Nitrogen (N) is an integral component of nucleic acids and amino acids, and as such an essential macronutrient for living organisms. Plants take up inorganic N mainly as nitrate from the soil through the roots, and transform it into organic N in roots and leaves. Consequently, N availability in the soil influences plant development and morphology, and is a major determinant of crop yield (Frink et al. 1999). Substantial yield increases have been obtained through application of fertiliser, but at high environmental costs, such as nitrate pollution of water resources (Sutton et al. 2011). The extensive use of N fertilisers in crop production also causes high costs for farmers (Robertson and Vitousek 2009). To implement low-input regimes, increasing the nitrogen use efficiency (NUE) of crop plants is of paramount importance (Gutiérrez 2012). NUE is a complex trait that emerges from integration of metabolic, physiological and developmental processes, and depends mainly on two factors: N uptake efficiency and N utilisation efficiency (Masclaux-Daubresse et al. 2010; Xu et al. 2012). A large proportion of the total N in the plant is bound in photosynthetic proteins; a crucial component of N utilisation efficiency is, therefore, likely to be the ability to detect changes in light intensity and quality, and reallocate N to zones with the most favourable light regimes (Hirose and Bazzaz 1998).

Quantitative trait loci (QTL) for NUE were identified in maize (Agrama et al. 1999), barley (Kindu et al. 2014), rice (Wei et al. 2012) and Arabidopsis (Loudet et al. 2003). Molecular analyses of NUE have focused on individual genes whose properties or annotation suggested a role in the regulation of processes related to nitrogen uptake, assimilation or use, such as members of the NRT1 and NRT2 nitrate transporter families (reviewed in Krapp et al. 2014; O’Brien et al. 2016). Their relation to NUE, however, remains unclear.

In general, high nitrate supply promotes shoot growth (Scheible et al. 2004), while low nitrate availability induces enhanced root growth for foraging (Walch-Liu et al. 2006; Gruber et al. 2013). Moreover, the effect of nitrate on root morphology is independent of its assimilation (Forde and Lorenzo 2001). Therefore, nitrate has been considered not only as a major macronutrient, but also as a powerful signalling molecule. The nitrate signal has both systemic morphological effects, inhibiting the lateral root development when provided at a uniform high concentration, and local morphological effects, enhancing local lateral root elongation when provided in a restricted area of the root (Zhang and Forde 2000). More recently, natural variation in the nitrogen response of different Arabidopsis accessions has been described (North et al. 2009; De Pessemier et al. 2013). Several studies pointed out the interdependence between nitrogen and carbohydrate metabolism (Martin et al. 2002; Scheible et al. 2004; Tschoep et al. 2009; Huarancca Reyes et al. 2016).

The world-wide Arabidopsis population contains a diverse range of genotypes, which have adapted to many different growth conditions and show large variation in the expression of numerous traits including morphology, biotic and abiotic stress tolerance, flowering time, seed composition, water use, and nutrient efficiency (Koornneef et al. 2004; Weigel 2012). They show considerable variation of NUE (Chardon et al. 2010; Ikram et al. 2012). In the Bay x Sha RIL population, QTL were detected that affect NUE and other traits related to nitrogen metabolism and uptake (Loudet et al. 2003).

Many molecular studies of the transcriptomic response to N stress address the primary nitrogen response using either very short exposures to stress (2–30 min), or resupplying N after a longer stress (Kiba et al. 2012; Marchive et al. 2013; Vidal et al. 2013). These studies have shown that plants react within minutes to changes in the nitrate/ammonium level in their environment. The primary nitrogen response leads to transcriptomic reprogramming of up to several thousand genes. Fewer studies are concerned with morphological and transcriptomic adaptations to sustained nitrogen stress. Several differentially expressed genes were identified under mild or severe chronic N stress in Arabidopsis (Bi et al. 2007). Nitrogen remobilisation was identified as a means to cope with long-term N limitation (Masclaux-Daubresse et al. 2010).

We performed deep phenotyping and gene expression analyses on Arabidopsis accessions and mixed recombinant inbred lines to explore ways to detect morphological traits that could be used as biomarkers allowing identification of plants with good NUE at an early developmental stage, and to elucidate the genetic factors associated with good NUE. We show that the (growth) response to differing nitrate availability is highly diverse in Arabidopsis accessions. Transcription and epigenetic factors were identified as important players in the adaption to limited nitrogen in a global gene expression analysis. Five nitrate-responsive genes emerged as possible biomarkers for NUE in Arabidopsis.

Materials and methods

Biological materials

Two complementary populations of Arabidopsis thaliana that sample existing natural diversity were used. One population consisted of 101 Arabidopsis accessions selected based on genetic diversity estimated using 149 SNPs (Platt et al. 2010) and different geographical origins (Table S1). A mixed population of 123 recombinant inbred lines (mRILs) was assembled from 9 crosses involving 18 different parental accessions from various sources (Table S2). As mRILs originate from recent crosses, they offer the chance to uncover natural variation hidden in the accessions.

