Genetic Screening for Mutants with Altered Seminal Root Numbers in Hexaploid Wheat Using a High-Throughput Root Phenotyping Platform

Roots are the main channel for water and nutrient uptake in plants. Optimization of root architecture provides a viable strategy to improve nutrient and water uptake efficiency and maintain crop productivity under water-limiting and nutrient-poor conditions. We know little, however, about the genetic control of root development in wheat, a crop supplying 20% of global calorie and protein intake. To improve our understanding of the genetic control of seminal root development in wheat, we conducted a high-throughput screen for variation in seminal root number using an exome-sequenced mutant population derived from the hexaploid wheat cultivar Cadenza. The screen identified seven independent mutants with homozygous and stably altered seminal root number phenotypes. One mutant, Cadenza0900, displays a recessive extra seminal root number phenotype, while six mutants (Cadenza0062, Cadenza0369, Cadenza0393, Cadenza0465, Cadenza0818 and Cadenza1273) show lower seminal root number phenotypes most likely originating from defects in the formation and activation of seminal root primordia. Segregation analysis in F2 populations suggest that the phenotype of Cadenza0900 is controlled by multiple loci whereas the Cadenza0062 phenotype fits a 3:1 mutant:wild-type segregation ratio characteristic of dominant single gene action. This work highlights the potential to use the sequenced wheat mutant population as a forward genetic resource to uncover novel variation in agronomic traits, such as seminal root architecture.

The wheat root system is comprised of two main root types, seminal (embryonic) and nodal (post-embryonic) roots, that develop at different times (Manske and Vlek 2002). As the first root type that emerges, seminal roots are entirely responsible for nutrient and water uptake in seedlings. Seminal roots are therefore important for seedling vigor and early plant establishment which also determines competitiveness against weeds. Nodal roots on the other hand are shootborne and develop soon after tillering to provide anchorage and support resource uptake especially during the reproductive stage of wheat growth (Manske and Vlek 2002).
Despite their early establishment, seminal roots remain functionally active through to the reproductive stage and may grow up to 2 m in length (Manschadi et al. 2013;Araki and Iijima 2001). They have also been shown to have similar nutrient uptake efficiency as nodal roots in wheat and contribute to yield potential especially under conditions of low soil moisture where nodal roots may not grow (Weaver and Zink 1945;Sebastian et al. 2016). Given their importance, seminal root traits, such as number and angle, have been linked to adaptive responses under water limiting conditions (Manschadi et al. 2008;Canè et al. 2014;Golan et al. 2018). Steep seminal root angle has been associated with increased soil water exploration at depth which is beneficial in drought condition where topsoil moisture is depleted (Richard et al. 2015;Olivares-Villegas et al. 2007;Manschadi et al. 2008).
Seminal roots develop from the root primordia in the embryo of a germinating wheat seed. There is genetic variation among wheat genotypes for the number of seminal roots that develop, this can range from three to six seminal roots per plant dependent on cultivar (Araki and Iijima 2001;Robertson et al. 1979;Golan et al. 2018). Typically, the seminal root system consists of a primary root that emerges first followed by two pairs of secondary seminal roots that emerge sequentially. A sixth seminal root may also develop in some wheat varieties. A few quantitative trait loci (QTL) have been identified to underlie variation in seminal root number in wheat germplasm (Atkinson et al. 2015;Maccaferri et al. 2016;Ren et al. 2012;Sanguineti et al. 2007;Ma et al. 2017;Iannucci et al. 2017). However, many of these QTL are defined to broad intervals which makes their genetic dissection difficult and their use in breeding limited. Unlike other cereals (e.g., rice, maize), only one gene controlling root system architecture (RSA), VERNALIZATION1 (VRN1; Voss-Fels et al. 2018), has been identified in wheat.
This delay in identifying genetic loci controlling root traits is most likely due to a series of factors which makes genetic analyses in wheat difficult. Bread wheat is a hexaploid plant with a relatively large (16 Mb) and repeat-rich (.85%) genome comprised of three homeologous subgenomes (B, A and D). High sequence similarity in the coding regions of these sub-genomes results in high levels of genetic redundancy that mask the phenotypic effects of underlying natural variation for many traits, including RSA traits (Uauy et al. 2017;Borrill et al. 2015). Also, the "out-of-sight" nature and extreme phenotypic plasticity of roots under native field conditions makes root phenotyping difficult, cumbersome and time-consuming (Atkinson et al. 2019).
