Ploidy analysis of novel saplings and control diploid saplings was performed using flow cytometry analysis and chromosome counts. Results from flow cytometry analysis are shown in Figure 1a,b. Figure 1a represents the DNA content of the control diploids, with a main peak at channel 100. Consequently, the saplings with DNA content peaks only at channel 200 were defined as autotetraploids (Figure 1b). The numbers of autotetraploid in the 5 × 3, 5 × 9, 5 × 11 and 6 × 5 families were 31, 22, 23 and 23, respectively, and the total number of autotetraploids was 99.

The phenotypes of autotetraploid trees differed significantly from diploids in four full-sib families (Figure 2 and Table 1). The breast-height diameter, volume, leaf, fruit and stoma of autotetraploid individuals were significantly larger than those observed in the diploids of each family. The mean breast-height diameter, volume, leaf area, fruit length, fruit diameter and stoma lengths of autotetraploids were 25.09%, 37.31%, 81.43%, 12.37%, 36.97% and 89.72% larger than those of the diploids, respectively. Nevertheless, the heights of the autotetraploid individuals were significantly lower than those observed for the diploids in each family. The mean height of the autotetraploids was only 5.58 m, which was 18.61% lower than that of diploids. The maximal height of the autotetraploid in four families was only 6.51 m, which was 4.96% lower than the mean height observed for the diploids. These results indicate that the vertical growth of the autotetraploids was inferior to that of diploids, but the increase in breast-height diameter and volume and the leaf, fruit and stoma sizes of autotetraploids were superior to those of diploids. Consequently, the autotetraploid trees were stouter, and the diploid trees were more slender.

The apical meristems in plant shoot tips contain a pool of pluripotent stem cells that can self-maintain and produce cells that can differentiate into different cell and tissue types [26], making this tissue a key target for the investigation of plant development and organ formation. Consequently, we constructed a transcriptome from the tender shoot tips of an individual autotetraploid tree and one from its diploid sister from the 6 × 5 family using Illumina paired-end sequencing technology. The transcriptomes of the diploid and autotetraploid produced 4,612,611,600 nt and 4,944,254,580 nt raw data, respectively, and the Q20 percentages were both over 96% (Table 2).

Short reads from two transcriptomic libraries were assembled into unigenes, taking the distance of pair-end reads into account. A library produced from the diploid contained 78,213 unigenes, while a library produced from the autotetraploid contained 88,607 unigenes. The mean lengths of the diploid and autotetraploid unigenes were 625 nt and 622 nt, respectively. All of the sequences were assembled, yielding 84,788 non-redundant unigenes with a mean length of 762 nt, and the N50 was 1294 nt (Table 3).

Approximately 56% of the unigenes (47,324) were annotated by Blastx and Blastn, with a threshold of 10⁻⁵, to six public databases (Nr, Nt, Swiss-Prot, KEGG, COG and GO). Among them, Unigenes 44,187, 38,305, 26,574, 24,472, 15,105 and 17,807 could be annotated to the Nr, Nt, Swiss-Prot, KEGG, COG and GO database, respectively (Figure 3a). Based on Nr annotation and E-value distribution, 74.67% of the mapped sequences showed strong homology (E-value ≤ 10⁻²⁰), and 51.36% showed very strong homology (E-value ≤ 10⁻⁵⁰) to available plant sequences (Figure 3b).

The differentially expressed unigenes (DEUs) were selected based on expression profiles. A total of 7052 unigenes were found to have been upregulated in the autotetraploids compared to the diploids. Moreover, the expression of 3658 unigenes was downregulated in the autotetraploids (Figure 4).

