Unraveling Genomic Regions Controlling Root Traits as a Function of Nitrogen Availability in the MAGIC Wheat Population WM-800

An ever-growing world population demands to be fed in the future and environmental protection and climate change need to be taken into account. An important factor here is nitrogen uptake efficiency (NUpE), which is influenced by the root system (the interface between plant and soil). To understand the natural variation of root system architecture (RSA) as a function of nitrogen (N) availability, a subset of the multiparent advanced generation intercross (MAGIC) winter wheat population WM-800 was phenotyped under two contrasting N treatments in a high-throughput phenotyping system at the seedling stage. Fourteen root and shoot traits were measured. Subsequently, these traits were genetically analyzed using 13,060 polymorphic haplotypes and SNPs in a genome-wide association study (GWAS). In total, 64 quantitative trait loci (QTL) were detected; 60 of them were N treatment specific. Candidate genes for the detected QTL included NRT1.1 and genes involved in stress signaling under N−, whereas candidate genes under N+ were more associated with general growth, such as mei2 and TaWOX11b. This finding may indicate (i) a disparity of the genetic control of root development under low and high N supply and, furthermore, (ii) the need for an N specific selection of genes and genotypes in breeding new wheat cultivars with improved NUpE.


Introduction
Wheat (Triticum aestivum) is one of the most important crops in human nutrition [1] but currently also requires a high level of nitrogen (N) fertilization, because N plays a central role for plant growth and the formation of yield and quality in winter wheat [2]. On the downside, intensive N fertilization is known to have negative effects on biodiversity [3][4][5], and leads both to the degradation of land and water [6,7] (causing eutrophication and hypoxic zones in marine ecosystems) and to the emission of greenhouse gases, which contributes to ozone damage and global warming [8,9].
At this point, nutrient use efficiency (NUE) enters the equation, being defined as grain dry matter yield per unit N available in the soil, consisting of N uptake efficiency (plant N uptake/available N; NUpE) and N utilization efficiency (grain dry matter yield/N plant uptake; UTE) [10].
Efficient N uptake could prevent these negative effects of intensive N fertilization, and a well-developed root system (already at the seedling stage) plays a crucial role in the NUE of plants [2,[11][12][13], acting as an interface between the plant and water and nutrients in the soil [13,14]. Fortunately, winter wheat has one of the fastest growing and prolific root systems of all arable crops [15]. The size of a root system as well as its architecture is known to exhibit a very high phenotypic plasticity in response to environmental factors [16][17][18][19], particularly with respect to nitrogen availability and distribution [20]. However, in addition to environmental effects, there is also genotypic variation in the development of the root system architecture (RSA) of wheat [21], providing the baseline for later adaptation to the environmental conditions. To investigate seedling roots, soil-free methods, such as hydroponic or paper-based approaches, are often preferred [22][23][24], because root characteristics can be examined more conveniently. However, a drawback of these soil-free methods is the lack of the complex environmental interaction effects just mentioned, which can only be represented to a limited extent in such setups.
Studies have made it evident that a well-developed root system can, in general, improve N uptake, but it must be taken into consideration that the formation and maintenance of the root system is a carbon-costly process [14,25]. Under N deficiency, wheat invests particularly in root length in order to acquire N resources [26]. Especially the growth of lateral roots is of importance here [27], which is regulated by N signaling via a NO 3regulated repression [13,28]. In this regard, NRT1.1, a dual-affinity nitrate transporter, plays a key role not only in the transport of N but also in the signaling, which leads to N dependent changes in downstream gene expression. NRT1.1, in its function as a nitrate sensor, was proposed to be involved in the expression of high-affinity nitrate transporters of the NRT2 gene family, such as NRT2.1 [29], that drive nitrate uptake at low N availability. Differential phosphorylation of NRT1.1 in dependency of the N availability and associated Ca 2+ signaling mechanisms affect root growth through the activation of auxin transport activity and Ca 2+ -ANR1 signaling [30]. In addition, plant hormones are known to play an important role in root architecture formation [31]. Cytokinin balances the division and differentiation of stem cells [32] and its concentration in the root tip (together with auxin) defines the size of the meristem [33].
RSA was shown to be highly heritable in wheat [34][35][36][37] and several studies, investigating RSA under contrasting nitrogen levels, revealed significant differences among genotypes and between N treatments, as well as significant genotype x N treatment interactions [12,36,37] and correlations with agronomic traits such as maturity and yield [12,34,35,38]. To gain an understanding of this genetic variation and its underlying regulation of RSA, quantitative trait loci (QTL) studies are an effective tool. In recent years, many QTL for RSA traits under N deficiency have already been characterized in wheat [34,[36][37][38].
To increase the power of QTL detection by a genome-wide association study (GWAS), the concept of multi-parent advanced generation intercross (MAGIC) populations is a valuable tool. MAGIC populations, with their increased number of founders, combine the statistical power of bi-parental populations with the genetic diversity of association panels. In wheat, a number of MAGIC populations have already been successfully used [39][40][41]. To circumvent the limitation of bi-allelic SNPs in a MAGIC population, two or more SNPs in strong linkage disequilibrium (LD) can be combined into a haploblock [42]. This allows to trace back a detected QTL effect to the parent that inherited it. Haplotype-based GWAS has already proven its ability to detect both major QTL and small-effect QTL in several studies [43][44][45]. Within the frame of this study, the WM-800 population (a MAGIC population derived from eight European winter wheat varieties) was utilized, which was shown to be a valuable mapping population for different agronomic traits [46][47][48].
To investigate the genetic regulation of RSA traits and their plasticity to N availability, a subset of 350 lines of the winter wheat MAGIC population WM-800 was non-destructively phenotyped in a germination paper-based high-throughput phenotyping system called GrowScreen-PaGe [24]. Fourteen root and shoot traits were phenotyped at three time points under two contrasting N treatments to quantify variation within the WM-800 population and variation between the N treatments. Finally, we conducted a haplotype-based GWAS to answer the question whether the genetic control of those traits depends on N availability and whether it differs between time points.