The use of induced variation has proven useful to uncover novel phenotypes and dissect genetic pathways underlying complex phenotypes in plants (Parry et al. 2009). Our current understanding of the genetic determinants regulating root development in many cereals have almost entirely stemmed from the isolation and characterization of mutants defective in one or more RSA traits (Reviewed in Coudert et al. 2010;Hochholdinger et al. 2018;Marcon et al. 2013). Despite this potential, the use of mutant populations to study the genetic control of root development in wheat has not hitherto been exploited. The recent development of an in-silico platform for the rapid identification of mutations in 1,200 ethyl methanesulfonate (EMS) mutagenized lines in the UK hexaploid wheat cultivar 'Cadenza' now makes large-scale reverse and forward genetic investigation of traits more feasible in wheat (Krasileva et al. 2017). Progress has also been made on the root phenomics front, with the development of fast, low-cost, and flexible two-dimensional (2D) root phenotyping pipelines with sufficient throughput for phenotyping large populations (Selvara et al. 2013;Atkinson et al. 2019;Adeleke et al. 2019).
Taking advantage of these new developments, we implemented a relatively high-throughput root phenotyping pipeline to conduct a forward genetic screen for variation in seminal root number using a subset of the exome-sequenced Cadenza mutant population. From this work, we describe the identification and characterization of novel hexaploid bread wheat mutants with decreased and increased numbers of seminal roots.

Germplasm
Mutant population for primary screens: A hexaploid wheat mutant population was previously developed by EMS treatment of the UK bread wheat cultivar Cadenza (Krasileva et al. 2017;Rakszegi et al. 2010). In this study, we used 645 exome-sequenced mutants from this population (Krasileva et al. 2017). These mutant lines were selected based on the criteria that they show greater than 90% germination rate during a seed multiplication that was conducted in the field. To obtain homogenous phenotypes and reduce the variation from segregating mutations, single spikes harvested from field-grown M 4 plants were individually threshed and derived M 5 seeds (40 -50 seeds) from each spike were divided for use in forward genetic Screen A and Screen B described below (Figure 1).
Germplasm for mutant validation and characterization: Three additional M 4 spikes (containing M 5 seed) from independent spikes, but from the same field grown samples as those used for the primary screens (Figure 1), were used to further validate the phenotypes of the seven mutants identified in the primary screens (described in results). The spikes were individually threshed and M 5 seeds from each spike were phenotyped separately. Two validated mutants (Cadenza0900 and Cadenza0062) were further selected for genetic characterization: M 5 plants were grown to maturity for cross-pollination with wild-type Cadenza to generate F 1 hybrids which were subsequently self-pollinated to generate F 2 progenies. M 5 plants of the selected mutants were also self-pollinated to generate M 6 seeds to characterize the stability of their phenotypes in the subsequent generation.
High-throughput seminal root phenotyping Two similar and independent screens (Screen A and B) were conducted concurrently on subsets of the Cadenza mutant population to identify lines with altered seminal root numbers ( Figure 1). All 645 lines were phenotyped at high-throughput in Screen A, while only the first 385 of the 645 lines (in numerical order) were phenotyped in Screen B due to more limited throughput. For mutant lines phenotyped in both screens, M 5 seeds from the same field-grown M 4 spike were used in both screens as described in the germplasm section above.
Screen A: This screen was carried out at the John Innes Centre, UK, using a custom 2D root phenotyping platform based on the protocol described by Atkinson et al. (2015) with some modifications to increase the throughput from 360 to 1,800 seedlings per run. This screen also took seed size effect on RSA traits into consideration. For each mutant line, M 5 seeds from a M 4 single spike were first size-stratified into large, medium or small seed by passing the seeds through two sets of calibrated graduated sieves with 2.8 mm and 3.35 mm mesh sizes. This allowed us to examine the relationship between grain size and root architecture. Large-sized seeds were collected above the 3.35 mm sieve, medium-sized seeds collected between the 2.8 mm and 3.35 mm sieves, and small-sized seeds were collected below the 2.8 mm sieve. Each mutant line was thus designated as either being Large (524 lines), Medium (119 lines) or Small (2 lines) based on the size group where most of its seeds were defined. Only seeds representative of each mutant line size classification were used. The two mutants with small seed size were phenotyped for