Background

Adaptation to various abiotic stresses is critical for the survival and biomass accumulation of sessile plants and is particularly true for perennial tree species due to their relatively long-life cycle. Populus, a model tree species due to the availability of a near complete set of experimental resources such as easy propagation, transformation methods, and abundance of genetic and genomic materials [1, 2], provides an ideal system to uncover how perennial trees adapt to abiotic stresses.

Recent studies in Arabidopsis thaliana and other model plants show that plants experiencing sub-lethal abiotic stress may memorize that stress at physiological and transcriptional levels to promote better performance when they encounter the same stress again [3,4,5,6,7,8]. For example, Arabidopsis and maize plants that have experienced one or more cycles of dehydration stress and watered recovery exhibit greater ability to retain leaf relative water content (RWC) compared to plants that have no experience of dehydration [9,10,11]. At the transcriptional level, stress response genes can be divided into two types based on their responsiveness to successive stress: ‘memory genes’ that show significantly different levels of up- or down-regulation in subsequent stress than the previous one, and ‘non-memory genes’ that show similar responses to each stress [12]. Comparative studies between Arabidopsis, maize, and switchgrass found that there are conserved dehydration memory genes but also remarkable differences in the total number and homologs of dehydration memory genes [10, 13], suggesting the existence of both evolutionarily conserved and species-specific mechanisms regulating plant responses to repeated abiotic stresses.

Transcriptional regulation is dynamic, and different types of genes usually respond to perturbations with different kinetics and patterns. Thus, it would be much more informative to profile plant transcriptional responses to abiotic stress with multiple time points of gene expression data rather than with a single time point [14,15,16]. The increasing capacity of high-throughput sequencing makes it feasible to characterize whole genome transcriptional dynamics with time-series RNA-seq experiments. Coupled with computational analysis, it is possible to identify key gene modules, hub genes or infer the hierarchical structure of the regulatory network using this kind of time-series expression data [16,17,18,19,20,21,22,23], and provide a better overview of how underlying biological processes are regulated.

High salinity, usually presented by accumulation of NaCl in soil, causes osmotic and ionic stresses, and is one of the most widely spread abiotic stresses that limit plant growth and distribution [24, 25]. Therefore, it is practically important to investigate the mechanisms of how plants adapt to salt stress. In this study, we established a precisely controlled liquid culture-based experimental procedure to investigate plant adaptation to salt stress in Populus. We first evaluated the phenotypic responses to salt stress and then performed time-course RNA-seq to characterize transcriptome dynamics during salt treatments. Our results showed that Populus plants displayed quick adaptation to salt stress phenotypically and transcriptionally. Key gene modules were identified through co-expression analysis, and the biological relevance of these gene modules were analyzed.

Results

Phenotypic adaptation to salt stress in Populus

In order to study plant adaptation to salt stress, we adopted the liquid culture system of Populus that allows precisely controlled NaCl treatments. We set up two groups of experiments: for the first group, we used 200 mM NaCl to do a high salt treatment directly; for the second group, we used 100 mM NaCl to do 1 day low salt treatment, then followed by 3 days of recovery culture, and finally high salt treatment with 200 mM NaCl. In the first group, plants quickly showed a significantly severe stress phenotype, with strong shoot apical bend and leaves dropping. By contrast, in the second group, plants treated with low salt acted similar to the control plants, indistinguishable in phenotype to the control plants after 3 days of recovery and showed a mild response following 200 mM NaCl salt treatment (Fig. 1a). Time-course analysis further supported improved plant tolerance to NaCl treatment in the second group: plants responded to 200 mM NaCl treatment after 0.5 h and started to recover after 6 h while plants directly treated with 200 mM NaCl showed a much stronger phenotype after 0.5 h and started to recover after 12 h (Additional file 1: Figure S1). Detailed inspection of the leaf phenotype showed consistent results with whole plant responses (Fig. 1b), with severe leaf damage in the first group, and mild damage in the second group. Overall, these results indicated that Populus plants could physiologically adapt to salt stress quickly after a pre-treatment.

