Maize maintains growth in response to decreased nitrate supply through a highly dynamic and developmental stage‐specific transcriptional response

Summary

Elucidation of the gene networks underlying the response to N supply and demand will facilitate the improvement of the N uptake efficiency of plants. We undertook a transcriptomic analysis of maize to identify genes responding to both a non‐growth‐limiting decrease in ${\text{NO}}_3^{-}$ provision and to development‐based N demand changes at seven representative points across the life cycle. Gene co‐expression networks were derived by cluster analysis of the transcript profiles. The majority of ${\text{NO}}_3^{-}$‐responsive transcription occurred at 11 (D11), 18 (D18) and 29 (D29) days after emergence, with differential expression predominating in the root at D11 and D29 and in the leaf at D18. A cluster of 98 probe sets was identified, the expression pattern of which is similar to that of the high‐affinity ${\text{NO}}_3^{-}$ transporter ( NRT2) genes across the life cycle. The cluster is enriched with genes encoding enzymes and proteins of lipid metabolism and transport, respectively. These are candidate genes for the response of maize to N supply and demand. Only a few patterns of differential gene expression were observed over the entire life cycle; however, the composition of the classes of the genes differentially regulated at individual time points was unique, suggesting tightly controlled regulation of ${\text{NO}}_3^{-}$‐responsive gene expression.

Introduction

Approximately only 40%–50% of N fertilizer applied each year to cereal crops is estimated to be used for maximal yield (Sylvester‐Bradley and Kindred, 2009). Development of germplasm with improved N uptake and utilization traits would help mitigate N loss from the soil, which eventually improves financial margins for agricultural producers and reduces the impacts of unused N on the environment (Garnett et al., 2009; Hawkesford, 2011). The N uptake and transport pathways in plants have been extensively studied, leading to the identification of multiple transport proteins and regulatory pathway components (Gutiérrez, 2012; Wang et al., 2012). However, efforts to improve N uptake and utilization efficiency by crop plants through manipulation of transporter or assimilatory enzyme genes (via breeding or transgenic approaches) have had limited success to date (McAllister et al., 2012). This suggests that improvements may only be possible through the identification and modification of the regulators of the nitrate uptake and assimilation systems (Canales et al., 2014; Simons et al., 2014).

Several studies have been undertaken in Arabidopsis to identify global transcriptional responses to ${\text{NO}}_3^{-}$ supply. Looking at a single point in development, a number of these studies led to the identification of Arabidopsis genes responsive to ${\text{NO}}_3^{-}$ deprivation and resupply (known as the ‘primary ${\text{NO}}_3^{-}$ response’). Wang et al. (2003) found nearly 10% of genes with measurable transcription were ${\text{NO}}_3^{-}$ responsive with over six times as many genes being responsive in the root compared to the shoot. Along with the genes involved in ${\text{NO}}_3^{-}$ uptake and reduction, those encoding proteins involved in glycolysis, trehalose‐6‐phosphate metabolism and iron and sulphur uptake and metabolism were especially predominant (Wang et al., 2003). Scheible et al. (2004) found a similar widespread, but coordinated list of genes responsive to ${\text{NO}}_3^{-}$ deprivation and subsequent resupply. Genes involved in ${\text{NO}}_3^{-}$ transport and reduction responded after 30 min of ${\text{NO}}_3^{-}$ resupply after deprivation, while genes involved in amino acid biosynthesis responded after 3 h (Scheible et al., 2004). More recently, Krouk et al. (2010) examined ${\text{NO}}_3^{-}$‐responsive transcription within 20 min of ${\text{NO}}_3^{-}$ resupply after deprivation and identified a transcription factor (SPL9) for which transcript abundance increases within minutes of ${\text{NO}}_3^{-}$ resupply. Overexpression of SPL9 altered the ${\text{NO}}_3^{-}$ response of key ${\text{NO}}_3^{-}$ uptake and reduction genes such as nitrite reductase (NIR) and NRT1.1 (Krouk et al., 2010). Overall, these studies led to the identification of large numbers of genes responding to ${\text{NO}}_3^{-}$ starvation and resupply at individual points in development. However, determining the transcriptional response of these genes to N supply and demand across different growth stages would greatly broaden our understanding of N‐responsive genes and their associated regulatory networks.

An alternative approach taken to understanding the global transcriptional response of plants to N provision is to describe longer term effects of ${\text{NO}}_3^{-}$ deprivation at one or two points later in development. This approach has been used to discover the transcriptional and metabolic responses of leaves and roots at a point following several days or weeks of limited N supply (Bi et al., 2007, 2014; Krapp et al., 2011). The response of maize at the V6 stage of vegetative growth to limiting N provision was used to develop leaf transcriptional biomarkers which may be used to determine the N status of maize plants in the glasshouse or field (Yang et al., 2011). Another study in maize focussed on the transcriptional and metabolic response of the source leaf to N deprivation during two points of vegetative development and showed shifts in regulation of N and C metabolism occurred as a result of increased N deprivation between the time points (Schlüter et al., 2012). Amiour et al. (2012) undertook a microarray analysis of leaves from N‐deficient maize plants at the vegetative (10–12 leaves) and mature (55 days after silking) stages and found that genes involved in carbon assimilation were differentially expressed in response to plant N status, a response which was also found in proteomic and metabolomic analyses of the same experiment. Integration of existing lists of N‐responsive genes appears a difficult task as the lists developed in previous studies can be widely different and highly dependent on the specific N treatment employed and also the N status and developmental stage of the plants. Further, identification of controllers of ${\text{NO}}_3^{-}$ transport through network analysis may only indicate controllers unique to each individual study as transcription has been measured at one or two points in development.