variation in seminal root number compared to wildtype Cadenza, but were not included in the analysis of seed size effect on RSA due to the small sample size. Seeds (15 per mutant line) were surface sterilized by rinsing in 5% (v/v) Sodium Hypochlorite (Sigma Aldrich, UK) for 10 min and were rinsed with water three times before being imbibed in 1.75 mL of water for 5 days at 4°to ensure uniform germination. Of these, 10 seeds per mutant were placed crease facing down into individual growth pouches made from a sheet of germination paper (21.5 cm · 28 cm; Anchor Paper Company, St Paul, MN, USA) clipped to a black polythene sheet (22 cm · 28 cm, 75 mm thickness, Cransford Polythene LTD, Suffolk, UK) using an acrylic rod and 18 mm fold clip ( Figure 2A). The growth pouches were suspended in an upright position in plastic storage boxes (120 cm · 27 cm · 36 cm, Really Useful Product, West Yorkshire, UK) with 60 pouches per box ( Figure 2B). The sides of the box were covered in black plastic sticky back cover film to block out light from the roots of the developing seedlings. Each box was filled with 10 L of half-strength Hoaglands growth solution containing (Hoagland and Arnon 1950): NH 4 H 2 PO 4 , 0.6 g; Ca(NO 3 ) 2 , 3.3 g; MgSO 4 , 1.2 g; KNO 3 , 1.0 g; H 3 BO 3 , 14.3 mg; Cu 2 SO 4 , 0.4 mg; MnCl 2 (H 2 O) 4 , 9.1 mg; MoO 3 , 0.1 mg; ZnSO 4 , 1.1 mg; KHCO 3 , 2.0 g, Ferric Tartrate, 2.8 g. The base of each pouch was suspended in the growth solution to supply nutrients to the developing seedling through capillary action. A randomized complete block design was adopted with each line replicated across 10 different boxes (blocks). The phenotyping boxes were placed in a controlled environment room under long day conditions with 16h light (250-400 mmol) at 20°, 8h darkness at 15°and at 70% relative humidity ( Figure 2C). After 7 days of growth, pouches were taken out of the phenotyping box; placed on a copy stand and the black plastic sheet covering the germination paper was gently pulled back to reveal the roots. Images of the roots were taken with a Nikon D3400 DSLR Camera fitted to the copy stand ( Figure 2D). Phenotyping of the mutant population was done over 12 experiments with 60 lines (59 mutants and a Cadenza control) phenotyped per experiment. Mutants of the same seed-size group were phenotyped togetherlarge sized mutants in experiment 1 -9 and small/medium sized mutants in experiment 10 -12. Cadenza seeds representative of the seed-size group for each experiment were used as controls. The 12 experiments were completed across five rounds of phenotyping with 2 -3 experiments (120 -180 lines) set up per round. For 96% of the mutants examined, we successfully imaged the roots of 8 -10 plants. In the remain 4% of the mutants, only 3 -7 plants could be images due to poor seed germination or poor seedling growth on the pouch. In total, 6,240 (6,127 mutant and 113 Cadenza) plants were phenotyped. The same phenotyping set-up was used for the M 5 validation experiments and to characterize M 6 , F 1 and F 2 progenies of the two mutants selected for genetic characterization.
Screen B: A second screen was conducted in parallel at the University of Leeds, UK, using commercial CYG seed germination pouches (Mega-International, Minnesota, USA). Due to the more limited throughput of this screen, only the first 385 of the 645 mutant lines (in numerical order) were examined in this screen. For each mutant line, 10 visually uniform M 5 seeds were selected and placed onto moist filter paper in a 90 mm round petri dish. Petri dishes were wrapped in aluminum foil to exclude light and placed at 4°for 2 days, seeds were placed crease side down into individual CYG seed germination pouches with the bottom removed to allow wicking of growth solution from a reservoir of media. Pouches were wrapped in aluminum foil in batches of five to exclude light from the roots and were placed upright in a reservoir of full strength Hoaglands No 2 growth solution (NH 4 H 2 PO 4 , 115.03 mg; Ca(NO 3 ) 2 , 656.4 mg; MgSO 4 , 240.76 mg; KNO 3 , 606.6 mg; H 3 BO 3 , 2.86 mg; Cu 2 SO 4 , 0.08 mg; MnCl 2 (H 2 O) 4 , 1.81 mg; MoO 3 , 0.016 mg; ZnSO 4 , 0.22 mg; Ferric Tartrate, 5 mg. per Liter). Pouches were placed in long day conditions (as in screen A) and were grown for 5 days before roots were imaged. For imaging, pouches were placed onto a copy stand; the front of the pouch carefully removed, and the root system imaged using a Sony Cybershot DSC-RX100. Plants were screened in rounds of twenty lines with a total of 200 plants per round.

Image analysis
High-resolution images captured from the phenotyping were preprocessed (rotated, cropped and compressed) using ImageJ (https:// imagej.nih.gov/index.html) and Caesium image compressor before being processed in RootNav (Pound et al. 2013). Captured root architectures were imported into a RootNav viewer database for measurement of RSA traits using standard RootNav functions ( Figure 2E).

