responses to chilling stress and have important roles in plant cold acclimation 8 . In alfafa, low temperature lead to rapid inhibition of PP2A activity, and in turn lead to phosphorylation of proteins involved in cold tolerance acquisitions 56 57 . Transcriptional activation of genes of several kinase families were also found under low temperature stress, such as MAP kinase family genes MKK2 58 , OsMEK1 and OsMAP1 59 , CDPK family genes OsCDPK7 60 61 , OsCDPK13 62 and PaCDPK1 63 , and CIPK family genes CIPK3 64 and CIPK7 65 . Although many studies reported that certain protein kinases were activated and their transcriptional expression increased in response to cold stress, few studies reported that the expressions of protein kinases as a whole increased at transcriptome level. In our study, a large number of genes whose products were involved in protein phosphorylation were over-represented in chilling up-regulated DEGs in C. bungeana. Given the habitats of C. bungeana, in which the daytime temperatures fluctuate frequently and during almost the whole plant growing seasons, our results suggest that protein phosphorylation may be an important mechanism for rapid and flexible regulation of cold tolerance of C. bungeana.

Protein autoubiquitination may play similar roles as protein phosphorylation. In Arabidopsis, ubiquitination of ICE1 by HOS1 which leads to ICE1 degradation is vital for the activation of CBF pathways 66 . In this study, eight chilling up-regulated unigenes of C. bungeana were associated with protein ubiquitination, six of which might be involved directly in protein ubiquitination (Table 7). However, POGs of HOS1 was not on the list. Therefore, the roles of protein ubiquitination in chilling responses of C. bungeana need further investigations.

Comparison of the molecular function networks of chilling up-regulated DEGs showed that only one term/node, “sequence-specific DNA binding transcription factor activity”, was in common in both C. bungeana and Arabidopsis (Figure 4, Table 6). It was consistent with the over-representative term of “regulation of transcription, DNA-dependent” in network of biological process. However, only a small amount of TF POGs of the three experiments were overlapped (Figure 5A), including ZAT12/RHL41, COL1, TOC1 and RAP2.7 orthologs (Table 8) which were reported to be involved in plant cold responses 33 34 67 68 . Surprisingly, none of the CBFs (CBF1/DREB1b, CBF2/DREB1c and CBF3/DREF1a) was on the list of overlapped TF genes though CBF2 and CBF3 were chilling up-regulated in Arabidopsis as was shown by both ATH-SR and ATH-AR data (Additional file 4). In fact, no ortholog of Arabidopsis CBF1 or CBF2 was found in the transcriptome of C. bungeana, while there were orthologs of CBF3 and CBF4 (data not shown). The results suggest that the transcriptional activation mechanism of C. bungeana differs greatly from that of Arabidopsis in chilling responses although they share some common mechanisms. Given the important roles of CBFs in plant cold acclimation, lack of CBF orthologs suggests that cold acclimation mechanisms may be weak in or absent from C. bungeana, consisting with the finding that genes involved in cold acclimation was not enriched in chilling up-regulated DEGs of C. bungeana. Classification results showed that MYB, AP2/ERF, WRKY and NAC family members represent the most abundant TFs in chilling up-regulated DEGs of C. bungeana (Figure 5B). The data are insightful for further investigation of specific tolerance mechanisms of C. bungeana.

Ten terms/nodes in the network of C. bungeana were not in the networks of Arabidopsis (Figure 4, Table 9). Again, the over-representation of “protein serine/threonine kinase activity” was overlapped with “protein phosphorylation” in the network of biological process. The most abundant protein kinases in chilling up-regulated DEGs encoded cysteine-rich receptor-like protein kinases (CRK), whose roles in plant cold responses were largely unknown (Figure 6, Additional file 5). Genes for leucine-rich receptor-like protein kinases (LRR RLK) ranked the second. A small number of POGs of CDPKs, CIPKs, MPKs, MKKs and MKKKs, some of which have been reported to be involved in plant cold responses 58 59 60 61 62 63 64 65 , were found in chilling up-regulated DEGs of C. bungeana.

* Only GO terms of the end nodes in the network were presented.

KEGG pathway network analysis showed that “Biosynthesis of Other Secondary Metabolites” and “Environmental Adaptation” were enriched in chilling up-regulated DEGs of C. bungeana (Figure 7). The over-representation of “Biosynthesis of Other Secondary Metabolites” was due to biosynthesis of three kinds of secondary metabolites: flavonoids, glucosinolates and phenylpropanoids; and the over-presentation of “Environmental Adaptation” was due to enrichment of genes involved in “plant-pathogen interaction” and “circadian rhythm” regulation. Besides, genes involved in alpha linolenic acid metabolism were also enriched. The phenylalanine/tyrosine/tryptophan biosynthesis pathway was overlapped with phenylpropanoid biosynthesis. In Arabidopsis, genes involved in flavonoids biosynthesis and circadian rhythm pathways were also enriched in chilling up-regulated DEGs.

