Nitrogen deprivation promotes Populus root growth through global transcriptome reprogramming and activation of hierarchical genetic networks


  • Hairong Wei,

    1. School of Forest Resources and Environmental Science, Michigan Technological University, Houghton, MI, USA
    2. Biotechnology Research Center, Michigan Technological University, Houghton, MI, USA
    3. Computer Science, Michigan Technological University, Houghton, MI, USA
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Yordan S. Yordanov,

    1. School of Forest Resources and Environmental Science, Michigan Technological University, Houghton, MI, USA
    Search for more papers by this author
    • These authors contributed equally to this work.
  • Tatyana Georgieva,

    1. School of Forest Resources and Environmental Science, Michigan Technological University, Houghton, MI, USA
    Search for more papers by this author
  • Xiang Li,

    1. Computer Science, Michigan Technological University, Houghton, MI, USA
    Search for more papers by this author
  • Victor Busov

    Corresponding author
    1. School of Forest Resources and Environmental Science, Michigan Technological University, Houghton, MI, USA
    2. Biotechnology Research Center, Michigan Technological University, Houghton, MI, USA
    • Author for correspondence:

      Victor Busov

      Tel: +1 906 487 1728


    Search for more papers by this author


  • We show a distinct and previously poorly characterized response of poplar (Populus tremula × Populus alba) roots to low nitrogen (LN), which involves activation of root growth and significant transcriptome reprogramming.
  • Analysis of the temporal patterns of enriched ontologies among the differentially expressed genes revealed an ordered assembly of functionally cohesive biological events that aligned well with growth and morphological responses. A core set of 28 biological processes was significantly enriched across the whole studied period and 21 of these were also enriched in the roots of Arabidopsis thaliana during the LN response. More than half (15) of the 28 processes belong to gene ontology (GO) terms associated with signaling and signal transduction pathways, suggesting the presence of conserved signaling mechanisms triggered by LN.
  • A reconstruction of genetic regulatory network analysis revealed a sub-network centered on a PtaNAC1 (P. tremula × alba NAM, ATAF, CUC 1) transcription factor. PtaNAC1 root-specific up-regulation increased root biomass and significantly changed the expression of the connected hub genes specifically under LN.
  • Our results provide evidence that the root response to LN involves hierarchically structured genetic networks centered on key regulatory factors. Targeting these factors via genetic engineering or breeding approaches can allow dynamic adjustment of root architecture in response to variable nitrogen availabilities in the soil.


Nitrogen is an essential macronutrient and therefore represents a major determinant of crop health and productivity. Adoption of nitrogen (N) fertilization over the past 50 yr has led to dramatic increases in yields and has averted looming food shortages and hunger. However, this approach for increasing agricultural productivity has reached critical environmental and economic limits (Pretty, 2008). Thus, there has been a growing interest in developing crop varieties that can more efficiently assimilate and use N and thus require lower N inputs (Good et al., 2004; Hirel et al., 2007). This is particularly important for lignocellulosic biofuel crops such as poplar that are intended to be grown on marginal lands to avoid competition with food crops (Robertson et al., 2008). Developing such varieties would entail understanding the mechanisms of N assimilation, utilization and recycling.

Plants respond to N deficiency through a number of morphological and physiological responses (Remans et al., 2006b; Zhang et al., 2007; Coll et al., 2011). Because N is assimilated from the soil, root morphology dynamically adjusts in relation to spatial and temporal heterogeneity in N availability (Remans et al., 2006b; Zhang et al., 2007). Under low nitrogen (LN) conditions, nitrate-rich patches stimulate lateral root (LR) growth into the patch (Drew, 1975). High N concentrations typically inhibit LR growth (Zhang et al., 2007). Amino acid sources of N such as glutamate have a highly suppressive effect on LR development (Filleur et al., 2005; Walch-Liu et al., 2006). The response of root architecture to N can also be conditional on sugar supply (Malamy & Ryan, 2001).

The regulation of root development, particularly in Arabidopsis thaliana, under optimal growth conditions has been dissected in detail (reviewed in Petricka et al., 2012). The factors controlling root development include auxin (Tian & Reed, 1999; Casimiro et al., 2001; Grunewald et al., 2012), cytokinin (Eckardt, 2006; Gonzalez-Rizzo et al., 2006; Riefler et al., 2006; Bishopp et al., 2009; Muraro et al., 2011; An et al., 2012), phytochrome (Salisbury et al., 2007; Costigan et al., 2011; Warnasooriya & Montomery, 2011), temperature (Bielach et al., 2012), phosphorus (Bates & Lynch, 2000a,b; Remy et al., 2012; Cerutti & Delatorre, 2013), and N (Remans et al., 2006b; Sanchez-Calderon et al., 2006; Zhang et al., 2007; Krapp et al., 2011; Cerutti & Delatorre, 2013). Previous studies on N have primarily focused on the effect of N oversupply rather than starvation. However, the molecular mechanisms that control the response of root architecture to N, including N sensing and the signal transduction pathway, are still poorly understood. In A. thaliana, the MADS (MCM1, AGAMOUS, DEFICIENS, Serum Response Factor)-box gene ARABIDOPSIS NITRATE REGULATED1 (ANR1) controls LR proliferation into nitrate-rich patches (Zhang & Forde, 1998). Nitrate transporters NRT1.1 and NRT2.1 mediate nitrate sensing as well as the cross-talk with hormonal signaling. Knockout of NRT1.1 leads to decreased LR growth into N-rich patches and this effect appears to be mediated by ANR1 (Remans et al., 2006a). NRT1.1 is the nitrate sensor in plants (Ho et al., 2009; Krouk et al., 2010a) and has also been shown to integrate auxin signal and nitrate response (Krouk et al., 2010b). Another nitrate transporter (NRT2.1) has been linked to regulation of LR initiation in response to LN (Remans et al., 2006b) as well as to LR repression under high carbon (C):N conditions (Little et al., 2005). Auxin has long been suspected as a modulator of root architecture in relation to N availability (Walch-Liu et al., 2006; Zhang et al., 2007; Tian et al., 2008). Integration of N responses and auxin signaling is mediated via an miRNA regulatory mechanism (Gifford et al., 2008; Vidal et al., 2010a). Abscisic acid (ABA) represses LR growth under high nitrate concentrations (Signora et al., 2001; De Smet et al., 2003).