Anatomical characterization of seminal root primordia
Embryos of mutants validated to have reduced seminal root number phenotypes were examined using the method described by Golan et al. (2018). In brief: embryos from mature dry grains were fixed in FAA solution (10% formaldehyde, 5% acetic acid, 50% ethanol, and 35% distilled water by volume) overnight and dehydrated at room temperature in a graded ethanol series (30 min each, in 50%, 70%, 90%, 95%, and 100% ethanol). Then, embryos were cleared in xylene, embedded in paraffin and sectioned (5 mM) using a microtome (Leica Biosystems, Germany). Cross sections were de-paraffinized with histoclear, rehydrated and stained with Harris Hematoxylin. A stereo microscope (SZX16, Olympus, Tokyo, Japan) was used for imaging.

Statistical analysis
All statistical analyses were performed in R 3.5.1 (R Core Team 2018) and Minitab 17 statistical software. Statistically significant seminal root number difference in the primary screening experiments (Screen A and B) was determined by ANOVA using a Dunnett's comparison within each phenotyping experiment with the Cadenza plants in each experiment used as control. Adjusted probability values of P , 0.05 were considered statistically significant. We estimated broad-sense heritability (H 2 ) for seminal root number across Screen A and B using the formula: H 2 = sg 2 / (sg 2 + se 2 /n) where sg 2 represents the genotypic variance component, se 2 represents the residual variance component and n is the number of replications. Statistically significant root architectural difference in the validation experiment as well as in the M 6 and F 1 phenotyping experiments were based on Student's t-test comparison of individual spike/line to the Cadenza control. A Chi-square test was used to examine the goodness of fit of the segregation pattern observed in the F 2 progenies to patterns expected for single recessive or dominant gene action.

Data availability
Seeds of mutant lines reported in this study can be ordered through the SeedStor site at www.seedstor.ac.uk. Table S1 contains information on mutants with significantly different seminal root number phenotypes to Cadenza from Screen A. Table S2 contains information on all the mutants phenotyped in both Screen A and B. Figure  S1 shows the relationship between seminal root number and total root length in Screen A. Figure S2 and Figure S3 show representative images and embryo size measurement of mutants with validated altered seminal root number phenotypes, respectively. All the root images from Screen A (6,240 images) including the original RootNav measurements for different root traits are publicly available on Zenodo (https://doi.org/10.5281/zenodo.3270726) for download and reuse. Supplemental material available at FigShare: https://doi.org/10.25387/g3.8796785.

Identifying induced variation for seminal root number in hexaploid wheat
We implemented a 2D root phenotyping pipeline suitable for large-scale phenotyping at a throughput of 1,800 seedlings per run (Figure 1 and 2). Using this platform, we performed a forward genetic screen for variation in seminal root number using 645 seed-size stratified (small, medium and large; see Methods) M 5 mutants from the exomesequenced Cadenza mutant population (Krasileva et al. 2017). In our screen, Cadenza mainly displayed five seminal roots (4.9 6 0.05) including a primary seminal root (SR 1 ), as well as first (SR 2,3 ) and second (SR 4,5 ) pairs of seminal roots ( Figure 3A). We observed variation in seminal root number in the mutant population, with seminal root number ranging from 1 to 7 in individual plants and mean seminal root number per mutant (with n $ 4 plants per mutant) ranging from 2.9 to 5.9.
Within the mutant lines, seed size groups (large and medium) showed significant difference in seminal root number (P , 0.0001) and total root length (P , 0.0001). Mutants with large and mediumsized grains had an average seminal root number of 4.7 and 4.4, respectively, and total root length of 481 mm and 324 mm, respectively ( Figure 3B-C). In the Cadenza wild-type, significant differences between seed size groups were only observed for total root length. There were only two mutant lines with small seed size and these were not included in the analysis of seed size effect on seminal root traits. Across all mutant lines, we observed a significant positive correlation (R = 0.56, P , 0.0001) between the number of seminal roots and the total root length in the population ( Figure S1).
Parallel to this, we phenotyped a subset of these Cadenza mutants for similar root traits in an independent screen (Screen B; Figure 1). Given the lower throughput of this approach (See Screen 2 in Methods), only 385 out of the 645 mutant lines were screened. We observed a significant positive correlation between seminal root number measurements in Screen A and Screen B (R = 0.63; P , 0.0001; Figure 3D). The heritability estimate of the seminal root measurement across the two screens was 0.77 suggesting a strong heritable genetic effect in the determination of seminal root number in the Cadenza mutant population.