Gene ontology analysis was performed by mapping each DEU to the GO database. In total, 2651 DEUs matched genes in Blast2GO were separated into gene ontology classes according to biological process, cellular component and molecular function (Figure 5). With regard to biological process, metabolic processes (929 unigenes, 35.04%) and cellular processes (866 unigenes, 32.67%) were prominently represented. For cellular component, cell (1422 unigenes, 53.64%), cell part (1255 unigenes, 47.34%) and organelle components (960 unigenes, 36.21%) represented the majorities of this category. Catalytic activity (1195 unigenes, 45.08%) and binding (1125 unigenes, 42.44%) represented a high percentage of the molecular function category. Moreover, cell death, the aminoglycan catabolic process, response to stimulus, the polysaccharide catabolic process and stress differed significantly (corrected p-value < 0.05) between autotetraploids and diploids.

To further investigate biological behavior, KEGG pathway analysis was used to identify the biological pathways of DEUs. There were 3437 DEUs mapped into 122 KEGG pathways, some of which were consistent with biological processes already revealed by GO analyses. The pathway that differed most significantly between the autotetraploids and diploids was plant-pathogen interactions (Ko04626), followed by stilbenoid, diarylheptanoid and gingerol biosynthesis (Ko00945), limonene and pinene degradation (Ko00903), phenylpropanoid biosynthesis (Ko00940), amino sugar and nucleotide sugar metabolism (Ko00520), flavonoid biosynthesis (Ko00941), biosynthesis of secondary metabolites (Ko01110), plant hormone signal transduction (Ko04075) and phenylalanine metabolism (Ko00360).

The DEUs could be divided into two groups related to phenotype and transcriptome differences between the autotetraploids and diploids based on previous knowledge about plant growth. The first group of DEUs were involved in the biosynthesis and signal transduction of indoleacetate (IAA), including eleven genes involved in aldehyde dehydrogenase (EC: 1.2.1.3), indole-3-acetaldehyde oxidase (EC: 1.2.3.7), AUX1, AUX/IAA and GH3 (Figure 6 and Table 4). The second group were genes involved in the biosynthesis and signal transduction of ethylene, including six genes involved in aminocyclopropanecarboxylate oxidase (EC: 1.14.17.4), MAPK, ERF1 and ERF2 (Figure 7 and Table 5). All of these genes were upregulated in the autotetraploid.

To test the reliability of RNA-Seq further, q-PCR was performed with specific primers for the 17 differentially expressed genes involved in the biosynthesis and signal transduction of IAA and ethylene.

Expression patterns revealed by q-PCR analysis were similar to those revealed by RNA-Seq for the same genes (Figure 8). In addition, a significantly positive correlation (r = 0.833; p < 0.01; n = 17) was found between RNA-Seq and q-PCR results from these genes.

The experiment was initiated in April 2004. Seeds from four full-sib families (5 × 3, 5 × 9, 5 × 11 and 6 × 5; all parent plants were located in an intensive seed orchard in Harbin, China) were soaked in 0.1% (w/v) colchicine for 48 h in the dark. A number of seeds from these four full-sib families were soaked in distilled water for the controls. The seeds were sown in a greenhouse after colchicine treatment, and a total of approximately 7000 saplings were obtained. In 2005, 128 novel saplings and 120 control diploid saplings (30 saplings chosen from each family at random) were transplanted into plastic pots in an intensive seed orchard.

The DNA contents of leaves of novel saplings were evaluated by flow cytometry (PA, Partec, Germany) using the methods of Zhang et al. [27]. Briefly, approximately 2 cm² leaf material was chopped with a razor blade in a plastic petri dish containing 1.5 mL nuclei extraction buffer (10 mmol/L MgSO₄·7H₂O, 50 mmol/L KCl, 5 mmol/L HEPES, 1% (w/v) PVP-40, 0.25% (v/v) Triton X-100, pH 8.0). The homogenate was filtered through a 30 μm Partec Celltrics™ nylon filter to remove cell debris, and 1 mL staining solution (Partec) was added. The analysis was performed automatically. The total DNA content of each sample was measured, and the percentages of the cells that showed varied DNA content were processed using DPAC software. Meanwhile, the DNA content of diploid leaves was measured in the control group.