All over-represented pathways in C. bungeana, regardless of whether they were enriched in Arabidopsis, had proved to be important in plant cold tolerance. For instance, circadian rhythm regulates the expression of CBFs 28 69 , the core identified TFs that involved in plant cold tolerance. As another example, under chilling stress, plants preferentially accumulate polyunsaturated fatty acids such as linoleic and linolenic fatty acids 70 71 72 , and genetically increasing of unsaturated fatty acids or lipids could enhance cold tolerance of transgenic plants, probably by maintaining membrane fluidity under cold stress 73 74 . Our previous findings indicated that cold tolerance of C. bungeana was correlated with changes in membrane lipids and membrane-associated enzymes 3 . Under chilling treatment, the proportion of unsaturated fatty acid in the plasma membrane increased significantly in callus of C. bungeana 75 . Paralleling to these results, KEGG analysis in this study showed that unigenes involved in "alpha-Linolenic acid metabolism" were enriched significantly in chilling up-regulated DEGs, suggesting that lipid metabolism, especially linolenic acid metabolism, might play a role in chilling tolerance of C. bungeana.

In chilling stress down-regulated DEGs of both C. bungeana and Arabidopsis, there were several over-represented terms in every biological process networks (Figure 8). However, no over-represented term was in common in C. bungeana and Arabidopsis. Furthermore, none of the over-represented term was the same between two networks of Arabidopsis, although both of them were related to chilling stressed down-regulated DEGs. Similar results were also found in the networks of molecular function (Figure 9). The huge discrepancy among the networks implied that the gene members of chilling stress down-regulated DEGs were highly variable, which might be affected by some subtle experimental details other than chilling temperatures only. It was hard to deduce an unbiased mechanism from their networks analysis. Therefore, no further analysis was performed for the down-regulated DEGs.

Plant regeneration of C. bungeana via somatic embryogenesis was performed as described by Wang et al. 76 . Callus was induced from matured seeds of C. bungeana on MS medium containing 4.0 mg l^-1 GA3, 2.0 mg l^-1 NAA, and 2.0 mg l^-1 2,4-D. Seedlings were regenerated from callus on MS medium containing 3% sucrose in about 3 weeks. Regenerated plants were transferred to new MS medium containing 3% sucrose and grown at 22°C with a 14 h photoperiod under 80 μmol m^-2 s^-1 fluorescent light for further 7 days before treatments. For each treatment, ten plants (roots, shoots and leaves) were randomly pooled and treated in MS liquid medium containing 3% sucrose at 22°C or 2°C. Chilling stress was initiated 4 hours after dawn (zeitgeber time 4; ZT4). Upon the treatment time reaching 24 hours, both control and chilling stressed samples were collected at the same time point and frozen immediately with liquid nitrogen.

For RNA sequencing, total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA). The quality of total RNA was checked using the NanoDrop Spectrometer (ND-1000 Spectrophotometer, Peqlab) and the Agilent 2100 Bioanalyzer (RNA Nano Chip, Agilent). High quality RNA samples (20 μg each) were sent to Beijing Genomics Institute (BGI, Shenzhen) for cDNA libraries construction and sequencing using Illumina HiSeq™ 2000. The cDNA library construction method and Illumina deep-sequencing processes were the same as described by Xu et al. 45 .

The Trinity method 42 was used for de novo assembly of the clean reads to generate Trinity unigenes, with optimized k-mer length of 25. Then, the Trinity unigenes of both libraries were clustered with TGICL software 43 to get sequences (final unigenes) that cannot be extended on either end. De novo assembly was also conducted with SOAPdenovo software 44 with optimized k-mer length of 41.

Files containing the raw read sequences and their quality scores are available from the National Center for Biotechnology Information (NCBI) Short Read Archive with the accession number: SRA054354. The Trinity unigenes have been deposited in the Transcriptome Shotgun Assembly Sequence Database (TSA) at NCBI [GenBank: JW988067-JW999999, KA000001-KA089547].

Clean reads were mapped back to assembled unigenes with SOAPaligner (version 2.21) 44 allowing maximum 2 mismatches. The reads with unique best hits were counted for each unigene. The expression level of C. bungeana unigene was normalized by the number of RPKM (reads per kilobase exon region per million mapped reads) 77 . Since Illumina sequencing method is highly sensitive, we only used a subset of unigenes which presented in both sequencing libraries with a minimal RPKM of 1 for DEGs analysis. Unigene expressions were analyzed using DEGseq R package 78 with MARS method. Chilling-regulated DEGs were identified with Benjamini q < 0.001 79 and normalized fold change > =2.