We first assessed the statistical significance from the mutants studied in Screen A. Dunnett's multiple comparison identified 52 mutants (8% of mutants in Screen A) with significantly different number of seminal roots relative to the Cadenza control across the three seed size groups (Table S1). Five of these mutants had significantly higher number of seminal roots with mean seminal root number ranging from 5.7 to 5.9 and modal seminal root number of 6 per mutant (Table S1). The higher seminal root phenotype is mainly driven by the development of an extra root, hereafter referred to as SR 6 . The remaining 47 mutants showed significantly lower number of seminal roots with mean seminal root number per mutant of 2.9 to 4.1 and modal seminal root number between 3 and 5.
Thirty three out of the 52 significant mutants identified in Screen A were also phenotyped as part of Screen B. This included four of the five higher root number mutants and 29 lower root number mutants. Details of individual mutant lines phenotyped in both screens are presented in Table S2. In Screen B we confirmed the significant phenotype of three of the four higher root number mutants in common with Screen A; these lines displayed a mean seminal root number of 5.1 to 5.5 in Screen B. In the case of the 29 lower root number mutants which were investigated in both screens, we confirmed 20 mutants which displayed fewer numbers of seminal roots (less than 4 roots) than Cadenza. Based on the results from the two screens, we selected the three higher and 20 lower root number mutants with consistent phenotypes in both screens for further phenotypic evaluation.

Altered root number mutants show stable homozygous seminal root number phenotypes
To validate the 23 selected mutants, we phenotyped M 5 seeds from three additional M 4 spikes (containing M 5 seeds) from the same field-grown bulk as the spike used in the primary screen ( Figure 1). These M 4 spikes originate from successive bulking of multiple M 3 and M 4 plants. Selecting three separate spikes increases the probability of phenotyping plants with independent background mutations thereby providing robust biological replications to examine the stability of the mutations effects and segregation patterns (homozygous or heterozygous).
For seven of the selected mutants including one higher root number (Cadenza0900) and six lower root number mutants (Cadenza0062, Cadenza0369, Cadenza0393, Cadenza0465, Cadenza0818, and Cadenza1273), we observed the altered seminal root number phenotype in the three additional spikes phenotyped (Figure 4, Figure S2). This suggests that the phenotypes of these mutants are consistent across sibling lines and controlled by mutations that were most likely homozygous in the original single M 2 plant from which the population was derived. For the rest of the 16 mutants, we did not consistently observe the altered seminal root number phenotypes in all three additional spikes. These might represent lines with segregating phenotypes arising from heterozygous mutations in the initial M 2 plants or false-positive selection in the primary screens; these lines were not studied further.
We further characterized the seven validated mutants from 1 to 7 days post germination (dpg) to examine when the phenotype was first visible and to identify the seminal root type (primary, first pair or second pair) defective in these mutants ( Figure 5). Cadenza showed fully emerged primary root (SR 1 ) at 1 dpg, while the first (SR 2,3 ) and second pairs (SR 4,5 ) of seminal roots emerged at 3 and 5 dpg, respectively. Similar to Cadenza, Cadenza0369, Cadenza0393, Cadenza0465, Cadenza0818, and Cadenza1273 with lower root counts also developed the primary and first pair of seminal roots at 1 dpg and 3 dpg, respectively, but are defective in the development of the second pair of seminal roots (SR 4,5 or only SR 5 ) . Contrary to this, Cadenza0900 with higher root count showed a faster rate of seminal root development relative to Cadenza with the primary , first and second pair of seminal roots having emerged by 3 dpg and an extra sixth root emerged at 7 dpg. Cadenza0062 showed a strong dormancy phenotype and was not included in this experiment.
Embryo imaging of the lower root number mutants points to defects in primordia development and growth Seminal roots emerge from the activation of root primordia that form in the embryo of a developing seed. The primary (SR 1 ) and secondary (SR 2,3 and SR 4,5 ) seminal roots emerge from primordia formed in the central portion and sides of the embryo, respectively. Although uncommon, the sixth seminal roots (SR 6 like in Cadenza0900) are known to emerge from a primordium located in the midpoint between the primordia of the second pair of secondary seminal root (Hoshikawa 1964). Importantly, differential activation of root primordia of SR 4,5 has been shown to account for lower numbers of seminal roots in some wild wheat species (Golan et al., 2018). We therefore examined if the seminal root phenotypes of the lower root number mutants (Cadenza0062, Cadenza0369, Cadenza0393, Cadenza0465, Cadenza0818, and Cadenza1273) lacking either SR 5 or SR 4,5 originate from defects in primordia development and/or the failure of developed root primordia to activate to become seminal roots. Cadenza (WT) consistently develops five fully formed root primordia ( Figure 6). All the lower root number mutants examined showed altered root primordia development compared to Cadenza, with SR 4 , 5 (or only SR 5 ) primordia either absent or reduced in size ( Figure 6). In addition, primordia activity was altered in the mutants, as all mutants had greater number of primordia compared to the number of roots observed in the M 5 seedlings ( Figure 5, Table 1). These measurements provide an initial indication that the defects in the lower root count mutants most likely originate from a combination of both lower number and activity of root primordia in these lines. Embryo size was also significantly lower in four of the six mutants (Cadenza0393, Cadenza0465, Cadenza0818, and Cadenza1273; Figure S3), but it is not clear from these results if the smaller embryo size of these mutants contributes to their lower seminal roots number phenotypes.