Species and interspecific hybrids from the genus Populus (poplars) are among the fastest growing trees in temperate climates and are thus considered as a premiere bioenergy crop. Poplar has also emerged as a model system for understanding the molecular mechanisms of tree growth, development and response to environment (Brunner et al., 2004). N signaling, metabolism and storage in poplar shoot growth and development have been extensively studied (Zhu & Coleman, 2001; Canton et al., 2005; Pascual et al., 2008; Hacke et al., 2010; Man et al., 2011). However, little is still known about the mechanisms of how N affects poplar root growth and development. The adjustment of root architecture to spatiotemporal heterogeneity in N supply is particularly important for poplar and other trees that can occupy a site for hundreds of years. Localized N appears to stimulate growth of fine roots and decrease their mortality in hybrid poplar (Pregitzer et al., 1995) and cottonwood (Populus trichocarpa × Populus deltoides, Populus deltoides) (Friend et al., 2000; Woolfolk & Friend, 2003). This stimulatory effect was also found to be genotype-specific (Friend et al., 2000). No information is available about what the genetic and genomic underpinnings associated with these responses are.

Here we used a transcriptome approach coupled with genetic network analysis to study the responses of the poplar root system to LN. We show distinct morphological and transcriptional changes involving a highly interconnected hierarchically structured network.

Materials and Methods

Plant material, treatments and root measurements

All experiments, including the transgenic manipulations, were performed in the Populus tremula (L.) × Populus alba (L.) INRA 717-IB4 clone (referred to as ‘wild type’ (WT)). The plants were maintained in vitro on half-strength Murashige and Skoog (MS) medium with 20 g l−1 sucrose (Caisson), 0.1 mg l−1 IBA (Sigma-Aldrich), vitamins (Han et al., 2000) solidified with 2.5 g l−1 Gelrite (Sigma) and 4 g l−1 Phytablend agar (Caisson Laboratories, North Logan, UT, USA), with a 16 : 8 h day : night photoperiod (20 μm E).

For the microarray analysis treatments, the top three internodes of in vitro propagated WT plants (leaves removed) were placed on filter paper bridges in glass tubes (2 cm width and 25 cm height, VWR International, filled with 15 ml of liquid half-strength MS medium (without IBA) for 3 wk to allow development of the root system. Only uniformly developed plants were used in the subsequent experiments. For the control (normal N) treatment we used freshly added half-strength MS medium containing 20 mM KNO3. For the LN treatment we used a half-strength MS formulation with 0.05 mM KNO3 (the only source of N) and accordingly adjusted the K+ concentration by the addition of K2SO4. To maintain a constant concentration of N in the solution, the medium was changed after 2, 5 and 15 d. Roots were sampled at 6, 12, 24, 48, and 504 h after transfer to control and LN media (see above) and stored at −80°C until further processed for RNA extraction.

The response of WT and PtaNAC1 transgenics to LN was studied essentially as described above with slight modifications. Stem cuttings of one internode from both WT and PtaNAC1 transgenics were first synchronized for root development by cultivation for 1 wk on half-strength MS solid medium containing 2 mg l−1 IBA in the dark. The IBA-treated cuttings were then directly transferred to tubes with liquid control and LN media (as described above for the microarray analysis treatments), and allowed to grow for 21 d (504 h), before data were collected.

Roots were scanned and different parameters measured using the ImageJ software ( After scanning, roots were dried in an oven at 60°C until they were fully dehydrated and dry biomass was measured.

Production of binary vectors and transformation

The PtaNAC1 open reading frame was amplified using primers with attB sites (PtaNAC1-B1, GGGGACAAGTTTGTACAAAAAAGCAGGCTATGAGCAACATAAGCTTTGTGGAG and PtaNAC1-B2, GGGGACCACTTTGTACAAGAAAGCTGGGTTCAATAATGATTCCATAAGTTGGGC) and HiFi Taq polymerase (Invitrogen). The ET304 3-kbp promoter region (Filichkin et al., 2006) was amplified using primers with attB sites (ET304-B4, GGGGACAACTTTGTATAGAAAAGTTGGAACACTCCCCAGATTCACAAGTACTTTAGAG and ET304-B1R, GGGGACTGCTTTTTTGTACAAACTTGGGTTGTCAAAGGATCACAAACAAGATGCGTC) and HiFi Taq polymerase. The cloning procedure for PtaNAC1 in pDONR221 and ET304 in pDONR4-1 was as previously described (Yordanov et al., 2010). The ET304 promoter was cloned upstream of PtaNAC1 in the Gateway binary vector pK7m24GW,3 (Karimi et al., 2002). Following sequence verification, the binary plasmid was transformed into Agrobacterium tumefaciens strain GV3101/pMP90 (Koncz & Schell, 1986) using the freeze/thaw method (Holsters et al., 1978). The transformation procedure was as previously described (Han et al., 2000).