Plant culture

Population screens

All accessions and mixed RILs were screened for biomass production in an agar plate-based high-throughput procedure with limited supply of nitrogen. The cultivation medium was adapted from Estelle and Somerville (1987). Severe nitrogen stress was applied for the selection. For sowing, the seeds, which had been stored in the refrigerator at − 20 °C, were removed from the Eppendorf tubes with a toothpick and transferred onto agar plates, with limited supply of nitrogen (0.05 mM KNO3). In total, approximately, 15–30 seeds were distributed horizontally on each plate (12 × 12 cm). After sowing, plates were subjected to stratification for 2–4 days in the dark at 4 °C. After stratification, the plates were transferred into growth cabinets and plants were grown under a 16-h-light, 8-h-dark rhythm at 20 °C. The daytime light intensity was set to 100 µmol PAR m−2 s−1 (Philips TL5 54 W/840 HO). Excessive plants were removed from the plate 10–11 days after sowing, leaving up to 8 plants per plate for further screening and biomass assessment. Plant fresh biomass was assessed by weighing of shoots and roots 23 days after stratification.

In a complementary approach, analysis of shoot growth of the populations on low N soil was performed according to Tschoep et al. (2009). Shoot fresh weight was determined 35 days after germination.

Validation of candidate biomarker genes was performed using independent accessions with contrasting growth on low nitrogen, with Ak-1, Cvi-0, Gy-0 showing low NUE, and El-0, Col-0, Appt-1 demonstrating good NUE. Plants were grown for 40 days on a mixture of 70% vermiculite (Kakteen Schwarz, Nürnberg, Germany) and 30% soil substrate 2 (Klasmann-Deilmann GmbH, Geeste, Germany) and irrigated every day with liquid Estelle and Somerville solution containing 0.4 mM nitrate.

Assessment of morphological traits

Thirteen lines with contrasting NUE in the population screens were selected for deep phenotyping in vertical square petri dishes (120 × 120 × 17 mm). The medium (Estelle and Somerville 1987) was supplemented with 8% agar and 1% sucrose. The control medium (9 mM nitrate) contained 5 mM KNO3 and 2 mM Ca(NO3)2, the low nitrogen medium (0.4 mM nitrate) contained 0.2 mM Ca(NO3)2 and 5 mM KCl. The nitrogen stress applied, less severe than in the population screen, balances the need for a strong selection with the need to produce enough material for molecular analyses. Seeds were surface-sterilised for 1 min in 70% ethanol followed by 1 min in 2% NaClO and four washing steps in sterile distilled water. 12 or 6 seeds were regularly spaced 1 cm from the upper border of the dish. Seeds were stratified for 24 h at 4 °C, and then transferred to a growth cabinet set to 16-h light (100 µmol PAR m−2 s−1)/8-h dark, 20°/18 °C for 21 days.

Plants were scanned on an Epson Expression 10000XL scanner 14 days after sowing. Background and contrast of the images were adjusted in Adobe Photoshop CS3. Root architecture traits, namely number of roots, main root length in mm, and total lateral root length in mm, were extracted using WinRhizo Pro with root morphology and link analysis in developmental mode. Lateral root density was calculated as number of roots/main root length, lateral root ratio as total lateral root length/(total lateral root length + main root length).

After 21 days, plants were harvested, separated into root and shoot, weighed and immediately quenched in liquid nitrogen. The plant material was stored at − 80 °C until further use. Shoot and root fresh weight, in mg plant−1, were determined using pools of 3–10 plants per plate, and fresh weight was calculated per plant. The shoot–root ratio was derived from the calculated values as shoot fresh weigh/root fresh weight.

Gene expression analysis

Six lines showing contrasting growth characteristics on low N in the deep phenotyping screen were selected for gene expression analysis, the accessions El-0 and Fei-0, and the mRILs KB006, NG064, NG087, SG063. For each line, root and shoot samples were taken from plants grown on 0.4 mM and 9 mM nitrate. Three biological replicates were used per organ and treatment, in total 72 samples. One replicate consisted of a pool of 30 plants harvested from three plates. Total RNA was isolated using the adapted hot-borate method (Wan and Wilkins 1994). Hybridisation to the Affymetrix ATH1 array and readout of data in the form of CEL files were performed by Atlas Biolabs (Berlin, Germany). The ATH1 gene expression datasets generated were submitted to the Gene Expression Omnibus (GEO) repository (

Candidate biomarker genes were analysed in six accessions with four biological and three technical replicates using quantitative real-time PCR. RNA was extracted from shoots using the Spectrum Plant Total RNA Kit (Sigma–Aldrich, Munich, Germany) according to the manufacturer’s instructions. cDNA was synthesised using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). AtEF1α (AT1G07940) was used as housekeeping reference gene. The following primer pairs were designed using Primer3web version 4.1.0 (Untergasser et al. 2012) ( AtAMT1;2 (5′-tgg gtg acg gta act atg gg-3′, 5′-gtg tca tgt cca ttc ctg cc-3′), AtGATA21 (5′-tgt aac acc acc aag act cc-3, 5′-aga cac gag cca aac aca-3); AtGLU1 (5′-tgc tcg ata ttg gcg gac ta-3′, 5′-act ttt gcc tca tca act ggt -3′); AtNPF6.4 (5′-gac gct aaa ctc ggt cgc ta-3′, 5-tgg ctc cat gag atg atg aca-3′); AtEF1α (5′-tga gca cgc tct tct tgc ttt ca-3′, 5′-ggt ggt ggc atc cat ctt gtt aca-3′). qPCR was performed on a Quantstudio 6 Real-Time PCR System (Applied Biosystems) using the SYBR Green PCR Master Mix (Thermo Fisher Scientific). ΔCT were calculated as CT (marker gene) − CT (EF1alpha reference gene) and compared by ANOVA using Fisher’s LSD (alpha = 0.05) with Bonferroni correction.

