Introduction

Brassica napus L. is an amphidiploid species (AACC) developed from natural hybridization of two other diploid Brassica species, B. rapa (AA) and B. oleracea (CC) (UN 1935). B. napus is one of the few Brassica species that are categorized as canola/rapeseed if its oil has less than 2% erucic acid and meal contains less than 30 µm/gm glucosinolate. Rapeseed/canola oil is now regarded as one of the healthiest edible oils worldwide (Lin et al. 2013). It is the second largest oilseed crop in terms of global production after soybean (Foreign Agriculture Services, USDA, 2017). B. napus canola includes three different growth types, spring, winter and semi-winter based on growing season and vernalization requirement for flowering (Ferreira et al. 1995; Rahman and McClean 2013). According to Kebede et al. (2010), the spring and winter canola are genetically diverse from one another. Besides their flowering time difference, spring and winter types are distinct in terms of other morphological characteristics. Winter type canola plants have larger leaves, wider stem diameters, higher plant height, and more vigorous root system with higher yield compared to spring canola (Rahman and McClean 2013; Arifuzzaman et al. 2016; Arifuzzaman and Rahman 2017). The phenotypic variation in root system between the canola growth types and genetic diversity among them provides a huge scope of identifying markers and/or loci controlling root system difference in canola.

The root system in plants has several major functions such as supplying water and nutrients, acting as food and water storage unit, and anchoring the plant to the soil (de Dorlodot et al. 2007). The spatial arrangement of the whole root system in the soil is referred to as its root system architecture (Lynch 2007). Different components of a root system, i.e., root length, primary root branch, lateral root branch, root density, root diameter, root angle, total root surface, etc. (Kuijken et al. 2015) define the root system architecture of that root system. Different root architectural traits respond to the soil microenvironment differentially and can be modified according to the plant needs. Therefore, the root system plays a significant role in adaptation of plants under abiotic stress conditions. According to Lynch (2007), greater focus should be implemented on the modification of root system to enhance crop adaptation under different environmental stresses to increase the crop production. Manipulation of root system architectural traits of crops to grow in nutrient deficient or drought condition can significantly enhance the yield (Wasson et al. 2012; Kuijken et al. 2015).

The root system has remained the “the hidden half” throughout the last century in the field of plant breeding and crop improvement. Plasticity of the root system and difficulties in phenotyping root traits has discouraged breeder from focusing on root traits for crop improvement. Recent advancement in genetic mapping and molecular marker technologies encouraged breeders to emphasize on the root architectural traits for crop improvement through marker assisted selection (Clark et al. 2011). Nevertheless, large-scale, highly accurate and efficient phenotyping of root traits is essential for an ideal mapping study (Canè et al. 2014). Recent technological advances in the field of phenomics allow plant breeders to successfully evaluate the root architecture of different crops from younger seedlings in controlled environment. There are very few methods for large-scale high-throughput phenotyping for the root architectural traits in the field. Shovelomics is one of the popular methods of root system phenotyping in the field. Shovelomics involves digging up the root system with shovel from the field, washing them and measuring the root traits with the help of a phenotyping board and/or imaging technology (Trachsel et al. 2011).

Genome-wide associate studies (GWAS) were first implemented to study human diseases as an alternate to family based mapping strategies (Begum et al. 2015). Eventually, it becomes very popular among plant scientists to identify quantitative trait loci. A GWAS is performed with a set of diverse genotypes which enable us to capture all the historical recombination events within a germplasm panel (Zhao et al. 2014; Rahaman et al. 2017) and is highly efficient in identifying common alleles controlling a complex trait. Single nucleotide polymorphism (SNP) markers are now the most popular marker due to their low cost and ease of development. Researchers are now able to develop genetic maps with higher marker density with the help of abundantly generated SNPs (Luo et al. 2016).

Genome-wide association mapping is performed to successfully identify QTL and molecular markers associated with different root traits in different crops. However, most of these studies were performed under controlled condition. There are only few examples of phenotyping for root architectural traits in the field; i.e., rice with PVC pipe method (Li et al. 2017), rice with basket method (Lou et al. 2015), and cowpea with shovelomics (Burridge et al. 2017) to use in genome-wide association mapping studies. In the current study, we adopted the shovelomics approach described by (Trachsel et al. 2011) to phenotype genetically diverse B. napus germplasms for different root architectural traits, and to identify genomic regions associated with different root architectural traits of rapeseed/canola (B. napus L.)