Microarray analysis

Collection and analysis of data were compliant with MIAME (Minimum Information About a Microarray Experiment) standards (Brazma et al., 2001). For each treatment, two independent biological replicates were used. Total RNA from the roots was extracted as previously described (Busov et al., 2003) and on-column DNase I-treated (Qiagen). Before labeling, RNA quality was assessed using the Agilent Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and 0.2 μg of total RNA was used to prepare biotinylated complementary RNA (cRNA). The labeling, hybridization, and imaging procedures were performed according to Affymetrix protocols at the Center for Genomics Research and Biocomputing, Oregon State University (, using the Affymetrix Poplar GeneChip (Affymetrix). Microarray data resulting from this study were deposited in the Gene Expression Omnibus (GEO) database at NIH ( with the accession number GSE43162.

Microarray normalization and identification of differentially expressed genes

The quality of microarray data sets from the experiment described in the previous section was first checked by examining the distribution of the Studentized deleted residuals using a previously described procedure (Persson et al., 2005). Only high-quality microarray data were normalized using the robust multichip average (RMA) (Irizarry et al., 2003). The rank product (RP) was used to identify differentially expressed genes (DEGs) (Breitling et al., 2004). RP outputs up-regulated and down-regulated genes separately, and we sorted all genes according to the absolute difference between their two rankings in up-regulated and down-regulated gene lists before we applied a correction for multiple testing using the Benjamini and Hochberg false discovery rate (FDR) (Benjamini & Hochberg, 1995). Genes with corrected P-values (< 0.05) were selected as DEGs.

Protein domain enrichment analysis

Protein domains were analyzed with InterproScan (Zdobnov & Apweiler, 2001). We downloaded and installed InterproScan and associated databases in our Linux server, and performed the standalone analysis to identify protein domains of all target sequences provided by The enrichment of each domain in the DEG list was compared with the occurrence of the respective domain in the background of all genomic genes, and two parameters were introduced to show the enrichment of each domain: (1) the enrichment factor, EF = k/(nM/N); and (2) Escore, which is the hypergeometric probability of identifying at least k domains from the DEG list. It is calculated using the following formula (Wei et al., 2013):

display math

(N, the total number of domains associated with all genomic genes; M, the total number of specific domains for all genes in the genome; n, the number of all domains associated with the DEGs; k, the number of a specific domains present in the DEG list.) We applied a multiple testing correction using the Benjamini and Hochberg FDR (Benjamini & Hochberg, 1995), with a significance cut-off P value < 0.001.

Gene ontology analysis

The DEGs of each time-point were used for gene ontology (GO) analysis using AmiGO's Term Enrichment tool ( This tool uses the Perl module GO:TermFinder available at CPAN ( to identify the enriched GO terms associated with a DEG list via hypergeometric probability, as described in the previous section for protein domain enrichment analysis. We applied a multiple testing correction using the Benjamini & Hochberg (1995) FDR. GO terms with a corrected P-value < 0.001 were considered to be significantly enriched.

Gene reduction and network construction

We used a software pipeline, TF-Cluster (Nie et al., 2011), to identify key transcription factors involved in the stress response under N starvation conditions. TF-Cluster is designed to identify key regulatory genes controlling a trait or a biological process from a time series or compendium microarray data set from a specific condition. The underlying hypothesis is that regulatory genes that control the same trait or biological process are coordinated, and thus can be identified through co-expression network construction followed by decomposition. Detailed procedure is described in our earlier publication (Nie et al., 2011). With the identified key regulatory genes from TF-Cluster as the backbone, we constructed a local gene regulatory network (GRN) using the Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE) (Margolin et al., 2006), which has been shown to be able to identify both linear and nonlinear regulatory relationships and in most cases outperforms relevance networks and Bayesian network methods (Basso et al., 2005; Margolin et al., 2006). The local GRN built was then searched with a Perl script to identify Transciption factors (TFs) that were connected to three to 15 hub genes. GO term enrichment analysis was also applied to the genes linked to each of eight hub genes in the same way as described above for domain enrichment analysis.

First-strand cDNA synthesis and quantitative RT-PCR analysis

Reverse transcription was performed on 1 μg of DNAase I-treated total RNA in a final reaction mixture of 20 μl using an M-MuLV (Moloney Murine Leukemia Virus) reverse transcriptase (Fermentas) following the manufacturer's protocol. Quantitative RT-PCR (qRT-PCR) was performed using the StepOnePlus Real Time System (Applied Biosystems). Gene expression was analyzed quantitatively using the SYBR Green detection system. Each PCR reaction contained 1× Maxima SYBR Green qPCR master mix (Fermentas), 0.1 μM of each forward and reverse primer (Eurofins MWG Operon), 1 μl of 10× diluted cDNA solution and nuclease-free water. The final volume of each PCR reaction was 20 μl. The qRT-PCR cycling stages consisted of initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min, and a final melting curve stage of 95°C for 15 s, 60°C for 1 min and 95°C for 15 s. Samples for qRT-PCR were run in two biological replicates and two technical replicates. Relative gene expression was calculated according to the method described previously (Tsai et al., 2006). For the normalization of gene expression, Ubiquitin (Ubq) was used as an internal standard. Primers used in gene expression analysis are listed in Supporting Information Table S5. They were designed from the Populus refseq mRNA sequences using the Primer-BLAST web resource at NCBI (National Center for Biotechnology Information;

Sequence analyses

Homology searches were performed using BLAST searches on the NCBI site ( or Phytozome ( Identification of micro-RNAs and their target sites was performed using miRBase ( The PtaNAC1 and pta-MIR164e amplified products were fully sequenced and the sequences were deposited in GenBank with accession numbers JX534240 and JX560538, respectively. PtaNAC1 (98.3% amino acid identity to POPTR_0007s08420.1) is the closest ortholog to NAC1 (58.7% amino acid identity to AT1G56010.2) from A. thaliana (based on the protein-BLAST search). The secondary structure of pta-MIR164e was generated using mfold (


Nitrogen deprivation promotes poplar root proliferation and elongation

We studied the morphological and transcriptomic changes in poplar roots under LN (see 'Materials and Methods' for more details). Nitrate (NO3) was the only source of N in the media. We were interested mostly in the developmental changes that occur with respect to root growth and development. We found that N starvation elicited increased elongation of both the primary roots and LRs as well as LR proliferation (Fig. 1). The increased elongation growth was the first response and could be detected as early as 1 d after treatment initiation (Fig. 1c). Significant changes in LR formation that require de novo organ initiation and growth were detected later at 2–4 d under LN (Fig. 1d,e) but were most pronounced at the end of the studied period, at 504 h (21 d) (Fig. 1f). Significant increases in root biomass were measured only at the last sampling time of 21 d (Fig. 1f). This suggests that LN stimulates LR proliferation. The stimulatory effect could be detected as early as 2–4 d after treatment initiation but the most vigorous LR growth and development occurred at between 4 and 21 d.