Statistical analyses

The biomass data were log10-transformed before further analysis to achieve an approximate normal distribution. Univariate statistics of the log10-transformed data were calculated by ANOVA with a linear model in SAS (proc mixed) with line (accession/mixed RIL) as a fixed effect. Each line was tested against the remaining population. Significance level was set to an alpha-error of 5%.

Further statistical analyses were performed using GenSTAT 17th edition (VSNi, Hempstead, UK). Adjusted mean values and associated standard errors for fresh weight, root architecture traits were estimated by ANOVA. The variation explained was extracted from sums of squares. Significant differences between lines within treatments were determined by least significant difference (LSD at 5%) with Bonferroni correction. Significant differences between the treatments within lines were analysed by Mann–Whitney U test with Bonferroni correction. Pearson correlations were performed using the procedure FCORRELATION, with FDRMIXTURES for correlations between genes and morphological traits, and FDRBONFERRONI for correlations among morphological traits. PCA was performed in R using package pcaMethods (Stacklies et al. 2007) on scaled and centred data.

Microarrays were analysed using the R packages affy (Gautier et al. 2004), limma (Ritchie et al. 2015), simpleaffy (Wilson and Miller 2005), and affyPLM (Bolstad et al. 2005). Quality was controlled using fitPLM (affyPLM package) and qc (simpleaffy package). Data were normalised with GC-RMA algorithm (affy package). Differential expression was detected by fitting linear models using lmFit (limma package) to the contrasts between plus and minus nitrogen samples. Resulting p values were controlled for multiple error correction using false discovery rate (FDR).

Mean transcript levels were used for a pattern matching search using Pavlidis Template Matching embedded in the MultiExperiment Viewer 4.9 (Saeed et al. 2006) ( The patterns followed the distribution of the morphological traits (shoot fresh weight, root fresh weight) across the lines. VENN diagrams were constructed with InterActiVenn (Heberle et al. 2015). GO term overrepresentation analyses were performed using the web-based application GOrilla (; Eden et al. 2009).


Large-scale phenotypic screening for plant lines with contrasting NUE

Good NUE in the context of the population screens was defined as higher than average aerial biomass (fresh weight) on N depleted synthetic media, and determined in two complementary populations. The first population consisted of a collection of 123 mixed recombinant inbred lines (mRILs) and their 18 parental lines. Adjusted mean fresh weight was estimated using ANOVA with line as fixed effect (Table S3). 55 mRILs from the first screen with high, average or low biomass were selected for two further rounds of screening. To identify lines with reproducible growth characteristics on depleted N for deep phenotyping in synthetic media, a reproducibility measure based on rank categories was implemented (see below). From this set, three top-ranking mRILs (KB006, SG063, KT130) and three low-ranking mRILs (NG064, NG087, WC056) were selected for deep phenotyping. Transgressive segregation was observed in most mRIL combinations, with individual RILs exceeding the fresh weight of the worst or best parent by 38% and 118%, respectively (Table 1). The extent of transgression is negatively correlated with the phenotypic difference between the parental lines (r = − 0.733, P  = 0.006 and r = − 0.817, P  = 0.002, respectively, for lines larger than the high parent or smaller than the low parent), confirming the segregation of ‘hidden’ alleles.

Table 1 Transgressive segregation for shoot fresh weight in mRILs

Phenotypic screening was also performed using a population of 101 Arabidopsis accessions. Growth of the accessions was assayed in two consecutive experiments in synthetic (agar) media, and furthermore in two consecutive experiments in soil. Adjusted mean fresh weight was estimated using ANOVA (line = fixed effect). Mean fresh weight ranged between 0.55–1.67 (experiment 1) and 0.51–1.50 (experiment 2) mg plant−1 in agar, and between 27.08–152.50 and 37.41–106.63 mg plant−1 in soil (Table S4), respectively, revealing considerable (genetic) variation. For each experiment, lines were assigned the classes A, B, C, or D, according to their ranking in the upper, second, third or lower quartile. Lines that did only switch between adjacent classes were considered to exhibit reproducible behaviour on low N, and therefore suitable for further analyses. From this set, three top-ranking accessions (El-0, Rak-2, Lu) and three low-ranking accessions (Fei-0, Ts-1, C24) were selected for deep phenotyping, supplemented by the reference accession Col-0. Correlations between the different experiments were significant (P < 0.05) within growth conditions, with r = 0.537 for soil and r = 0.337 for agar media experiments, but no significant correlation between growth in agar and in soil was found. Twenty-one accessions showed similar behaviour in agar and in soil on limited N (Fig. 1a), whereas 14 accessions showed contrasting behaviour (Fig. 1b).

