Plant Material

Following the approach used previously for P. euphratica [24], transcripome analyses of P. pruinosa were based on three sets of samples representing control callus, salt-stressed callus and desert-grown trees. For the desert-grown tree samples, we collected three replicate samples of roots, leaves, flower buds, flowers, xylem and phloem from two males and one mature female in the Tarim Basin desert in Xinjiang. Despite the fact that P. pruinosa always occur together with P. euphratica in their natural distributions, three collected P. pruinosa trees used for the present analyses occur far away from the nearby P. euphratica. These trees in the present study were selected in the same physiological state and age class as those used for P. euphratica [24]. No specific permits were required for the described field studies. In addition, we cultured P. pruinosa calli using the method described by Zhang et al. [25] under the same conditions with P. euphratica, replaced the growth medium of one set with fresh, unamended medium and that of another set with fresh medium supplemented with 100 mM NaCl (to impose salt stress), and then harvested both sets 24 h later. The callus from P. euphratica and P. pruinosa has the same passage number and they were highly comparable in terms of origin, physiological state, age, cell types present, morphology, and growth rate within the experimental error. We stored all samples at −80°C prior to RNA extraction.

RNA Extraction and Quality Determination

Total RNA was extracted three times from each of the sample sets, using a CTAB procedure [26]. A260/A280 ratios of the RNA samples dissolved in 10 mM Tris (pH 7.6) ranged from 1.9 to 2.1. The integrity of the RNA samples was examined with an Agilent 2100 Bioanalyzer and their RIN (RNA integrity number) values ranged from 8.6 to 10.0, with no sign of degradation. RNA from each replicate was pooled (in equal volumes) to obtain a single RNA sample for cDNA preparation and RNA-Seq, and equal amounts of mRNA from different tissues of the desert-grown trees were pooled to make single samples.

cDNA Library Construction and Illumina Sequencing

cDNA library construction and sequencing were performed by BGI (Shenzhen, China). The entire process followed a standardized procedure and monitored by BGI's Quality Control System. For cDNA synthesis and Solexa sequencing, 20 µg of total RNA was used, at a concentration of ≥400 ng/µl. Poly(A) mRNA was first isolated using oligo dT beads. Then, mRNA was fragmented into small pieces using divalent cations for 5 min at an elevated temperature (70°C) according to the manufacturer's instructions (Illumina). Based on these cleaved RNA fragments, we used random hexamer-primer and reverse transcriptase (Invitrogen) to synthesize first-strand cDNA. Second-strand cDNA was synthesized using RNase H (Invitrogen) and DNA polymerase I (New England BioLabs). We constructed three paired-end cDNA libraries with insert sizes of 200 base pair (bp), and then sequenced the cDNA using an Illumina (San Diego, CA, USA) Genome Analyzer platform according to the manufacturer's protocols with a read length of 75 bp in two lanes. Image output data from the sequencer was transformed into raw sequence data by base calling.

De Novo Assembly of the P. Pruinosa Transcriptome

We first cleaned raw sequence reads by removing exact duplicates from both sequencing directions. We further cleaned reads by removing adapter sequences as well as reads with too many (>8) unknown base calls (N), low complexity, and low-quality bases (>50% of the bases with a quality score ≤5). Cleaned reads from each library were assembled separately.

De novo assembly of the clean reads was performed using SOAPdenovo software [27] (http://soap.genomics.org.cn), which applies de Bruijn graph algorithm and a stepwise strategy. Briefly, the clean reads were firstly split into smaller pieces, the ‘k-mers’, for assembly to produce contigs using the de Bruijn graph. Next, the reads were realigned to the contig sequences, and the paired-end relationships between the reads were used to construct scaffolds between contigs. Then gap fillings were carried out using pair-end information to retrieve read pairs with one read well aligned on the contigs and another read located in the gap region. The resulting scaffolds with the least Ns were defined as unigenes. The unigenes assembled by short reads from three samples were further clustered into a non-redundant all-unigene set in a comprehensive transcriptome using the TGI Clustering Tool (TGICL) [28] with 50 bp overlap and a minimum of 90% identity.