Figure 1.

Nitrogen (N) deprivation increases root growth in Populus tremula × Populus alba (poplar). (a, b) Poplar roots grown on (a) low nitrogen (LN) and (b) control (normal N) media. (c) Length of main roots. (d) Lateral root (LR) density, measured as the number of LRs per main root. (e) Combined length of all LRs divided by the length of the main roots in cm. (f) Root biomass. Bars show the means of four biological replicates ± standard error. Scale bars in (a) and (b) correspond to 2 cm. * and ** indicate significant differences between LN and control medium determined by Student's t-test at < 0.05 and < 0.01, respectively.

Global changes in the root transcriptome in response to N deficiency

To better understand the root morphological and growth changes, we performed a microarray time-series experiment under the same experimental conditions (see previous section and ‘Materials and Methods’ for more details). A total of 9198 genes were found to be significantly differentially expressed at the six time-points (Table 1, S1; see also, and the overlapping genes between any two time-points are shown in Table 2. We successfully validated the expression changes for 12 genes (six up-regulated and six down-regulated; Fig. S1). Although a high number of DEGs were found at all times, the highest numbers were discovered at 6, 48 h and 21 d (Table 1). The early (6 h) peak suggests that N deficiency led to significant and rapid reprogramming of the poplar root transcriptome. Interestingly, the 48-h peak in the number of DEGs coincided with the initiation of significant increases in lateral root proliferation, while the 21-d peak corresponded to significant changes in biomass accumulation (Fig. 1).

Table 1. Differentially expressed genes (DEGs) in wild-type (WT) poplar (Populus tremula × Populus alba) roots in response to low nitrogen (LN) conditions
Time6 h12 h24 h48 h96 h504 h
  1. For a complete list of the genes, annotation and expression values, see Table S1.

Total DEGs370418831309380824833328
Table 2. Significantly enriched gene ontology (GO) terms in wild-type (WT) poplar roots in response to low nitrogen (LN)
  1. BP, biological process; MF, molecular function; CC, cellular component. For a complete list of the terms, see Table S2.

6 h301634641025.3
12 h1252941589.7
24 h11226131519.3
48 h260351931419.4
96 h243234330919.0
504 h200483328117.3

Temporal changes in the root response to LN

To gain further insight into the biological processes associated with the observed temporal changes, we performed GO enrichment analysis at each of the studied time-points (Table 2). We first identified a core set of biological processes that were affected under N limitations across all time-points. A total of 28 biological processes were significantly enriched during the whole studied period (Table S3). These 28 common ontologies are ‘descendants’ of three higher hierarchical nodes in the GO tree: metabolic processes (GO:0008152), response to stimulus (GO:0050896) and localization (GO:0051179). More than half of the enriched ontologies (15 of 28) belonged to response to stimulus (GO:0050896), five to metabolic processes (GO:0008152) and eight to localization (GO:0051179) (Fig. 2 and Table S3). The response to stimulus category which dominates the common set includes not only genes associated with the response to exogenous environmental stimuli but also genes involved in internal signaling (e.g. the response to endogenous stimulus – GO:0009719). This suggests significant cross-talk between LN signaling and perception and the plant endogenous developmental programs. The majority of genes involved in metabolism (GO:0008152), and two sub-processes of response to stimulus, response to chemical stimulus (GO:042221) and response to abiotic stimulus (GO:000926), showed an expression pattern of initial (6 h) down-regulation followed by elevated expression levels throughout the rest of the studied period (Fig. 2). Genes involved in establishment of localization (GO:0051234) and transport (GO:0006810) showed a more complex expression pattern – early and late peaks of expression and a dip toward the 48-h time-point (Fig. 2).

Figure 2.

Biological processes significantly affected by nitrogen (N) deficiency in Populus tremula × Populus alba throughout the whole studied period. The significance of effects was determined by examining enrichment of gene ontology (GO) terms associated with differentially expressed genes (DEGs) versus the respective gene ontology (GO) terms in the whole genome distribution using a hypogeometric test and a threshold of Benjamini and Hochberg false discovery rate (FDR) corrected < 0.001. The average fold change (AFC) indicates the fold change in expression between the control and low nitrogen (LN) treatments for all genes that belong to the respective GO term. Note that only 16 ontologies from the top three layers of the GO tree are shown in Fig. 2. The 28 conserved ontologies are shown in Supporting Information Table S3.

We also studied ontology terms enriched specifically in one or more time periods (Table S3, Fig. 3). A number of enriched categories associated with signal transduction were found in the period between 6 and 24 h. The biological processes represented by these ontologies were essentially absent in the remainder of the studied period (Fig. 3), implying that LN triggered a number of signal transduction pathways and this response occurred within hours under LN. The spike of signaling, in the early 24 h into LN, gave a way to enrichment of ontologies associated with cell growth, development and proliferation (Fig. 3). For instance, 19 ontologies representing DNA replication and chromatin modification and 15 ontologies representing cell wall biosynthesis and modification were enriched at 48 h and/or 96 h (Table S3). Many categories associated with organ morphogenesis/development were enriched throughout the 48 h–21 d period. The last time-point (21 d = 504 h) was specifically enriched in ontologies associated with root development and morphogenesis (Fig. 3). The temporal trends in GO term enrichment corresponded very well with the observed root morphological responses under LN (Fig. 1).

