Genetic variation underlying differential ammonium and nitrate responses in Arabidopsis thaliana

Abstract

Nitrogen is an essential element required for plant growth and productivity. Understanding the mechanisms and natural genetic variation underlying nitrogen use in plants will facilitate the engineering of plant nitrogen use to maximize crop productivity while minimizing environmental costs. To understand the scope of natural variation that may influence nitrogen use, we grew 1,135 Arabidopsis thaliana natural genotypes on two nitrogen sources, nitrate and ammonium, and measured both developmental and defense metabolite traits. By using different environments and focusing on multiple traits, we identified a wide array of different nitrogen responses. These responses are associated with numerous genes, most of which were not previously associated with nitrogen responses. Only a small portion of these genes appear to be shared between environments or traits, while most are predominantly specific to a developmental or defense trait under a specific nitrogen source. Finally, by using a large population, we were able to identify unique nitrogen responses, such as preferring ammonium or nitrate, which appear to be generated by combinations of loci rather than a few large-effect loci. This suggests that it may be possible to obtain novel phenotypes in complex nitrogen responses by manipulating sets of genes with small effects rather than solely focusing on large-effect single gene manipulations.

Analysis of developmental and metabolic responses using a large collection of natural Arabidopsis genotypes revealed numerous genes involved in the plant nitrogen response.

In a Nutshell.

Background: Nitrogen is an essential element for plant growth and development. Nitrogen availability and form affects plant productivity and fitness, especially in changing environments. Plant roots absorb nitrogen from the soil in different forms. Each plant has an ideal nitrogen source and concentration that depends on the plant species, genotype, developmental stage, and the surrounding environment. Shifts in the ideal nitrogen condition will be quickly reflected in root development but will also affect multiple other processes in the plant, including nutrient uptake, cell metabolic homeostasis, signaling and hormone pathways, defense responses, and responses to elevated carbon dioxide.

Question: We wanted to understand how plant genetic variation shapes the responses to different nitrogen conditions, and identify the mechanisms involved in those responses. More specifically, we wanted to understand if different environments and different sets of traits identify similar or disparate mechanisms/genetic variation.

Findings: To answer this question, we grew a large collection of Arabidopsis thaliana natural genotypes on two nitrogen sources, and measured a set of developmental traits, and a set of defense metabolite traits. We identified a wide array of different nitrogen responses that are controlled by variation in a large number of genes. Different combinations of these small-effect genes appear to be responsible for the observed responses. A small portion of these genes are shared between nitrogen conditions or traits (developmental versus metabolite traits), while most of the detected genes are specific to a developmental or defense trait under each unique nitrogen condition.

Next steps: Our results, specifically the list of genes that we created, will facilitate engineering or breeding plants that use nitrogen more efficiently under different environmental conditions. Future validation of the detected genes will take our results to the next step.

Introduction

Organisms obtain nitrogen, an element required for growth and productivity, from different organic and inorganic compounds. The organic compounds include amino acids and nucleotides, which are the preferred nitrogen forms of microorganisms, and inorganic forms including nitrite (${\text{NO}}_2^{–}$), nitrate (${\text{NO}}_3^{–}$), and ammonium (${\text{NH}}_4^{+}$). Vascular plants acquire most of their nitrogen in the form of nitrate or ammonium, which their roots absorb from the soil (Bloom, 2015). Within plants, these compounds are converted into different organic nitrogen compounds through several biochemical pathways. An alternative way that plants acquire nitrogen is through symbiotic relationships with bacteria. The bacteria fix nitrogen from the atmosphere (N₂) into ammonia (NH₃), a form that plants assimilate into organic nitrogen compounds, and in return the plants provide a habitat for the bacteria as well as fixed carbon (Mus et al., 2016). An individual plant’s optimal nitrogen source and concentration range depend on multiple parameters, including plant species, genetics, developmental stage, and the surrounding environment (Britto and Kronzucker, 2002; Fuertes-Mendizábal et al., 2013; Waidmann et al., 2020). The excess application of nitrogen not only affects plant growth, but it also leads to environmental degradation, such as groundwater contamination, eutrophication of freshwater, soil salinization, photochemical smog, and greenhouse gas production (Zhang et al., 2015). A better understanding of plant nitrogen acquisition (uptake, assimilation, and their regulation) will facilitate efforts to adjust nitrogen form and concentration to maximize crop productivity yet minimize environmental damage.

Because nitrate is the predominant nitrogen form in temperate agricultural soils, most genetic and physiological studies on plant nitrogen relations performed to date have focused on nitrate. This focus is based on the finding that most plants prefer nitrate as a nitrogen source because ammonium can become toxic if it accumulates in plant cells (Li et al., 2014; Esteban et al., 2016; Liu and von Wirén, 2017; Marino and Moran, 2019). Another potential reason for plant dependence on nitrate is that plants have fewer competitors for this nitrogen form, whereas they must compete with microorganisms for soil ammonium that microorganisms use as both a nitrogen and energy source (Matson et al., 1998). Nonetheless, the preference for nitrate is not ubiquitous and can shift due to genetic variation between and within plants and is sensitive to environmental variation, such as changes in carbon dioxide (CO₂) levels (Britto and Kronzucker, 2002; Bloom et al., 2012; Menz et al., 2018; Marino and Moran, 2019). This suggests that the balance between nitrate versus ammonium usage is not fixed within a plant species. Identifying the genetic mechanisms that influence this balance would allow plants to be reprogrammed by breeding or synthetic biology to achieve a better match with available nitrogen sources.

To optimize nitrogen acquisition, plants modify root growth and branching, rapidly responding to changes in nitrogen source or concentration. Any shift in soil nitrogen form or availability will typically lead to alterations in primary and lateral root growth and development that maximize the ability to search for more optimal soil nitrogen conditions (Drew, 1975; Zhang and Forde, 1998; Zhang et al., 1999; Epstein and Bloom, 2005; Tian et al., 2009; Lima et al., 2010; Rogato et al., 2010; Fuertes-Mendizábal et al., 2013; Gifford et al., 2013; Gruber et al., 2013; Liu et al., 2013; Waidmann et al., 2020). Although the mechanisms behind these responses are becoming understood, how many of these mechanisms are species specific and how many are shared among plants is still unknown.

Shifts in nitrogen source and concentration, in addition to altering root development, can lead to a global reprogramming of the plant’s nutrient uptake, cell metabolic homeostasis, signaling and phytohormone pathways, defense responses, and responses to elevated CO₂ levels (Bloom et al., 2010; Marino et al., 2016; Marino and Moran, 2019). One defense response that is highly dependent on nitrogen availability is the production of glucosinolates (GSLs), a class of specialized defense metabolites in Arabidopsis thaliana and other plants in the order Brassicales. Breakdown products of GSLs are toxic to herbivores and pathogens and thus play a central role in plant defense against attackers (Beekwilder et al., 2008; Hansen et al., 2008). GSLs display extensive variation across Arabidopsis genotypes, and their composition and accumulation depend on genetics, developmental stage, and external cues (Rodman, 1980; Daxenbichler et al., 1991; Kliebenstein et al., 2001a, 2001b; Wright et al., 2002; Benderoth et al., 2006; Halkier and Gershenzon, 2006; Bakker et al., 2008; Chan et al., 2010; Sønderby et al., 2010; Brachi et al., 2015; Kerwin et al., 2015; Katz et al., 2021). Because GSLs are derived from amino acids and contain nitrogen as part of their basic structure, GSL accumulation is positively correlated with nitrogen supply (Yan and Chen, 2007; Omirou et al., 2009; He et al. 2014; Marino et al., 2016). Furthermore, one plant strategy for dealing with excess ammonium and avoiding toxicity is to accumulate and store more GSLs (La, 2013; Marino et al., 2016; Coleto et al., 2017; Marino and Moran, 2019).