To obtain high-quality sequences for further annotation and analysis, we excluded unigene sequences that might represent non-coding RNAs. Unigene sequences assigned to known non-coding RNAs (Rfam database: http://www.sanger.ac.uk/Software/Rfam/, release 10.0), microbial (MBGD: http://mbgd.genome.ad.jp/), fungal and virus (based on data downloaded from the NCBI database, fungal: http://www.ncbi.nlm.nih.gov/genome/?term=fungi; virus: http://www.ncbi.nlm.nih.gov/genome/?term=virus) sources were filtered out.

Transcriptome Functional Annotation and Open Reading Frame Identification

We annotated the all-unigenes based on sequential BLAST searches [29] designed to find the most descriptive annotation for each sequence [24]. The assembled unique transcripts were determined by BLASTX against the NCBI non-redundant (Nr) protein database, the Swiss-Prot protein database (http://www.expasy.ch/sprot), the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways database [30], and the Cluster of Orthologous Groups (COG) database (http://www.ncbi.nlm.nih.gov/COG/), applying a E-value cutoff level of <1E − 5. Incongruent results from different databases were settled under a priority order of Nr, Swiss-Prot, KEGG, and COG. Proteins with the highest sequence similarity in blast results were taken to determine the open reading frames (ORF) with the given unigenes.

Based on their Nr annotations, the all-unigenes were assigned GO annotations using the Blast2GO program [31], followed by functional classification using the WEGO software [32], to identify the distribution of gene functions of the species from the macro level (the third GO level). Gene Ontology (GO: http://www.geneontology.org/) is an international standardized gene functional classification system which offers a dynamically updated controlled vocabulary and a strictly defined concept to comprehensively describe properties of genes and their products in any organism. Three ontologies, molecular function, cellular component and biological process, are provided in GO database.

Identification of Rapidly Evolved Genes

Identification of orthologous ESTs between P. euphratica and P. pruinosa was performed using the bidirectional best hit (BBH) method [33]. This method outperforms more complex orthology identification algorithms [34]. Reciprocal batch BLASTP searches were carried out setting the expected value cut off to 1E − 5 to minimize significant matches to paralogous sequences. Hits with a bit score >400 were retrieved for further analysis. Based on translated proteins of the orthologous sequences, ESTs were aligned by the Probabilistic Alignment Kit (PRANK) software [35].

Due to the frequent genome duplications, it is difficult to distinguishing orthologs from paralogs in certain situations when transcriptome sequences are analyzed. In order to establish orthologous relationships between the paired of sequences, we used a series of strict criteria to exclude paralogs. These filtering methods included (i) only those with the corresponding best hits of two ESTs that had a length overlap >150 bp were retrieved; (ii) orthologous pairs with a percent coverage of 90% were retrieved; (iii) aligned sequences with unexpected stop-codons and ambiguous alignments were excluded from further analysis; (iv) all candidate orthologs with a synonymous (Ks) substitution value greater than 0.1 were excluded due to the possibility of being paralogs [1], [12]; and (v) all alignments were manually checked.

Once the putative ortholog pairs were aligned, nonsynonymous (Ka) and synonymous (Ks) substitution ratios were calculated between orthologous coding regions in KaKs_Calculator v1.2 [36] by using the maximum-likelihood YN model [37]. We calculated synonymous and nonsynonymous sites, synonymous and nonsynonymous substitutions, GC contents, and the sequence length after removing gaps and stop codons, in addition to synonymous and nonsynonymous substitution rates and their ratio. Meanwhile, the Fisher's exact test for small sample was applied to justify the validity of Ka and Ks values calculated by this method. To determine if there was any enrichment of GO categories in the positively selected group versus the non-selected group of orthologs, Fisher's exact test was also used to test for over-represented functional categories among positively selected genes (PSGs). For each category C and set of PSGs S, a 2×2 contingency table was constructed for the numbers of genes assigned or not assigned to C and within or outside S. Then, (one-sided) P values for the independence of rows and columns were computed by Fisher's exact test.