Another factor limiting successful improvement of N uptake and assimilation may be our incomplete knowledge of the root N transport system and in particular how this system changes over the life cycle of crop plants. Previously, we found that ${\text{NO}}_3^{-}$ uptake capacity in maize grown with steady‐state N provision was highly variable across the life cycle and this capacity increased during times of peak ${\text{NO}}_3^{-}$ demand and further increased in response to reduced ${\text{NO}}_3^{-}$ (Garnett et al., 2013). The expression of genes encoding high‐affinity transport system (HATS) transporters (NRT2s) was highly correlated with the ${\text{NO}}_3^{-}$ uptake capacity across the life cycle and also ${\text{NO}}_3^{-}$ supply (Garnett et al., 2013). In particular, expression of ZmNRT2.1, ZmNRT2.2 and ZmNRT2.5 showed expression peaks at 18 and 29 days after emergence, which closely resembled the peaks in ${\text{NO}}_3^{-}$ uptake capacity. The orthologue of ZmNRT2.5 from rice was characterized as a high‐affinity ${\text{NO}}_3^{-}$ transporter localized to the root stele, and the authors hypothesize it is involved in the unloading of ${\text{NO}}_3^{-}$ from the symplasm into the apoplastic transpirational stream for transport to the shoot (Tang et al., 2012). Recently, AtNRT2.5 has been reported to mediate high‐affinity ${\text{NO}}_3^{-}$ uptake in coordination with AtNRT2.1, AtNRT2.2 and AtNRT2.4 under very low ${\text{NO}}_3^{-}$ provision (Lezhneva et al., 2014). Identifying the genes involved in regulation of NRT2‐mediated ${\text{NO}}_3^{-}$ transport is important for development of crops with improved ${\text{NO}}_3^{-}$ uptake and tissue redistribution efficiency.

In this study, we followed on from our previous work to determine whether other genes in the maize genome followed the same ${\text{NO}}_3^{-}$‐responsive expression pattern as NRT2 transporter genes. We also sought to identify genes that were coregulated with the NRT2 genes and likely to be involved in regulating the signalling cascade responsible for the NRT2 response observed previously. To accomplish this, we chose representative time points across the maize life cycle to analyse ${\text{NO}}_3^{-}$‐responsive transcription in leaves and roots via microarrays. We found that the number of genes exhibiting ${\text{NO}}_3^{-}$‐responsive transcription is highly variable across the life cycle and most of these genes are responsive at only one or two time points. Further, we identified a small number of genes, which are closely coregulated with NRT2 genes and found evidence for the involvement of lipid metabolism in the regulation of the response to limiting N supply and increased N demand.

Results

Previous work identified a correlation between the ${\text{NO}}_3^{-}$ uptake capacity in response to N demand and supply and root expression values for ZmNRT2 genes across the short‐life cycle of maize line, Gaspe (Garnett et al., 2013). Based on this finding, we chose seven key time points across the Gaspe life cycle to examine leaf and root transcription more completely in plants at most developmental stages (Figure S1). We chose one time point prior to the initial peak in ZmNRT2 expression (D11), one during the initial rise in expression (D15), one in the middle of the first expression peak (D18), one in the expression trough between peaks (D22), one in the middle of the second expression peak (D29) and two points after the second expression peak (D32 and D36) (Figure 1a,b). Anthesis occurred at D31. Expression analysis was conducted on tissue derived from the previous experiment using a custom 4 × 44K maize microarray that contained 43 843 features per array representing approximately 43 000 maize genes.

Figure 1.

Time points chosen for microarray analysis and comparison of ZmNRT2 transcript abundance from Q‐PCR and microarray data. The seven time points of Q‐PCR data for (a) ZmNRT2.1, (b) ZmNRT2.2 and (d) ZmNRT2.5 are adapted from (Garnett et al., 2013). Microarray data for (c) ZmNRT2.1/2 and (e) ZmNRT2.5 are presented in relative fluorescence intensity values. Open symbols are from roots grown in 0.5 mm ${\text{NO}}_3^{-}$ , filled symbols are from roots grown in 2.5 mm ${\text{NO}}_3^{-}$ .

Individual expression values for ZmNRT2.1 and ZmNRT2.2 were not able to be determined from the microarray data as the nucleotide sequence similarity between the genes is high (transcript—84.0%; coding sequence—98.5%), and the same 60‐mer probe set on this microarray measures expression of both genes together (ZmNRT2.1—perfect match; ZmNRT2.2—1‐bp mismatch). The expression pattern for this probe set did, however, bear similarity to the patterns of ZmNRT2.1 and ZmNRT2.2 expression as measured previously by Q‐PCR (Figure 1c). As the expression pattern of ZmNRT2.5 was very similar for Q‐PCR and microarray measurements it was chosen as a better representative of the two peaks of expression previously observed for ZmNRT2 genes (Figure 1d,e). We found further evidence of the high quality of the microarray data by comparing it to previously generated Q‐PCR data for ZmNRT1.2, ZmNRT1.5A and ZmNRT3.1A (Figure S2), which are highly expressed and variable across the life cycle (Garnett et al., 2013).