Figure 3.

Temporal trends in expression of differentially expressed genes and the associated biological processes during the response of poplar roots to low nitrogen (LN). The criteria for gene ontology (GO) enrichment analysis are described in the legend of Fig. 2. The minimal P-value associated with each GO at all time-points is reported (‘P-value’ heading). The average fold change (AFC) was obtained in the same way as described in Fig. 2. Note that, because of space limitation, 22 ontologies that represent the cell cycle and cell division are not shown but can be found in Table S3. All of these have AFCs between 2.0 and 3.0 at both 48 and 96 h, except for one that has an AFC of 1.76 at 48 h and 1.5 at 96 h. NG, number of genes.

Early activation of cell cycle- and growth-associated genes under N limitation

The clear indication that the 48–96-h time window corresponds to the initiation of root proliferation and growth in response to N starvation prompted us to more closely inspect genes of known function in the regulation of the cell cycle. Among all DEGs we found 39 genes encoding key regulators of the cell cycle in plants (Fig. 4a). The majority (35 of 39) of these were highly and specifically up-regulated during the 48–96-h period.

Figure 4.

Specific temporal activation of cell cycle-associated genes under low nitrogen (LN). (a) Cell cycle-associated genes are activated in poplar roots at between 48 and 96 h under LN. For the full names and expression and annotation data of the genes, see Table S1. FC, fold change of gene expression between the control and LN treatments. The number after the FC indicates the sampling time in hours. All genes in the list are differentially expressed (Table S1). (b) Protein domains associated with the cell cycle and DNA replication are enriched among differentially expressed genes (DEGs) at between 48 and 96 h under LN. EF, enrichment factor relative to the whole genome background. White color indicates that there is no DEG containing the domain in the respective time periods.

In addition, we performed a domain enrichment analysis among the DEGs. Again, analysis of the domains enriched 15-fold or more indicated significant increase in domains associated with cell division and replication (Fig. 4b). We found that 18 domain types associated with replication and cell division were enriched. This was the single most represented group; genes in this group were specifically up-regulated during the 48–96-h period.

Response to nitrogen deficiency involves hierarchically structured sub-networks

To understand the regulatory interactions and relationships during the response to N deprivation, we employed reconstruction of GRN analysis using our transcription profiling data. We found a highly ordered structure composed of sub-networks organized in a hierarchical pattern. One of these networks was centered on a transcription factor (PtaNAC1; GenBank: JX534240, POPTR_0007s08420) with the highest homology to NAC1 from A. thaliana (Fig. 5). Some of the connected genes in this network do suggest that it may be related to the increased root growth. For example, PtaNAC1 was connected to groups centered on expansin17 and the GA-induced gene GIBBERELLIC ACID-STIMULATED transcript in ARABIDOPSIS 1 (GASA1). Even more interestingly, PtaNAC1 was also connected to hub genes suggesting of even larger regulatory context regarding N assimilation through a homolog of NIA2 (Nitrate reductase in Arabidopis 2 and C/N metabolism through PPCK (Phosphoenolpyruvate carboxykinase which is involved in phosphorylation-mediated activation of a major enzyme involved in replenishing Krebs cycle intermediates and implicated in playing a key role in regulation of the C/N balance (Walker et al., 1999).

Figure 5.

PtaNAC1-centered genetic network in Populus tremula × Populus alba. The PtaNAC1 (P. tremula × alba NAM, ATAF, CUC 1) regulatory module was generated using ARACNE (see 'Materials and Methods' for details) with the following thresholds: a kernel width of 0.175, a Mutual Information (MI) threshold of 0.04 and a P-value of 0.01. The undirected graph shows significant expression dependence between the genes (dots) in the network. Abbreviations for the eight hub genes are as follows: GASA1 (XM_002300671), GIBBERELLIC ACID-STIMULATED transcript in ARABIDOPSIS 1; EXPA17 (XM_002301427), alpha-expansin 17; PPCK (XM_002319617), phosphoenolpyruvate carboxylase; WAK (XM_002297646), wall-associated kinase; NIA2 (XM_002300979), nitrate reductase 2; RLK1 (XM_002297635), receptor-like protein kinase 1; AIR3 (XM_002333256), AUXIN INDUCED IN ROOT CULTURES 3, a subtilisin-like serine endopeptidase; PK (XM_002324523), protein kinase. The number of genes in each module and significantly enriched gene ontology (GO) terms associated with these genes are shown in the table below the graph. Only the top six most significantly enriched GO terms in each regulatory module associated with each hub gene are shown; the full list of significantly enriched GO terms can be found in Table S4. The full list of enriched GO categories is also given in Table S4. The significance of effects was determined by examining enrichment of GO terms associated with the genes in each regulatory module versus the respective GO terms in the whole genome distribution using a hypogeometric test and a threshold of Benjamini and Hochberg false discovery rate (FDR) corrected < 0.001. Note that no significantly enriched GO terms were associated with the genes in the PK (XM_002324523; protein kinase) module.

Ontology analysis of the genes involved in this network (Fig. 5) suggests that it is an essential part of the overall response: 18 of 28 ontologies that were found to be enriched at all six time-points (Fig. 2 and Table S3) were also significantly overrepresented in the PtaNAC1-centered network (Table S4).