We randomly selected 10 pairs of ESTs showing apparent adaptive evolution (Ka/Ks >1) to examine further whether they are orthologs rather than paralogs. We designed common primers for each of these pairs of ESTs. We then used these primers to amplify, clone and sequence DNA from five other individuals of P. euphratica and P. pruinosa. If paired ESTs from two species were paralogs, DNA sequences amplified and sequenced according the conserved region of each pair of ESTs should comprise different gene copies in the same species. These DNA sequences should cluster into at least two groups and each group should include sequences from different species.

Gene Expression MeasurEment

Gene expression levels were measured in RNA-Seq analyses as numbers of reads per kilobase per million mapped reads (RPKM) on exon regions within a given gene [38]. For a given all-unigene, two RPKM values were generated by mapping the subtranscriptome reads of the control and salt-stressed callus to it using SOAP2 with a maximum of three mismatches [39].

De Novo Assembly of the Transcriptome of P. Pruinosa

After removing low-quality sequences and trimming adapter sequences, 28 million 75-bp paired-end clean reads were generated from each of the control-callus, salt-stressed callus and desert-grown tree cDNA libraries in the Illumina Genome Analyzer runs (Table S1). The percentage of Q20 bases for the clean reads in the three sub-transcriptomes were more than 95% (Table S1). In total, these clean reads constitute ∼12 GB of sequence data. The raw reads produced in this study have been deposited in the Short Read Archive (SRA) at the NCBI database under the project accession number SRP018875.

De novo assembly was carried out by SOAPdenovo, a genome assembly program developed specifically for next-generation short-read sequences [27]. The clean reads from the control-callus, salt-stressed callus and desert-grown tree sub-transcriptomes were assembled into 226 299, 200 532 and 1 497 985 contigs, respectively. The average contig size exceeded 101 bp in all three libraries (Table 1). The high heterozygosity of this wind-pollinated species may have resulted in the high number of contigs and the small length of contigs for desert-grown trees sample (Table 1). After using paired-end information to join the contigs into scaffolds and local assembly, we generated 119 471 scaffolds for control-callus samples, 106 812 scaffolds for salt-stressed callus samples and 161 908 scaffolds for desert-grown trees, with average lengths of 381, 392 and 226 nt, respectively. These scaffolds were assembled into 122 883 unigenes by TGICL clustering tools, more than those (94 196 unigenes) from P. euphratica [24]. Out of the 122 883 unigenes, 49 349 unigenes were ≥500 bp and 17 402 were ≥1,000 bp, with an average unigenes length of 526 bp and an N50 size of 681 bp (Table 1). The assembly of transcriptome sequencing reads produced more scaffolds than the expressed genes, suggesting the redundant assembled sequences (i.e., more than one sequence per gene). According to 73 013 transcripts in P. trichocarpa (based on Populus trichocarpa v3.0, DOE-JGI, http:://www.phytozome.net/poplar), the number of all-unigenes is clearly >1.5× more than the number of genes in a poplar species. The size distribution for these unigenes is shown in Figure S1, with the majority being shorter sequences. Numerous short sequences may result from the fragmented genes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Overview of the de novo assembly of the transcriptome of P. pruinosa.

https://doi.org/10.1371/journal.pone.0066370.t001

To evaluate the quality of the dataset, we analyzed the gap ratio of the assembly and the ratio of the gap's length to the length of all-unigenes (Figures S2 and S3). The majority of the unigenes showed gap lengths that were less than 5% of the total length, which accounted for 70.1% of the total unigenes (86 870 unigenes). In addition, sequencing bias was also analyzed by examining random distribution of reads in all-unigenes (Figure S4). All-unigenes were evenly covered by the reads from the control-callus and salt-stressed callus sub-transcriptomes with relatively fewer reads in the 3′ ends of them (Figures S4A and S4B). This observation is consistent with previous reports [40], [41], suggesting that the quality of our dataset was comparable to similar studies of the other non-model species despite a peak in the distribution of reads at 0.3 relative gene length (Figure S4C). However, this did not affect subsequent analyses after clustering the unigenes using the TGI Clustering Tool.