Interestingly, nitrogen and GSLs are bidirectionally coordinated, as GSLs influence nitrogen signaling and/or responses. A metabolic genome-wide association study (GWAS) measuring natural variation in amino acid content within Arabidopsis seeds found that two key GSL biosynthetic loci were causally associated with the level of free glutamine in the seed (Slaten et al., 2020). One possible explanation for this GSL-to-nitrogen connection comes from the observation that specific GSLs can function as regulators of growth and development through interactions with different mechanisms, including the target of rapamycin (TOR) complex, which integrates nitrogen availability into complex signaling networks (Katz et al., 2015; Malinovsky et al., 2017; Salehin et al., 2019; Katz et al., 2020). Similarly, MYB29, a transcription factor that regulates the expression of GSL biosynthesis genes, can influence root plasticity in response to changes in nitrate levels (Gaudinier et al., 2018). While these studies highlight the involvement of GSLs in nitrogen responses, the mechanisms behind the interplay between nitrogen and defense metabolism are largely unknown.

To develop a deeper understanding of natural variation in A. thaliana responses to different nitrogen sources and concentrations, we utilized a large collection of accessions to survey for novel nitrogen responses/behaviors and to potentially enhance the power to identify causal loci. We phenotyped 1135 A. thaliana natural genotypes (Kawakatsu et al., 2016; The 1001 Genomes Consortium, 2016). While natural variation in nitrogen responses in Arabidopsis has received some attention (Chardon et al., 2010; Meyer et al., 2019; Cheng et al., 2021), most work on nitrogen responses in Arabidopsis has focused on the reference Columbia-0 (Col-0) genotype (Menz et al., 2018), and the effectiveness and necessity of using populations with natural variation when querying for disparate responses, that is, genotypes that prefer ammonium, has been debated (Waddington, 1953; Schlichting and Pigliucci, 1993; Pigliucci and Murren, 2003; Pigliucci et al., 2006). One line of thinking suggests that Col-0 mutants can mimic the entire range of natural variation in phenotypes available within the species, as demonstrated in a study that identified genes involved in nitrogen-mediated root architecture plasticity in 69 Arabidopsis genotypes. This work suggested that the mechanisms controlling roots plasticity in Col-0 are not different from those of other natural genotypes and that Col-0 mutants can be used to mimic the natural variation in a limited population size (Rosas et al., 2013). A recent genomic analysis of pangenomic variation revealed extensive presence/absence variation in genes, including genes encoding enzymes and regulatory genes, suggesting the likelihood that other accessions have mechanisms not present in the Col-0 reference (Menz et al., 2018).

To address these questions, we phenotyped the growth responses of seedlings grown in the presence of nitrate or ammonium, the preferred nitrogen source of most plants, as the sole nitrogen source to determine if the genetics of natural variation to the two sources are similar or different. We measured the responses (phenotypic diversity) of the genotypes using both developmental phenotyping of root architecture and metabolic phenotyping of the defense metabolites GSLs in seedlings. This allowed us to assess if the two trait classes are associated with similar or disparate genetic mechanisms and test if the extensive genetic variation in GSLs may contribute to nitrogen responses. This analysis identified many genes involved in the nitrogen response that are predominantly specific to a trait class under a specific nitrogen source and in different genetic backgrounds. Having such an extensive collection of genotypes allowed us to identify unexpected behaviors in a polygenic trait like Arabidopsis nitrogen responses. These traits are likely caused by combinations of diverse alleles and as such would be difficult or impossible to observe in smaller populations or a single reference genotype. The use of this vast collection of natural genotypes to study both developmental and metabolic responses expands our understanding of the processes and mechanisms involved in nitrogen responses.

Results

Optimization of the phenotyping platform

To optimize our ability to measure and assess phenotypic diversity across Arabidopsis genotypes to variation in nitrogen levels, both source and concentration, we quantified seedling growth on a range of nitrate and ammonium concentrations. To maximize the potential range of measurable traits across the Arabidopsis genotypes, we identified matching concentrations of the two nitrogen sources that elicited substantial trait variation without constraining growth, for example, neither an oversupply nor an undersupply. We grew the reference Arabidopsis Col-0 on media with different concentrations of either nitrate (${\text{NO}}_3^{–}$ as KNO₃) or ammonium (${\text{NH}}_4^{+}$ as NH₄HCO₃) as the sole nitrogen source. After 12 days, we measured five developmental traits linked to differential nitrogen responses, including leaf area and root phenotypes (a full list is provided in Supplemental Figure S1 and Supplemental Data Sets 1 and 2). As previously reported (Bloom et al., 2012), Arabidopsis Col-0 seedlings at ambient CO₂ grew better with nitrate as the nitrogen source than ammonium at most concentrations, with higher average weight, larger leaf area, and longer primary roots (Supplemental Figure S1). On ammonium, Arabidopsis Col-0 seedlings performed well from 0.1 mM to 1 mM, but the extreme ammonium concentrations (0.05 mM and 5 mM) prevented seed germination or growth (Supplemental Figure S1). Although excess nitrate can affect plant viability, the seedlings were still viable on 10 mM nitrate, but since they did not germinate on this ammonium concentration, we excluded this concentration from the analysis and present concentrations ranging from 0.05 mM to 5 mM. Therefore, 0.1 mM and 1 mM, the widest pair of concentrations that allowed growth under both nitrogen sources, were chosen as the working concentrations for the experiments to assess genetic variation in Arabidopsis.

To test how genetic variation affects some of the nitrogen responses, we grew nine different natural genotypes on the same nitrogen conditions as above and measured their growth (Supplemental Data Set 3). Some of the measured traits, especially those involving lateral roots, were significantly affected by the interaction between the plant’s genotype and the nitrogen condition (Supplemental Data Set 4 and Supplemental Figure 2). This indicates that different genotypes respond to nitrogen conditions in different ways, as shown before (Chardon et al., 2010; Cheng et al., 2021). A wide range of nitrogen behaviors was observed in these nine accessions across the different nitrogen sources and concentrations. For either nitrogen source, at a concentration below 0.1 mM or above 1 mM, the phenotypic variation in the accessions diminished significantly, supporting the use of 0.1 mM and 1 mM for a broader survey. However, if resources were not limiting, the inclusion of 0.5 mM would have been informative.

Diversity of the responses of different Arabidopsis genotypes to ammonium and nitrate measured using developmental and biochemical traits

To measure natural genetic variation in the responses of Arabidopsis to nitrate and ammonium, we assessed developmental and defense metabolite traits for a population of 1,135 sequenced natural genotypes collected from geographical locations around the world (The 1001 Genomes Consortium, 2016). We included developmental and defense metabolite traits to test if the genetic variation in nitrogen responses had global effects on the Arabidopsis genotypes or if each trait is associated with different responses and genes.

Each genotype was sown in triplicate on four nitrogen conditions: nitrate as the sole nitrogen source at a concentration of 0.1 mM or 1 mM, or ammonium as the sole nitrogen source at a concentration of 0.1 mM or 1 mM (for more details, see Materials and Methods). Twelve days after planting, the developmental and defense metabolite traits of each seedling were measured (for more details, see Materials and Methods; Supplemental Data Sets 5–7).