Fig. 1
figure 1

Accessions with similar a and contrasting b growth behaviour on agar and soil on low N during population screens. Shown are scatterplots of adjusted mean shoot fresh weight of accessions grown for 21 days on synthetic agar medium (x-axis) or for 35 days on soil (y-axis). The corresponding linear regression is indicated by the black line

Deep phenotyping

The 13 lines with contrasting NUE selected from the population screens underwent deep phenotyping in agar plates on low (0.4 mM) and sufficient (9 mM) nitrate (Table 2). ANOVA revealed significant effects (P < 0.001) in shoot and root fresh weight and shoot–root ratio for line (genotype), treatment (nitrogen) and the interaction term, indicating that lines reacted differently to changes in the nitrogen supply (Table S5). The 13 lines tested differed in their growth at sufficient nitrogen, e.g. KB006 produced three times more shoot material than Col-0 (Table 2). However, the shoot growth behaviour at sufficient N did not correlate with growth at low N; the correlation of the respective ranks, r = − 0.11, is not significant. There is a significant correlation between shoot and root growth at sufficient N (r = 0.835, P = 0.0004), but this link is lost on low N (r = 0.106, P = 0.7304). The root response was more variable than the shoot response, with a CV of 39.7% vs. 21.1%. This is also reflected in the respective portions of variance explained by genotype (line), treatment and the interaction between genotype and treatment. The highest portion of the variation can be attributed to nitrogen treatment for all three traits (69.0, 50.3. 27.4% for shoot and root fresh weight and shoot–root ratio, respectively; Table S5).

Table 2 Fresh weights and ratios of 13 Arabidopsis accessions and lines grown on differing N regimes

Root biomass on low N was significantly and strongly reduced in three lines, Ts-1, KB006, KT130, and moderately in four lines, C24, El-0, Lu, SG063 (Table 2). The two lines KB006 and KT130 showing the strongest reduction in shoot growth also had the strongest reduction in root growth. At sufficient N, shoot–root ratio is greater than 1 in all lines, reflecting a greater investment in shoot growth. At low N, the ratio is generally lower, and a significant reduction was detected in seven lines (El-0, Fei-0, NG064, NG087, SG063, WC056), indicating a shift to increased root growth. Only Ts-1 displays a slight increase in shoot–root ratio, while the shoot–root ratio of KB006 remains virtually unchanged on low N. Interestingly, Fei-0 and SG063 on low N display a ratio below 1 (0.950 and 0.886, respectively); a significant decrease from 2.752 and 2.084, respectively, on sufficient N. Twelve lines reacted to low N with very strongly reduced shoot growth (average 19.9 ± 4.2% of biomass at sufficient N, ranging from 13.3 to 24.8%; Table 3), and significant differences between shoot biomass at low and sufficient nitrogen. The notable exception is Col-0 whose growth was only reduced to 77.2% (Table 3), leading to no significant differences in fresh shoot weight.

Table 3 Differences in plant growth normalised to performance at sufficient N

The analysis of biomass was accompanied by a deep phenotyping of root system architecture. Only lines with consistent shoot fresh weight were analysed in detail in these experiments. Col-0, Lu, NG087, NG064 were defined as ‘good’ NUE lines, and Ts-1, Fei-0, El-0 and KT130 as ‘bad’ NUE lines, based on the shoot weight at low N (Table 2). Six plants were grown per plate, and scanned after 14 days in a growth cabinet. Root morphology, i.e. number of roots, main root length, total lateral root length and lateral root density were determined using WinRHIZO Pro and subjected to the same analyses as the biomass traits described above. Results are summarised in Table 4. ANOVA (Table S5) revealed significant effects (P < 0.001) in all root system architecture traits for line (genotype) and the interaction term, indicating that lines reacted differently to changes in the nitrogen supply. The nitrogen treatment effect was significant (P < 0.001) for main root length, number of roots and total lateral root length, but not lateral root density (P = 0.06). The genotype explained 22.2, 8.3, 15.2 and 20.9% of the variation, respectively, the treatment 16.8, 11.7, 13.8 and 0.3%, and the interaction 5.6, 4.2, 4.0 and 8.8%. Overall, the percentage variation explained by these factors is lower than for the biomass traits. As for biomass, root architecture traits are generally reduced on low N compared to sufficient N. Main and total lateral root length are reduced by 60% and 52% on low N compared to sufficient N, respectively. KB006 has the strongest reduction in main and lateral root length and number of roots, but the strongest induction in lateral root density (Table 4). In contrast to root length and number, the lateral root density, i.e. the number of lateral roots per cm main root, is significantly increased in C24, El-0, and KB006, but decreased in SG063 and Ts-1.

Table 4 Root system architecture traits at low and sufficient N

Differentially expressed genes

Six lines with contrasting NUE selected in the deep phenotyping screen were used for expression analysis. NG087, NG064 and KB006 were defined as ‘good’ NUE lines, and Fei-0, El-0 and SG063 as ‘bad’ NUE lines, based on the shoot weight at low N. Three biological replicates (30–50 plants each) were grown on agar plates. Shoot and root material was harvested separately after 21 days. Signal and fold change (logFC) data are available in Table S6.