Materials and methods

Plant materials and experimental design

A total of 216 B. napus genotypes of diversified origin (Supplementary Table 1) were planted in North Dakota State University Agricultural Experiment Station at Carrington, N. D. in mid May of 2015 (47.516998, − 99.148217) and mid May of 2016 (47.517740, − 99.144523). The soil type of the area is Heimdahl silt loam with pH of ~ 6.5 and OM of ~ 2.7. Each genotype was grown in a single row plot with 3 cm plant spacing and 61 cm row spacing. All the three growth habit types of canola/rapeseed were included in the germplasm panel: spring (n = 85), winter (n = 90) and semi-winter (n = 39), rutabaga (n = 2) and grown in the same field within a year to maintain the uniformity of the growing condition. The germplasm accessions were obtained from Germplasm Resource Information Network (GRIN) of USDA-ARS. Germplasms were planted as randomized complete block design with two replications in each year.

Phenotyping

Root architectural traits were phenotyped with the shovelomics approach described by (Trachsel et al. 2011). Phenotyping was done at 50 days after planting. The total phenotyping procedure is divided into three major steps (Fig. 1). Three representative plants per plot were selected based on plant height and general appearance and excavated them with a shovel. The root system in 50-day-old canola plant is very complex and grows deep in the soil horizon. Therefore, we excavated the root system only to capture the top 10 cm of the root system and phenotyped the root traits. The root sample is then washed. All three excavated plants with their root systems were zip tied together, tagged and brought to the washing area. The roots were washed first with soapy water followed by a wash with normal water. After 5–10 min of air drying, data measurements were taken. Data were collected on the number of primary root branches (PRB), lateral root angle (RA), taproot diameter at soil level (R1Dia), belowground taproot diameter at 10 cm below the soil level (R2Dia) and a visual scoring of the root system (RS) (Table 1). RA and the first 10 cm length of the taproot were measured with a phenotyping board labeled with angle and length measurement. Diameters were taken with slide calipers at the soil surface position and 10 cm down from the first data point. RS was recorded with a visual scoring system based on general appearance of the root system and scored from 1 to 5, where score “1” represents the weakest root system with the thinnest taproot and fewest branch roots, score “2” represents weak root system with thin taproot and less root branches, score “3” represents root system with intermediate taproot diameter and intermediate root branches, score “4” represents root system with higher root vigor, thick taproot and root branches and score “5” represents very vigorous root system with the thickest taproot and higher number of root branches (Fig. 2).

Fig. 1
figure 1

Shovelomics in phenotyping root system architectural traits of canola. a Work station in the field, b selecting representative plants and excavating, c excavating, d clearing dart, e tagging, f washing, g phenotyping board, h data collection

Full size image
Table 1 Root architectural traits, their abbreviation and description used in this study
Full size table
Fig. 2
figure 2

Visual root scoring with 1–5 scale, where (1) is the weakest root system with low taproot diameter and primary root branches and (5) is the strongest root system with high taproot diameter

Full size image

Statistical analysis

Data from three representative plants per plot were averaged and were analyzed 2015 (E1) and 2016 (E2) separately in SAS 9.3 (SAS Institute Inc., USA). Means were used to construct the analysis of variance (ANOVA) for all traits except RS, for which the medians were calculated and used. If the ratio of the effective error variance for each trait is less than tenfold, data from both environments were combined and used for the GWAS analysis (Tabachnick and Fidel 2001; Elias and Manthey 2016). Pearson correlation coefficients were calculated in SAS 9.3 for all traits except RS for which spearman correlation coefficient was calculated. Shapiro–Wilk test was conducted in SAS 9.3 for all traits in both E1 and E2 and combined dataset. Entry mean basis heritability was calculated in SAS 9.3 for all traits in both E1, E2, and combined dataset. RS data were available for only one replication in E1 and, therefore, the heritability was not calculated for RS in E1.