Genetic characterization of the altered root number phenotypes
To understand the transgenerational stability and mode of inheritance of the altered root number mutant phenotype, we characterized M 6 plants of the higher root number mutant (Cadenza0900) and one lower root number mutant (Cadenza0062), as well as F 1 hybrids derived from crosses of these mutants to Cadenza. M 6 progenies of Cadenza0900 showed significantly (P , 0.0001) increased number of seminal roots compared to Cadenza and no significant difference to the phenotypes of it M 5 parents (Figure 7) with an average root number of 5.73 and more than 73% of the plants having six roots. F 1 hybrids of Cadenza0900 x Cadenza all had five seminal roots like Cadenza (P = 0.19), but significantly lower than their M 5 plants (P , 0.0001; Figure 7). This suggests that the Cadenza0900 phenotype originates from a recessive mutation or is caused by a combination of loci segregating independently.
The M 6 plants of Cadenza0062 showed a significantly lower number of seminal roots (P , 0.0001) compared to Cadenza with a mean root number of 2.6 ( Figure 7). The seminal root numbers of the M 6 progenies of Cadenza0062 were also slightly but significantly (P = 0.04) lower than the M 5 plant, most probably due to smaller variation in the seminal root number phenotype in the M 6 lines. With an average root number of 3.0 (Figure 7), the F 1 hybrid of Cadenza0062 x Cadenza showed a significantly (P , 0.0001) lower number of seminal roots compared to Cadenza but a nonsignificant difference to the original M 5 plants (Figure 7). This suggests that the Cadenza0062 phenotype is caused by a dominant mutation(s). Unlike the M 5 plants, the Cadenza0062 x Cadenza F 1 did not show any reduced germination, suggesting that the dormancy phenotype of Cadenza0062 segregates independently of its altered root number.
To gain further insight into the genetic architecture underlying higher and lower root count phenotypes, we further characterized F 2 progenies of the Cadenza0900 and Cadenza0062 crosses to Cadenza. We used the chi-square test-statistic to test the goodness of fit of the inheritance pattern of Cadenza0900 and Cadenza0062 phenotypes to those consistent with segregation of single recessive and single dominant traits, respectively. The phenotype of Cadenza0900 F 2 progenies (238) was not consistent with the expected 3:1 wild-type:mutant phenotype segregation ratio of a single recessive gene (x 2 = 52.71, P , 0.0001), suggesting that multiple genes may be responsible for the Cadenza0900 phenotype. In contrast, the segregation pattern of the Cadenza0062 F 2 population (51 plants) is consistent with the 3:1 mutant:wildtype segregation ratio expected of a single dominant gene (x 2 = 0.53, P = 0.4669), suggesting that the Cadenza0062 phenotype is caused by a single dominant gene.

DISCUSSION
Use of sequenced mutant population to characterize RSA genes in wheat Mutant analyses have played a key role in the identification of genes controlling key stages of root development. For instance, most of the genes identified to control root architecture in maize were identified via mutant analyses (Hochholdinger et al. 2018;Marcon et al. 2013). These include RTCS, RTCL, RUM1 and BIGE1 which display seminal root phenotypes (Taramino et al. 2007;Suzuki et al. 2015;von Behrens et al. 2011;Xu et al. 2015). Despite the buffering effect of genetic redundancy that often masks single homeolog mutations in polyploid wheat (Borrill et al. 2015), our study highlights the usefulness of forward genetic screens to identify heritable variation for root development traits in wheat. These results are also consistent with recent examples of dominant mutations being identified in forward screens of the exomesequenced populations (Harrington et al. 2019;Mo et al. 2018).  The use of a sequenced mutant population in this study also provided the opportunity to examine the presence of mutations in candidate genes from other species. For example, the phenotypes of the lower root number mutants (Cadenza0062, Cadenza0369, Cadenza0393, Cadenza0465, Cadenza0818, and Cadenza1273) are similar to the phenotypes of maize rtcs and rtcl mutations (Taramino et al. 2007), and their orthologous rice mutants (Liu et al. 2005;Inukai et al. 2005).