We used a linear model to parse the influence of genotype (genetic variation amongst the genotypes that is independent of nitrogen), environment (trait variation solely influenced by the nitrogen source, and nitrogen source by concentration), and the interaction of genotype and environment (trait variation where genotype variation alters the response to nitrogen variation) on the variation in developmental traits (Figure 1; Supplemental Data Sets 8 and 9). The linear model also included terms for random effects of experiment and culture plate (for more details, see Materials and Methods). Across all genotypes and nitrogen conditions, genotype (accession) influenced the majority of explained developmental trait variation. Nitrogen source and the interaction of nitrogen source and concentration significantly altered developmental traits, but these effects were secondary to the interaction of the nitrogen terms (source or concentration) with genotype (Figure 1). This indicates that for developmental traits such as lateral root formation, primary root length, and leaf area, the difference between ammonium and nitrate at these concentrations is less influential than the interaction between nitrogen source or concentration and the genotype.

Figure 1.

Broad-sense heritability. The percentage of variance explained by each variable was estimated for each trait based on a linear model followed by analysis of variance with the indicated effects upon the measured trait (Acc = accession, conc = nitrogen concentration, Exp*Plate = experiment block, Ind = individual).

Using the same linear model, we estimated the level of variation due to each of the indicated parameters for each defense metabolite trait. For all individual defense metabolite traits, the variation was determined equally by genotype and the interaction of nitrogen and genotype (Figure 1). For some of the GSL summation traits (e.g. the C3 versus C4 ratio = the number of carbons in the GSL backbone, alkenyl ratio, and glucosinolate hydroxylase (GSOH) activity), genotype explained most of the variation, while the interaction of nitrogen and genotype explained a small portion of the variation. This high genotypic variance for these three traits was expected, as GSL structure is determined by structural variation at three major effect loci encoding the enzymes responsible for these traits (methylthioalkylmalate (MAM) Alkenyl Hydroxypropyl (AOP); and GSOH) (Brachi et al., 2015; Katz et al., 2021). These results indicate that diverse nitrogen responses occurred across genotypes for both the defense metabolite and developmental traits.

Genotype and nitrogen interactions create diverse changes in shoot and root development

The linear models uncovered a high level of diversity in the responses of different genotypes to the four nitrogen conditions. To visualize the potential range of responses across the genotypes, we plotted for the distribution of each trait using each genotype’s adjusted mean phenotypes for each nitrogen condition (Figure 2, A–D; Supplemental Figure S3).

Figure 2.

Nitrogen conditions affect different traits across natural accessions. Phenotypes of Arabidopsis accessions grown under four different nitrogen conditions (nitrate 0.1 mM, nitrate 1 mM, ammonium 0.1 mM, ammonium 1 mM). A, Leaf area. B, Primary root length. C, Lateral root length. D, Number of lateral roots. E, Aliphatic GSL contents. F, Indolic GSL contents. Gray lines between each pair of nitrogen concentrations connect the means of each individual accession grown under each nitrogen concentration. Red lines between the violin plots connect the mean phenotype between two concentrations of the same nitrogen source. Aliphatic and indolic GSL contents were normalized by total root length for seedlings grown on ${\text{NO}}_3^{-}$ 1mM and ${\text{NH}}_4^{+}$ 1 mM (units are µmol/mm), and by total root length and leaf area for seedlings grown on ${\text{NO}}_3^{-}$ 0.1 mM and ${\text{NH}}_4^{+}$ 0.1 mM (units are µmol/mm + mm²).

On average, the population of Arabidopsis genotypes had significantly more leaf area, longer primary roots, and more lateral roots on nitrate than ammonium (Figure 2, A–D) (post-hoc Tukey HSD test shown in Supplemental Data Set 10). Furthermore, the population average for these traits increased with increasing nitrate concentration and decreased with increasing ammonium concentration (Figure 2, A–D). Thus, the average Arabidopsis genotype tended to present better growth on nitrate than ammonium and was constrained by nitrate availability, aligning with previous findings (Britto and Kronzucker, 2002; Fuertes-Mendizábal et al., 2013; Li et al., 2014; Jian et al., 2018; Marino and Moran, 2019).

As anticipated from the high fraction of variance attributed to the genotype × nitrogen interactions, the range in the responses of different genotypes to nitrogen source dwarfed the average population response. This includes genotypes that had very different behaviors from the average. We chose a few genotypes that showed some unique behaviors to demonstrate this point and plotted the results (Figure 3). For example, while most genotypes showed enhanced growth on 1 mM compared to 0.1-mM nitrate, several genotypes, including IP-Con-0, IP-Ses-0, and Ven-1, performed better on the lower nitrate concentration for most developmental traits (Figures 2 and 3). These genotypes also tended to perform better on higher ammonium than on higher nitrate concentrations. Thus, an extensive range of genetically dependent nitrogen responses was observed amongst these genotypes, including the potential to perform better with ammonium as a nitrogen source.

Figure 3.

Different Arabidopsis accessions show diverse phenotypes under different nitrogen conditions. Phenotypes of various Arabidopsis accessions grown under four different nitrogen conditions (nitrate 0.1 mM, nitrate 1 mM, ammonium 0.1 mM, ammonium 1 mM, with up to three replicates per accession per condition). A, Leaf area. B, Primary root length. C, Lateral root length. D, Number of lateral roots. E, Aliphatic GSL contents. F, Indolic GSL contents. Aliphatic and indolic GSL contents were normalized by total root length for seedlings grown on ${\text{NO}}_3^{-}$ 1 mM and ${\text{NH}}_4^{+}$ 1 mM (units are µmol/mm), and by total root length and leaf area for seedlings grown on ${\text{NO}}_3^{-}$ 0.1 mM and ${\text{NH}}_4^{+}$ 0.1 mM (units are µmol/mm + mm²). Significance was tested via two-way analysis of variances, which were done for each measured trait under each nitrogen condition against the accession (^*P < 0.05, ^**P < 0.01, and ^***P < 0.0001). Detailed statistics are shown in Supplemental Data Set 9. The lines provide visual continuity between the same accessions under different nitrogen concentrations. The red lines are the population averages.

To test if the origin of the plants might affect those responses, we created linear models to test the effects of different parameters related to the plant’s origin (genomic group, original location, and a variety of soil parameters in that area) on the different traits. The traits that were tested were different log ratios of the developmental and GSL traits, providing an indication of the responsiveness of the plants to the nitrogen conditions. Of all the tests, only one crossed the false discovery threshold (longitude by latitude, for leaf area); otherwise, there was no association in the data (Supplemental Data Set 11). Thus, this analysis allowed us to identify genotypes with unexpected responses compared to the Col-0 reference genotype.

Genotype and nitrogen interactions lead to diverse defense metabolite profiles

To evaluate whether defense metabolites exhibit similar diversity in response to nitrogen, and to compare metabolic and developmental responses, we assayed seedlings for the presence and amounts of individual GSLs from two families based on the amino acids from which they are derived: indolic GSLs derived from tryptophan, and aliphatic GSLs derived from methionine. As individual GSLs are derived from amino acids, the sums of GSL contents provide a sensitive indication of the availability of these amino acids within the plant and how they are affected by the nitrogen condition provided to the seedling.

We used these traits to conduct a similar analysis as above. For this analysis, we focused on two defense metabolite traits, the sum of all indolic GSLs (GSLs derived from tryptophan) and the sum of all aliphatic GSLs (GSLs derived from methionine) in each seedling, because these traits are quantifiable in all the genotypes (Figure 2, E and F; Supplemental Figure S4–S6). By contrast, most individual aliphatic GSLs have significant presence/absence variation, limiting the ability for direct comparison across the whole population. Plotting the genotype’s adjusted mean phenotypes showed that GSL contents were 10-fold higher under ammonium than under nitrate conditions, with no significant differences across the different nitrogen concentrations (Supplemental Data Set 10; Figure 2, E and F). This result agrees with previous reports, which suggest that this represents a strategy to avoid ammonium toxicity by diverting ammonium to GSL production (La, 2013; Marino et al., 2016; Coleto et al., 2017; Marino and Moran, 2019). This pattern differs from the findings for developmental traits, suggesting that GSLs have different nitrogen responses.