As a first step, we looked at the expression of genes known to be involved in nitrogen metabolism to verify the experimental set-up and detect possible differences between lines (Table S6). On low N compared to sufficient N, the expression of AtNRT2.4 (AT5G60770), encoding a high-affinity nitrate transporter, is increased in roots in all lines; while the expression of AtNRT2.5 (AT1G12940), also encoding a high-affinity nitrate transporter, is significantly and substantially increased in the shoot. Its expression level differs between the lines grown on low N, with a fold change of − 1.25 between the shoots of El-0 and NG087, and a twofold difference in the roots of El-0, Fei-0, SG063 vs. KB006, NG087. The expression of high-affinity nitrate transporter AtNRT2.6 (AT3G45060) is decreased in the root in low N. The expression of AtCLCa (AT5G40890) is significantly down-regulated in both shoot and root. The transcription factors (TF) LBD37/38/39 (AT5G67420, AT3G49940, AT4G37540) are significantly down-regulated in limited N in shoots of all lines, and LBD38/39 are also significantly lower expressed in roots of all lines (Table S6). The bZIP transcription factor gene TGA1 (AT5G65210), but not TGA4 (AT5G10030), is up-regulated in the shoots of all lines but NG087, and down-regulated in the roots of El-0 only. TGA4 is only up-regulated in the shoot of KB006.

In total, 12428 loci displayed at least one significant logFC. Of these loci, 18.6% were detected in both shoot and root with the majority of genes (87.4%) behaving similarly in shoot and root, and only 12.6% showing contrasting behaviour in the two organs. Overall, 5374 genes showed a higher expression (FC > 1, P < 0.01) in shoots on low nitrogen, 4748 genes a lower expression. The number of differentially expressed genes was lower in the roots, with 2428 genes up-regulated on low nitrogen, and 1950 genes down-regulated in low nitrogen. The overlap between these groups is low (Fig. 2). Overlaps between ‘up’ and ‘down’ within an organ may indicate genotypic differences, i.e. the lines may react differently to the nitrogen regimes. In all lines, more genes are up-regulated on low nitrogen in the shoot; while in the root, in all lines but KB006, more genes are down-regulated on low nitrogen. No significant GO term enrichments were found in the different classes.

Fig. 2
figure 2

VENN diagrams of gene expression on low N using two accessions and four mixed RILs selected for contrasting and stable growth during deep phenotyping. Depicted are four conditions: expression increased on low N in shoot a or root c, and expression decreased on low N in shoot b or root d

Linking phenotype and gene expression

A Principal Component Analysis (PCA) of the Affymetrix signal data from 12428 genes with significant expression changes between low and sufficient N was performed. The first principal component (PC1) accounted for 40% of the variation and separated shoot and root, while PC2 explained 14% of the variation and separated low and sufficient N (Fig. 3a). PC4 accounts for 4.1% variance and partially separates accessions and mRILs, PC5 accounts for 3.2% and partially separates the lines (Fig. 3b), with the related mRILs NG064 and NG087, exhibiting ‘good’ NUE, clustering together as expected, as do the accessions El-0 and Fei-0 with ‘bad’ NUE. The unrelated mRILs KB006 (‘good’ NUE) and SG063 (‘bad’ NUE) are distinguished by PC5. We used the 30 highest and 30 lowest loadings of each PC (Table 5) and performed a GO term overrepresentation analysis (Table 6). As expected, genes associated with chloroplasts are significantly overrepresented in PC1. A total of 23 genes (38.3%) present in the top loadings are located in the chloroplast. Surprisingly, the only GO term associated with PC2 is phosphate ion homeostasis, represented by two genes (Table 6). Inspecting the loci contributing to PC2 in more detail, six genes linked to nitrogen metabolism or signalling are found in the top loadings (Table 5).

Fig. 3
figure 3

Principal Component Analysis (PCA) of differentially expressed genes. a Scatterplot of PC1 and PC2, explaining 40% and 14% of the variation, respectively, and separating the samples according to organ and nitrogen treatment. The colour of the data points indicates the organ (black = shoot; red = root); the shape indicates the nitrogen treatment (circle = low N, triangle = high N). b Scatterplot of PC4 and PC5, explaining 4.1% and 3.2% of the variation, respectively, which separate the lines, as indicated by the colour code. Good NUE lines are represented by a triangle, low NUE lines by a circle

Table 5 Top loadings of the principal components PC1, PC2, PC4 and PC5
Table 6 GO term enrichment analysis of Principal Component Analysis (PCA) top loadings

Pearson correlations between morphological traits of plants grown on low or sufficient N (Table 7) reveal different correlation patterns between the nitrogen treatments. On low N, root fresh weight is negatively correlated with shoot–root ratio and lateral root density, but positively with total fresh weight, while lateral root density is positively correlated with shoot–root ratio. On sufficient N, only positive significant correlations are detected, between total fresh weight and root fresh weight, between number of roots and main root length, and between number of roots and lateral root length. Shoot fresh weight, the criterion used for classification of the lines, does not show a significant correlation with any other trait on low N. We also determined Pearson correlations between transcript levels and morphological traits in shoot (Table S7) and in root (Table S8). For shoot, 13590 significant correlations were detected, ranging from 5428 (53.6% of tested genes) for shoot fresh weight to 83 (0.82%) for main root length (Table 8). For root, 4208 significant correlations were detected, ranging from 1303 (27.0%) for shoot fresh weight to 54 (1.1%) for shoot/root ratio (Table 8). No significant correlations were found for lateral root density in neither shoot nor root. No significant GO term enrichment was found.