Differential expression between ${\text{NO}}_3^{-}$ treatments

The total number of probe sets showing significant (P ≤ 0.01) differential expression values between the ${\text{NO}}_3^{-}$ treatments was calculated for both leaf and root samples at all seven time points (Figure 2). A large number of probe sets were differentially regulated at D11, D18 and D29, and a comparatively small number at D15, D22, D32 and D36. Further, of D11, D18 and D29, the root had a much higher number of differentially regulated probe sets than the leaf at D11 and D29, while the opposite was true at D18 (Figure 2).

Figure 2.

Total number of differentially expressed probe sets in leaf and root over the maize life cycle. The total number of differentially regulated probe sets (P ≤ 0.01) between plants grown at 0.5 and 2.5 mm ${\text{NO}}_3^{-}$ . The area of the circle for each day represented is proportional to the number of differentially regulated probe sets on that day of sampling. The four colours represent: leaf—0.5 > 2.5 mm (light green); leaf—2.5 > 0.5 mm (dark green); root—0.5 > 2.5 mm (yellow); and root—2.5 > 0.5 mm (red). The total number of differentially regulated probe sets is provided for each circle.

As D11, D18 and D29 clearly had the most differentially regulated probe sets, Venn diagrams were used to identify the differentially regulated probe sets that were similarly affected at the three time points. Generally, common differentially regulated probe sets could be identified between two and three of the time points, but with some notable exceptions. For example, no differentially regulated probe sets were in common in the leaf between D11 and D29; thus, there were no differentially regulated probe sets in common in the leaf between all three time points (Figure 3). Of the 19 differentially regulated probe sets in the root among all 3 days, only one had higher expression values in the 0.5 mm ${\text{NO}}_3^{-}$ (annotated as a copper amine oxidase) and 2.5 mm ${\text{NO}}_3^{-}$ treatments (annotated as an aspartate aminotransferase), respectively (Figures 3 and S3). The other 17 probe sets on this list are regulated in opposite directions between D11 and D18, generally with a higher expression in 0.5 mm at D11 and in 2.5 mm at D18 (Figure S3). It is evident that this stems from a larger trend of there being only two common differentially regulated probe sets (out of a total of 183 common differentially regulated probe sets) regulated in the same direction in the root between D11 and D18 (Figure 3), one higher in 0.5 mm and the other in 2.5 mm.

Figure 3.

Analysis of differentially regulated probe sets at D11, D18 and D29. Venn diagrams depict the number of differentially regulated probe sets (P ≤ 0.01) found in the leaf (green) and root (orange). Values found in one circle are the number of differentially regulated probe sets unique to the particular day of sampling. Values located in two or all three circles are differentially regulated probe sets that are in common between the 2 or 3 days of sampling. In (a), the total number of differentially regulated probe sets is shown, in (b) the number of differentially regulated probe sets where expression is higher at 0.5 mm ${\text{NO}}_3^{-}$ is shown and in (c) the number of differentially regulated probe sets where expression is higher at 2.5 mm ${\text{NO}}_3^{-}$ is shown.

The common differentially regulated probe sets in the root between D11 and D29 were largely probe sets with higher expression in the 0.5 mm treatment (329) compared to those with higher expression in the 2.5 mm treatment (69) (Figure 3). This was unusual, for when comparing lists of common differentially regulated probe sets between time points, there was usually an even split between common differentially regulated probe sets being higher in the 0.5 mm or 2.5 mm treatments (Figure S4).

A comparison of the differentially regulated probe set lists between leaf and root at D11, D18 and D29 revealed small numbers (65, 76 and 14) of common differentially regulated probe sets for all three time points (D11, D18 and D29, respectively). However, no probe sets were regulated in the same direction (0.5 mm or 2.5 mm expression higher) between the leaf and root at D11 (Figure 4). The list of 65 common differentially regulated probe sets between the leaf and the root at D11 shows that at D11, all 65 probe sets are regulated in opposite directions between the leaf and root, but by D18, the direction of the root differential expression had reversed to that of the leaf at D11 and D18 (Figure S5).

Figure 4.

Analysis of differentially regulated probe sets at D11, D18 and D29. Venn diagrams depict the number of differentially regulated probe sets (P ≤ 0.01) found in the leaf (green) and root (orange). Values found in one circle are the number of differentially regulated probe sets unique to the particular tissue. Values located in both circles are differentially regulated probe sets that are in common between the tissues. In (a), the total number of differentially regulated probe sets is shown; in (b), the number of differentially regulated probe sets where expression is higher at 0.5 mm ${\text{NO}}_3^{-}$ is shown; and in (c), the number of differentially regulated probe sets where expression is higher at 2.5 mm ${\text{NO}}_3^{-}$ is shown.