WM-800 Shows Broad Variation for Root Traits
Phenotyping in the GrowScreen-PaGe system has allowed a dynamic analysis of root system growth in the winter wheat MAGIC population WM-800 under two different N treatments. The soil-free setup enabled phenotyping root systems of a large set of genotypes in a high-throughput process and provided high resolution time series data for a broad range of traits describing early root and shoot development as a function of N availability. At the same time, it must be considered that root growth and NUpE in the field is affected by a wide range of interaction effects between the root system and the soil and its microbiome. Those effects can only be represented to a limited extent in such a phenotyping system.
In total, 14 traits were examined (Table 1), which form four different trait groups and can thus comprehensively represent the development of both the root system and the shoot of the seedlings. Seminal root length (SRL) and lateral root length (LRL) consider root length, while branching angle lateral (BAL), seminal root angle (SRA) and root system radius angle left (RRL) and right (RRR) describe the angle of root growth. Root system depth (RSD), root system width (RSW) and convex hull area (CHA) are composed of root system length and angle. Finally, harvest parameters' root dry weight (RDW), shoot dry weight (SDW), root to shoot ratio (R:S), SPAD (measurement for leaf chlorophyll content) and leaf length (LLE) were recorded 12 days after transplantation (DAT).
Significant differences were observed between genotypes and between treatments for most traits and time points, as well as significant genotype × treatment interaction effects (Table 2). An exception is LRL at 3 DAT, where no significant genotype, treatment or genotype × treatment interaction effects were observed because lateral roots had not yet been formed. Therefore, also the branching angle of the lateral roots (BAL) could not be determined 3 DAT. A broad variation in phenotypes of the analyzed 350 WM-800 lines, higher than in the founders and check varieties, was observed for all traits ( Figure S1). This is an indication for transgressive segregation in the WM-800 population, which is beneficial in later genetic analysis and selection in plant breeding. The phenomenon of transgressive segregation for root traits has also been described in other wheat studies [22,49]. Studies related to plant height, yield and yield parameters in the WM-800 population already demonstrated the broad variation within this population and its power to map new QTL affecting a variety of important agronomic traits [46][47][48].
The coefficient of variation (CV) ranged between 6.67% for SPAD (N+, 12 DAT) and 95.69% for LRL (N-, 3 DAT). The root length traits and traits composed of root length and root angle, with exception of SRL, showed overall higher CV than the root angle and harvest traits. The exceptional high CV for LRL might be explained by the origin of development of lateral roots in comparison with seminal roots. Lateral roots are developed post-embryonic and thus might be more strongly affected by environmental factors [50,51].
For 19 out of 31 traits by time point combinations the mean phenotype values were significantly lower under N− compared with the N+ treatment ( Figure 1 and Table S5), indicating that N deficiency had a negative effect on the root and shoot development. The only exceptions from this rule were observed for R:S at 12 DAT and SRA at all time points, where higher values under N− were observed. Wang et al. [52] and Guo et al. [53] observed similar responses of plant growth and chlorophyll content studied under contrasting N treatments. However, they also observed a significantly increased root growth under N deficiency, which was not observed in the experiment conducted here. Studies showed that nitrate stimulates not only lateral root growth [54] but also primary root growth through an increased cell division and elongation in Arabidopsis [55]. Since N is supplied in the form of nitrate in this experiment, this might explain the observations of longer roots under N+ compared with N−. A study in barley described the phenomenon of root growth promotion under high N supply, in case this N supply is concentrated in a specific zone and not evenly distributed [56]. Since the plants in our experiment were supplied via a nutrient solution at the bottom of the growth paper, this might also explain the observed trends. The improved root growth under N+ might also have been caused by optimal nitrate availability in the N+ treatment.   Table 1.
Over the course of the experiment, the values for traits related to root length and the combination of root length and angle (SRL, LRL, RSD, RSW, CHA) increased, while the traits related to root system angle changed little over time and showed no upward or downward trend.

Correlations Indicate N Dependent Genotypic Plasticity
Over the three phenotyping time points, correlations between traits remained rather stable ( Figure S2). The traits based on root length and root length x angle (SRL, LRL, RSD, RSW, CHA) formed a block of significant positive correlations similarly in both N Plants 2022, 11, 3520 7 of 26 treatments ( Figure 2). The strong positive correlations between root traits have also been observed in other studies [22,49] and might suggest that they are under shared control.  Table 1. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.

GWAS Reveals a Different Genetic Control of Root Traits under Both N Treatments
GWAS was performed separately for the two N treatments, which allowed us to investigate the genetic regulation of NUE depending on N availability. A total of 102 significant marker-trait associations (Table S6), grouped into 64 QTL, were detected for 12 of the 14 traits (Table 3). No significant QTL were found for the traits BAL and RDW. The detected QTL were distributed equally over all three wheat genomes with 24, 22 and 28 on genomes A, B and D, respectively.   Table 1. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001.
Correlations of LRL with the other traits were close to zero at 3 DAT ( Figure S2), because lateral roots just started to form. Furthermore, RDW showed a strong and significant positive correlation with SDW and LLE under both treatments, suggesting a dependency of root and shoot growth and biomass accumulation. On the other hand, RDW was just moderately correlated with the two root length parameters SRL and LRL, suggesting that there is a second parameter next to root length determining the root biomass, such as root diameter, that showed significant genotypic differences in another study investigating the WM-800 population [48]. The SPAD value was positively but only weakly correlated with SRL, RDW and SDW in both treatments.
In contrast, the auto-correlations of the traits between the two N treatments were rather low, except for LLE (r = 0.61, 12 DAT), indicating that the tested wheat genotypes developed roots as well as shoots differently under the two N treatments. This plasticity of the root system in response to environmental conditions has been widely characterized in the literature [57,58].