PtaNAC1 shows specific expression patterns

To better understand the involvement of PtaNAC1 in root development and the response to LN, we performed an expression analysis. The gene was highly and predominantly expressed in roots (Fig. 6). In A. thaliana, NAC1 is downstream of auxin-regulated signaling leading to LR formation (Xie et al., 2000). We therefore tested whether PtaNAC1 is induced by auxin. We could not detect any induction (Fig. 7a), while an auxin-inducible PtaPIN1 (P. tremula × alba Pin-formed 1) gene was highly up-regulated (Fig. 7b). However, we could not preclude the possibility that the response of PtaNAC1 to auxin does not manifest at the transcription level, as research in A. thaliana has shown posttranslational control of PtaNAC1 protein abundance (Xie et al., 2002). In A. thaliana, NAC1 has been found to be regulated by miR164 (Guo et al., 2005) and a miR164 target site is present in PtaNAC1 (Fig. S2). We therefore studied the expression of pta-MIR164s in response to N limitation. The expression of pta-miR164e (GenBank: JX560538.1, scaffold_16:12 939 102–12 939 122, Populus genome v2.2; see also Fig. S2) was high in the first 12 h under LN treatment followed by a gradual decrease to near-undetectable levels by 48 h (Fig. 7c). The expression of pta-miR164e was approximately inverse to that of PtaNAC1 (Fig. 7d). This suggests that this may be one mechanism, if not the main mechanism, for regulation of PtaNAC1 levels during the LN response.

Figure 6.

PtaNAC1 is predominantly expressed in roots of Populus tremula × Populus alba. Quantitative real-time RT-PCR analysis of PtaNAC1 expression in different tissues of wild-type (WT) plants grown on standard control medium was performed; see ‘'Materials and Methods'’ for details. Bars show the means of three biological replicates ± standard error of the mean. Ubq was used as a normalization control.

Figure 7.

PtaNAC1 expression in response to auxin and low nitrogen (LN). (a, b) Expression of (a) PtaNAC1 and (b) PtaPIN1 genes in auxin-treated wild-type (WT) poplar roots. Expression was studied at 24 h after treatment initiation using quantitative real-time RT-PCR. Bars show the means of four replicates ± standard error. * and ** indicate significant differences between treated and untreated (control) plants as determined by Student's t-test at < 0.05 and < 0.01, respectively. Ubq was used as a normalization control. (c) Inverse expression patterns of (c) PtaNAC1 and (d) pta-MIR164e in response to LN in WT plants. Bars represent the mean of four replicates ± standard error. Expression was studied using quantitative real-time RT-PCR.

PtaNAC1 up-regulation in poplar transgenics increased root growth specifically under LN

Because the GRN and expression analysis indicated that PtaNAC1 may play a major role in the root response to N deficiency, we investigated whether the gene can indeed affect root growth and development under LN conditions. To avoid any confounding pleitropic effects, and because PtaNAC1 was predominantly expressed in roots, we up-regulated PtaNAC1 using a previously described Populus root-predominant (ET304) promoter (Filichkin et al., 2006). The ET304-PtaNAC1 transgenic plants showed no significant differences compared with WT with respect to shoot development (Fig. S3). A slight increase in root biomass was observed under normal (optimal N) conditions but the differences were not significant (Fig. 8a). However, the N-limiting conditions caused a significant increase in root proliferation in ET304-PtaNAC1 transgenics compared with WT (Fig. 8b,c). This indicates that PtaNAC1 can specifically promote root growth under LN conditions. To explain this outcome, we compared the expression of PtaNAC1 in the WT and transgenics under both control and LN conditions at the end of the sampling period (21 d/504 h) when the phenotypic differences were greatest. PtaNAC1 was only slightly up-regulated in the transgenic plants under the control treatment, and its expression was also slightly, but significantly, higher in the transgenics compared with the WT under LN (Fig. 8d). Note that the PtaNAC1 sequence used to generate the ET304-PtaNAC1 transgenics did have the miR164 site and thus was still susceptible to miRNA-directed degradation (Fig. S2). We conclude that the small differences in transcript level cannot fully explain the observed positive changes in many root traits in the ET304-PtaNAC1 plants. Thus, the phenotype can be associated with changes in protein level, posttranslational modification and stability of the PtaNAC1 protein. In support of this hypothesis, the NAC1 protein abundance in A. thaliana is regulated through controlled proteolytic degradation (Xie et al., 2002).

Figure 8.

Increased root growth and PtaNAC1 expression in ET304-PtaNAC1 Populus tremula × Populus alba transgenics under low nitrogen (LN). (a) Root biomass. (b) Lateral root (LR) length. (c) LR number. (d) PtaNAC1 expression in WT and ET304-PtaNAC1 transgenics. Bars represent the mean of two independent transgenic lines ± standard error from two independent trials each with 10 replicates of wild type (WT) and each transgenic line. Expression of the PtaNAC1 gene (d) was analyzed in roots of WT and transgenic lines. Bars represent the means of at least four replicates ± standard error. Ubq was used as a normalization control. In all experiments transgenic plants were compared with WT plants in control and LN conditions. *Indicates significance compared with WT plants as determined by Student's t-test at < 0.05.

PtaNAC1 affects the expression of network genes specifically under LN conditions

To further understand the roles of the GRN in the LN response, we studied the expression of eight of the hub genes under control and LN conditions in both WT and ET304-PtaNAC1 transgenic roots (Fig. 9). Again, this analysis was performed at the end of the sampling period (21 d/504 h) when the phenotypic differences were greatest. Only two of the eight hub genes were significantly differentially expressed between WT and transgenic plants under normal (optimal N) conditions (Fig. 9). In contrast, six of the eight hubs were differentially expressed in the ET304-PtaNAC1 transgenics compared with WT under LN. Some of the genes not only changed but reversed their expression patterns. A homolog of GASA1 was perhaps the most extreme example. This gene was significantly down-regulated in ET304-PtaNAC1 transgenics under normal N conditions but significantly up-regulated under LN (Fig. 9). This finding is consistent with the significantly higher expression of PtaNAC1 in the transgenic plants under LN conditions at the same sampling time (Fig. 8d).