Defense metabolite traits, like developmental traits, displayed a large diversity of genotype by nitrogen interactions (Figure 2, E and F). This is supported by the linear model showing that the genotype by nitrogen interaction explained most of the trait variation. While the average genotype showed minimal responses to the different nitrogen concentrations for both nitrogen sources, certain genotypes exhibited enhanced nitrogen sensitivity. Both the IP-Ara-4 and IP-Mah-6 genotypes accumulated significantly higher amounts of both aliphatic and indolic GSLs when grown on the higher concentration of ammonium (Figure 3; Supplemental Data Set 12). By contrast, IP-Ara-4 seemed to decrease GSL (aliphatics and indolics) production when grown on higher nitrate concentrations. This further demonstrates the potential of this collection of genotypes for identifying nitrogen responses that differ from the average genotype or from the Col-0 reference genotype.

Partial correlation between developmental and metabolic traits

Because mechanistic links connecting nitrogen, development, and defense metabolism are possible, we tested for such links in the population. We calculated genetic correlations between the traits under the assumption that mechanistic links would show up as shared genetic causality and result in correlated trait variation. To minimize the influence of any major effect of nitrogen on the analysis, we examined correlations across each nitrogen condition, source, or concentration separately. All developmental traits were significantly positively correlated (based on Spearman correlation; Figure 4). By contrast, the defense metabolites showed more diversity in their correlations both within and across nitrogen conditions (Figure 4). For example, the indolic GSL contents were significantly positively correlated with all the developmental traits when the genotypes were grown on 0.1-mM nitrate or 1-mM ammonium. However, this correlation was largely nonexistent when the accessions were grown on 1-mM nitrate or 0.1-mM ammonium. By contrast, the aliphatic GSL contents showed a low correlation to developmental traits when the genotypes were grown on 1-mM nitrate or 0.1-mM ammonium but a higher correlation when the genotypes were grown on 0.1-mM nitrate or 1-mM ammonium. Specific metabolites showed negative correlations to each other. For example, Allyl was negatively correlated to OH-3-Butenyl across all nitrogen conditions, and 6MSO was negatively correlated to 3 across most conditions. These negative correlations are expected, as different alleles at the MAM locus in each accession create alternative GSL structures. We then analyzed the correlations between the different conditions. Although there were some (predominantly positive) correlations among nitrogen conditions and traits, the strongest correlations were those within the same nitrogen condition (Figure 4E). This suggests that there is some shared genetic causality across the developmental and defense metabolite traits but that it is highly conditional to the specific nitrogen condition.

Figure 4.

Different correlations between traits occur depending on nitrogen conditions. Correlation matrix between the traits under the different nitrogen conditions (A–D), or all conditions together (E). Colors represent the correlation values. White squares represent nonsignificant correlations (method: Spearman confidence level = 0.95).

The combination of developmental and defense metabolite traits also allowed us to determine if there is modularity amongst the genotypes whereby groups of genotypes showed similar responses to the nitrogen conditions. If the Arabidopsis genotypes are evolving to adapt to a limited range of nitrogen conditions that are distinct from each other, a modular structure might be expected to reflect this structure. The alternative option is that nitrogen responses exist on continuums with no grouping in the genotypes. To test these possibilities, and to check for some relationship between the traits and/or nitrogen conditions, we took the developmental and defense metabolite traits across all four nitrogen conditions and used principal component analysis (PCA) to assess the structure of the phenotypic variation. The PCA showed that there was no readily identifiable genotype grouping and that phenotypic variation in the genotypes exists on a multidimensional continuum. Furthermore, the PCA largely did not separate the values by nitrogen conditions or genomic group, supporting the importance of genotype × nitrogen interactions (Supplemental Figure S7). We then applied a t-distributed stochastic neighbor embedding (tSNE) analysis on the data but did not find any distinct grouping, and this analysis was not informative enough to draw any conclusions (Supplemental Figure S7). When conducting these analyses on each nitrogen condition separately, we obtained similar results.

Candidate genes associated with nitrogen responses

To identify loci underlying genetic variation in these nitrogen responses, we performed GWAS (with EMMAX algorithms) using the dense single-nucleotide polymorphism (SNP) data available for this genotype collection. We focused on four developmental traits (leaf area, primary root length, lateral root length, and number of lateral roots) and the two defense metabolite traits (aliphatic and indolic) under each nitrogen condition as traits for GWAS. We also included the log₁₀ ratio for each trait in each genotype across the two concentrations of the same nitrogen source as additional traits to approximate the nitrogen responsivity (Supplemental Figure S8). This resulted in 36 different GWAS: six traits across four nitrogen conditions and two ratios per the six traits (Supplemental Figure S9). Each trait yielded a diverse group of significant SNPs (we observed low overlap between SNPs from each GWAS, Supplemental Figures S10–S13). Amongst this data set, no large effect loci were detected, which suggests a poly- or oligo-genic architecture.

To filter for core genes and other trends, and to deal with false-positive GWAS hits, we combined each group of traits (development, aliphatic, and indolic) under each nitrogen source as a set (Supplemental Figures S10–S13, all future references of “set” refer to those groups of traits under each nitrogen source). We combined leaf and roots traits into one set of developmental traits because they showed a high correlation under all nitrogen conditions (Figure 4), and previous mechanistic work showed that these traits can have a coordinate genetic control; for example, the SNF1-related protein kinase (SnRK) gene family controls both root and shoot growth through the activation of abscisic acid (Fujii and Zhu, 2009) and TOR kinase (McCready et al., 2020). Furthermore, factor analysis conducted with all the traits suggested a clear separation between the developmental traits and the GSL traits, and together with the correlation results, supported grouping the developmental traits and GSL traits into two separate groups. For each of these sets, we identified the genes with the most consistent effect across all individual genome-wide associations within a set using the Multivariate Adaptive Shrinkage (MASH) method. The MASH method uses all the SNPs in the set and ranks them based on their shared and consistent effects across the set. This ranked set is then used to identify the SNPs with the most consistent effects on traits to shrink the number of SNPs/genes for further interrogation (for details, see Materials and Methods). For example, we took all the EMMAX output for GWAS for all developmental traits under all nitrate conditions (total of 12 trait GWAS datasets), combined them into a set, and used MASH to find the shared SNPs that influence development across nitrate conditions. This analysis still found 1,281 genes as being at the core of this set of genome-wide associations (Supplemental Data Sets 13 and 14).

Testing for overlap between the six gene lists from the six trait sets from the MASH highlighted candidate genes that were mutual to the nitrogen sources and/or trait types (Figure 5A). Overall, ∼30% of the genes in each set were found in at least one other set (Supplemental Table S1). This indicates that most of the genes in most of the sets are specific to one set and are not mutual across other nitrogen sources and/or traits (Figure 5; Supplemental Table S1). We then tested for overlap between genes identified using GWAS on the same trait type across the different nitrogen sources. This test for more shared causality within the developmental traits across nitrate and ammonium. A Fisher’s exact test to measure the overlap of genes commonly found from different sets showed significant enrichment for common genes influencing developmental traits across nitrate and ammonium (Figure 5). By contrast, we obtained no evidence for enrichment of either defense metabolite trait across the two nitrogen sources, suggesting that largely different genetic networks may influence defense metabolism across nitrogen sources. Next, we shifted the analysis to test for overlap between trait sets on a specific nitrogen source; for example, did ammonium have more commonality across aliphatic and indolic GLS? There was a significant enrichment in gene overlap in the indolic and aliphatic GLS trait sets under ammonium conditions, but almost none in the sets under nitrate conditions. This suggests that when seedlings are grown on ammonium, shared genes influence variation in both the indolic and aliphatic GLS pathways (Figure 5;Supplemental Data Set 14).