GWAS Reveals a Different Genetic Control of Root Traits under Both N Treatments
GWAS was performed separately for the two N treatments, which allowed us to investigate the genetic regulation of NUE depending on N availability. A total of 102  (Table S6), grouped into 64 QTL, were detected for 12 of the 14 traits (Table 3). No significant QTL were found for the traits BAL and RDW. The detected QTL were distributed equally over all three wheat genomes with 24, 22 and 28 on genomes A, B and D, respectively. A total of 24 of the QTL were N− specific, 36 N+ specific and only 2 QTL were significant under both N treatments. The finding that the traits in the two N treatments were largely regulated by different QTL suggests that selection for NUE under N deficiency is more beneficial than passive selection under standard N availability. Studies in maize and wheat could confirm the advantageousness of direct selection for N efficiency [59,60]. These individual QTL explained between 4.83% (LLE N+ 12 DAT) and 21.03% (LRL N− 7 DAT) of the genotypic variance (R 2 ) ( Table 4); the overall GWAS models could explain between 19.28% (SRL N− 3 DAT) and 57.53% (LLE N+ 12 DAT) of the genotypic variance per trait, treatment and time point. The R 2 values were on average slightly higher under N+ than under N−.

Thirteen QTL Hotspots Showed Pleiotropic Effects on Multiple Traits
The GWAS analysis revealed 13 QTL hotspot regions of either equal or genetically linked HBs or singular SNPs having pleiotropic effects on two or more traits simultaneously. These QTL hotspots could be detected within one treatment as well as across treatments and affected between two and five traits. The five N non-specific, four N− specific and four N+ specific QTL hotspots are discussed in more detail below, complemented by the discussion about further interesting QTL regions.
Effects of the Rht Genes Rht-B1 and Rht-D1 and Further N Non-Specific QTL Hotspots The GWAS revealed two QTL hotspots on chromosomes 4B and 4D associated with the well-known reduced height (Rht) genes Rht-B1 and Rht-D1. These QTL showed effects exclusively on aboveground traits and not on root traits. While some studies, just as in this study, found no significant effects of the semi-dwarf variants of Rht-B1 and Rht-D1 on root traits [48,63,64], Li et al. detected significant effects on root length of a QTL mapped to Rht-B1 [65].
The semi-dwarf phenotypes of the Rht genes, Rht-B1b and Rht-D1b, are a result of a mutation in the N-terminal DELLA domain of Rht-B1 and Rht-D1, which prevents the two proteins being repressed by gibberellic acid (GA), leading to a reduced plant growth [66]. The founders Tobak and Safari carry the semi-dwarf allele Rht-B1b, while the founders Patras, Linus, JB Asano and Julius carry the semi-dwarf allele Rht-D1b. The founders Meister and Bernstein carry the long straw alleles at both loci.
The QTL hotspot on chromosome 4B, associated with Rht-B1, showed a significant effect on LLE in both N treatments with R 2 values of 10.13 and 13.17% under N− and N+, respectively. The HT derived from Tobak and Safari reduced LLE by 10.21 and 13.19% under N− and N+, respectively.
The QTL hotspot on 4D, associated with Rht-D1, affected LLE across treatments, SDW under N+ and SPAD under N− at 12 DAT. For the founders Patras, Linus, JB Asano and Julius carrying the semi-dwarf allele Rht-D1b LLE was reduced by 11.13% under N− and 12.33% under N+, SDW under N+ was reduced by 9.45% and SPAD value under N− was increased by 4.11%. The R 2 values ranged from 6.82% (SPAD N−) to 15.75% (LLE N+). The N non-specific genetic regulation of LLE by Rht-B1 and Rht-D1 is also reflected in the high auto-correlation between the two N treatments.
A next hotspot was localized on chromosome 2A. It significantly affected LRL (N-7 DAT) and R:S (N+ 12 DAT) and could explain 21.03 and 8.90% of the variance, respectively. Another hotspot was located on 2D and consisted of two QTL in close genetic distance. They affected SRL (N+ 7 DAT) and RSW (N− 7 DAT) and explained 8.54 and 11.96% of the variance, respectively. For both hotspots on chromosome 2A and 2D a candidate gene has not yet been identified.
A last N non-specific QTL hotspot was located on chromosome 7A and affected RSW under N+ at 3 DAT and RRL under N− at 3 DAT with a relative effect of 18.17 and 12.74% for the QTL allele derived from Meister, Linus and Julius, respectively. A candidate gene has not yet been identified.