Figure 9.

PtaNAC1 transgenic modification affects the expression of key network genes in Populus tremula × Populus alba under low nitrogen (LN) conditions. Expression was analyzed in roots of wild type (WT) and ET304-PtaNAC1 transgenics after 21 d/504 h under control and LN conditions. Bars represent the means of at least four replicates ± standard error. Transgenic plants were compared with WT plants in control and LN conditions. *Indicates significance compared with WT plants as determined by Student's t-test at < 0.05. The abbreviation of genes is as follows: GASA1 (XM_002300671), GA-responsive GAST1 (GIBBERELLIC ACID-STIMULATED transcript in ARABIDOPSIS 1) protein; EXPA17 (XM_002301427), alpha-expansin 17; PPCK (XM_002319617), phosphoenolpyruvate carboxylase; WAK (XM_002297646), wall-associated kinase; NIA2 (XM_002300979), nitrate reductase 2; RLK1 (XM_002297635), receptor-like protein kinase 1; AIR3 (XM_002333256), AUXIN INDUCED IN ROOT CULTURES 3, a subtilisin-like serine endopeptidase; PK (XM_002324523), protein kinase.


Here we describe the response of poplar roots to LN. N deprivation led to distinct morphological and corresponding transcriptome changes indicative of temporally staged and hierarchically structured responses. We first found that poplar roots responded to N deficiency by rapidly increasing primary root elongation followed by LR proliferation, resulting ultimately in a significantly higher root biomass in LN compared with control conditions. Very similar responses were reported in maize (Zea mays) (Chun et al., 2005; Gaudin et al., 2011). A number of morphological root responses in relation to N were described in A. thaliana (Walch-Liu et al., 2006). However, information about how roots respond to LN conditions in this species is scarce. High C : N conditions led to decreased LR proliferation (Malamy, 2005). Only recently, it was shown that the biomass of A. thaliana roots transiently increased under LN conditions, but this was very short-lived (2–4 d after initiation of LN treatment) and root growth rapidly decreased thereafter (Krapp et al., 2011).

The molecular responses to N deprivation, particularly in roots, are even less well understood and poorly characterized (Krapp et al., 2011; Liang et al., 2012). Studies on a fine temporal scale (h) of the root response to LN are virtually absent, thus precluding understanding of signal perception, transduction and response to LN. The majority of efforts to date have been concentrated on N sensing (Coruzzi & Zhou, 2001; Vidal et al., 2010b). These studies are extremely valuable, as N sensing is an essential part of the response, but probably represent only a fraction of the LN response network involved. Here, we provide for the first time a glimpse into these responses and networks.

We first demonstrated that a set of 28 biological processes were enriched during the whole studied period and thus probably represent the core of the poplar root response to LN. We also found that 21 of these 28 processes were also enriched in A. thaliana roots during the response to LN (Table S3; Krapp et al., 2011). This suggests that there could be a conserved set of processes that are characteristic of the general interspecific plant response to LN. However, this will need to be demonstrated in future work with evidence from other plant species. These 28 conserved biological processes belong to three larger groups: metabolic processes, response to stimulus, and establishment of localization, involving five, 15, and eight ontologies, respectively (Table S3; Fig. 2). Clearly, this core set is dominated by the response to stimulus ontologies (15 processes), strongly suggesting that intricate and multiple sensing and signal transduction mechanisms are involved in the response to LN. Inspection of the downstream hierarchies suggests that these sensing and signaling mechanisms involve putative cross-talk between exogenous/environmental and endogenous/developmental signaling. Our study provides significant evidence for the existence of conserved but virtually unknown and unstudied signaling mechanisms in the roots governing the response to LN.

Analysis of temporal trends in GO enrichment during the studied period showed pronounced patterns that corresponded with the observed morphological changes. The response to LN in poplar roots started with an upsurge in signaling and signal transduction in the first 24 h under N deprivation. These signaling events were immediately followed in the 48–96-h time window with activation of genes involved in cell proliferation and the growth machinery. This activation set the stage for the intensified LR growth that was most evident at the last time-point (21 d/504 h), where we observed highly significant enrichment of biological processes associated with root outgrowth, development and morphogenesis. In summary, the transcriptomics results closely parallel the phenotypic observations and changes and strongly suggest signaling pathways in poplar that lead to activation of LR proliferation and growth in response to LN.

To reconstruct the regulatory events involved, we used genetic network analysis which revealed a highly ordered and hierarchical genetic network. We focused on a network centered on the PtaNAC1 transcription factor. This network shows a high integration of metabolic, developmental and growth processes. For example, the individual hubs of the network represent very different functional categories involved in hormonal signaling, primary metabolism and N assimilation. This suggests that a single transcription factor can provide coordinated and simultaneous regulation of a number of very diverse processes. We still do not know the nature of this regulation (e.g. direct or indirect gene targets). Several lines of evidence suggest that PtaNAC1 is unlikely to regulate directly all of the connected hub genes. The trends in the expression of the hub genes in the network are diverse and sometimes opposite to each other. Because most transcription factors act as either repressors or activators, it is unlikely (although possible) that the diversity of expression patterns results from the activity of only one transcription factor. Although binding sites for some NAC TFs have been identified (Tran et al., 2004), it is still unclear if this is a canonical binding site for all members of this TF family. Unfortunately, the attempt to delineate the NAC1 binding sites in A. thaliana was unsuccessful (Xie et al., 2000). Analysis of the promoter regions of all hubs for known NAC binding sites indicates the presence of some but not all genes. One of the hub genes (an ortholog of AUXIN INDUCED IN ROOT CULTURES 3 (AIR3)) has been shown in A. thaliana to be the direct target of NAC1 (Xie et al., 2000), suggesting that one or more of the hubs are indeed probably under direct regulation by PtaNAC1. However, other regulatory assemblages observed in this network are probably outcomes of more complex and indirect cascading changes and effects. It is very likely that we are missing some intermediate steps in the observed regulatory processes because they operate exclusively at the protein level (stability, localization, and posttranslational modifications) and result in no detectable change in the expression of the respective genes. Nevertheless, our analysis indicates that employing GRN analysis can reveal the existence of such networks and represents a good starting point for their unraveling.