Figure 5.

Overlaps between genes behind the different nitrogen responses. A, Flow chart of the steps taken to detect and analyze genes associated with nitrogen responses. B, SNPs with mash score >0.98 were transformed to genes, and the overlap of these genes between the different sets was analyzed. N genes are a set of Arabidopsis genes known to be related to some nitrogen process. Fisher’s exact test was performed for each pair of sets (Supplemental Data Set 14). Total number of genes = 28,496.

We then assessed whether the candidate genes we identified were already known to be related to some nitrogen process (assimilation, signaling, transport, etc.) or whether inclusion of defense metabolites and ammonium identified new possible nitrogen-associated genes. Arabidopsis has a large catalog of genes with previously ascertained mechanistic links to nitrogen, which were typically derived based on root traits and nitrate. Using annotations from the literature, we created a set of 302 genes that are known in Arabidopsis to be related to some nitrogen process (Alvarez et al., 2014; Brooks et al., 2019; Castaings et al., 2009; Cheng et al., 2021; Gaudinier et al., 2018; Gifford et al., 2008; Konishi and Yanagisawa, 2013; Krouk et al., 2010; Medici et al., 2015; Obertello et al., 2010; Rubin et al., 2009; Varala et al., 2018; Vidal et al., 2014; Xu et al., 2016; Zhang and Forde, 1998) (For details, see Materials and Methods; Supplemental Data Sets 13 and 15). Using this list of known mechanistic genes, we checked for overlap with the six gene sets from the MASH analysis (Figure 5). Among the genes that appeared in more than one set that are known to be involved in some nitrogen response are genes involved in development (AT4G18390, teosinte branched 1), nitrate regulation (AT3G60320), and nitrate reduction (AT2G15620), and hormonal regulation (AT5G51190 is a member of the ERF/AP2 domain transcription family, and the B-Box gene BZS1 [AT4G39070] is regulated by brassinosteroids through brassinazole-resistant1) (Supplemental Data Set 16). Gene ontology (GO) analysis of the full list of genes that appeared in three or more sets did not yield any enrichment (Supplemental Data Set 16).

A test of the scope of overlap between the six sets and the known nitrogen genes showed that <1% of the genes in each trait set overlapped with the set of known nitrogen-related genes, regardless of whether nitrate or ammonium was the nitrogen source. Fisher’s exact test indicated that this overlap was not significant, indicating that known nitrogen-associated genes were not enriched (Supplemental Data Set 14). Therefore, most of the candidate genes were not previously linked to nitrogen responses in Arabidopsis.

To obtain insights about the biological processes in which the associated candidate genes are involved, we conducted GO analysis (Figure 5A). Analysis of each MASH-derived gene list for the six trait sets showed that some high-level GO terms were enriched, such as metabolic processes and defense responses. However, these were only higher-level GO terms with no especially informative connections (Supplemental Data Set 17). To test for a more limited gene set common to a few trait sets, we used genes that appeared in more than one set. Using both the developmental and metabolic trait sets, the biological processes that were enriched based on these genes include categories involving both GSL and general nitrogen metabolism and categories involving cell processes, such as regulation of gene expression and transcription, and regulation of metabolic and biosynthetic processes (Supplemental Data Set 16).

Candidate nitrogen-related genes are co-expressed

Previous studies showed that identifying co-expression modules of candidate genes identified by GWAS can identify genes that are more likely to be casually connected to the traits of interest (Chan et al., 2011; Wisecaver et al., 2017). To test if co-expression filtering of the candidate genes would identify more mechanistically insightful gene sets, as previously suggested, we developed co-expression gene modules using expression data from seedlings of all the genotypes (Supplemental Data Set 18) (Kawakatsu et al., 2016; Wisecaver et al., 2017). Previous studies showed that co-expression modules do not have to be derived from transcriptomic experiments designed around the specific conditions being tested and that any transcriptomic experiment can provide highly informative modules. This extends to the observation that informative co-expression modules can be readily obtained from stochastic interindividual variation within a single genotype (Liu et al., 2021). This analysis resulted in 2,864 modules, with a median of seven genes in a co-expression module (ranging from 3 to 257 genes). Of these co-expression modules, 436 had no GWAS candidates.

Using these co-expression modules, we tested if our GWAS candidates clustered into specific co-expression modules. We assigned each gene in each set to a co-expression module(s), thereby creating a list of modules for each set based on the genes in each set. Candidate genes not assigned to a co-expression module were discarded. By filtering for likely causal genes using these co-expression modules, we found more overlap across gene modules of trait sets than previously found using the original gene lists (Supplemental Data Set 19). Whereas analysis using the simple gene lists found little overlap between traits sets, the co-expression modules were highly shared among trait sets: 90% of the identified co-expression modules were associated with at least two different trait sets, and only 10% of the modules were specific to one trait set (Supplemental Table S1). For example, one co-expression module (module 104) was found to be associated with candidate genes from all trait sets, suggesting that this module of 119 genes may play a common role in nitrogen responses across development and defense. These results are in agreement with the results of a previous study that compared lists of genes regulated in different cell types and found that although the lists contained different genes, they were regulated by common modules (Walker et al., 2017).

To learn about the biological processes involving the modules that were associated with at least five of the trait sets (24 modules; Supplemental Data Set 18), we subjected the genes in these modules to GO enrichment analysis. The co-expression modules associated with most nitrogen traits were enriched for genes involved in root development and cell division (Supplemental Data Set 19). These processes are expected to be involved in altering plant development and metabolism. This means that while most of the candidate genes showed specificity to distinct nitrogen responses (a specific set), they appear to be related to a few commonly identified modules, as previously suggested (Gaudinier et al., 2018). This suggests that natural variation might function through a defined set of modules that can be found using the combined GWAS/co-expression approach.

Discussion

How a plant uses nitrogen is critical to the plant’s productivity and fitness, especially in a changing environment. Developing a better understanding of the genetics and mechanisms that govern plant nitrogen use will facilitate the engineering or breeding of plants to use nitrogen more efficiently. To map the potential diversity of processes involved in plant nitrogen responses and use, we grew a large population of Arabidopsis genotypes under four different nitrogen environments and measured a set of developmental traits and a set of defense metabolite traits.

The population showed a diversity of responses to the different nitrogen conditions, which was associated with a large number of genes. While most of those genes were specific to distinct nitrogen conditions (a particular nitrogen source or concentration), it was possible to identify genes potentially involved in responses across nitrogen conditions using a combined GWAS/co-expression filtering method.

The large population size allowed us to identify rare and unusual nitrogen responses amongst genotypes

Using a large number of genotypes, we found that the population had a clear average response, while individual genotypes had a wide range of responses to nitrogen across the different traits. The large population size allowed us to identify extended phenotypic tails that include responses of Arabidopsis to nitrogen. For example, while most of the genotypes showed better root and shoot growth on nitrate versus ammonium, several genotypes exhibited the opposite behavior, with increased growth on ammonium.

Extended phenotypic tails in natural populations can be caused by variation in either a few major loci or multiple small-effect loci. Several lines of evidence suggest that we mainly observed variation in small-effect genes that blend to create these unique events. This means that each of these genes by itself will probably create a small (if any) phenotypic effect. However, when a few genes appear together, they will result in the creation of these unique events. First, most of the traits showed a unimodal distribution of phenotypes, with a long continuous tail of phenotypic variation and the lack of any evidence of multimodality or extreme outliers (Schraiber and Landis 2015). These distributions imply that these traits are controlled by different combinations of small-effect alleles.