N− Specific QTL Hotspots
The hotspot that affected the most traits under N-was located on chromosome 5A at 688-691 Mbp. This HB (Chr5A_HB137, see Table S6) affected SRL (12 DAT), RSD (7 DAT), RSW (7 DAT), SRA (7 DAT) and RRL (7 DAT). Five different HTs could be differentiated, but only the HT of the founder JB Asano was significant and showed strong relative effects ranging from −16.41 to 50.05% on RRL and RSW, respectively. A candidate gene is NRT1.1 (IWGSC RefSeq v1.1: TraesCS5A02G537100.1 [61]) in close proximity to the detected HB. NRT1.1 belongs to the gene family NRT1, which codes for low-affinity nitrate transporters, acting under high nitrate concentration. However, in Arabidopsis, NRT1.1 is known to change its function under low nitrate availability by being phosphorylated, acting as a nitrate sensor instead [30]. An expression study in wheat demonstrated a higher expression of NRT1.1 under low nitrate availability [53]. The cascade downstream of the phosphorylated NRT1.1 activates phosphatidylinositol, resulting in a Ca 2+ influx that then changes the expression of transcription factors and genes involved in nitrate transport (highaffinity transporters of the gene family NRT2) and assimilation. The activated high-affinity transporters in the end affect lateral root initiation and growth [30].
A QTL hotspot on chromosome 3B (492-497Mbp) showed significant effects on RSD (QRSD3_N-_WM-800_3B) and RSW (QRSW3_N-_WM-800_3B) at 3 DAT under N-. The HT derived by the founders Patras, Tobak and Safari reduced RSD and RSW by 25.67 and 51.85% with an R 2 value of 9.00 and 17.74%, respectively. Within this HB, a candidate gene is located coding for an SKP1-like protein (IWGSC RefSeq 1.1: TraesCS3B02G308600.1 [61]). These proteins have been described to affect seed germination and seedling growth in Arabidopsis [67,68] and are upregulated under abiotic stress [67]. SKP1 is a S-phase kinaseassociated protein 1 and a component of the SCF-type E3 ligase, where it regulates the signaling pathways of several phytohormones [68]. A study in wheat found that a SKP1 gene was highly expressed in the elongation zone of young roots and an overexpression of TSK1 in Arabidopsis mutants resulted in changes in the auxin response and auxin-related root phenotypes [69].
One further hotspot that was significant exclusively under N-was located on chromosome 4A and affected RSD and RSW at 3 DAT. The significant HT was carried only by the founder Safari, increasing RSD and RSW by 18.71 and 27.54%, respectively. In direct proximity to this hotspot, a candidate gene is located (IWGSC RefSeq v1.1: TraesCS4A02G312300.1 [61]) that is highly expressed in root tissue [70,71] and codes for a calcium-dependent lipid-binding domain protein. Phospholipids, precisely the phosphoinositide/phospholipase C (PI/PLC) system, and Ca 2+ are both known to be involved in stress signaling in plants [72,73]. Several studies described the role of calcium-dependent lipid-binding proteins in abiotic stress response [74,75]. A study in rice proposed that the calcium-dependent lipid-binding domain protein might affect Ca 2+ influx by binding to the phospholipid. An overexpression of the coding gene led to a more drought stress resistant phenotype with longer roots [75]. This cascade of PI/PLC activation and subsequent Ca 2+ influx also play a key role in the NO 3 − dependent changes in the expression of transcription factors and genes involved in nitrate transport and nitrate assimilation downstream of NRT1.1. NRT1.1 is a candidate gene we already discussed for a previous N-specific hotspot on 5A that, in the end, affects lateral root initiation and growth [30].

N+ Specific QTL Hotspots
We found that an N+ specific hotspot on chromosome 1B had a significant effect on RSW (QRSW3_N+_WM-800_1B) and CHA (QCHA3_N+_WM-800_1B) at 3 DAT. The HT, derived from the founders JB Asano, Meister, Patras, Safari and Tobak, increased RSW and CHA by 23.73 and 25.00% with an R 2 value of 10.56 and 8.08%, respectively. A potential candidate gene is a gene coding for a mei2-like 2 protein (IWGSC RefSeq v1.1: TraesCS1B02G164600.1 [61]) at 287.5 Mbp. The mei2 gene family was first described for its role in meiosis, but newer studies have observed a role in seedling development and meris-tem activity as well [77]. They observed an alternated root phenotype, suggesting alternated meristem activity for mei2 mutants and an expression of mei2-genes in the roots, which is consistent with the expression pattern for our candidate gene TraesCS1B02G164600.1 [70,71]. A potential explanation for the QTL just being significant under N+ is that the N deficiency masks the effect of this QTL under N-.
Another N+ specific QTL hotspot on chromosome 2D had strong significant effects on CHA at 3 DAT and on RSW at 7 DAT. The founder Linus carried the significant HT that increased CHA and RSW by 44.80 and 28.14%, respectively. Two further QTL hotspots were located on chromosomes 6B and 6D. The QTL on 6B affected LRL at 3 DAT and RSD at 7 DAT and explained 14.36 and 7.80% of the variation, respectively, while the QTL on 6D affected R:S and RSW at 12 DAT with an R 2 value of 6.05 and 12.74%, respectively. It remains open which genes are responsible for the observed effects on 2D, 6B and 6D.
Additional QTL of interest showing a significant effect on only one trait are sorted by trait categories and discussed in the following.

Number of Detected QTL Suggests High Plasticity of Lateral Roots to Environmental Influences
Eight QTL have been detected for the root length related traits SRL, while only four for LRL. The QTL were distributed evenly over time points and N treatments. The significantly higher amount of detected QTL for SRL than for LRL might be explained by the fact that seminal roots of plants are established during embryogenesis while lateral roots are established post-embryonically [51] and, thus, might be affected by environmental influences stronger than seminal roots.

Seminal Root Length (SRL)
In addition to the two QTL on chromosome 2D and 5A, which have already been discussed as hotspots, six further QTL were detected for SRL. One of these QTL was located on chromosome 2D and the significant HT, derived from Patras, JB Asano, Tobak and Julius, increased SRL under N-at 12 DAT by 12.01%. Two candidate genes within the HB that are highly expressed just in root tissue [70,71] code for ABC transporters of the subfamily B (IWGSC RefSeq v1.1: TraesCS2D02G010400.1 and TraesCS2D02G010500.1 [61]). The role of ABC subfamily B transporters in root development has been described in multiple studies [78,79]. ABC transporters are known to be located in the plasma membrane and function in auxin transport and, thus, auxin regulated development [80,81]. Variation in the auxin transport could explain the observed effects on SRL.
The next QTL on chromosome 7B affected SRL (QSRL3_N+_WM-800_7B) under N+ at 3 DAT. The significant HT is present in the founders JB Asano, Julius, Meister, Patras and Tobak and showed a relative effect of 15.77%. Close to this QTL a gene is located coding for a glycosyltransferase (IWGSC RefSeq v1.1: TraesCS7B02G340100.1 [60]) with high expression in root tissue [70,71]. Glycosyltransferases are described as a required component for normal cell-cycle regulation and a reduced expression can result in an extended cell cycle and reduced growth [82]. This observation can be explained, most likely, by the effect of glycosyltransferases on the expression of many flavonoids, which function as auxin transport inhibitors [83]. Natural allelic variation might, thus, explain the observed effect on root growth in our study.