Fig. 1
figure 1

Plant phenotypic responses during repeated NaCl treatments in Populus. a Whole plants responses to salt stress. Photos were taken 1 h after each NaCl treatment. Scale bar, 20 cm. b Leaf phenotype during salt stress. Photos were taken 24 h after each NaCl treatment. Scale bar, 20 cm

Full size image

RNA-seq reveals massive transcriptional reprogramming in repeated salt stress

We next performed RNA-seq to profile the dynamic changes of genome-wide transcript abundance during repeated salt treatments (Methods). We first treated the plants with 100 mM NaCl for 1 day, followed by 3 days of recovery culture, and then repeated 100 mM NaCl treatment again. Samples were collected after 0 h (control samples, designated as CK1 and CK2, respectively), 1 h, 3 h, 6 h, and 12 h (designated as T1H1, T1H3, T1H6, T1H12, T2H1, T2H3, T2H6, and T2H12, respectively) of each salt treatment (Fig. 2a). Notably, for RNA-seq sample preparation, we used 100 mM NaCl for both salt treatments to avoid the influence of different NaCl concentrations on transcription, and all time points for each treatment were collected within 2 h to minimize the influences of circadian rhythms or other environmental factors. High quality total RNAs were isolated and submitted for RNA-seq library construction and sequencing. At least 45–76 million 150 bp paired-end clean reads were obtained for each library (Additional file 7: Table S1).

Fig. 2
figure 2

Overview of the RNA-seq data. a Workflow for RNA-seq sample preparation. b Venn diagrams show overlap of differentially expressed genes (DEGs) between the same time point of repeated NaCl treatment. T1: the first salt treatment; T2: the second salt treatment after 3 days recovery following the first salt treatment; H1/H3/H6/H12: 1/3/6/12 h of salt treatment

Full size image

To identify differentially expressed genes (DEGs) during each salt treatment, we performed pair-wise edgeR analysis between individual time point and the corresponding control samples. The number of DEGs changed dramatically along the time-course of each treatment (Additional file 8: Table S2; Additional file 9: Table S3). Moreover, DEGs at the same time point of the first treatment (T1) and the second treatment (T2) were dramatically different. For example, there were 2171 DEGs in the T1H1/CK1 group whereas there were only 179 in the T2H1/CK2 group. Overlapping studies found that there were 145, 836, 473, and 1551 common genes for each time point (Fig. 2b), which account for 81, 41, 54, and 73% of the smaller dataset, respectively. Overall, the differences in the total number of DEGs and the low overlapping rates indicate that massive transcriptional reprogramming during repeated salt treatments occurred. Notably, there were 1051 DEGs between CK1 and CK2, with 649 up-regulated and 402 down-regulated DEGs. GO function annotation showed that 43 out of 1051 DEGs were involved in response to salt stress. Among those 43 DEGs, 19 were up-regulated, including membrane transporter proteins and MYB transcription factors; 24 were down-regulated, including tonoplast intrinsic protein TIP1, drought-induced protein Di19, gibberellin-regulated protein GASA14, and calcium signaling proteins CDPK and GAP1 (Additional file 9: Table S3).

Co-expression analysis identifies key gene modules in response to repeated salt stress

To better understand the correlation of gene expression between time-course treatments, weighted gene correlation network analysis (WGCNA) was performed. In total, 10 co-expressed gene modules, which showed different expression curves and peaks during NaCl treatments, were identified (Fig. 3a; Additional file 2: Figure S2; Additional file 10: Table S4). Module size ranged from 692 to 8341 genes, with the green module containing a moderate amount (2071 genes), while the blue (7173 genes) and turquoise (8341 genes) modules contained the largest number of genes (these three modules showed significant correlation with salt stress response or adaptation as discussed later) (Table 1).

Fig. 3
figure 3

Co-expressed gene modules identified by Weighted Gene Co-expression Network Analysis (WGCNA). a Module dynamic eigengene expression in response to NaCl treatments (T1, T2). The numbers in parentheses represent the number of genes in the module. b Correlation of individual gene module with Treatment (T1, T2) and Time (0 h, 1 h, 3 h, 6 h, 12 h). The number stand for correlation value and related P value (in parentheses) between the module and “Treatment” or “Time”. P value < 0.05 was the statistical significance threshold

Full size image
Table 1 Summary of all co-expression gene modules
Full size table

Gene expression profile was visualized with eigengene values (the first principal component of transcript profiles) for each module and showed distinct co-expression patterns across modules (Fig. 3a): the blue module increased along the time-course and showed similar changes between the first and the second treatment, the green module showed a significant peak at 1 h of the first treatment compared to the second treatment, and the turquoise module showed a significant peak at 3 h of the second treatment compared to the first treatment, the other seven modules did not have significant changes across the time-course. Consistently, Module-Trait relationships showed that the green and turquoise modules highly correlated with “Treatment” (T1 and T2), whereas the blue module significantly correlated with “Time” (1 h, 3 h, 6 h, and 12 h) (Fig. 3b). Other seven modules showed relatively low correlation to any variable. These results indicated that the green and turquoise modules may be critical for adaptation to repeated salt stress, whereas the blue module appears to be essential for each salt stress response. Therefore, we focused on analyzing the green, turquoise, and blue modules.