Genotyping

A total of 366 germplasms from diverse origin and sources were planted in the greenhouse in 2013 (Michalack et al., Unpublished). Young leaf tissues were collected from each germplasm and freeze dried. DNA was extracted using Qiagene DNeasy kit (Qiagen, CA, US) following the manufacturer’s protocol. After extraction, DNA was quantified, checked for quality, optimized and sent to Institute of Genomic Diversity (IGD), Cornell University for Genotyping by Sequencing (GBS). GBS libraries were prepared using ApeKI single cutter enzyme following the protocol described by (Elshire et al. 2011). Libraries were sequenced with Illumina GAII sequencer, and 100 bp single end sequenced GBS data were aligned using BWA-MEM (Li 2013). VarScan (Koboldt et al. 2012) was used for SNP calling. SNPs were imputed for missing allele using FastPHASE (Scheet and Stephens 2006) and finally 42,575 SNPs were obtained. For the current experiment, we used the SNP data with minor allele frequency greater than 0.05 on 216 genotypes. The name of each SNP marker consists of chromosome number and physical position of that marker. For example, an SNP marker located on 123456 bp of chromosome C04 was named as “C04_123456”.

Marker-trait association

Association mapping was performed using GAPIT R-based program (Lipka et al. 2012). A number of principle components (PC) that collectively explain 25% of the variation in the population were determined by principle component analysis (PCA) in R 3.4.1 using prcomp () function (Price et al. 2006). A pairwise kinship coefficient matrix was calculated to account for individual relatedness using the EMMA algorithm (Kang et al. 2008) embedded in GAPIT. A total of four models were tested to detect the marker-trait association: (1) a null general linear model or naïve model accounting for neither population structure nor kinship (EMMA); (2) a general linear model accounting for population structure as a fixed effect using the number of PC accounting for 25% of the variation (PC-25%); (3) an efficient mixed model association or EMMA accounting for kinship in the population; and (4) a mixed linear model accounting for both population structure and kinship, PC-25% + EMMA. All the models were compared on a rank-based mean square difference (MSD) value, and the model with the lowest MSD value was selected as the best model (Mamidi et al. 2011). Significant markers were declared based on the P value of the markers at lower 0.01 and 0.1 percentile tail of an empirical distribution obtained by 10,000 bootstraps (Mamidi et al. 2014). Manhattan plots were created using mhtplot () function in R statistical software package gap (Zhao 2007). Log likelihood ratio-based R2 or \({\text{R}}^{ 2}_{LR}\) (Sun et al. 2010) was calculated in genAble package in R.4.1 (Aulchenko et al. 2007) for the most significant markers in the best models to determine the phenotypic variation explained by those markers.

Candidate gene

B. napus gene models within 100 kbp upstream and downstream of the significant markers were taken into account for candidate gene search. Protein sequences from the gene models were blasted against TAIR 10 protein database to determine the gene annotation. Genes associated with root development were identified based on the Gene Ontology terms (GO terms) from TAIR website and gene functions found in previous literatures.

Results

Phenotypic distribution

Phenotypic variation was observed among the germplasm panel in both environments and when the data were combined (Fig. 3). The range between maximum and minimum observations in 2015 for R1Dia (22.88), R2Dia (8.15), PRB (13.80) and RA (96.79) was higher than that of 2016 (Table 2). The CVs of R1Dia (27%), R2Dia (44%), PRB (31%) and RS (31%) in 2015 were higher than the CVs for the respective traits in 2016. Relatively higher CV was observed for R2Dia across both environments and the combined dataset (31–44%). The lowest CVs were observed for RA across both years and with the combined datasets ranging 6–9%. High family based mean heritability was detected for R1Dia in both years and for combined dataset (0.60–0.79) (Table 2). High heritability was detected for R1Dia in 2015 (0.83), but moderate heritability was detected in 2016 (0.60) and the combined dataset (0.67). A similar trend of high heritability in 2015 (0.72) and moderate heritability in 2016 (0.56) and the combined dataset (0.39) was found for PRB. Very low heritability was observed for RA in combined dataset (0.33). RS heritability in 2015 and combined dataset found moderate ranging from 0.46 to 0.58.

Fig. 3
figure 3

Phenotypic distribution of different root architectural traits in 2015, 2016 and combined dataset

Full size image
Table 2 Phenotypic variation in different root architectural traits
Full size table

Correlation among the traits

A significant positive correlation (r = 0.57–0.65, P < 0.001) was detected between R1Dia and R2Dia in all datasets (Fig. 4). Both the root diameters, R1Dia and R2Dia, were significantly (P < 0.001) positively correlated with RS in all datasets. PRB and R1Dia were also significantly (P < 0.001) correlated with each other in all environment. Very low or negative nonsignificant correlation was detected between RA and all other traits.