Lateral Root Length (LRL)
For LRL, only four QTL were detected, of which the two on chromosome 2A and 6B also showed an effect on other traits. One N− specific QTL was detected on 2A at 7 DAT, which is consistent with the low repeatability under N− at the two other time points. The very high CV for LRL, which decreased slightly over time, may explain the high relative effects.
The QTL on chromosome 3D was detected under N+ at 12 DAT and the significant HT, carried by Safari, increased LRL by 59.69% and could explain 8.87% of the variation. A poten-tial candidate gene upstream of this QTL (IWGSC RefSeq v1.1: TraesCS3D02G474800.1 [61]) is highly expressed specifically in roots [70,71] and encodes expansin, a cell wall-loosening protein [84] involved in numerous developmental processes in shoots and roots during which cell wall modifications occur [85][86][87]. A study in Arabidopsis proved that expansin plays a role in lateral root initiation by acting in radial pericycle cell expansion [88].

The Traits
Describing the Spatial Distribution of Root Systems Are Mainly Controlled by QTL Hotspots RSD, RSW and CHA reflect the area encompassed by the root system of the plants. A total of 26 QTL were detected, equally distributed across both N treatments. Among them, only eight QTL had an effect on one trait only; the others were all assigned to one of the before-mentioned hotspots.

Root System Depth (RSD)
Four of the five detected QTL were part of a hotspot and only one QTL on chromosome 3A had an effect exclusively on RSD. The significant HT, which originated from the founders JB Asano, Bernstein and Julius, had a relative QTL effect of −13.67% and an R 2 value of 6.99% under N+ at 3 DAT, respectively. A candidate gene could not be identified yet.

Root System Width (RSW)
Twelve QTL showed a significant effect on RSW, nine of which were part of a QTL hotspot. One of the three remaining QTL on chromosome 2B had a significant effect of −18.91% with an R 2 value of 7.98% under N+ at 7 DAT. The HT with the significant effect was derived from the founders Bernstein, Linus, Meister, Safari and Tobak. In close proximity to this hotspot is the candidate gene TaWOX11b [89] (IWGSC RefSeq v1.1: TraesCS2B02G117900.1 [61]), coding for a WUSCHEL-related homeobox transcription factor. TaWOX11 showed an increased expression in roots in the vegetative state [70,71,89], suggesting a role in root development that would explain the observed QTL effect. A putative rice orthologue of TaWOX11, OsWOX11, was described to be involved in the regulation of crown root development [90] by affecting the expression of auxin-and cytokinin-responsive genes.

Convex Hull Area (CHA)
In addition to the two QTL already discussed as hotspots, six additional QTL affected CHA. These six QTL showed strong relative effects between −43.90% (QCHA7_N+_WM-800_7D) and −38.79% (QCHA12_N+_WM-800_7A). The QTL on 7A, affecting CHA under N+ at 12 DAT, is surrounded by four NRT1 genes and four PTR genes. The candidate gene (IWGSC RefSeq v1.1: TraesCS7A02G054000.2 [61]) is coding for a NRT1/PTR family 2.3-like protein that is highly expressed specifically in root tissue [70,71]. Transporters of the NRT1 gene family are low-affinity transporters, acting under sufficient N availability, which explains the QTL detection solely under N+. The Arabidopsis orthologue NPF2.3 is expressed in root pericycle cells and the transporter itself is located in the plasma membrane. Since N is known to be an important compound in the regulation of root growth [18], NPF2.3 is a plausible candidate gene for the detected QTL.
On chromosome 7D, a QTL was located that had a significant effect on CHA under N+ at 7 DAT, explaining 9.32% of the genotypic variation. The significant HT is solely present in the founder Meister and increased CHA by 43.90%. As such, this QTL showed one of the strongest relative effects of this study. Next to the significant HB, two genes are located (IWGSC RefSeq v1.1: TraesCS7D02G000700.1 and TraesCS7D02G000800.1 [61]) that are highly expressed, especially in root tissue [70,71], and coding for a receptor-like protein kinase and a lectin-receptor kinase, respectively. Both belong to the Receptor-Like Kinase (RLK) protein family that forms a group of signaling molecules on the surface of cells, which are responsible for cell-to-cell communication and are, among others, described to control differentiation of the root meristem [91]. Lectin receptor kinase is one class of the RLK protein family and is involved in the regulation of plant hormones such as abscisic acid [92], which is known to affect root growth [93,94].

QTL for Root System Angle-Related Traits Are Mainly Located in Close Proximity to Nitrate Transporter Coding Genes
For the root system angle traits, with the exception of BAL, 13 QTL were identified that could explain between 6.20 and 11.91% of the variance and had relative effects ranging from −22.09 to 30.04%. It is striking that five of the six N-specific QTL were located in hotspots, whereas this was not true for any N+ specific QTL. Four of these five QTL were located in close proximity to nitrate transporters, which, by their function of N sensing [30], presumably influence the root angle so that the roots can reach N rich regions.
In general, root system angle traits are little studied in wheat so far. VRN1 is one of the few genes in wheat whose effect on root system angle has already been characterized [95]. Furthermore, genes affecting gravitropism are strong candidate genes for root system angle traits [96]. However, the QTL detected in this study are not localized near such genes.