In-silico examination of the coding regions in these mutants on the Ensembl Plant database revealed that Cadenza1273 contains a functional mutation in TraesCS4D01G312800, one of the three wheat homeologs of RTCS, RTCL and ARL1/CRL1. Cadenza1273 harbors a G765A mutation in TraesCS4D01G312800 producing a premature termination codon which results in a truncated 265 amino acid (aa) protein instead of the 289 aa native protein. Further molecular and genetic characterization will be required to test if the G765A mutation in Cadenza1273 is responsible for its phenotype. However, this exemplifies the power of combining the sequenced mutant information with known candidate genes and a now fully annotated wheat genome (Appels et al. 2018). The other lower root number mutants do not contain any functional EMS mutations in the three wheat homeologs of RTCS, RTCL and ARL1/CRL1 and therefore likely represent new variation controlling seminal root development in cereals.
Similarly, the higher seminal root number phenotype of Cadenza0900 is similar to the phenotype of maize bige1 (Suzuki et al. 2015) mutants. However, unlike bige1 which is monogenic, we show that the phenotype of Cadenza0900 is most likely conditioned by more than one mutation. Also, in-silico examination of mutations in Cadenza0900 show that it does not harbor any mis-sense or non-sense mutations in the coding sequences of the three wheat orthologs of BIGE1 (TraesCS4A02G350200, TraesCS5B02G522900 and TraesCS5D02G521600) thereby reducing the likelihood that the Cadenza0900 phenotype originates from mutations of the wheat BIGE1 gene. It is, however, important to note that these in-silico investigations are restricted to mutations in the coding region of the wheat genome and we cannot rule out that mutations in promoter regions of these candidate genes might be responsible for some of the altered root number mutants identified.
Relationship between grain size and seminal root traits Size stratification of seeds in our study allowed an examination of the relationship between grain size and root architecture. We observed a positive effect of grain size on root length and number. Irrespective of the genetic background (wild-type Cadenza or mutant background), largesized grain showed longer root length compared to medium-sized grain consistent with the rationale that bigger grains have greater carbohydrate reserves in their endosperm to support faster root elongation. We also noticed a weak but positive effect of grain size on the number of seminal roots developed in the mutant population. Grain size effect on seedling traits, including seminal root traits, can be attributed to constituent componentsembryo and/or endosperm (Bremner et al. 1963;Meyer 1976). Interestingly, four of the lower root number mutants (Cadenza0393, Cadenza0465, Cadenza0818 and Cadenza1273) have smaller embryos compared to Cadenza. However, without further genetic analyses, it is premature to conclude that the small embryo of these mutants directly affects their seminal root number phenotypes. Indeed, grain size only accounts for a small proportion (3.4%) of the total variance in root number in our study, indicating that seed size per se is not a major determinant of root number. This is further underlined by the fact that differently sized wild-type Cadenza seeds show similar root number averages ( Figure 3B).
Although informative, the qualitative stratification of grain size (large, medium and small) adopted in this study does not allow a quantitative modeling of grain size effect on root traits. We propose that a finer calibration and partitioning of grain size measurement into constituent parameters (width, length, height) and tissue (embryo and endosperm) components (Brinton and Uauy 2019) will allow for a finer understanding of the effects of these seed size components on root architecture.

High-throughput platforms enables populationscale screening
The high-throughput afforded by the 2D platform used in Screen A in this study enabled the screening of a large mutant population (645 lines). Based on a previous 2D phenotyping pipeline implemented by Atkinson et al. (2015), our platform increases the throughput from 360 to 1,800 seedlings while maintaining a similar phenotyping rate ($4.5 min per plant). A similar high-throughput phenotyping platform was also recently adapted by (Adeleke et al. 2019) with a maximum phenotyping capacity of 672 plants. It is also worth noting that our high-throughput platform is relatively inexpensive and easy to implement using widelyavailable low-cost materials: storage boxes, germination paper, plastic sheets and paper binder clips. The entire phenotyping pipeline costed £5,570 to phenotype 6,240 plants: £3,280 to set-up the infrastructure and £2,090 recurring cost (paper and nutrient solution reagents). Besides the initial set-up cost, this represents a recurring cost of £0.33 per plant, making it cheaper than commercially available CYG germination papers. Also, we believe an important feature of this set-up is the modularity of phenotyping it offers, that is, the ability to set up experiment in small individual phenotyping units (boxes, Figure 2B and C) which can be used as experimental blocks. This modularity also allows flexibility of phenotyping scale as the experiments can be scaled up or down by adjusting the number of boxes used.
More than half of the processing time in Screen A was spent on image analysis. This included semi-automated analysis of RSA using RootNav as individual root tips need to be selected first before the root outline is automatically extracted. This processing time can be further improved by adopting the latest deep machine-learning algorithm recently applied to root image processing allowing fully automated extraction of root outline. This innovation will drastically reduce the time taken for image processing without compromising on accuracy (Pound et al. 2017).