Gene ontology enrichment

To determine whether the differentially expressed probe sets at specific days featured overrepresentation of particular functional classes of related genes, we conducted a gene ontology (GO) enrichment analysis using AgriGO (Du et al., 2010). As the maize genome annotation is much less complete than the traditional model species (Arabidopsis, rice), we first checked that the number of GO annotated probe sets was proportional to the total number of differentially regulated probe sets (Figure 2) so as not to bias the analysis (Figure S6a). The two sets of data were proportional, but interestingly, the number of GO‐enriched terms found within each list of probe sets was not. At D11 and D29, there were no GO‐enriched terms identified in the leaf (P ≤ 0.01); however, there were a larger number of GO‐enriched terms identified in the 0.5 mm than in the 2.5 mm treatment in the root at D11, while the opposite was true at D29. At D18, a larger number of GO‐enriched terms were identified in the leaf than the root; however, in the leaf, the enriched terms were identified almost exclusively in the 2.5 mm higher than 0.5 mm set, while the opposite was true in the root (Figure S6b). The lists of differentially regulated probe sets from D15, D22, D32 and D36 yielded no enriched GO terms, presumably because the lists did not contain enough probe sets to produce statistically significant enriched GO terms.

Inspection of the lists revealed three groups of GO‐enriched terms in particular contained noticeably related descriptors. The D11 root list where 0.5 mm had higher expression than 2.5 mm contained 29 enriched terms that were predominantly regulatory in nature (e.g. transcription, regulation of biosynthetic process and regulation of nitrogen compound metabolic process). The D18 leaf list where 2.5 mm had higher expression than 0.5 mm contained 26 enriched terms that were largely RNA‐ and ribosome‐related terms (e.g. RNA processing, translation and helicase activity). The D29 root list where 2.5 mm had higher expression than 0.5 mm contained 53 enriched terms many of which were related to N, amine and amino acid pathways and transport (e.g. cellular N metabolic process and glutamine family amino acid metabolic process) (Table S1).

Graphical cluster decomposition

To mine the microarray data for dominant expression patterns over the life cycle and to look for potential networks of probe sets with similar expression patterns, we employed a cluster decomposition of the relevant gene co‐expression networks as described in Methods. A co‐expression network for the expression data of the 0.5 mm ${\text{NO}}_3^{-}$‐treated roots defined by edges drawn between probe set (node) pairs exhibiting a Pearson correlation greater than 0.99 was developed (Figure 5a). In total, this network contains 14 530 nodes connected by 90 942 edges. The largest component contains 11 774 probe sets, and interestingly, the second largest component, cluster 2, contains only 98 probe sets and is the expression pattern observed for the ZmNRT2 genes in 0.5 mm ${\text{NO}}_3^{-}$‐treated roots including ZmNRT2.5 (Figure 5b). The largest component was subsequently further divided into a set of subclusters by increasing the correlation coefficient cut‐off to 0.996 (Figure S7).

Figure 5.

Description and example cluster decomposition of co‐expression networks in 0.5 mm‐treated roots. (a) The largest cluster connects 11, 774 probe sets, followed by smaller clusters with 98, 58 and 34 nodes. (b) Expression profiles for the probe sets in the seven largest clusters. Cluster 2 (marked in red) is the expression pattern which best represents the ZmNRT2 expression pattern. All probe sets that did not show more than a 0.5 (log₂) change in expression over the life cycle were excluded from the analysis.

The cluster decomposition was also repeated for the co‐expression networks of the 2.5 mm ‐treated roots (Figure S8a), and the 0.5 and 2.5 mm‐treated leaves (Figures S9a and S10a). Again, the largest cluster for each of these was further decomposed into subclusters as before (Figures S8b, S9b, S10b and S10c). Interestingly, the expression pattern in cluster 2 from the 0.5 mm‐treated roots (resembling ZmNRT2 expression) (Figure 5a) is not found in the 2.5 mm‐treated root or leaf clusters from the first or second decompositions. In the 0.5 mm‐treated leaves, there is a cluster with two peaks (cluster 1_1) as found for the ZmNRT2 genes (Figure S9b); however, the second peak is smaller than the first and there is a general decline in expression over time in this cluster (which is dissimilar to the cluster 2 from the 0.5 mm‐treated roots) suggesting these probe sets would not cluster with the cluster 2 probe sets from the 0.5 mm‐treated root.

Gene cluster composition

We compared the lists of probe sets in the clusters from all tissues and treatments as well as from the clusters derived from differential expression patterns in the leaf and root. This analysis revealed the clusters containing the largest number of probe sets in each respective data set were comprised of highly similar probe sets. This is visually represented by the group of clusters connected by darker lines (Figure S11), indicating the level of similarity between these clusters is often >50%. Despite the similarity among the rest of the clusters, the relationships among them are less defined. The largest clusters from the 0.5 and 2.5 mm‐treated leaf and root data sets were highly similar with 23.6%–79.1% similarity between them (Figure 6a). A total of 260 probe sets were found in each of the largest clusters across tissue and treatment. A GO analysis of the genes assayed by this list of probe sets revealed an overrepresentation of genes involved in abiotic stress response, in particular the response to low temperature (Figure S12a). The promoters (1 kb upstream of transcriptional start site) of this list of genes were analysed for common motifs using Promzea (www.promzea.org) which showed the CCAAT motif was overrepresented (Figure S12b). We found that the majority of the differential expression patterns clustered into two groups: in the leaf and in the root. Further, the differential expression clusters containing the largest number of probe sets for the leaf and root contained 67.3% of the same probe sets, while the second largest clusters contained 50.3% the same probe sets (Figure 6b).

Figure 6.