The network was centered on an NAC transcription factor with the highest sequence homology to A. thaliana NAC1 (Xie et al., 2000). The putative A. thaliana ortholog was discovered as a central regulator of LR proliferation downstream of auxin signaling (Xie et al., 2000, 2002; Guo et al., 2005). The gene is subject to posttranscriptional miR164-mediated repression (Guo et al., 2005). The NAC1 protein also appears to be subject to ubiquitin-mediated controlled proteolysis facilitated by SINAT5 (Arabidopsis homologue of the RING-finger Drosophila melanogaster protein SINA; a ubiquitin protein ligase) (Xie et al., 2002). More recent work, however, showed significant regulatory and functional differences between A. thaliana NAC1 and the apparent Medicago truncatula ortholog (MtNAC1) (D'haeseleer et al., 2011). Most importantly, MtNAC1 did not appear to be induced by auxin and did not have any effects on LR formation. Furthermore, MtNAC1 did not interact with the M. truncatula SINAT5 putative homologs. Nevertheless, the miR164-mediated mechanism for MtNAC1 repression was conserved. Similarly, the poplar PtaNAC1 showed significant deviations from the developmental roles and regulatory functions described in A. thaliana. First, as in M. truncatula, the gene was not induced by auxin. Second, similar to the work in M. truncatula, up-regulation of the gene did not lead to modification of LR formation under control conditions (optimal N concentrations). However, it should be noted that an increase in LR density was measured under control conditions, but the differences were not statistically significant. In contrast to both M. truncatula and A. thaliana, where the gene was expressed in leaves, PtaNAC1 in poplar was predominantly expressed in roots, suggesting a mainly root-associated function(s). As mentioned in Fig. 8 above, the effect of PtaNAC1 on LR development was conditionally manifested only in the LN environment. In WT plants, PtaNAC1 expression was low in the first few days under LN, and then there was an increase in transcript abundance toward the end of the studied period (21 d/504 h). Interestingly, and in support of the findings in A. thaliana and M. truncatula, we found that PtaNAC1 was probably regulated by miR164 and this regulation was linked to the LN response. We first identified an miR164 site in the PtaNAC1 sequence and then showed that pta-MIR164e was highly expressed in the first 24 h under LN, and this exactly corresponded to the low expression level of PtaNAC1. Further into the treatment, the decline in pta-MIR164e expression corresponded to the recovery of PtaNAC1 transcript abundance (albeit with some expected lag). Experimental validation of poplar PtaNAC1 cleavage has already been demonstrated by high-throughput sequencing (Li et al., 2011). The sum of all evidence strongly suggests that PtaNAC1 temporal regulation under LN conditions is partly if not exclusively mediated via microRNA-mediated posttranscriptional degradation. Involvement of miRNAs in the regulation of root responses to nitrate and to N deprivation has been recently described (Gifford et al., 2008; Vidal et al., 2010a; Liang et al., 2012). This suggests that miRNAs provide a major mechanism for modulation of the root architecture in response to changes in N availability. The involvement of miRNAs in other stress responses has been shown in several plant species, including poplar (Lu et al., 2008; Li et al., 2009, 2011).

The exact role of PtaNAC1 and the associated network in the observed phenotypic response is still difficult to understand. The PtaNAC1 expression pattern and the GO categories enriched in the various groups of the network suggest that it is not directly involved in the regulation of the cell proliferation and growth response that occurs between 48 and 96 h. We did not identify any of the cell cycle and growth GO categories within this network (Fig. 6, Table S4) and the gene was expressed at very low levels during this period. This suggests that PtaNAC1 is probably not involved in the signaling steps leading to LR initiation but rather plays a role in the later stage of root growth and development. Its higher expression at the end of the studied period (21 d/504 h) aligns well with the increased root growth and development at this time-point (Fig. 1). One possible role of the network is coordination of the various metabolic processes associated with the expansive LR growth, as suggested by significant enrichment of various ontologies associated with metabolic processes (Table S4).

In summary, we show that poplar roots respond to LN by activation of growth, and this response is underpinned by significant transcriptome reprogramming (9198 regulated genes). By employing advanced genetic network analysis we have demonstrated that these responses involve hierarchically structured network modules with highly integrative function that encompass a variety of apparently unrelated processes. Knowledge of the key regulators of the networks allowed modulation of root architecture specifically under LN conditions.


We thank Hang Zhang for data analyses and Dr Minsheng Wu for the production of the transgenic plants. This work was supported in part by grants from the US Department of Energy (DOE), Poplar Genome Based Research for Carbon Sequestration in Terrestrial Ecosystems (DE-FG02-06ER64185 and DE-FG02-05ER64113), the Consortium for Plant Biotechnology Research, Inc. (GO12026-203A), the United States Department of Agriculture (USDA) CSREES, the USDA-NRI Plant Genome program (2003-04345) and USDA CSREES, the Biotechnology Risk Assessment Research Grants Program (2004-35300-14687), and Plant Feedstock Genomics for Bioenergy: A Joint Research Program of USDA and DOE (2009-65504-05767 and ER65454-1040591-0018445).