For comparisons, two public available data sets of Arabidopsis were used in our study. One data set (referred to ATH-SR, means Arabidopsis short reads) was RNA sequencing data downloaded from NCBI Sequence Read Archive (SRA) database (http://www.ncbi.nlm.nih.gov), including a chilling-treated sample (4°C; SRA accession: SRX006193) and a control (21°C; SRA accession: SRX006704) sample 80 . After removing low quality reads (polyA/T/G/C sequences) and trimming off four NTs of both ends, all clean reads (28 NTs long) were mapped to Arabidopsis cDNAs (TAIR10) with SOAPaligner. DEGs identification was the same as described above. The DEGs and indentified gene with RPKM > =1 were listed in Additional file 6.

The other data set (referred to ATH-AR, means Arabidopsis array) was Affimetrix microarray data set (Expression Set: ME00325) 81 downloaded from TAIR (http://www.arabidopsis.org). Only cel files for 4 chilling-treated samples (2 for roots and 2 for shoots, 24-hour chilling-treated) and 4 control samples were used here. The cel files were imported into R and analyzed with Affy package 82 . Root and shoot arrays were analyzed separately. Probes expressed in all root or shoot arrays were considered to be presented probes (by mas5 present calls). Differential expressed probes were identified using mas5 method of with FDR corrected p < 0.05 and fold change > =2 and mapped to Arabidopsis transcripts. The gene lists of roots and shoots were combined together to get chilling regulated DEGs and all expressed genes for further analysis (Additional file 7).

We used two methods for functional categorization of unigenes.

To get general gene ontology (GO) annotations for all unigenes, sequences longer than 200 bp were aligned to three public databases (NR, Swiss-Prot and KEGG) by BLASTX with E-value < =1e-5. The GO annotations for the top blast hits were retrieved with Blast2GO program 83 and used to annotate the C. bungeana transcripts. GO functional classification was performed by WEGO website tool 84 .

For GO terms and KEGG pathways enrichment analysis, we used the Arabidopsis annotation systems. Briefly, the sequences of all unigenes were aligned against Arabidopsis peptide database (TAIR10) using BLASTX program with E-value < =1e-5. The top blast hits were considered to be putative orthologous genes (POGs). Then the C. bungeana unigenes were annotated with GO (downloaded from TAIR) and KEGG annotations (ath00001.keg, from http://www.kegg.jp/) for Arabidopsis POGs, respectively. The ontology (GO and KEGG) enrichment was analyzed with BiNGO plugins 49 for Cytoscape 85 , using hypergeometric test for statistical analysis. For p value correction, we used rigorous Bonferroni correction method. The cutoff p value after correction was 0.05. For ATH-SR dataset, since the stressed sample was pooled from seedlings subjected to various periods of chilling-treated (1, 2, 5, 10, 24 hours of stressed) 80 , the expressions of DEGs specific to a certain stage might have been “normalized”. Therefore, to get more information, we used FDR method instead of Bonferronic method for p value correction to find over-representative terms with BiNGO.

The gene-specific primers for real-time PCR analysis were designed using Primer Premier (version 5.0) software (PREMIER Biosoft). The specifities of primer pairs were confirmed by BLASTN with non-redundant unigene set of C. bungeana transcripts and the PCR products were checked by agrose electrophoresis to ensure single band amplifications. The primer sequences for all unigenes used in this study were listed in Additional file 8.

For qPCR analysis, total RNAs were extracted from control or chilling stressed C. bungeana seedlings (two biological repeats) with TRIZOL reagent and treated (20 μg RNA) with 1U DNase (TAKARA, Japan). cDNA was transcribed reversely from 1 μg of DNase-treated RNA with 200U M-MLV reverse transcriptase (Promega, USA) and analyzed with Platinum SYBR green qPCR supermix-UDG reagents (Invitrogen).

Before quantification of unigenes, the geNorm method was applied to select stable expressed unigenes in the four samples as reference genes 86 . A total of 8 candidate reference unigenes were selected for reference genes screening, including an ACTIN2 ortholog, 3 unigenes showed stable expression levels in RNA-seq analysis and the other 4 unigenes were orthologs of recommended Arabidopsis reference genes 87 . The information of reference gene candidates and the geNorm analysis results were shown Additional file 8. Three unigenes (CBT10872/AT3G60800, CBT28565/AT5G27630 and CBT12464/AT2G28390) expressed most stably in control and chilling-treated samples were selected and used for all qPCR analysis.

qPCR analysis was performed with three technical repeats for each sample. The relative expression levels of unigenes were normalized with the three selected reference genes with Pfaffl method 86 88 .