The salt overly sensitive (SOS) pathway is essential for salt tolerance in plants [26]. There are three homologs of SOS1, two of SOS2, and five of SOS3 in Populus genome. We checked our co-expression analysis and found that two homologs of SOS1 were classified into turquoise module, two of SOS2 were classified into blue module, and one of the SOS3 was in green module and one was in turquoise module (Additional file 10: Table S4).

Transcription factors (TFs) are important regulators in plant development and response to stress [27,28,29,30,31,32,33]. Therefore, we investigated the distribution of TF genes in these three gene modules. Compared to the whole genome, the green and blue modules contained a significantly higher percentage of TF, whereas the turquoise module contained a significantly lower percentage of TF (Table 1; Additional file 11: Table S5). Additional file 12: Table S6 shows a summary of individual TF families in the green, turquoise, and blue modules. Statistical analysis found that the green module was significantly enriched in the MYB/MYB-related, ERF, bHLH, and HD-ZIP families of TFs, whereas the blue module was significantly enriched in the NAC, bZIP, WRKY, C3H, B3 and HSF families of TFs. These results provide valuable information for identifying key regulators that underly plant responses to salt stress in general, but more importantly, adaptation to salt stress.

Gene ontology enrichment analysis of the key gene modules

To further explore the functional significance of the key gene modules, we performed gene ontology (GO) enrichment analysis to identify significantly enriched biological pathway (BP), molecular function (MF), and cellular component (CC) GO terms of each gene module. In total, 399, 154, and 681 enriched BP GO terms were identified from the blue, green, and turquoise modules, respectively (Additional file 13: Table S7).

Plant hormones are important regulators for abiotic stress responses. In this study, we found enriched GO terms for all classical plant hormone pathways in the selected gene modules (Table 2). Interestingly, genes functioning in auxin polar transport and signaling are mainly enriched in the green and turquoise modules, whereas genes functioning in abscisic acid (ABA) and salicylic acid (SA) signaling pathways are only enriched in the blue module. Jasmonic acid (JA) signaling pathways are highly enriched in the blue module. Genes functioning in responses to plant hormone gibberellins (GA), brassinosteroid (BR), and ethylene (ET) are also enriched in the green module. These results indicate that plant hormones, especially auxin may be critical for plant adaptation to repeated salt stress while ABA, SA, and JA are essential for plant responses but not sufficient for plant adaptation to each salt stress.

Table 2 Enrichment of plant hormone and cell wall related gene ontology (GO) categories in each gene module
Full size table

Detailed analysis confirmed that each module has specific enriched GO terms. For example, the green module showed enrichment in negative regulation of growth (Additional file 3: Figure S3; Additional file 13: Table S7); the turquoise module is enriched in GO terms involved in RNA transport and catabolic, protein modification and transport, and meristem development (Additional file 4: Figure S4; Additional file 13: Table S7); and the blue module is enriched in regulation of mitotic cell cycle, protein dephosphorylation, programmed cell death, and response to abiotic stress (Additional file 5: Figure S5; Additional file 13: Table S7). Remarkably, cell wall related GO terms are highly enriched in green module, such as plant-type cell wall biogenesis, plant-type cell wall modification, cell wall loosening, and lignin/pectin metabolic or catabolic process (Table 2); and the turquoise module is significantly enriched in epigenetic modification pathways such as histone methylation, DNA methylation, and chromatin organization (Additional file 4: Figure S4; Additional file 13: Table S7).

Verification of gene expression profile of the green, turquoise and blue module

To validate the gene expression profiles derived from RNA-seq analysis, we randomly selected six genes from green, turquoise, and blue module with relatively higher module membership value (MM value) and expression level for quantitative PCR (qPCR) test. Similar expression patterns were found for all selected genes (Fig. 4; Additional file 6: Figure S6), suggesting that our results of co-expression analysis with RNA-seq data are reliable for analyzing gene expression in response to repeated salt stress in Populus.