The Majority of the QTL Detected for Harvest-Related Traits Were N+ Specific
At the third phenotyping time point (12 DAT), the plants were harvested. For the parameters SDW, R:S, SPAD and LLE, 13 QTL were detected, four under N− and nine under N+. As mentioned above, for LLE, significant QTL linked to the Rht genes Rht-B1 and Rht-D1 on chromosome 4B and 4D were detected in both treatments; all remaining QTL were N-treatment specific. In contrast to other studies [97,98], no QTL were detected for the trait RDW in this study. Only one QTL, part of the QTL hotspot linked to Rht-D1, was detected for SDW.

Root to Shoot Ratio (R:S)
Two of the four QTL were part of the QTL hotspots on 2A and 6D. The remaining two QTL on 5B and 6B were detected under N− and N+ and explained 7.61 and 14.92% of the variance, respectively. Candidate genes could not be identified yet.

SPAD
In addition to the QTL hotspot on chromosome 4D, another QTL on 4D at 211 Mbp showed a significant relative effect of 4.33% and explained 8.28% of the variance under N+. Upstream of this QTL are three genes coding for photosystem II reaction center proteins (IWGSC RefSeq v1.1: TraesCS4D02G155100.1, TraesCS4D02G155200.1 and TraesCS4D02G155500.1 [61]. The SPAD value provides information about the chlorophyll content of the plant and chlorophyll is a central component of the photosystem II reaction center, which renders these three genes potential candidates to explain the observed effect. One further QTL on chromosome 7A affected SPAD under N+ at 12 DAT. The significant HT derived from Meister, Linus and Safari reduced SPAD by 4.30%. In close genetic distance are two genes coding for peptide transporters (PTR) (IWGSC RefSeq v1.1: TraesCS7A02G381500.1 and TraesCS7A02G381800.1 [61]). PTRs transport peptides within the plant and, consequently, play a role in the distribution of N within the plant [99]. Genetic variation in a PTR coding gene could explain differences in SPAD values, which are closely linked to the N content in the leaves through differences in the distribution of N within the plant.

Leaf Length (LLE)
Only one further QTL on chromosome 3D affected LLE under N+ at 12 DAT other than the two hotspots linked to Rht-B1 and Rht-D1. Tobak is the only founder carrying the significant HT, which reduced LLE by 10.12%. A candidate gene (IWGSC RefSeq v1.1: TraesCS3D02G358900.2 [61]) codes for a O-fucosyltransferase, which is described in Arabidopsis as mono-O-fucosylate DELLA, enhancing its activity [100]. DELLA is a negative regulator of gibberellin signaling and, thus, of gibberellin-responsive growth [101], which could explain the observed effect on LLE.

Plant Material
Due to capacity limitations of the phenotyping platform, a randomly selected subset of 350 lines of the WM-800 (a winter wheat MAGIC population) was used. Eight modern German elite varieties (released between 2008 and 2017; Table S1) were crossed according to a crossing scheme adapted from Cavanagh et al., [102]. In the F 4:6 generation, 1323 recombinant inbred lines (RILs) were randomly selected in a first step. In a second step, all double-dwarf lines, carrying Rht-B1b and RhtD1b, were excluded and the final WM-800, consisting of 800 lines in generation F 4:7, was established. More details on the WM-800 population can be found in Sannemann et al., [47].

Plant Cultivation and Experimental Set-Up
Plants were grown and phenotyped using the GrowScreen-PaGe phenotyping platform ( Figure 3) [24,103], specializing in the non-invasive study of root morphology.

Leaf Length (LLE)
Only one further QTL on chromosome 3D affected LLE under N+ at 12 DAT other than the two hotspots linked to Rht-B1 and Rht-D1. Tobak is the only founder carrying the significant HT, which reduced LLE by 10.12%. A candidate gene (IWGSC RefSeq v1.1: TraesCS3D02G358900.2 [61]) codes for a O-fucosyltransferase, which is described in Arabidopsis as mono-O-fucosylate DELLA, enhancing its activity [100]. DELLA is a negative regulator of gibberellin signaling and, thus, of gibberellin-responsive growth [101], which could explain the observed effect on LLE.

Plant Material
Due to capacity limitations of the phenotyping platform, a randomly selected subset of 350 lines of the WM-800 (a winter wheat MAGIC population) was used. Eight modern German elite varieties (released between 2008 and 2017; Table S1) were crossed according to a crossing scheme adapted from Cavanagh et al., [102]. In the F4:6 generation, 1323 recombinant inbred lines (RILs) were randomly selected in a first step. In a second step, all double-dwarf lines, carrying Rht-B1b and RhtD1b, were excluded and the final WM-800, consisting of 800 lines in generation F4:7, was established. More details on the WM-800 population can be found in Sannemann et al., [47].