Mitotic chromosomes were isolated from young leaf buds using the protoplast dropping method of Anamthawat-Jónsson [28]. Briefly, buds were collected in ice water and treated for 24 h to arrest the cells at metaphase before fixing the buds in a mixture of three parts absolute ethanol to one part glacial acetic acid. The fixed buds were digested in an enzyme mixture of 2.5% (w/v) cellulase Onozuka R10 (Merck, Darmstadt, Germany) and 2.5% (v/v) pectinase (Sigma-Aldrich, St. Louis, MO, USA) for 5 h at room temperature. After hypotonic treatment with 75 mmol/L KCl for 5–10 min and repeated fixation to clear the cytoplasm, the protoplasts were placed onto microscope slides using a dropper. The chromosomes were stained with carmine and visualized under a microscope (Axioimanger A1, ZEISS, Germany) using 1000× magnification. Chromosomes were counted from 10 to 20 cells in metaphase from each individual tree.

Height, breast-height diameter, volume, leaf, fruit and stoma of diploid and autotetraploid trees were measured in 2009. The areas of mature leaves were measured using a portable area meter (LI-3100C, LI-COR, USA). At least ten healthy, undamaged and fully expanded leaves from the top of each individual tree were measured at random.

Stoma lengths were evaluated using a scanning electron microscope (S-4800, Hitachi, Japan). Briefly, approximately 1 cm² from the center of each leaf was fixed in 2.5% (v/v) glutaral for 24 h and then transferred to 70% (v/v) ethanol for storage. Prior to evaluation, the material was dehydrated through a graded series of alcohol-isoamyl acetate, critical point dried in carbon dioxide for 15 min with a critical point dryer (HCP-2, Hitachi) and coated with gold palladium at 15 nm with an ion coater (EIKO IB-3, Hitachi). At least 30 stomata from each individual tree were viewed and measured at 15 kV using 800× magnification.

The lengths and diameters of the fruits were measured using a vernier caliper. At least ten healthy, undamaged fruits from each tree that had flowered were selected at random and measured. Height and breast-height diameter measurements were taken after the vegetation period ended. Individual tree volumes were calculated following the methods of Meng [29].

For transcriptomic analysis, a single autotetraploid tree and its diploid sister from the 6 × 5 family, both of which were healthy, had flowered and possessed erect stems, were chosen. In June 2011, shoot tips from diploid and autotetraploid trees were collected, immediately frozen in liquid nitrogen and stored at −80 °C until use. Total RNA from each sample was isolated using a modified CTAB method and digested with DNaseI (RNase free) to remove contaminating DNA. Enrichment of mRNA, fragment interruption, addition of adapters, size selection and PCR amplification and RNA-Seq were performed by staff at the Beijing Genome Institute (BGI) (Shenzhen, China). Total mRNA was isolated with oligo (dT) cellulose and broken into short fragments. Using these short fragments as templates, the first-strand cDNA and second-strand cDNA were synthesized. Sequencing adapters were ligated to short fragments after purification using a QiaQuick PCR extraction kit, which were used to distinguish different sequencing samples. Fragments that were 200 ± 25 bp in length were then separated by agarose gel electrophoresis and selected as sequencing templates for PCR amplification. Finally, the two transcriptomic libraries were sequenced using Illumina HiSeq™ 2000. The remaining RNA was used for real-time quantitative RT-PCR verification.

The raw reads were first filtered by removing the adapter sequences and low quality sequences, which included reads with N percentages (i.e., the percentage of nucleotides in a read that could not be sequenced) of over 5% and sequences containing more than 20% nucleotides with a Q-value ≤ 10. The Q-value represents the sequencing quality of related nucleotides. Clean reads were used in de novo assembly and read-mapping to the transcriptome. RNA-Seq data was de novo assembled using the Trinity assembling program [30]. For each library, short reads were first assembled into longer but gapless contigs. Then the reads were mapped back to contigs, taking the distance of paired-end reads as frame. The contigs were connected to access the sequence that could not be extended on either end, and the sequence of the unigene was then produced. Next, unigenes from two libraries were further spliced and assembled to obtain maximum length nonredundant unigenes using TGICL clustering software [31] with a minimum overlap length of 100 bp. After clustering, the unigenes were divided to two classes: clusters, with the prefix CL and singletons, with the prefix Unigene. Finally, unigenes were aligned with the NCBI Nr, Swiss-Prot, KEGG and COG protein databases using Blastx with an E-value ≤ 10⁻⁵. The best results were used to further determine the sequence orientations of the unigenes. If results from different databases conflicted with each other, a priority order of Nr, Swiss-Prot, KEGG and COG was followed to determine the sequence orientation. When a unigene did not align with any of the above databases, ESTScan software [32] was used to predict the coding regions and orientation of the sequence. Blast2GO [33] was used to obtain GO annotation of the unigenes based on Blastx hits against the NCBI Nr database. Blastn was used to align the unigenes to the NCBI Nt nucleotide database, retrieving proteins with the highest sequence similarity with the given unigenes along with their protein functional annotations.