Fig. 4
figure 4

Verification of the expression profile of representative genes from green (a), turquoise (b), and blue (c) module by quantitative real-time PCR. Relative gene expression level was calculated using actin as the internal control. The results were means ± SE with three biological replicates for each sample

Full size image

Co-expression network exploration

Co-expression network analysis shows the correlation of genes based on their expression patterns, showing possible causality in expression changes of co-expressed genes under certain experimental conditions. WOX4-CLE41 signaling pathway has been shown as the central regulator of cambium cell fate controlling secondary growth rate in Populus [34, 35], therefore, we used the WOX4b and CLE41a genes from green module as an example to further explore the co-expression network derived from this study. Top 20 co-expressed genes were displayed for each gene (Fig. 5; Additional file 14: Table S8). In WOX4b co-expressed genes, MIOX4, bHLH TF, two peroxidases, several kinds of transporter genes, and cell wall biogenesis and lignification related genes such as pectin lyase-like superfamily protein, plant invertase/pectin methylesterase inhibitor superfamily, and xyloglucan endoxyloglucan transferase XTH6 were found. Among CLE41a co-expressed genes, actin depolymerizing factor ADF5, LOL1, MLP28, gene encoding peptidase C15, protein kinase, and notably, WOX4b were found. Moreover, CLE41a was reported to be regulated by auxin and there was one SAUR-like auxin-responsive gene and IAA16 among the CLE41a top 20 co-expressed genes. These co-expressed genes are possibly important for the regulation of plant growth during salt stress.

Fig. 5
figure 5

Illustration of network study with our co-expression data, using WOX4b and CLE41a genes from green module as an example and displaying their top 20 co-expressed neighbor genes

Full size image

Discussion

Transcriptional regulation has been shown to play important roles in plant response and adaptation to abiotic stresses [5, 36, 37]. Therefore, identifying key transcriptional regulatory genes and signaling pathways that participate in the regulation of plant response to abiotic stresses is critical for genetic or biotechnological improvement of plant survival rate and biomass production. Recently, there have been several comprehensive studies on plant adaptation to abiotic stresses. However, our knowledge of how perennial trees respond to repeated abiotic stress is still limited. In this study, we used hybrid poplar clone 84 K (Populus alba X Populus glandulosa) as a model system to investigate plant responses to repeated salt stress, and demonstrated the adaption of perennial trees to abiotic stresses.

Populus plant could adapt to salt stress quickly at both physiological and transcriptional levels

We developed a liquid-culture system, which enabled us to precisely control the NaCl dosage and time of treatments, to investigate how Populus plant’s responds to repeated salt stress. Observation of whole plant responses and leaf phenotype to repeated salt stress demonstrated that plants experiencing 1 day of low NaCl pre-treatment followed by 3 days recovery perform much better in high NaCl treatment than plants treated directly with high NaCl (Fig. 1). This suggests Populus could adapt to salt stress quickly at the physiological level.

We also performed time-course RNA-seq to profile transcriptional dynamics in response to NaCl treatments, which provide a better resolution than single time point RNA-seq data. The total number of DEGs over the time-course and DEGs pair-wise overlapping study revealed that there were dynamic and massive transcriptional reprogramming in successive salt treatments (Fig. 2; Additional file 8: Table S2). Notably, there were less DEGs in T2 at all time points except 3 h, in which T2H3/CK2 has more DEGs than T1H3/CK1. Furthermore, the turquoise module from our co-expression analysis, which was the largest module and contain 8341 genes, has a relatively higher expression peak in T2H3 than T1H3. The turquoise module was enriched in GO categories such as auxin polar transport and signaling, RNA transport and catabolic, protein modification and transport, meristem development, and epigenetic modification pathways that were possibly important for salt adaptation. Therefore, we thought this could be reflective of plant response to repeated salt stress. Another interesting point from DEG analysis was that, there were still over 1000 genes that showed significantly differential expression after 3 days recovery culture (Additional file 8: Table S2) even though plant phenotype returned to pre-treated status, suggesting that transcriptional regulation is more sensitive than physiological regulation and 3 days recovery culture is not enough to completely reverse the transcriptional changes. This phenomena is consistent with previous results reported by others that plants retains the transcriptional memory of abiotic stress for up to 5 days [8, 9]. It would be interesting to investigate how their transcriptional levels are retained during the recovery period.