Plant Cultivation and Experimental Set-Up
Plants were grown and phenotyped using the GrowScreen-PaGe phenotyping platform ( Figure 3) [24,103], specializing in the non-invasive study of root morphology. Seeds of the 350 WM-800 lines, eight founders and four check varieties (Bonanza, Elixer, Genius, RGT Reform) were pre-germinated between two moist filter papers at 22 • C/18 • C (day/night) temperature in a darkened petri dish for 24h.
To prevent fungal infection, seeds were treated in advance with the fungicide prothioconazole (33 mL/100 kg seed). For each genotype, 40 seeds were pre-germinated and, after 24h hours, twelve uniformly germinated seedlings were selected and transferred, seminal root facing downwards, to the germination paper (size: 37 cm × 25 cm; smooth dark blue 194 grade paper; Ahlstrom Germany GmbH, Bärenstein, Germany) and attached with transparent self-adhesive tape (Opsite Flexifix, 7478029, Smith & Nephew GmbH, Hamburg, Germany). Using transparent shirt clips (shirt clip 38 mm, Georg Scharf GmbH, Balingen, Germany), a seedling on its germination paper was attached to each PVC plate (RAL 7011, Max Wirth GmbH, Braunschweig, Germany) on the front and back side [24].
The germination paper was then sprayed until saturation with a modified Hoagland nutrient solution with two different nitrogen concentrations for the different nitrogen treatments with N− = 0 µM Ca(NO 3 ) 2 × 4H 2 O and N+ = 2000 µM Ca(NO 3 ) 2 × 4H 2 O (for details see Table S2). The PVC plates were positioned vertically with the attached plants in opaque PVC containers, with a maximum of 25 plates spaced 2 cm apart. To further shield the roots from light, adjustable black strip brushes (Mink GmbH & Co. KG, Göppingen, Germany) were attached in front of the top opening of the box through which the shoot grows (for details see Pariyar et al., [24]). The containers were each filled with 12 L of modified Hoagland nutrient solution so that the bottom 5 cm of the germination paper was submerged in the solution. The nutrient solution was replaced seven days after transplanting from the petri dishes.
A total of 350 lines and eight parents were studied in two nitrogen treatments, with six replicates each arranged in a completely randomized block design. For capacity reasons, the plants were divided into eight experimental runs. Each experimental run consisted of two blocks with six boxes each (three N-and three N+ boxes). In each run, 46 lines (WM-800 and founder) and four checks were tested. The six repetitions of the 46 lines were divided among the six boxes. The four checks were included in each box and randomly placed. Plants were grown in a controlled growth chamber at 22 • C/18 • C (day/night) air temperature, 12 h/12 h light/dark, 60% relative air humidity and~150 µmol m −2 s −1 photosynthetically active radiation (PAR) at leaf level (Johnson Controls Systems & Service GmbH, Leipzig, Germany).

Phenotypic Data
High-resolution images (74 µm per pixel) of the root system were taken 3, 7 and 12 days after transplanting (DAT) to the germination paper using 16 MP cameras (IPX-16M3-G, monochrome, Imperx Inc.; combined with Zeiss Planar T 2.0/50 ZF-I lens) in a mobile imaging box, described in Gioia et al., [103]. At the last phenotyping time point, plants were harvested after imaging was finished. Leaf length and SPAD value of the plants were measured manually before roots and shoots were separated above the seed. Dry weights of roots and shoots were determined after drying plant materials at 65 • C in an oven until constant weight was reached.
For the analysis of the images, the workflow of the GrowScreen-PaGe system was used, which first segmented the roots from the background and labeled seminal and lateral roots based on their origin (from the grain or from the seminal roots, respectively) [24]. The segmentation and the labeling were manually checked in a second step and corrected if necessary. In the final step, the images were processed to quantify a total of ten root system architectural traits. Five root-and shoot-related harvest parameters, collected at 12 DAT, completed the overall picture (Table 1 and Table S3). In addition, tags were set (such as "not germinated", "damaged root", scored as true or false, Table S4) to assess the data quality and were subsequently used for data cleaning.
The 14 root-and shoot-related traits can be summarized into four trait groups. Seminal root length (SRL) and lateral root length (LRL) consider root length, while branching angle lateral (BAL), seminal root angle (SRA) and root system radius angle left (RRL) and right (RRR) describe the angle of root growth. Root system depth (RSD), root system width (RSW) and convex hull area (CHA) are composed of root system length and angle. Finally, harvest parameters' root dry weight (RDW), shoot dry weight (SDW), root to shoot ratio (R:S), SPAD and leaf length (LLE) were recorded 12 days after transplantation.

Statistical Analysis
Statistical analysis was performed with the software RStudio [104]. A first analysis of variance (ANOVA), modeling genotype, treatment and experimental run (batch) as fixed effects revealed significant batch effects, which is why we have normalized the data. In a first step, the least squares means (LSmeans) of the four check varieties across both N treatments and per batch were calculated using the package "emmeans" [105] and each WM-800 line value was divided by the mean LSmean of the checks of the corresponding batch. A mean LSmean was then calculated for the checks across the batches and both treatments, by which each normalized WM-800 value was multiplied. A second ANOVA after normalization revealed no more significant batch effects. Afterwards, ANOVA was recalculated to test for genotype and treatment effects and the two-way interaction effect.
In a first data cleaning step, all measurements that were scored as "true" for a tag that showed a significant effect on the trait, based on an ANOVA modeling genotype, treatment and tag as fixed effects, were discarded. In addition, outliers, where for instance the root length-related traits were close to zero (reflecting strong growth disorders), were removed.
Variance components were estimated separately for both N treatments using the package "blme" [106] with genotype modeled as a random effect. In a second step, a linear model containing genotype as fixed effect was used to estimate the standard error (SE) of the adjusted means, which, in the next step, was used for calculation of LSmeans and repeatability. Repeatability was estimated using the genetic variance and the standard error from the two previously described steps as described in the following formula [107]: where σ 2 G is the genotypic variance and SE 2 is the squared standard error of the difference between the LSmeans.
Then LSmeans for each genotype of the WM-800 over the six replications were calculated and the Spearman correlation coefficients between the traits and between the treatments were calculated using the package "psych" [108].