Fig. 4
figure 4

Correlation among the root architectural traits in 2015, 2016 and combined dataset

Full size image

Genotypic data and principle component analysis

Originally 366 genotypes were genotyped from which 42,575 single nucleotide polymorphism (SNP) markers were obtained as described by Michalack et al. (Unpublished). In this current study, a subset of 216 genotypes was used. From the original panel, 30,262 markers were retained for the panel evaluated here after correcting for MAF > 5% and were used finally in this current study. Principle component analysis was performed to control for population structure. The first two principle components grouped the population into three continuous subpopulations (Fig. 5). The first three principle components accounted for 25% of the population variation and were included in the association analysis as general linear model.

Fig. 5
figure 5

Principle component graphs showing the distribution of the population used in this study

Full size image

Marker-trait association

Four models were tested for each trait in each year and with the combined over the two years to detect best SNP marker/trait association. Among the models, the best model was selected based on the lowest MSD value (Mamidi et al. 2014; Soltani et al. 2017). The P values of the best model for a trait in a particular dataset (2015, 2016 or combined) were bootstrapped 1000 times, and the P value at 0.01 percentile of the empirical distribution was set as the cutoff for that particular trait in the respective dataset. Highly stringent P value cutoff increases the chance of eliminating false positive or the false association. However, it also increases the chance of excluding false negative or true association. To reduce the chance of excluding true association or falls negative, we fixed another less stringent P value cutoff at 0.1 percentile of the empirical distribution (Moghaddam et al. 2016). Only the markers that were found to be appear repeatedly for a trait in both years and the combined dataset were considered to be a significant association and were reported here. Candidate genes were searched 100 kbp at either side of a significant marker at 0.01 percentile cutoff.

Soil level taproot diameter (R1Dia)

The mixed linear model accounting for relatedness in the population was the best fitted model in both years and combined over the 2 years (Supplementary Table 2). Three markers, chrA01_rand_2039614, chrC01_rand_171065, and chrC04_4933647, were found to be associated with R1Dia in 2015 at the 0.01 percentile P value cutoff level (Table 3). The markers together explained 39.2% of the phenotypic variation. Marker locus chrA01_rand_2039614 from 2015 was also detected in 2016 and the combined dataset. Marker locus chrC01_rand_171065 from 2015 was also detected in combined dataset. In 2016, two other markers chrC02_30907736 and chrA09_32034416 explaining 16.2%, and 16.8% of the phenotypic variation, respectively, were detected. Two root-related candidate genes, STRUBBELIG-receptor family 3 (SUB), were detected nearby the marker locus chrA09_32034416 (2016). Marker chrC05_rand_678307 was detected in combined dataset and two candidate genes, Cyclin-dependent kinase 2;3 (CYC2; 3) and P-glycoprotein 11 (PGP 11), were detected nearby physical location of the marker.

Table 3 Significant markers for different root architectural traits in 2015, 2016 and combined datasets at 0.01 percentile
Full size table

Bottom taproot diameter (R2Dia)

A general linear model using the first three principal components was selected as the best model for R2Dia in 2015 and combined data (Supplementary Table 2). Three markers loci, chrA08_18428869, chrA08_18428839 and chrA08_18441370, within a very close physical distances on chromosome A08 were identified as the significant markers at 0.01 percentile P value cutoff in both 2015 and the combined dataset (Table 3). The markers altogether explained 13.2% and 9.8% of the phenotypic variation in 2015 and combined dataset, respectively. The fact that the three loci only marginally account for more of the phenotypic variation suggests that they are linked to the same causative locus. Two candidate genes, AUXIN RESISTANT 3 (AXR3) and Cryptochrome 2 (CRY2), were identified in the same region of chromosome A08 (Table 4). A mixed linear model, PC3 + EMMA, accounting for both population structure and relatedness was selected as the best model in 2016. Three unique marker loci, chrC05_7043010, chrA10_5618818 and chrA09_6247199, were found to be significantly associated with R2Dia in 2016. They altogether accounted for 33.8% of the phenotypic variation (Table 3).