The second line of evidence that small-effect genes mainly control these traits in the population comes from the GWAS we conducted using the different traits under the different nitrogen environments. GWAS did not provide any evidence of any large-effect loci associated with the responses to nitrogen. Instead, the loci had small effects, suggesting that the unusual phenotypic responses arose from combinations of alleles at a large number of loci. The potential for novel phenotypes to arise via the accumulation of alleles at multiple loci within a network was recently represented by a model showing that trait values can be nonlinearly created by a network of interactions with different weights (Milocco and Salazar-Ciudad, 2022). To check for potential rare large-effect loci, we queried the genotypes at the distribution tails for a preponderance of large-effect natural knockouts in known nitrogen genes but were unable to find any overlap.

Together, these results suggest that using a large population of genotypes with natural variation is a preferable strategy compared to using a single reference genotype, for example, Col-0, because it improves the detection of responses that result from different combinations of multiple alleles. Therefore, using a vast collection of natural genotypes and combining both genomic and phenotypic techniques can be a good strategy to study complex traits, such as the response to nitrogen conditions. Furthermore, these unique phenotypes (e.g. those showing better growth on ammonium than nitrate) may be generated by stacking small-effect loci to create a large phenotypic variance. Identifying the exact combinations of genes that create these unique phenotypes, for example, better growth on ammonium (as observed in several accessions), will allow plants to be created with desired phenotypes under specific environments.

Metabolic and defense metabolites provide different readouts of the plant nitrogen response

To test if the genetic basis of nitrogen responses is uniform across a range of traits or if it is somewhat trait specific, we analyzed two sets of traits that are associated with nitrogen responses. First, we used classical developmental traits that include shoot and root descriptors associated with nitrogen responses. We also measured the responses of defense metabolites including two families of GSLs (indolic and aliphatic) that depend on amino acid availability for their biosynthesis. Together, we used these two sets of responses to compare the effects of nitrogen on different processes in Arabidopsis seedlings. Phenotypic analyses revealed that each set of traits showed different patterns of nitrogen responses and preferences across the population. Furthermore, the use of GWAS based on these two sets of traits allowed us to detect loci that are associated with each set of traits and a set of loci that is shared among the traits. This demonstrates the importance of studying different types of responses (e.g. metabolic responses) under the same conditions when studying complex responses, because each set of traits highlights different aspects in the plant response to nitrogen and can potentially identify different genetic mechanisms involved in nitrogen responses.

Nitrate and ammonium induce different responses in plants

To test if there is a universal nitrogen response in Arabidopsis or if the responses to nitrate and ammonium (the preferred nitrogen forms of most plants) involve different genetic mechanisms, we measured trait variation using both nitrate and ammonium as the sole nitrogen source. GWAS using traits of genotypes that grew on each of these nitrogen sources revealed two sets of loci, each of which is associated with one of the nitrogen sources, and a set of loci that are shared between the two nitrogen sources. Most of the genes in this analysis are specific to one nitrogen source (68% of nitrate-related genes, and 93% of ammonium-related genes), and only a small portion of the genes is shared across the two sources. Analyzing the correlations between the different traits revealed that while the developmental traits showed strong correlations to each other across all nitrogen conditions, the GSL traits showed different patterns of correlations between each other and the developmental traits, depending on the nitrogen condition provided to the plants. Together, these results suggest that while there is a shared set of genes that will influence the plant across all nitrogen sources, there is an additional (potentially larger) set of genes that will be specific to the nitrogen condition (source and concentration) that the plants experience. In this case, we presented four nitrogen conditions; however, this can apply to other sources and concentrations as well. Therefore, the exact nitrogen condition (source and concentration) has a dramatic effect on the genes and mechanisms used by the plant to optimize its performance.

New gene identification

Another benefit of combining multiple traits, multiple environments, and a larger population is the potential to identify a large collection of potential candidate nitrogen-response genes. Interestingly, the overwhelming majority of these candidate genes have not been previously associated with nitrogen responses. However, many are associated with expected mechanisms such as root development, cell-cycle regulation, and metabolic processes. The new genes identified in these analyses represent potential new players involved in plant responses to nitrogen. For example, out of the five genes that appeared across four sets, only two had informative annotations (AT4G08950 and AT5G35410). Future validation efforts of these genes will help provide a better understanding of the mechanisms shaping a plant’s response to a specific nitrogen source. Overall, our experiment suggests that using such a large population of genotypes and analyzing diverse responses under different environments can reveal new genes that are potentially associated with nitrogen responses.

Materials and methods

Plant material and experimental design

Seeds for 1135 A. thaliana genotypes were obtained from the 1001 genomes catalog of A. thaliana genetic variation (The 1001 Genomes Consortium, 2016; https://1001genomes.org/). The seeds were surface sterilized with 4.125% (v/v) sodium hypochlorite and 0.01% (w/v) Tween 20 for 15 min and rinsed five times with distilled water.

The seeds were sown on 13 cm² plastic culture plates containing the following: 0.1-mM or 1-mM KNO₃ or NH₄HCO₃, 4-mM MgSO₄, 2-mM KH₂PO₄, 1-mM CaCl₂, 10-mM KCl, 36-mg L⁻¹ FeEDTA, 0.146-g L⁻¹ 2-morpholinoethane sulfonic acid, 1.43-mg L⁻¹ H₂BO₃, 0.905-mg L⁻¹ MnCl₂·4H₂O, 0.055-mg L⁻¹ Zinc sulfate heptahydrate, 0.025-mg L⁻¹ copper sulfate, 0.0125-Na₂MoO₄·2H₂O, 1% sucrose, 0.8% (w/v) agar, pH 5.7. The Petri dishes with seeds were placed at 4°C in the dark and incubated for three days. The plates were placed vertically, and the seedlings were grown at 22°C/24°C (day/night) under long-day conditions (16-h white light/8-h darkness, with light at 100–120 µEi). Germination timing was homogenous across the collection, and seedlings were collected 12 days following planting, when the seedlings had developed a first set of true leaves and started to develop an additional set of true leaves. For the initial concentration selection experiments, additional concentrations of KNO₃ and NH₄HCO₃ were analyzed.

Col-0 seeds were used for the preliminary calibrations, and an additional nine accessions were used for further analysis, including accessions CS77081, CS78814, CS77359, CS77353, CS76468, CS77340, CS76435, CS77113, and CS77387. These accessions were sown on medium with either ${\text{NO}}_3^{-}$ or ${\text{NH}}_4^{+}$ as the sole nitrogen source, in concentrations ranging from 0.05 mM to 5 mM.

Each genotype was sown on all four N conditions with three replicates per condition. The replicates were sown on separate plates. Each plate contained six different seeds. Each experimental block contained 60 different genotypes and was referred as “Experiment.”

Root and leaf measurements

The developmental traits that were measured to describe the plant responses to different nitrogen conditions were leaf area, primary root length, total lateral root length (the sum of the length of all the lateral roots in a seedling), and the number of lateral roots. Images of the seedlings were taken using a digital SLR camera (18–55 mm lens, Cannon) in a camera stand with a fixed base to keep the focal distance fixed, with an internal size standard for each photograph. Root traits were measured using Rootnav (Pound et al., 2013) or ImageJ software (https://imagej.nih.gov/ij/) when the roots were tangled with each other.

Leaf area was measured using an automated image processing workflow developed in Python, using functions from the OpenCV and PlantCV packages (Gehan et al., 2017), with the following steps: cropping the agar plate region, leaf identification, pixel counting, and scale identification.