Co-expression analysis identify putative gene modules critical for salt stress adaptation

With this time-series RNA-seq data, gene co-expression analysis was performed using the WGCNA package and 10 modules were identified. Each module has different gene expression profiles along NaCl treatments. Based on the analysis of module-trait correlation we decided to focus on the green, turquoise, and blue modules which showed significantly high correlation to repeated “Treatment” or “Time” (Fig. 3). Overall, genes in the green module reached their expression peak at 1 h of the first treatment but not the second treatment, whereas genes in the turquoise module showed an expression peak at 3 h of the second treatment but not the first treatment. This suggested that genes from these two modules could discriminate the first and repeated salt stresses and possibly lead to salt stress adaptation. On the other hand, expression levels of genes in the blue module increased along the time-course and showed similar patterns between the first and the second salt treatment, indicating that genes from this module are important for the plant response to salt stress in general.

Study of genome-wide TF distribution found that the green and blue modules contained significantly more TF genes whereas the turquoise module contained significantly less, indicating that the green and blue modules may act at a higher hierarchical level of transcriptional regulation that initiates a transcriptional signaling cascade to regulate downstream gene expression. The green module is significantly enriched in MYB/MYB-related, ERF, bHLH, and HD-ZIP families of TFs and the blue module is significantly enriched in NAC, bZIP, WRKY, C3H, B3 and HSF families of TFs. All of these TF families have been shown to be important regulators of abiotic stress responses [27, 28, 30,31,32,33, 38, 39]. However, whether they are participating in the regulation of stress adaptation still needs further investigation. This analysis provides putative candidate genes for genetic studies of plant transcriptional responses to abiotic stress for future studies.

Maintaining ion homeostasis is crucial for plant tolerance and adaptation to salt stress. The salt overly sensitive (SOS) pathway, including SOS1, SOS2, and SOS3, has been shown as a central player in sodium ion efflux under salt stress [40,41,42], and is conserved across different plant species such as Arabidopsis [26], rice [43], tomato [44], and poplar [45]. In our co-expression analysis, two homologs of SOS1 were classified into turquoise module, two of SOS2 were classified into blue module, one of SOS3 was in green module and one was in turquoise module (Fig. 4; Additional file 10: Table S4). The H+ electrochemical gradient across membranes could drive the Na+ compartmentalization into vacuoles or exclusion out cells, and is generated by H+ pump such as H+-pyrophosphate VP1 [46]. Two VP1 homolog genes were belong to turquoise module in our co-expression analysis (Additional file 10: Table S4), one of which have been functionally characterized in poplar and could improve plant growth under salt stress [47]. Together, these results suggest that Populus tree may adapt to repeated salt stress through modifying ion sensing and transporting pathways.

Functional analysis of the gene modules suggested plant hormones and epigenetic modifications are critical for Populus plant’s adaptation to salt stress

Plant hormones are important regulators of plant responses to abiotic stress and complex crosstalk occurs among the signaling pathways of different hormones for coordinating plant growth [29, 41, 48,49,50,51,52]. In general, ABA, SA, JA, and ET are characterized as “stress hormones” which are responsible for a plants’ quick response to biotic or abiotic stresses, whereas auxin, GA, BR, and cytokinin are recognized as “growth promoting hormones” which modify plant developmental processes in the long-term [49, 53]. Interestingly, our GO enrichment analysis with co-expression gene modules suggested that plant hormones play important roles in Populus response and adaptation to repeated salt stress. Moreover, the above two types of plant hormones may function at different aspects of plant response and adaptation to abiotic stresses. For example, we found that the ABA signaling pathway was only enriched in the blue module, suggesting that ABA signaling is necessary for plant response but not sufficient for plant adaptation to repeated salt stress, which is consistent with previous reports [9, 54]. JA and SA signaling pathways were also specifically enriched in the blue module. On the other hand, auxin, GA, and BR signaling or transport pathways were highly enriched in the green or turquoise modules, which indicated that these plant hormones may play critical roles in plant adaptation to salt stress through modulating the plant growth pattern in the long-term. There are also highly enriched GO terms referring to negative regulation of growth and cell wall biogenesis/modifications in the green module. It would be important to analyze how these plant hormones change and identify their crosstalk signaling pathways in the plant response and adaptation to abiotic stress, thus uncovering key signaling genes for improving plant growth during abiotic stresses.

Epigenetic regulation has been proposed as one of the mechanisms underlying transcriptional stress memory [37, 55,56,57,58]. For example, histone trimethylation markers H3K4me3 and H3K27me3 correlate to the target gene’s transcriptional memory in dehydration stress [9, 59, 60]. Also, DNA methylation changes in particular regions of the Arabidopsis genome are responsible for hyperosmotic stress memory caused by salt treatments [61]. In our results, we found epigenetic pathways, such as histone methylation, histone ubiquitination, DNA methylation, and chromatin organization, were highly enriched in the turquoise module which correlated with “Treatment” and showed a peak expression level in T2H3 (Fig. 3). To further reveal the mechanisms of plant adaptation to abiotic stress, it would be important to investigate when do these epigenetic modifications establish, the effects of these epigenetic modification on the expression of target genes, and their dynamics during the recovery period.