Despite the high plasticity (e.g., dynamic response to varying water, nutrient and environmental conditions) associated with root traits, we obtained a high heritability estimate for seminal root number measurements across the two screens, similar to previous estimates (Maccaferri et al. 2016;Ma et al. 2017). While this high heritability may be due to the controlled hydroponic environment used in these experiments (Figure 2), it nonetheless demonstrates that seminal root number is a stable phenotype under strong genetic control and can be targeted for selection to improve RSA in wheat breeding programs. There is also evidence to suggest that seminal root number phenotypes observed in hydroponic conditions are transferrable to soil conditions (Richard et al. 2015;Watt et al. 2013) and might therefore be useful under field conditions especially during the vegetative phase. We are currently evaluating the mutant lines under field conditions to determine this.
Developmental and genetic characterization of seminal root formation Natural variation in wheat seminal root number has been shown to originate from defects in root primordia development in the embryo and/or differential activation of developed primordia to form fully emerged seminal roots (Golan et al. 2018). Our anatomical characterization of intact non-imbibed seeds of the lower root number mutants highlight a tendency for these mutants to develop less than five root primordia, whereas wild-type Cadenza plants consistently develop five root primordia. In addition, these mutants form even fewer numbers of seminal roots than the root primordia they developed, suggesting that some of the primordia developed in these mutants might be dormant/ inactive or arrested soon after activation. Together these results highlight both primordia development and activation as two important development check points in the formation of seminal root in wheat.
All the altered root number mutants develop SR 1 and SR 2,3 but show defects in the development of SR 4,5 as in the lower root number mutants, or develop an extra root (SR 6 ) as in Cadenza0900. This suggests that the development of the different seminal root types is under distinct genetic control. We did not recover mutant lines defective in either SR 1 nor SR 2,3 : this could be due to the fact that defects in these root types might have severe negative or lethal effects on seedling growth. Golan et al. (2018) show that SR 1 and SR 2,3 contribute more to water uptake in durum wheat than SR 4,5 under well-watered conditions. It is however possible that SR 4,5 and indeed SR 6 may contribute significantly to nutrient and water uptake under resource-limiting conditions where an increase in root surface area maximizes soil exploration. Detailed field physiological evaluation will be required to better understand the cost-benefit relationship of the altered seminal root phenotypes.
The lower root number mutant Cadenza0062 shows a dominant mode of action with a 3:1 segregation ratio in the F 2 population, characteristic of a monogenic trait. Until this and the other mutants are independently characterized, we cannot conclude that this dominant monogenic behavior is representative of all the lower root number mutants, nor can we rule out the possibility of some of these mutations being allelic. Unlike Cadenza0062, the extra seminal root number mutant Cadenza0900 shows a recessive, multigenic phenotype that suggests a more complex genetic regulation of additional seminal roots in wheat. More detailed genetic characterization and mapping will be required to better dissect the genetic control of these phenotypes.

Outlook
Our work provides a complementary approach to the use of natural variation in dissecting the genetic control of seminal root development in wheat. The isolation of these mutants represents an important step in identifying the genetic determinants controlling wheat seminal root development and growth. These will be followed by extensive genetic characterization to map these mutations to defined chromosomal positions and identify the causal gene(s) underlying their phenotypes. Given the high mutation rate in the Cadenza mutant population ($33 mutation per Mb; Krasileva et al. 2017), it will also be necessary to reduce the mutation load in these mutants to enable specific characterization of each mutation. This can be achieved through backcrossing each mutant to wild-type Cadenza and selection of F 1 progenies (for dominant mutations like in Cadenza0062) or F 2 progenies (for recessive mutations like Cadenza0900) that retain the altered seminal root number phenotype (Uauy et al. 2017). These backcrossed mutants will also provide the background for developing crosses to examine interactions between the different seminal root number mutations.

ACKNOWLEDGMENTS
We would like to thank Dr Jonathan Atkinson, Prof Malcolm Bennett, Chloe Riviere and Guiditta Giordani for technical advice and assistance in the setting up the high-throughput 2D root screen and Andrew Davis for help with root imaging. This research is supported by the UK Biotechnology and Biological Sciences Research Council (BBSRC) Designing Future Wheat program (BB/ P016855/1), a Royal Society FLAIR award (FLR\R1\1918500) to OS, a BBSRC DTP award to RK, the Chief Scientist of the Israel Ministry of Agriculture and Rural Development grant (#12-01-0005), and the U.S. Agency for International Development Middle East Research and Cooperation grant (# M34-037).