GSL extraction and analyses

We measured the contents of 2 individual indolic GSLs and 15 individual aliphatic GSLs and calculated several traits derived from these individual GSLs (Supplemental Data Sets 6 and 7). GSLs were measured as previously described (Kliebenstein et al., 2001a, b, c). Briefly, each seed was harvested in 200 μL of 90% methanol. The samples were homogenized for 3 min in a paint shaker, centrifuged, and the supernatants transferred to a 96-well filter plate with DEAE Sephadex. The filter plate with DEAE Sephadex was washed with water, 90% methanol, and water again. The Sephadex-bound GSLs were eluted after an overnight incubation with 110 μL of sulfatase. Individual desulfo-GSLs within each sample were separated and detected by high-performance liquid chromatography with diode-array detection (HPLC-DAD), identified, and quantified by comparison to standard curves from purified compounds. A list of the GSLs and their structure is given in Supplemental Data Set 7. Raw GSL data are given in Supplemental Data Set 5.

$$\begin{array}{l}\text{GSL ratios were calculated as follows}:\hspace{0.17em}\mathrm{C}3 \text{ratio}=\frac{\mathrm{C}3}{\mathrm{C}3+\mathrm{C}4},\hspace{0.17em}\\ \text{Alk}=\frac{\text{OH}-3-\text{Butenyl}+3-\text{Butenyl}+\text{Allyl}}{3\text{OHP}+3\text{MSO}+\text{Allyl}+3\text{MT}+\text{OH}-3-\text{Butenyl}+4\text{MSO}+3-\text{Butenyl}+4\text{MT}+4\text{OHB}},\hspace{0.17em}\\ \text{GSOH}=\frac{\text{OH}-3-\text{Butenyl}}{\text{OH}-3-\text{Butenyl}+3-\text{Butenyl}}.\hspace{0.17em}\end{array}$$

Normalization of GSL levels

GSL contents were normalized for each nitrogen condition separately based on leaf area and total root length. For each nitrogen condition, a linear model was utilized to analyze the effect of leaf area and total root length on seedling weight. The intercept and the slope of leaf area and the total root length were used to normalize the GSL contents. This was done to provide a common unit of comparison when it was not possible to measure the mass of all the seedlings.

For seedlings grown on KNO₃ 0.1 mM: ${9.902}^{-6}\times \left(\mathrm{total}\mathrm{ }\mathrm{root}\mathrm{ }\mathrm{length}\right)+{7.706}^{-5}\left(\mathrm{leaf}\mathrm{ }\mathrm{area}\right)+{1.169}^{-3}$.

For seedlings grown on KNO₃ 1 mM: ${8.601}^{-5}\times \left(\mathrm{total}\mathrm{ }\mathrm{root}\mathrm{ }\mathrm{length}\right)-{1.313}^{-3}$.

For seedlings grown on NH₄HCO₃ 0.1 mM: ${6.352}^{-6}\times \left(\mathrm{total}\mathrm{ }\mathrm{root}\mathrm{ }\mathrm{length}\right)+{1.491}^{-4}\times \left(\mathrm{leaf}\mathrm{ }\mathrm{area}\right)-{1.528}^{-4}$.

For seedlings grown on NH₄HCO₃ 1 mM: ${4.768}^{-6}\times \left(\mathrm{total}\mathrm{ }\mathrm{root}\mathrm{ }\mathrm{length}\right)+\mathrm{ }{1.124}^{-3}$.

Statistics, heritability, and data visualization

Statistical analyses were conducted using R software (https://www.R-project.org/) with the Rstudio interface (http://www.rstudio.com/). For each trait, a linear model followed by analysis of variance was utilized to analyze the effect of genotype, nitrogen condition, the experiment, and the culture plate upon the measured trait ($\mathrm{Trait}\mathrm{ }\sim \mathrm{ }\mathrm{Accession}+\mathrm{N}\mathrm{ }\mathrm{source}+(\mathrm{N}\mathrm{ }\mathrm{source}÷\mathrm{N}\mathrm{ }\mathrm{concentration})+\mathrm{Accession}\times \mathrm{N}\mathrm{ }\mathrm{source}+\mathrm{Accession}\times (\mathrm{N}\mathrm{ }\mathrm{source}÷\mathrm{N}\mathrm{ }\mathrm{concentration})\mathrm{ }+(\mathrm{Experiment}÷\mathrm{Culture}\mathrm{ }\mathrm{plate})$). In these models, the nitrogen concentration is nested within the nitrogen source. Models where the nitrogen concentration is not nested within the nitrogen source were analyzed as well and provided similar outcomes. Broad-sense heritability (Supplemental Data Set 8) for the different metabolites was estimated from this model by taking the variance due to genotype (Accession) and dividing it by the total variance.

To test the effect of environmental parameters on the trait ratios, the following linear models were created:

$$\mathrm{Trait}\mathrm{ }\sim \mathrm{ }\mathrm{group}+\mathrm{latitude}+\mathrm{longitude}+\mathrm{latitude}\times \mathrm{longitude}+\mathrm{GAEZ}\_\mathrm{Nutrient}\_\mathrm{availability}+\mathrm{GAEZ}\_\mathrm{Nutrient}\_\mathrm{retention}\_\mathrm{capacity}+\mathrm{GAEZ}\_\mathrm{Rooting}\_\mathrm{conditions}+\mathrm{GAEZ}\_\mathrm{Oxygen}\_\mathrm{availability}\_\mathrm{to}\_\mathrm{roots}+\mathrm{GAEZ}\_\mathrm{Excess}\_\mathrm{salts}+\mathrm{GAEZ}\_\mathrm{Toxicity}+\mathrm{Plant}\_\mathrm{Extractable}\_\mathrm{Water}\_\mathrm{Capacity}\_\mathrm{of}\_\mathrm{Soil}+\mathrm{IGBP}\_\mathrm{DIS}\_\mathrm{Soil}\_\mathrm{pH}+\mathrm{Suitability}\_\mathrm{for}\_\mathrm{Agriculture}.$$

Estimated marginal means (emmeans) for each accession were calculated for each trait based on the genotype and nitrogen treatment using the package emmeans (CRAN—Package emmeans) (Supplemental Data Set 4). Differences between pairs of treatments were estimated for each trait using a Tukey HSD test. Data analyses and visualization were done using R software with the tidyverse (Wickham et al., 2019) and ggplot2 (Kahle and Wickham, 2013) packages. Fisher exact test was performed using the library “stats.”

PCAs were done with the FactoMineR and factoextra packages using default settings (Abdi and Williams, 2010). tSNEs were done with the Rtsne package using default settings (Krijthe, 2015). Factor analysis was done with the psych package; 4 factors were chosen based on initial analysis (Revelle 2022).

Correlations were analyzed using the R package corrplot (https://github.com/taiyun/corrplot.), based on Spearman correlation, using default settings.

Genome-wide association studies

For each trait, GWAS was implemented with the easyGWAS tool (Grimm et al., 2017) using the EMMAX algorithms (Kang et al., 2010) and a minor allele frequency cutoff of 5%. The results were visualized as Manhattan plots using the qqman package in R (Turner, 2014).

Environmental and demographic data

Environmental and demographic data (referred to as “genomic group”) were obtained from the 1001 genomes website (https://1001genomes.org/, for geographical and demographic data) and from the Arabidopsis CLIMtools (http://www.personal.psu.edu/sma3/CLIMtools.html, Ferrero-Serrano and Assmann, 2019). Each of the above variables (including genomic group) was assigned to each one of the accessions.