Table 7 Pearson correlation between morphological traits
Table 8 Real-time qPCR analysis of biomarker candidate genes

Using Pavlidis Template matching in Mev4.9, transcript levels were clustered according to the pattern observed for shoot growth on low nitrogen. The respective transcript levels of 1068 and 436 genes expressed in shoot and root, respectively, corresponded to the growth pattern. Genes involved in translation are overrepresented in these sets (Fig. S1). Looking at genes significantly changed in expression in relation to their occurrence in lines with low (Fei-0, El-0) or high (NG087, NG064) shoot fresh weight on low N, 3187 genes are expressed in the shoot unique to the low-weight class vs. 986 genes in the high-weight class. The inverse ratio is observed in the root, with 457 genes unique to low shoot weight, and 1067 genes to high shoot weight. The expression of AT3G21670, encoding the NPF6.4/NRT1.4 low-affinity nitrate transporter, is lower in roots of lines with high shoot fresh weight. The transcription factor GATA21 (AT5G56860) shows higher expression in shoots of lines with high shoot weight, while one of its targets, GLU1 (AT5G04140), shows the opposite expression pattern. Similarly, the expression of the ammonium transporter gene AMT1;2 (AT1G64780) and the aspartate amino transferase gene ASP3 (AT5G11520) is lower in shoots of lines with high fresh weight.

Four of the potential biomarkers could also be analysed using real-time quantitative PCR in two independent experiments using six different Arabidopsis accessions with contrasting growth behaviour on low nitrate. Only shoot material from plants grown on limiting nitrogen was included in the analysis. The resulting ʌCT values (CT values normalised to the housekeeping gene AtEF1α) are listed in Table 8. The expression of AMT1;2, GATA21 and GLU1 shows highly significant differences between accessions. AMT1;2 and GATA21 differentiate good and bad NUE lines to a limited extent, as accession Ak-1 is misclassified in both cases. Only the expression of GLU1 completely follows the pattern predicted from the array data (the same is true for NPF6.4, but the differences between the accessions are not significant).


Validation of experimental setup using nitrogen metabolism marker genes

Several markers genes for nitrogen metabolism were analysed to verify the experimental set-up. The results demonstrate the expected overall nitrate response of gene expression, thus validating our experimental set-up. As expected, the expression of the high-affinity nitrate transporter genes AtNRT2.4 (AT5G60770) and AtNRT2.5 (AT1G12940) is increased under limiting nitrogen. A similar increase in transcript abundance upon exposure to N limitation for both genes has been reported by Lezhneva et al. (2014). AtNRT2.4 is induced in long-term starvation and has been linked to nitrate uptake in the roots at very low nitrate concentration (Kiba et al. 2012), which fits to our findings of a significant and substantial increase of AtNRT2.4 expression in the root in all lines after 21-day growth on low N. The expression in the shoot, while slightly increased, is not significantly changed. The opposite situation is found with AtNRT2.5, whose expression is significantly and substantially increased in the shoot. AtNRT2.5 is required to support the growth of nitrogen-starved adult plants by ensuring the efficient nitrate uptake in the root and efficient remobilisation of nitrogen in the leaves (Lezhneva et al. 2014). In contrast to the other high-affinity transporters, expression of AtNRT2.6 (AT3G45060) is decreased in the root in low N. AtNRT2.6 expression has been described to be induced by high nitrogen levels (Dechorgnat et al. 2012). Its over-expression failed to complement the nitrate uptake defect of an nrt2.1-nrt2.2 double mutant. AtCLCa (AT5G40890) encodes a nitrate/proton exchanger that transports excess nitrate into the vacuole for storage and maintenance of nitrate homeostasis in the cytosol (De Angeli et al. 2006), and consequently its expression is down-regulated on low N, consistent with our findings. Three members, LBD37/38/39 (AT5G67420, AT3G49940, AT4G37540), of the LATERAL ORGAN BOUNDARY DOMAIN transcription factor gene family are induced by nitrate and suppress N starvation responses such as anthocyanin biosynthesis and key genes required for NO3− uptake and assimilation (Rubin et al. 2009). As expected, their expression was down-regulated under our low N condition. The bZIP transcription factor gene TGA1 (AT5G65210), but not TGA4 (AT5G10030), is up-regulated in the shoots of all lines but NG087, and down-regulated in the roots of El-0 only. TGA4 is only up-regulated in the shoot of KB006. Expression of TGA1 and TGA4 was shown to be induced by nitrate in roots of Col-0, to regulate the expression of NRT2.1 (AT1G08090) and NRT2.2 (AT1G08100) and to affect lateral root formation in response to nitrate (Alvarez et al. 2014). We did detect expression differences between Arabidopsis accessions, demonstrating the diversity of the N stress response.

Phenotypic plasticity

The phenotypic analysis of two complementary populations of genetically diverse Arabidopsis lines highlighted the plasticity of plant growth in response to differing nitrate availability in the substrate. Plasticity of root morphology in response to abiotic factors has been described on the inter- and intra-species level (Forde and Lorenzo 2001; Gruber et al. 2013), and natural variation for foraging under nutrient-deficient conditions has been observed (North et al. 2009; Chardon et al. 2010; Ikram et al. 2012). However, these studies did not analyse natural variation of the growth response to long-term limiting nitrogen. The lack of correlation between growth in two complementary experimental setups, synthetic agar medium and soil, respectively, indicates that distinct processes contribute to NUE in the two different growth conditions. This could be due to the difference in plant age at the time of sampling (23 days in agar vs. 35 days in soil). Differential stress responses during development have been described for drought tolerance in barley (Szira et al. 2008).