We excluded possible known non-coding RNAs, microbial, fungal and virus sequences identified by comparing our unigene sequences against entries in databases listed in the Materials and methods section. Finally, we identified a total of 114 866 high-quality unique sequences from P. pruinosa when 8 017 contaminated or confounded sequences were excluded.

Functional Annotation

The entire unigene sets were then annotated on the basis of similarities to known or putative sequences in public databases. Among the 114 866 high-quality unique sequences, 67 400 (58.7%) had at least one significant match to an existing gene model in BLASTX searches (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Functional annotation of high-quality unique sequences by sequence similarity (e-value<1e-5).

https://doi.org/10.1371/journal.pone.0066370.t002

First, the all-unigenes of P. pruinosa were assigned putative gene descriptions based on the BLAST search against the NCBI non-redundant (Nr) protein database. Out of the 114 866 all-unigenes, 65 219 (56.8%) showed significant similarity with proteins in the Nr database (Table 2). The proportion of all-unigenes with BLAST hits increased markedly for those with larger sizes (Figure 1). It seemed that longer all-unigenes were more likely to have Nr annotations. The E-value distribution of the top hits in the Nr database also revealed that a larger proportion of all-unigenes longer than 500 bp had strong homology (Figure S5). Following their Nr annotations, we mapped all-unigenes into the records of the GO database and retrieved GO annotations for 11 587 ones (10.1%) (Table 2). This observation of low proportion of all-unigenes assigned to a GO term was comparable to the previous reports [24], [41]. These all-unigenes were assigned into GO terms with three main functional categories, including 6 194 all-unigenes in ‘Biological process’, 7 409 in ‘Cellular component’, and 7 818 in ‘Molecular function’ (Figure 2 and Table S2) while 3 381 all-unigenes had an assignment in all three categories. The remaining all-unigenes failed to obtain a GO term, largely due to their uninformative (e.g. ‘unknown’, ‘putative’, or ‘hypothetical’ protein) descriptions. Within the ‘Biological process’, there were 21 GO categories and the two most abundantly represented lineages were ‘metabolic process’ and ‘cellular process’. There were also a large number of all-unigenes being involved in ‘biological regulation’, ‘localization’, ‘pigmentation’, and ‘response to stimulus’. In the ‘Molecular function’ division, there were 13 GO categories and the top two categories were ‘binding’ and ‘catalytic activity’. The former was mainly represented by genes for ‘nucleotide binding’ and ‘protein binding’, while the latter was mainly represented by genes with ‘transferase activity’, ‘hydrolase activity’, and‘kinase activity’. Finally, in the ‘Cellular component’ division, there were 11 categories and the top two categories being ‘cell’ and ‘organelle’. The matched proportions of GO categories in P. pruinosa ESTs are similar to those for P. euphratica [24], suggesting that our library and Illumina sequencing have adequately sampled the species' total transcriptome.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Length distribution of the all-unigenes with Nr annotations.

More than 80% of the all-unigenes over 500 bp had BLAST hits in the Nr database.

https://doi.org/10.1371/journal.pone.0066370.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. GO categories of the P. pruinosa all-unigenes.

WEGO was used to produce the graph. The results are summarized in ‘biological process’, ‘cellular component’, and ‘molecular function’. The percentage (left y-axis) and total number (right y-axis) of all-unigenes in each category (the third GO level) are shown. In total, 11 587 all-unigenes have been assigned. Y axis is in log(10) scale.

https://doi.org/10.1371/journal.pone.0066370.g002

Identification of Putative Orthologs Between P. Pruinosa and P. Euphratica

We recovered 68 526 CDSs for P. pruinosa, more than those recovered (59 721) from P. euphratica [24]. The average CDSs length for the P. pruinosa and P. euphratica were 437 and 532 respectively (Table S3). We identified 23 167 pairs of orthologs between P. pruinosa and P. euphratica, through a reciprocal best hit blast search. The orthologous distributions (Figure S6) suggested that most non-orthologous pairs were excluded. After a series of strict filtration, we then obtained 2859 pairs of transcripts that are highly orthologous between species. The median length of these transcripts is 1079 bp, ranging from 573 to 3555 bp (Table S3).