An alignment package, SOAPaligner [34], was used to map reads back to the transcriptome. The number of mapped clean reads for each unigene was counted and normalized into an RPKM value (reads per kb per million reads), which is commonly used to calculate unigene expression [35]. The fold change for each transcript was then calculated using the log₂ formula of autotetraploid RPKM/diploid RPKM. If the value of either autotetraploid RPKM or diploid RPKM was zero, 0.001, rather than 0, was used to calculate the fold change. The method of Audic et al. [36] was used to analyze differential gene expression. The formula used to calculate the probability of a specific gene being expressed equally between the two samples was as follows:

where N₁ and N₂ indicate the total number of clean reads in a diploid and autotetraploid, respectively, and x and y indicate the mapped clean read counts of the transcript in each sample, respectively. The FDR (false discovery rate) method [37] was used to determine the threshold of p-values in multiple tests. In this study, FDR ≤ 0.001 and |log₂ratio| ≥ 1 were the thresholds employed to judge the significance of differentiated gene expression.

The analysis mapped all DEUs to GO terms in the database to calculate the numbers of genes for every term. A hypergeometric test was then performed to find significantly enriched GO terms in DEUs compared to the transcriptome background. The formula used for this calculation was as follows:

where N is the number of all unigenes with GO annotation, n is the number of DEUs in N, M is the number of all unigenes annotated to the specific GO terms and m is the number of DEUs in M. The calculated p-value was subjected to Bonferroni Correction using the corrected p-value of 0.05 as the threshold. GO terms fulfilling this condition were defined as significantly enriched GO terms in DEUs. This analysis was able to recognize the main biological functions that DEUs exercised.

The Blastall program was used to annotate the pathways of DEUs against the KEGG database. The formula used to calculate p-value was similar with that used in the GO analysis. In this formula, N represents the total number of unigenes with KEGG annotation, n is the number of DEUs in N, M is the total number of unigenes annotated to specific pathways and m is the number of DEUs in M.

The expression of candidate genes was confirmed by q-PCR. Primers for these genes were designed manually (Table 6). RNA extraction and DNase treatment were performed as described above, and the first-strand cDNA was synthesized using ReverTra qPCR RT Master Mix with gDNA Remover (Toyobo, Osaka, Japan) according to the manufacturer protocol. The cDNA was diluted tenfold and used as the template for q-PCR.

The q-PCR mixture comprised a 3 μL template of the RT reaction mixture, 10 μL of 2× SYBR Green Master Mix (Toyobo, Osaka, Japan) and 0.5 μL of forward and reverse primer (10 μmol/L) brought to a final volume of 20 μL with water. The reactions were performed in an MJ Opticon™⁻² machine (Bio-Rad, Hercules, CA, USA) using the two-step method that was initiated for 30 s at 94 °C followed by 44 cycles of 94 °C for 12 s, 55 °C for 30 s, 72 °C for 30 s and 78.5 °C for 1 s for plate reading. A melting curve was performed from 55 °C to 95 °C to check the specificity of the amplified product. All experiments were conducted with three biological replicates for each sample. The expression was calculated by 2^−ΔΔCt and normalized to values obtained from the 18S rRNA and α-tubulin controls.