Comparison of common probe sets between the largest clusters derived from cluster decomposition of co‐expression networks. Arrow colour indicates degree of similarity of probe set lists between clusters 0%–24% (white), 25%–49% (light blue), 50%–74% (dark blue) and 75%–100% (black). Numbers in the ovals refer to the cluster number and the number of probe sets found within the cluster. Numbers beside the lines connecting the ovals are the number of similar probe sets between the clusters and the percentage similarity between the clusters ((# similar probe sets/probe sets in smaller cluster) × 100). Graphs depicting the expression patterns of the probe sets in each cluster are provided next to their respective clusters. (a) Similarity of probe sets between the largest clusters in terms of probe sets from leaves at 0.5 mm (light green) and 2.5 mm (dark green), roots at 0.5 mm (light red) and 2.5 mm (dark red). (b) Similarity of probe sets between the largest clusters of differentially expressed probe sets (0.5–2.5 mm expression values) found in the leaf (light green) and the root (yellow).

To delve further into potential relationships between the probe sets comprising cluster 2 from the 0.5 mm‐treated roots, a heat map with hierarchical clustering of the expression values in all tissues and treatments was developed (Figure 7). The values utilized for the heat map were derived by subtracting the average expression value from a tissue/treatment from the expression value at the individual days of the tissue/treatment set. This allowed visualization of common expression patterns as seen from the green expression peaks in the 0.5 mm‐treated roots at D18 and D29 (orange column). It was evident that the clustering of expression patterns was tightest for the 0.5 mm‐treated roots and the same pattern was not shared by the other tissue/treatment sets (red, light green and dark green columns). Minor subclustering appeared to exist based primarily on similarity in expression patterns in the leaf; however, both treatments shared similar expression profiles.

Figure 7.

Hierarchical clustering analysis of the probe set expression values from cluster 2 of the 0.5 mm ${\text{NO}}_3^{-}$‐treated roots. Values are the average expression value from each tissue/treatment data set subtracted from the individual expression value at each time point. Green squares represent positive values or greater than the average value for the tissue/treatment; red squares represent negative values or less than the average value for the tissue/treatment. All tissue/treatment data sets are included and the four colours represent: leaf—0.5 > 2.5 mm (light green); leaf—2.5 > 0.5 mm (dark green); root—0.5 > 2.5 mm (yellow); and root—2.5 > 0.5 mm (red). Highlighted are the probe sets representing ZmNRT2.5 and the four probe sets representing dual AP2 domain‐containing transcription factors (red arrows).

A GO enrichment analysis of cluster 2 from the 0.5 mm‐treated roots was completed using AgriGO (Du et al., 2010). The enriched GO terms were lipid transport, lipid localization, lipid binding and transferase activity (transferring acyl groups other than amino acyl groups; transferring acyl groups) (Table S2). Based on the number of probe sets annotated as encoding particular proteins on the microarray, an expected percentage was derived for each of the protein types represented in cluster 2. The actual percentage found for genes encoding several proteins was significantly enriched in cluster 2 (Table 1). In particular, dual AP2 domain‐containing transcription factors were 96.9 times more prevalent in cluster 2 than on the array in general. The four probe sets also are represented in one small node of the heat map of cluster 2 (Figure 7). STRUBBELIG‐RECEPTOR FAMILY 6 receptor kinases were overrepresented 35.2 times. Lipid transfer proteins, glycerol‐3‐phosphate acyltransferases, 3‐ketoacyl‐CoA synthases, alpha/beta hydrolases and GDSL‐like lipase/acylhydrolases were enriched 37.8, 77.5, 44.7, 20.1 and 8.2 times, respectively (Table 1).

Table 1.

Enriched gene classes represented by probe sets in cluster 2 from roots treated with 0.5 mm ${\text{NO}}_3^{-}$

Class	Description	Expected %	Actual %	Fold enrichment
Transcription factor	Dual AP2 Domain	0.04	4.08	96.9
Kinase	STRUBBELIG‐RECEPTOR FAMILY 6	0.06	2.04	35.2
Lipid transfer	Lipid transfer protein	0.22	8.16	37.8
Lipid metabolism	Glycerol‐3‐phosphate acyltransferase	0.04	3.06	77.5
	3‐ketoacyl‐CoA synthase	0.07	3.06	44.7
	Alpha/beta hydrolase fold	0.20	4.08	20.1
	GDSL‐like lipase/acylhydrolase	0.25	2.04	8.2

Expected % = ((# or probe sets from the gene class on the microarray ÷ total probe sets on the microarray) × 100). Actual % = ((# of probe sets from the gene class in cluster 2 ÷ total probe sets in cluster 2) × 100).

Discussion

Nitrogen‐responsive transcription is predominant in the roots and is dynamic across the life cycle

Overall, transcription in the root across time points was 3.7 times more responsive to ${\text{NO}}_3^{-}$ provision than it was in the leaf (total number of differentially regulated root probe sets divided by the same number for the leaf). The number of differentially regulated probe sets in both tissues across time points and tissues more highly expressed under low ${\text{NO}}_3^{-}$ provision was slightly higher than the reverse (1.2 times). This finding is similar to a study in Arabidopsis which found 6.5 times the number of genes expressed differentially in response to ${\text{NO}}_3^{-}$ provision in the root compared to the shoot (Wang et al., 2003). It is possible that the differentially expressed genes in the root are related to an increase in the ${\text{NO}}_3^{-}$ uptake capacity of the roots in the low ${\text{NO}}_3^{-}$ treatment. However, this would be somewhat surprising because a concomitant up‐regulation of genes encoding the N‐metabolism machinery located in the shoot was not measured. It is important to note that plant growth did not differ between the two ${\text{NO}}_3^{-}$ treatments although there were differences in ${\text{NO}}_3^{-}$ uptake capacity, total leaf N, ${\text{NO}}_3^{-}$ and amino acids implying that the leaf N levels were above some threshold which allowed it to maintain N assimilation without a large transcriptional response to the low ${\text{NO}}_3^{-}$ treatment (Garnett et al., 2013).