Genotypic Data
The 800 WM-800 lines, their eight founders and the four checks were genetically characterized with the Infinium 15K iSelect SNP array [47] and the 135k Affymetrix SNP array [46], both from SGS TraitGenetics (Gatersleben, Germany), using bulked DNA from 12 F 4:5 seedlings per genotype.
All SNPs polymorphic in WM-800 have undergone a quality check (SNP calls for all founders, <5% missing calls, >5% minor allele frequency [109] and a known physical position in the wheat genome), of which 27,006 polymorphic SNPs passed. The physical positions of the SNPs, anchored to the Refseq v1.0 reference genome sequence of Triticum aestivum [61], were provided by SGS TraitGenetics (Gatersleben, Germany). For the SNPs of the 15K iSelect SNP array, genetic positions were provided and for the remaining SNPs, the genetic position was assigned to the wheat consensus map [62] using LD mapping as described in [110].
To perform regression analysis, the 27,006 polymorphic SNPs were transcribed into a numerical matrix based on identity by state (IBS) according to the Julius founder allele [47]. All genotypes with the same homozygous status as Julius were coded as 2, all genotypes showing a homozygous Non-Julius allele were coded as 0 and heterozygous genotypes were coded as 1. In a last step, missing genotype calls were imputed using the mean imputation approach (MNI), which replaces each missing SNP value by the mean value of the remaining genotypes [111].

Haplotype Building
The software Haploview 4.2 [112] was used to create a haplotype (HT) matrix for the WM-800 population, made of SNPs in high LD. For this, all SNP genotypes of the eight founders with a minor allele frequency MAF > 0.05 were used. SNP pairs with a distance of <500 kb and consecutive SNPs were combined in a haploblock (HB) if (i) at least one out of the four possible gametes was observed with a frequency of <0.01 and (ii) a strong LD of D' = 1 was estimated between the SNP pairs. SNPs that were not included in a HB were retained as so-called "singular SNPs".
A total of 2970 HBs with 92,734 HTs have been identified. A total of 8498 of these HTs met the quality criteria of (i) HT frequency > 5% in WM-800 and (ii) HT sequence without missing nucleotides. These HTs were supplemented with 4562 singular SNP alleles that could not be assigned to any HB. Finally, the genotype scores were coded into a 0-1 (absence-presence) matrix to enable GWAS in a multiple linear regression framework (see [113], File D).

GWAS
GWAS was performed using SAS 9.4 software (SAS Institute Inc., Cary, NC, USA, 2016). In the first step, all SNP alleles and HTs associated with the targeted trait were selected using a multiple linear regression model (SAS PROC GLMSELECT). This was accomplished by creating 100 repeated subsamples, each containing 80% of the lines, and selecting only the SNP alleles and HTs that improved the prediction of the remaining 20% (corresponding to the minimum mean squared error). Those SNP alleles and HTs that were selected more than once in this step were defined as potential cofactors. Potential cofactors were then used as input for the final selection of cofactors (SAS PROC GLMSELECT based on minimizing the Schwarz-Bayesian criterion) in the complete dataset. Selected cofactors were then modeled using SAS PROC REG in the background as part of a multiple linear regression model in which all SNP alleles and HTs were tested for significance. HTs and SNP alleles with a −log10 Bonferroni corrected p value < 0.05 were selected as significant. Thus, allele effects, R 2 and Bonferroni corrected p-value were estimated as a function of the cofactors that entered the model first according to their ranking in the previous step by applying the PARTIALR2 (SEQTESTS) model option. A relative QTL effect was calculated by dividing the absolute estimated effect by the corresponding population mean. If a QTL or two or more linked QTL showed an effect on more than one trait they were classified as QTL hotspots. Note that the first four principle components (explaining more than 5%), presented in Lisker et al., 2022 [46], were included in the model in all steps mentioned above to correct for potential population structure effects.

Candidate Genes
For the candidate gene search, the physical positions of the HBs and singular SNPs anchored to the Refseq v1.1 reference genome sequence of Triticum aestivum [61] were used. The gene annotation from the IWGSC was used to identify genes either within the significant HB or upstream and downstream until reaching the neighboring HBs (complete list of all genes within the respective regions can be found in Table S7). Subsequently, to define candidate genes, these genes were screened for (i) their function, (ii) expression profiles using the wheat RNA-seq expression database of polyploid wheat (http://www. wheat-expression.com/ (accessed on 15 August 2022)) and (iii) information given in the literature. Candidate genes, for which all three criteria fit, were discussed in more detail in the text.

Conclusions
Root growth during the first days after germination provides an important foundation for plant establishment and later plant development. Both the finding of significant main effects for genotype and N treatment as well as genotype by N treatment interactions and the detection of numerous N specific QTL confirm the effectiveness of the measurement method and the dependence of the regulation of RSA traits on N availability. These findings are promising for breeding towards an increased nitrogen uptake efficiency by allowing plants to adapt well to marginal soils and to lower N input levels by an optimized root system.
The GWAS revealed 64 significant QTL. Due to the use of a haplotype-based GWAS approach, the effects could be assigned to the individual parents of the MAGIC population. Of these QTL, 24 were N− specific, 36 N+ specific and only 2 QTL were found active in both N treatments. In this context, at 13 QTL hotspots, pleiotropic effects were observed. In addition to the prominent Rht genes, Rht-B1 and Rht-D1, candidate genes under N− were, among others, NRT1.1 and genes involved in stress signaling and hormone regulation. In contrast, candidate genes under N+ were more associated with regulation of general growth, examples being mei2 and TaWOX11b.
The clear trend that more N specific than N non-specific QTL were detected suggests the differential genetic regulation of traits as a function of N availability. This is an indication that a direct selection for tolerance of reduced N availability under N deficiency is more effective in breeding than indirect selection under standard N availability.
To further investigate the regulation of RSA traits and the effects of RSA on yield and quality traits under differing N availability under field conditions, heterogeneous inbred families (HIFs) [114,115] can be generated easily from WM-800 lines that were heterozygous at the loci of interest in generation F 4 . The concept of HIFs allows QTL validation and fine mapping to ultimately clone and verify the responsible genes via transformation or knock out.
Deciphering the genetic regulation of RSA under N stress and integrating the traitimproving QTL alleles in modern plant breeding programs might be an important step in the selection of stress-adapted varieties that will allow to feed the human population under increasing requirements for environmental sustainability and resource protection.