MASH

To reduce the number of SNPs in each set, we chose the 100,000 strongest SNPs in each GWAS, which included all the significant SNPs in a genome-wide association and other SNPs. To combine the GWAS results, MASH was conducted using the mashr package in R (Urbut et al., 2019). The P-values of SNPs and standard errors produced by GWAS were used in the MASH analyzes as Bhat (a matrix of effects) and Shat (a matrix of standard errors). A subset of 50K SNPs was used as a “random” set to learn the correlation structure among null tests and to fit the mashr model. The analysis was done with a subset of the 100,000 strongest SNPs across each set, which was chosen in two ways: (1) Choosing 50,000 SNPs with the maximum −log10(P-values) to account for strong effects that are specific to traits/concentrations (specific to each genome-wide association). (2) Summing the −log10(P-values) across the set for each SNP and choosing the 50,000 SNPs with the maximum values. SNPs with local false sign rates values higher than 0.98 were chosen for further analysis. The SNPs were connected to genes using a sequence window of 2,000-bp upstream of the transcription start site to 2,000-bp downstream of the stop codon.

Filtering SNPs from the six sets

SNPs from GWAS were grouped into six sets based on the traits and the nitrogen source: (1) developmental traits grown on nitrate; (2) developmental traits grown on ammonium; (3) aliphatic GSLs grown on nitrate; (4) aliphatic GSLs grown on ammonium; (5) indolic GSLs grown on nitrate; and (6) indolic GSLs grown on ammonium (Supplemental Figures S10–S13). Each set included the traits of plants grown under the two nitrate concentrations and the logged ratio of the concentrations of each trait under the same nitrogen source.

SNP analysis

To analyze the SNPs in each of the six sets, we used the MASH method and estimated the effect of SNPs in each set of GWAS. Using this method, we could estimate the effect sizes of both shared and condition-specific effects within each of our six sets (Urbut et al., 2019). This enabled us to detect genes (based on the detected SNPs) that are associated with specific traits (e.g. developmental/aliphatic/indolic), genes that are associated with a specific nitrogen source (nitrate or ammonium), and genes that are shared across the different nitrogen conditions and/or traits. For each set, we conducted MASH analysis by considering both specific effects (trait specific and concentration specific) and mutual effects across all the genome-wide associations in the same set (for more details, see below). Using this method, we were able to rank the SNPs based of their effect in each set and provide a score for each SNP ranging from 0 to 1, where 1 represents the strongest effect. We used these MASH scores of each SNP from each set to choose SNPs with the highest effect from each set. Here we chose SNPs with MASH scores higher than 0.98 (Supplemental Figure S14). Plotting these SNPs based on their location on the chromosome showed that the chosen SNPs from each set are spread across the genome, and we could not detect specific areas with high SNP density (Supplemental Figure S15). We then transformed these SNPs into genes by considering 2,000 bp before and after the start and end point of each gene. This analysis yielded six sets of genes: each set is associated with a different set of traits under a specific nitrogen source (Supplemental Data Set 13) and contains between 22 and 5,051 genes.

List of nitrogen-related genes in Arabidopsis

Using several literature sources, we created a set of genes that were reported to be associated with some nitrogen process (Supplemental Data Set 15). These processes include nitrogen assimilation, transport, metabolism, signaling, and transcription factors (Zhang and Forde, 1998; Gifford et al., 2008; Castaings et al., 2009; Rubin et al., 2009; Krouk et al., 2010; Obertello et al., 2010; Konishi and Yanagisawa, 2013; Alvarez et al., 2014; Vidal et al., 2014; Medici et al., 2015; Xu et al., 2016; Gaudinier et al., 2018; Varala et al., 2018; Brooks et al., 2019; Cheng et al., 2021)

GO enrichment analysis

GO enrichment analysis was performed using TAIR (https://www.arabidopsis.org/tools/go_term_enrichment.jsp) and agriGO (Du et al., 2010).

Co-expression modules

Co-expressed gene sets were identified using the mr2mods program, and previously reported data measuring 874 transcriptomes from natural Arabidopsis genotypes were utilized (Kawakatsu et al., 2016; Wisecaver et al., 2017). Briefly, these packages take the data and create an all-by-all Spearman correlation coefficient matrix for the complete transcriptomes. This matrix is then converted into mutual ranks for each transcript pair. Module groupings are then defined using the mutual ranks, which are converted to edge weights using an exponential decay function. Eight different exponential decay values ranging from 2 to 100 were evaluated, and a decay value of 50 was chosen to maximize the average module size.

Accession numbers

Sequence data from this article can be found in the GenBank/EMBL libraries under the following accession numbers: AGI numbers: AT4G18390, AT3G60320, AT2G15620, AT5G51190, AT4G39070, AT4G08950, and AT5G35410.

Supplemental data

The following materials are available in the online version of this article:

Supplemental Figure S1. Col-0 seedlings present different phenotypes when grown under different nitrogen conditions.

Supplemental Figure S2. Seedlings present different lateral root lengths when grown under different nitrogen conditions.

Supplemental Figure S3. Nitrogen conditions affect developmental traits across natural accessions.

Supplemental Figure S4. Nitrogen conditions affect different GSL traits across natural accessions.

Supplemental Figure S5. Nitrogen conditions affect different GSL traits across natural accessions.

Supplemental Figure S6. Nitrogen conditions affect different GSL traits across natural accessions.

Supplemental Figure S7. Principal component and tSNE analyses.

Supplemental Figure S8. Nitrogen conditions affect different traits across natural accessions.

Supplemental Figure S9. Traits used for GWAS.

Supplemental Figure S10. Genome-wide association analyses of developmental traits of plants grown under ${\text{NO}}_3^{–}$ concentrations.

Supplemental Figure S11. Genome-wide association analyses of developmental traits of plants grown under ${\text{NH}}_4^{+}$ concentrations.

Supplemental Figure S12. Genome-wide association analyses of aliphatic and indolic traits of plants grown under ${\text{NO}}_3^{–}$ concentrations.

Supplemental Figure S13. Genome-wide association analyses of aliphatic and indolic traits of plants grown under NH₄ concentrations.

Supplemental Figure S14. Distribution of the number of SNPs by MASH scores.

Supplemental Figure S15. Loci associated with the nitrogen response.

Supplemental Figure S16. Overlaps between gene modules from the different MASH sets.

Supplemental Table S1. Percentage of overlapping genes/modules in the different sets.

Supplemental Data Set 1. Col-0 data.

Supplemental Data Set 2. Col-0 analysis of variances.

Supplemental Data Set 3. Accessions’ data.

Supplemental Data Set 4 . Accessions’ analysis of variances.

Supplemental Data Set 5. Raw data.

Supplemental Data Set 6. Emmeans data.

Supplemental Data Set 7. List of GSLs and structures.

Supplemental Data Set 8. Heritability values.

Supplemental Data Set 9. P-values of the different variables.

Supplemental Data Set 10. Differences between pairs of treatments.

Supplemental Data Set 11. The effect of the origin of the plants on the traits.

Supplemental Data Set 12. Samples’ statistics.

Supplemental Data Set 13. Lists of genes yield from the different mash sets.

Supplemental Data Set 14. Fisher’s exact test for each pair of sets.

Supplemental Data Set 15. List of N genes.

Supplemental Data Set 16. Combined gene list and GO enrichment.

Supplemental Data Set 17. GO enrichment for the six sets.

Supplemental Data Set 18. Lists of genes in common modules.

Supplemental Data Set 19. Enriched GO categories using genes from the common modules.

Funding

This work was supported by the National Science Foundation, Directorate for Biological Sciences, Division of Molecular and Cellular Biosciences (grant no. MCB 1906486 to D.J.K.) and Division of Integrative Organismal Systems (grant no. IOS 1655810 to A.J.B. and D.J.K.).

Conflict of interest statement. None declared.

 

E.K., D.J.K. and A.J.B. designed the research and wrote the article, E.K., A.K., M.L., C.K.K., J.S., J-J.L., E.S., and K.T.L. performed the research and analyzed the data.

The authors responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plcell) are: Ella Katz (elkatz@ucdavis.edu) and Daniel J. Kliebenstein (kliebenstein@ucdavis.edu).