Our previous work showed large and coordinated variation in ${\text{NO}}_3^{-}$ uptake capacity, amino acid content, ${\text{NO}}_3^{-}$ content and expression levels for genes encoding NRT2 ${\text{NO}}_3^{-}$ transporters across the life cycle of maize (Garnett et al., 2013). This implied that there was likely to be large variation in N‐related transcription across the maize life cycle which was tightly regulated in response to ${\text{NO}}_3^{-}$ provision. It also indicated that N‐responsive transcriptional analyses could differ markedly, even from time points only a few days apart. It is well established that gene expression changes greatly between tissues and across the development of multiple plants including Arabidopsis (Ma et al., 2005; Schmid et al., 2005), soya bean (Libault et al., 2010), Medicago truncatula (Benedito et al., 2008), rice (Jiao et al., 2009; Wang et al., 2010), barley (Druka et al., 2006), bread wheat (Schreiber et al., 2009) and maize (Downs et al., 2013; Liseron‐Monfils et al., 2013a; Sekhon et al., 2011). Our analysis that builds on this work has revealed responsive transcription is highly dynamic across the life cycle, and proper interpretation of expression data requires knowledge of this variation under steady‐state ${\text{NO}}_3^{-}$ conditions.

Differential expression does not appear to be influenced by developmental differences between plants grown in the two N treatments

Although the ZmNRT2 genes were more highly and differentially expressed at later time points (D18 and D29) as judged from the microarrays and Q‐PCR, the largest number of differentially regulated probe sets in the root occurred at D11. It is possible that this is related to a switch from amino acid‐derived N being sourced from the seed to an increased reliance on ${\text{NO}}_3^{-}$ uptake from the growth solution. As seen previously, the root levels of free amino acids reach a minimum at D11 with the levels being lower in the 0.5 mm ${\text{NO}}_3^{-}$‐treated roots (Garnett et al., 2013). Previous work indicated that maize seed N reserves were available until at least D7 (Watt and Cresswell, 1987). In support of this idea, the enriched GO terms derived from the list of differential genes in the roots at D11 for 0.5 > 2.5 mm included terms such as ‘transcription’ and ‘metabolism’, while the 2.5 > 0.5 mm list includes terms such as ‘transporter’ and ‘ATPase’ (Table S1).

There were no differences in leaf developmental rate, biomass or yield between the plants grown at the two levels of ${\text{NO}}_3^{-}$; presumably this was the result of the increase in ${\text{NO}}_3^{-}$ uptake capacity in the plants grown in the low ${\text{NO}}_3^{-}$ treatment (Garnett et al., 2013). The shoot and root differences in tissue ${\text{NO}}_3^{-}$ and total amino acids differed only marginally between plants grown in the two treatments further supporting this notion. However, subtle differences in development between plants in the two treatments could account for the large number of differentially regulated probe sets. To answer this question, we identified all putative maize orthologues represented on our microarray of several Arabidopsis genes which regulate circadian rhythms and plant development, namely CCA1, ELF4, LHY and TOC1/PRR genes (Kolmos et al., 2011; Lu et al., 2012; Nakamichi et al., 2010). We could find no differential expression of any of these orthologous genes at any time point (Figure S13), implying the genes we identified are truly differentially expressed in response to ${\text{NO}}_3^{-}$ supply.

Only a few patterns of differential gene expression occur over time

We examined all the patterns of differential regulation of probe sets over the life cycle to determine to what degree common patterns in gene expression response to ${\text{NO}}_3^{-}$ provision occurred and how this might change across the life cycle. We measured expression at seven time points and two treatments; thus, there could be 3⁷ patterns of expression because there could be three expression outcomes at each time point, 2.5 > 0.5 mm, 0.5 > 2.5 mm or 2.5 = 0.5 mm. This means 2187 differential probe set patterns could exist across the life cycle if there was no control over transcription and the outcome at each time point was essentially random. In both the leaf and the root, we found just two differential expression patterns dominated the data sets (Figure 6b). Over half of the differentially regulated probe sets in both the leaf and root clustered into two expression patterns and could not be decomposed further using the decomposition technique. In the leaf, there were just six differential expression pattern clusters containing more than 20 probe sets, while in the root, there were only ten. Whereas our analysis involved response to ${\text{NO}}_3^{-}$ provision across multiple time points, Krouk et al. (2009) completed a comparable analysis of differentially regulated genes in response to ${\text{NO}}_3^{-}$ provision, light and sucrose in leaves and roots of Arabidopsis at one time point. They found that nearly 80% of the genes that were regulated by the various combinations between the treatments were represented in only 87 models of over 14 million possibilities. They further attributed this observation to ‘a major constraining structure in plant cell signalling pathways’ and hypothesize the ‘existence of a “code” governing signal integration at the organism level, which is responsible for the observed global gene expression reprogramming….’ This also appears to be the case in our data as we found evidence for a very tight regulation of ${\text{NO}}_3^{-}$‐responsive gene transcription when looking at response to nongrowth‐limiting N over the majority of the life cycle.