Conclusions

In this study, we reported that Populus plants could adapt to salt stress quickly in both physiological and transcriptional levels. We also did time-series RNA-seq to profile the transcriptome dynamics during repeated salt treatments and showed that there was significant reprogramming at the transcriptional level. Our co-expression analysis with these time-series RNA-seq data identified 10 co-expressed gene modules, including two modules which were highly correlated to repeated salt stress and one module which was highly correlated to general salt stress response. Gene ontology (GO) analysis suggested that plant hormones regulating plant growth, particularly auxin signaling pathway, are possibly play critical roles in plant adaptation to repeated salt stress, and ABA signaling is critical for general response to salt stress. Cell wall biosynthesis/modification and epigenetic regulation are also suggested as important regulatory pathways for plant adaptation to salt stress. In summary, our results provided a framework for dissecting signaling pathways and identifying key regulatory genes for plant adaptation to salt stress in Populus.

Methods

Plant cultivation and sample collection

Hybrid poplar (Populus alba X Populus glandulosa) clone 84 K, grown under 16 h light/8 h dark photoperiod condition and at 25 °C, was used for all experiments. All plants were propagated in Shandong Agriculture University (Taian, Shandong, China). 84 K seedlings were subcultured in magenta boxes under controlled growth chamber, and then transferred to soil 1 month later, finally transferred to full Hoagland nutrient solution for another 3 weeks growth before NaCl treatments. Defoliated stems from apex to fourth internodes were collected and immediately frozen in liquid nitrogen for RNA extraction. Three biological replicates were collected for each sample, and three plants were combined for each biological replicate.

RNA extraction and qPCR

Samples collected from apex to 4th internode were ground to fine powder in liquid nitrogen. Total RNAs were extracted with CTAB method, treated with DNase (TaKaRa, 2270) and then purified with column from TaKaRa MiniBEST Plant RNA Extraction Kit (TaKaRa, 9769). The RNA purity and integrity were assessed by NanoDrop 2000 and Agilent 2100 Bioanalyzer. 0.5 μg total RNA was used for cDNA synthesis using HiScript II Q Select RT SuperMix for qPCR (+gDNA wiper) (Vazyme, R233–01). Vazyme-ChamQ SYBY Color qPCR Master Mix (Vazyme, Q411–02) were used for qPCR. Gene-specific primers were listed in Additional file 15: Table S9. Relative gene expression level was calculated using actin as the internal control. Three biological replicates were prepared for each time point.

RNA-seq data analysis

High quality RNAs were submitted for mRNA sequencing library preparation and 150 bp paired-end sequencing on Hiseq X 10 platform (Illumina). Clean sequencing reads were mapped to P. trichocarpa v3.0 genome assembly using hisat2 [62, 63] with default parameters. The raw mapped reads for each sample were counted using htseq-count [64]. The edgeR package [65] was used to identify significantly differentially expressed genes (DEGs), with Fisher’s exact test false discovery rate (FDR) less than 0.05 as the statistical significance threshold.

Gene co-expression network analysis

The gene expression abundance was calculated and used for weighted gene co-expression network analysis (WGCNA) [66]. The soft threshold power of the adjacency matrix for co-expression relationship between genes is 12. Hierarchical clustering was executed with a minimum module size of 300 and a cut height of 0.994. The different modules were assigned to different colors. The Weighted Correlation Network Analysis (WGCNA) included method testing the correlation between co-expression module and sample traits. The gene expression matrix of the module was computed by PCA (Principal Components Analysis) to determine the PC1, Module Eigengene (ME), which represents the module. Then, the module-trait correlation was computed using all module MEs and trait data. P value < 0.05 was used as the statistical significance threshold. The co-expression network was analyzed using Cytoscape [67]. Transcription factor (TF) database is derived from Plant TFDB (http://planttfdb.cbi.pku.edu.cn/index.php?sp=Ptr) [68].

Gene ontology (GO) enrichment analysis

P.trichocarpa v3 Gene ontology (GO) annotation was used, and the GO enrichment was analyzed using GOstats and GSEABase packages [69] with P-value < 0.01. REVIGO [70] was used for visualization of enriched GO terms with default parameters.