The detailed characterisation of a selected set of lines provided clear evidence that different genotypes (lines) follow very different adjustment/survival strategies to cope with long-term nitrogen limitation. Both biomass and root architecture traits are generally reduced on low N compared to sufficient N. ANOVA revealed that the highest portion of the variation for the biomass traits shoot and root fresh weight and shoot–root ratio can be attributed to nitrogen treatment. Similar results were reported from a study of 22 Arabidopsis accessions grown hydroponically (Ikram et al. 2012). ‘Genotype’ explains almost as much variance as nitrogen treatment for shoot–root ratio (24.6%), but considerably less for shoot and root fresh weight (6.1 and 11.2%). This is in agreement with reports describing the shoot–root ratio as the trait most suitable to analyse natural variation in plant nitrogen response (Scheible et al. 1997; Ågren and Franklin 2003). Overall, the percentage phenotypic variation explained by the root architecture factors is lower than for the biomass traits. The experimental error is expected to be higher in the more complex measurements of the root architecture. Interestingly, the genotypic component is more important than the treatment, in relative terms, for the root architecture traits. To be useful as biomarkers, morphological traits which are influenced by both genotype and treatment seem the most promising, but overall, the respective contributions appear rather low for reliable indicators. In line with this observation, no clear pattern in root architecture can be distinguished between lines with good NUE (Lu, NG087, NG064; shoot fresh weight ≥ upper quartile) and bad NUE (Ts-1, Fei-0, El-0; shoot fresh weight ≤ lower quartile).

Many plant species show increased root growth at the expense of shoot growth under N limiting conditions, thus increasing their chances to reach additional sources of nitrogen to sustain their growth (grapevine (Grechi et al. 2007); maize (Brouwer 1962); tobacco (Scheible et al. 1997); Arabidopsis (Ikram et al. 2012)). On the other hand, plants need carbon substrates from photosynthesis to be able to fix the available nitrogen, and therefore, they must balance both demands. Several studies identified the C/N balance in the shoot as a trigger for changes in the biomass allocation (Stitt 1999; Martin et al. 2002), while sucrose, as indicator of the nutritional status, acts as a signal for increased root growth (Gibson 2005). Our findings expose the bandwidth of adaptive reactions to limiting N conditions within the species Arabidopsis thaliana. Different lines, with different genetic backgrounds, follow different ways of achieving a balance between growth reduction due to N limitation and foraging to detect nitrogen sources. We observed several genotype-specific modes of action to achieve growth under N deficiency: modification of root system architecture, general root growth reduction, maintenance of root system, or use of nutrients primarily for shoot growth. Fei-0 builds an overall increased root system on low N, while KB006 produces a very dense root system with a short main root and many short lateral roots. These findings support the notion that lines with different genetic backgrounds adopt different adjustment strategies to low nitrogen (Chardon et al. 2010; Ikram et al. 2012; De Pessemier et al. 2013). From an agricultural point of view, this means that there is not one optimal way to increase growth and yield under limiting conditions, which broadens the basis for selection, but also render the selection of optimal lines more difficult.

This diversity of reaction modes causes on the one hand difficulties to identify certain morphological and molecular features of root systems for good nitrogen use efficiency. On the other hand, it opens access to a multitude of genetic factors which can be used as tools to modulate and adjust root systems to a wide range of environmental (soil) conditions. It will, thus, be of great interest to identify the molecular genetic determinants of various responses to low N supply. The large diversity of response modes observed in the Arabidopsis population may preclude the use of the frequently applied Genome Wide Association Study (GWAS) approach for identification of causal genetic loci and may rather ask for the use of the more classical linkage mapping approach using biparental populations, e.g. recombinant inbred line (RIL) or introgression line (IL) populations, each of which would segregate for a specific set of NUE-related traits.

Linking phenotype and gene expression

Several avenues were explored to link phenotype and gene expression. Using a principal component analysis, we subjected the 30 highest and 30 lowest loadings of each PC to a GO term overrepresentation analysis. As expected, genes associated with chloroplasts are significantly overrepresented in PC1, which separated shoot and root samples. Surprisingly, the only significant GO term associated with PC2, which separated low and sufficient N samples, is phosphate ion homeostasis, represented by two genes. However, six genes linked to nitrogen metabolism or signalling are found in the top loadings of PC2. These include genes encoding the nitrate transporter NRT2.5 (At1g12940), the glutamine synthase GLN1;4 (At5g16570), the chloride channel CLC-A (At5g40890), the nitrate-responsive transcriptional regulator LBD39 (At4g37540), the NIN-like transcription factor NLP8 (At2g43500), and GS1 (At5g57440) encoding a HAD superfamily protein. SKU5 (At4g12420) encodes a cupredoxin superfamily protein involved in root tip growth, and differences in gene expression might contribute to different root architecture.