Comparison of the probe set lists from the largest leaf and root cluster from both treatments revealed that the four lists were highly similar (Figure 6a), again indicating that the expression of large groups of genes are responding in the same way to ${\text{NO}}_3^{-}$ treatment across tissues. This list was overrepresented for genes involved in response to cold (Figure S12a), indicating there may be significant similarity in the ‘core machinery’ responses of plants to low temperature and limiting N provision. Further, the promoters of this group of genes were found to contain an overrepresentation of the CCAAT motif (Figure S12b), which is bound by the nuclear factor Y (NFY) transcription factors (Laloum et al., 2013), and are important for plant response to multiple environmental and developmental stimuli (Petroni et al., 2012). Several of the NFY transcription factors are regulated by N provision in Arabidopsis (Bi et al., 2007; Zhao et al., 2011). A similar regulatory system may be important for our list of genes.

A lipid metabolism network responds to N supply and demand

The genes coregulated with NRT2s contained an enrichment for those encoding proteins involved in lipid trafficking and metabolism. These data suggest several possible scenarios. The first is that the regulation of the response to N supply and demand involves lipid signalling. This model of lipid transfer protein‐mediated signalling has been reported previously in relation to pathogen response signalling (Chanda et al., 2011; Lascombe et al., 2008; Maldonado et al., 2002). The second is that the genes may regulate root endodermal and exodermal suberin content, which is an important factor in controlling root uptake of water and minerals (Baxter et al., 2009; Chao et al., 2011; Pollard et al., 2008; Ranathunge et al., 2011). The third is that the up‐regulation of lipid metabolism is necessary for trafficking of N transport proteins into or from the plasma membrane or for assembly of the complex in the plasma membrane, as has been postulated for NRTs (Laugier et al., 2012; Yong et al., 2010). A fourth scenario is that the genes may control synthesis of extra storage lipids which would facilitate ‘mopping up’ of excess carbon skeletons not required for amino acid synthesis during periods of low N availability.

Conclusion

We have described here the first characterization of ${\text{NO}}_3^{-}$‐responsive transcription across seven key points in the life cycle of a crop plant when challenged with limiting levels of N (which affect N uptake, but not plant growth). This analysis revealed how the response of the plant to ${\text{NO}}_3^{-}$ provision is highly dynamic, with individual time points having unique compositions of ${\text{NO}}_3^{-}$‐responsive genes. It is clear that analysing transcription at single time points or studying short‐term responses to drastic changes in ${\text{NO}}_3^{-}$ provision will only provide a snapshot of the underlying response of many genes to N supply and demand. This may provide an explanation to the different lists reported by previous studies characterizing N transcriptional responses in plants. While this analysis looking at a range of time points provided many new insights into the transcriptional response of plants to ${\text{NO}}_3^{-}$ availability, it is clear that the timescale we used in this study will need to be finer to truly dissect the gene expression responses around key time points in the life cycle of the plant. In particular, daily measurements of transcription (if not more frequent) will be required around the most active ${\text{NO}}_3^{-}$‐responsive transcription days (D11, D18 and D29) to clarify ${\text{NO}}_3^{-}$ signalling and regulatory responses involved in key functional events controlling ${\text{NO}}_3^{-}$ uptake and metabolism. When used in combination with measurements of N‐related enzyme activities, proteins and metabolites across the life cycle, such an analysis will allow a systems biology approach to be employed as has been undertaken in Arabidopsis (Gutierrez et al., 2007; Gutiérrez et al., 2008; Krapp et al., 2011) and is beginning to emerge in maize (Amiour et al., 2012; Schlüter et al., 2012). Finally, transcriptomic data are traditionally an excellent source of candidate gene discovery and the added temporal aspect of the candidate gene discovery in this study provides an extra degree of information to the candidate genes we identified. An enrichment of genes encoding lipid metabolism‐related proteins were co‐expressed with those encoding NRT2 transporters, suggesting the existence of a lipid metabolism response to N supply and demand. These genes will be explored further with a transgenic approach in efforts to improve ${\text{NO}}_3^{-}$ uptake and N‐metabolism as part of endeavours to improve the NUE of crop plants.

Experimental procedures

The plants used in the experiments described here were the same dwarf maize (Zea mays L. var Gaspe Flint) which were grown in hydroponic systems in modified Johnson's solution (Johnson et al., 1957) containing either 0.5 or 2.5 mm ${\text{NO}}_3^{-}$ as described previously (Garnett et al., 2013). Plants were sampled between 5 and 7 h after the start of the light period (06:00). The whole root and the youngest fully emerged leaf blade were excised, snap‐frozen in liquid N and stored at −80 °C. Total RNA was extracted from frozen tissue (Chomczynski, 1993), and 10 μg of aliquots was prepared for microarray analysis. RNA integrity was checked on a 1.2% (w/v) agarose gel. Three biological replicates were analysed in all experiments.