Ka/Ks Between Pairs of Orthologs

Out of the 2859 ortholog pairs, divergence was sufficiently high for 2339 orthologs (81.8%) to allow both Ka and Ks rates to be calculated (Table S4). Some of them showed high Ka/Ks values. Of these, 84 pairs of orthologs (2.9%) have a Ka/Ks >1, indicating positive selection, and 355 pairs (12.4%) have a Ka/Ks between 0.5 and 1, indicating weak purifying selection (Figure 3). The average synonymous substitutions and nonsynonymous substitutions for these 84 ortholog pairs were 2 and 8, respectively. For the remaining pairs of orthologs, we could calculate either only Ka (332 pairs of orthologs, 11.6%), or Ks (99 pairs of orthologs, 3.5%), or incalculable estimations (89 pairs of orthologs or 3.1%) (Table S4). The functions of the homologous genes of those pairs of orthologs with Ka/Ks>1 were mainly involved in ‘responses to stress’, ‘ubiquitin-dependent protein catabolic processes’, and ‘biological regulation’ (Table S5 and Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Distribution of Ka and Ks for 2 859 pairs of the putative orthologs.

The orthologs with Ka/Ks>1 fall above the red line while those with Ka/Ks = 0.5−1 fall between the green and red lines.

https://doi.org/10.1371/journal.pone.0066370.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. GO categories over-represented among the homologous genes of the orthologs with positive selection (P-values from the Fisher's exact test).

https://doi.org/10.1371/journal.pone.0066370.t003

Gene ontology: F, molecular function, P, biological process, C, cellular components.

For those pairs of ESTs with the sign of adaptive evolution, we randomly selected 10 pairs and designed the starting and ending primers across two species. The adaptive mutations were confirmed by amplifying and sequencing five individuals of each species (Figure S7). In addition, for each locus, all sequences from each species clustered respectively, similarly suggesting that all pairs of ESTs for analyses are orthologous rather than paralogous. For each pair of orthologs, most mutations indicated by EST comparisons were confirmed as species-specific; however, a few of them were found to be ‘introgressed’ into the other from one species (for example, Unigene2702_All, site 190, Figure S7). In addition, we found that all of five individuals in one species possessed additive polymorphisms at some sites while those of the other species were fixed for one of two additive polymorphisms (for example, Unigene37561_All, site 607, additive polymorphisms W = A/T for P. euphratica and only ‘A’ for P. pruinosa, FigureS7).

Expression Changes of Rapidly Evolved Orthologs Under the Salt Stress

For the positively selected genes, gene expression levels were measured and RPKM values were generated by mapping the reads from the control and salt-stressed callus subtranscriptomes to it for P. euphratica and P. pruinosa (Table S6). Based on the log 2 ratio value, we found that only one pair of the orthologous ESTs (Unigene56801_All and Unigene25389_All in P. euphratica and P. pruinosa, homologous to Arabidopsis Proteinase inhibitor I4, serpin) has a higher expression level in salt-stressed callus when compared with control callus in both P. euphratica and P. pruinosa. Four pairs of orthologous ESTs (Unigene7552_All, Unigene65280_All, Unigene5742_All and Unigene34836_All in P. euphratica, homologous to Arabidopsis UDP-glucose:glucosyltransferase, glycosyltransferase, wax synthase isoform 3, and Ripening-related protein grip22) have higher expression levels in salt-stress callus when compared with control callus in P. euphratica but not in P. pruinosa. Three pairs of orthologous ESTs (Unigene45594_All, Unigene5696_All and Unigene18546_All in P. pruinosa, homologous to Arabidopsis MYB family transcription factor, WAK-like kinase, and NAC domain protein, IPR003441 (NAC036)) have higher expression levels in salt-stress callus when compared with control callus in P. pruinosa in but not in P. euphratica (Figure 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Gene expressions of 84 pairs of orthologs in P. euphratica and P. pruinosa under salt stress.

Red dot – higher expression levels in both P. euphratica and P. pruinosa; green dots – higher expression levels only in P. euphratica; and blue dots – higher expression levels only in P. pruinosa.

https://doi.org/10.1371/journal.pone.0066370.g004