We also determined Pearson correlations between transcript levels and morphological traits in shoot and in root. Significant correlations between genes expressed in the root with shoot fresh weight indicate that gene expression in the root is associated with the growth above-ground, and that crosstalk between the organs is likely bidirectional. Among the significantly correlated genes are several encoding nitrate transporters (NRT1.1, NRT1.6, NRT2.2, NRT2.5), several genes related to RNA regulation and stress or nutrient signalling, and the RDR2 gene encoding the RNA-dependent RNA polymerase 2, which is involved in the generation of endogenous siRNAs (Jia et al. 2009). Post-transcriptional and epigenetic regulation mechanisms have been proposed for several enzymes related to nitrate transport or nitrogen metabolism. The switch between modes of the dual affinity nitrate transporter NPF6.3/NRT1.1 (AT1G12110) is achieved by phosphorylation (Liu and Tsay 2003). Protein structure analyses suggest that the increase in transport rate is the result of altering the efficiency in the formation (Parker and Newstead 2014). Nitrate-responsive miRNAs have been identified in Arabidopsis (Vidal et al. 2010; Zhao et al. 2011) and several crop species (Zhao et al. 2012; Paul et al. 2015).

In a complementary approach, transcript levels were clustered according to the pattern observed for shoot growth on low nitrogen. A GO term analysis revealed that genes involved in translation/organonitrogen compound biosynthetic processes are overrepresented in these sets. This is consistent with a remodelling of the metabolism during severe nitrogen stress, such as down-regulation of genes involved in photosynthesis and of ribosomal genes involved in protein biosynthesis; while genes encoding enzymes involved in starch biosynthesis and protein degradation are up-regulated (Bi et al. 2007). These events correspond to nitrogen remobilization from senescing tissues (Havé et al. 2017) also described for Arabidopsis under low N conditions (Diaz et al. 2008). Protein degradation serves as source for amino acids which are transported into active tissues via the phloem.

Looking at genes showing significantly changed expression between lines with low or high shoot fresh weight on low N (low or good NUE), five possible biomarkers emerged from analysis of the array data. The expression of the ammonium transporter gene AMT1;2 (AT1G64780) and the aspartate amino transferase gene ASP3 (AT5G11520) is lower in shoots of lines with good NUE. In an independent experiment, qPCR data of AMT1;2 expression correctly classified four out of six Arabidopsis accessions as low or good NUE lines. AMT1 family members encode high-affinity ammonium transporters and have also been identified in maize (Gu et al. 2013), poplar (Wu et al. 2015), and rice (Yang et al. 2015). In tomato, Arabidopsis, and rapeseed, several AMT1 genes are differential regulated by nitrogen supply and time of day (Gazzarrini et al. 1999; von Wirén et al. 2000; Pearson et al. 2002). AMT1 family members have been linked to N remobilization from old leaves during senescence (Gregersen and Holm 2007; Havé et al. 2017), a similar role under N stress is conceivable. The expression of AT3G21670, encoding the NPF6.4/NRT1.3 low-affinity nitrate transporter, is lower in roots of lines with high shoot fresh weight. Up-regulation of NPF6.4 in shoots of Arabidopsis grown under limiting nitrogen supply has been reported (Okamoto et al. 2003; Bi et al. 2007). Similarly, the wheat orthologue TaNPF6.6 was significantly up-regulated by nitrate starvation (Buchner and Hawkesford 2014). The transcription factor GNC (AT5G56860), a nitrate-inducible, carbon metabolism-involved member of the GATA factor family of zinc finger transcription factors, shows higher expression in shoots of lines with high shoot weight. It has been described to modulate chlorophyll biosynthesis and glutamate synthase (GLU1/GOGAT) expression (Hudson et al. 2011) and to be induced by both nitrate and cytokinin (Bi et al. 2005; Kiba et al. 2005). In the validation experiment, four out of six accessions were correctly classified. The glutamate synthase gene GLU1 (AT5G04140), a target of GATA21, shows lower expression in shoots of lines with high NUE. The array data could be validated using qPCR data, which also associated lower GLU1 expression with better growth on low N. However, in a global expression study using only the Arabidopsis reference accession Col-0, no fluctuation in the response to differing N supply was detected (Bi et al. 2007). GLU1/GOGAT has been identified in several plant species, including maize, tomato, rice and wheat. GOGAT was identified as a major effector of NUE in wheat (Quraishi et al. 2011) using a ‘syntenic’ metaQTL approach. A significant positive correlation was found between global expression of the GoGAT gene and the Nitrogen Nutrition Index score of leaf samples in two wheat cultivars, indicating that an increased expression of GOGAT is associated with good NUE in wheat, contrary to our findings in Arabidopsis. This difference could be due to the fact that our definition of good NUE is solely based on good shoot growth on low N.

The global expression analysis performed here showed the expected patterns of N and C metabolism-associated genes and pointed to regulatory factors (transcription and epigenetic factors) as important players in the adaption to limited nitrogen. One gene, GLU1, emerged as a possible biomarker for NUE in Arabidopsis. These findings need to be validated in different genetic backgrounds of Arabidopsis (and other plant species) with a sequence-independent method such as RNA-Seq and should be deepened towards identification of genes specifically involved in different modes of adjustment to low N conditions.

Author contribution statement

TA, RCM, HS and AS designed the research. CG performed the sample collection and RNA extraction, and analysed data. RCM, CN, KWF and MM conducted the deep phenotyping and analysed data. AB processed the gene expression data. RCM performed statistical analyses and drafted the manuscript. TA and AS revised and edited the text. All authors have read and approved the final manuscript.