Total RNA was treated with DNase‐I followed by polyA RNA isolation (Illustra mRNA Purification Kit; GE Biosciences, Pittsburgh, PA) for all samples. The total RNA and polyA RNA samples were visualized and quantified on Agilent's Bioanalyzer 2100 (Palo Alto, CA) to check for degradation and to determine final concentration. Each mRNA sample was made into double‐stranded DNA, amplified by an in vitro transcription reaction and labelled with Cy3 fluorescent dye using Agilent's Low RNA Input Fluorescent Linear Amplification Kit. The cRNA product was purified with Agencourt's RNAClean Kit that utilizes SPRI (solid‐phase reversible immobilization) paramagnetic bead‐based technology. Overnight hybridizations were performed with equal amounts of labelled cRNA to a custom 4 × 44K Maize Oligo Microarray from Agilent Technologies according to Agilent's One‐Color Microarray‐Based Gene Expression Analysis protocol. After hybridization, the microarray slides were washed and immediately scanned with Agilent's G2505C DNA Microarray Scanner. The images were visually inspected for image artefacts, and feature intensities were extracted, filtered, and normalized with Agilent's Feature Extraction Software (v10.5.1.1). Further quality control and downstream analysis were performed using data analysis tools in GeneData Analyst (v2.2.2) (GeneData; Basel, Switzerland).

Cy3 median signal intensities were imported into R for further processing, omitting 4825 probes with no and very low fluorescent signals. The intensity values were log₂ transformed and quantile normalized. The average of the Pearson Correlation coefficients for the biological replicates in the leaf samples was 0.983 (range 0.967–0.996) and in the root samples was 0.991 (range 0.974–0.997). Differently expressed genes were identified by the moderated t‐statistic implemented in the LIMMA package (Smyth 2005). P‐values were adjusted employing the method by Benjamini & Hochberg (1995) to control the false discovery rate (FDR). Genes were considered differentially expressed between the two conditions when their adjusted P‐values were less or equal to 0.01.

All gene ontology analysis was undertaken using AgriGO (Du et al., 2010). A singular enrichment analysis (complete GO) was undertaken on the query lists using the ‘Zea mays ssp V5a’ species and the ‘Maize genome V5a transcript ID’ reference background. The Fisher statistical method was chosen with the Hochberg (FDR) multitest adjustment at a significance level of 0.05. Heat maps and hierarchical clustering analysis on mean‐centred probe set data were undertaken using Genesis (Sturn et al., 2002). Analysis of maize promoters for common motifs was completed using Promzea (Liseron‐Monfils et al., 2013b). Gene ID lists were entered and the 1000‐bp promoter size was chosen for analysis.

Individual gene co‐expression networks were constructed for the root 0.5 mm, root 2.5 mm, leaf 0.5 mm and leaf 2.5 mm data as follows. First, profiles were defined by averaging over repeats for each of the seven time points. Next, contributions from uninformative genes were eliminated by removing those probe sets whose expression relative to average did not deviate by more than 0.5 (log₂) for any one time point. A co‐expression network was then constructed using the software program Wolfram Mathematica 8 (Wolfram Research; Champaign, IL) by connecting those probe sets (nodes) by edges that had expression profiles with a Pearson Correlation coefficient greater than a predetermined cut‐off. Clusters emerge naturally as the disconnected components of such networks.

Microarray data reported in this study have been deposited in the Gene Expression Omnibus database under the accession number GSE61820.

Supporting information

Figure S1 Representative images of Gaspe maize throughout the lifecycle.

Figure S2 Comparison of transcript abundance from Q‐PCR and microarray data.

Figure S3 Heatmap and hierarchical clustering of 19 probe sets differentially regulated at D11, D18 and D29 in roots.

Figure S4 Analysis of differentially regulated probe sets between consecutive sample days.

Figure S5 Heatmap and hierarchical clustering of the 65 probe sets differentially regulated at D11 in both leaf and root.

Figure S6 Identification of Gene Ontology (GO) term enrichment among differentially regulated probe sets.

Figure S7 Sub‐clustering of the largest connected component of the co‐expression network for roots from plants grown in 0.5 mm ${\text{NO}}_3^{-}$.

Figure S8 Results of the cluster decomposition of co‐expression networks for roots from plants grown in 2.5 mm ${\text{NO}}_3^{-}$.

Figure S9 Results of the cluster decomposition of co‐expression networks for leaves from plants grown in 0.5 mm ${\text{NO}}_3^{-}$.

Figure S10 Results of the cluster decomposition of co‐expression networks for leaves from plants grown in 2.5 mm ${\text{NO}}_3^{-}$.

Figure S11 Comparison of common probe sets between the largest clusters derived from cluster decomposition of co‐expression networks.

Figure S12 Analysis of the probe set list common between the largest cluster from each tissue and treatment.

Figure S13 Expression values for putative maize orthologues of developmental genes (CCA1, ELF4, LHY and PRR/TOC1) in Arabidopsis.

Table S1 Gene Ontology (GO) term enrichment analysis of significant differentially regulated probe sets from leaf and root at D11, D18 and D29.

Table S2 Gene Ontology (GO) term enrichment analysis of significant differentially regulated probe sets from Cluster 2 of the cluster decomposition of co‐expression networks of roots from plants grown in 0.5 mm ${\text{NO}}_3^{-}$.