Table 4 Candidate genes for different root architectural traits within 100 kbp region at either side of the significant markers
Full size table

Primary root branches (PRB)

EMMA, accounting for kinship in the population, was found to be the best model for PRB in 2015 (Supplementary Table 2). On the other hand, PC3, accounting for population structure, was identified as the best fitted model in 2016 and combined dataset. In 2015, marker loci, chrC01_rand_171065, chrC04_24065408 and chrC03_3328679, were identified as the significant markers associated with PRB at 0.01 percentile P value cutoff and accounts for 37.4% of the phenotypic variation (Table 3). GAST1, a homolog of GASA4, was detected at 53 kbp upstream of the marker locus chrC03_3328679 (Table 4). One of the 2015 marker loci, chrC01_rand_171065, was also detected in combined dataset and explained 9.2% phenotypic variation. Two other unique markers, chrA06_2653673 and chrA08_4600588, were also found to be significantly associated with PRB in combined dataset. Marker locus chrA06_2653673 was also found to be associated with PRB in both E1 and E2 with a less stringent at 0.1 percentile P value cutoff. A lateral organ boundaries (LOB) domain family (LBD1) gene was detected at 30 kbp upstream of the marker locus chrA06_2653673. At 0.01 percentile P value cutoff, two markers, chrC05_7907786 and chrA08_6097421, were identified in 2016 and explained 16.9% of the phenotypic variation. Aminotransferase-like, plant mobile domain family protein (MAIN) was detected at 87 kbp downstream of the marker locus chrC05_7907786.

Root angle (RA)

The naïve model, accounting for neither the population structure nor kinship, was the best model in 2015 for RA (Supplementary Table 2). A mixed linear model including kinship and PC3 + EMMA was the best model in 2016 and combined dataset, respectively. Marker locus, chrA02_18466347, was detected as the most significant marker associated with root angle in both 2015 and combined dataset (Table 3). This marker explained 7.3% and 16.2% of the total phenotypic variation in 2015 and combined dataset, respectively. The other two marker loci found to be significantly associated with RA in 2015 are chrA09_8677526 and chrA03_25132124 and accounted 7.3% and 6.5% of the total phenotypic variation, respectively. Candidate gene SABRE-like protein (SABRE) and auxin-responsive GH3 family were detected at 58 kbp upstream and 34 kbp downstream, respectively, of the marker locus chrA09_8677526 (Table 4). Candidate genes, Cytokinin response factor 4 (CRF4) and Tetraspanin 7 (TET7), were detected at 5 kbp and 36 kbp downstream, respectively, of the marker locus chrA03_25132124. Three marker loci, chrC03_36049749, chrA03_18531995 and chrC09_8316001, were detected for RA in 2016, and they altogether explained 29.8% of the total phenotypic variation. Two unique marker loci, chrC06_35857438 and chrA02_22847077, were detected in combined dataset. Auxin-responsive factor 17 (ARF17) was detected at 99 kbp downstream of the marker locus chrC06_35857438 associated with RA in combined dataset. Another candidate GA requiring 3 (GA3) was detected at 36 kbp downstream of the marker chrA02_22847077 associated with RA in combined dataset.

Root score (RS)

A mixed linear model including kinship was found to be the best model in 2015 and the combined analysis (Supplementary Table 2). Three marker loci, chrC01_11345932, chrC09_19217839 and chrC05_13423686, were found to be significantly associated with RS in 2015 at 0.01 P value cutoff (Table 3). The markers together explained 28.5% of the total phenotypic variation. Candidate gene Cytokinin responsive factor 6 (CRF6) was detected at 89 kbp upstream of the marker locus chrC09_19217839 (Table 4). In combined dataset, three additional marker loci, chrA09_32034416, chrA09_32034411 and chrA01_rand_2039614, explained 20.4% of the total phenotypic variation were detected. A candidate gene STRUBBELIG-receptor family 3 (SUB) was located on chromosome A09, close to the marker loci chrA09_32034416 and chrA09_32034411. Marker locus chrA01_rand_2039614 was also appeared in 2016 and explains 5.8% of the total phenotypic variation. Two other marker loci, chrC09_23510860 and chrC05_rand_678307, were also detected in 2016, and explained 5.8% and 5.6% of the total phenotypic variation, respectively. Candidate gene, P-glycoprotein 11 (PGP 11), was detected only 10 kbp upstream of the marker locus chrC05_rand_678307. Naïve model was selected as the best fitted model for 2016 based on the lowest MSD value.

Discussion

Shovelomics in B. napus

Plant breeders have historically modified the above ground morphological plant architecture of a crop to improve productivity. One of the greatest examples is the green revolution that developed high yielding dwarf verities in wheat (Vergauwen and De Smet 2017). Modifying shoot features like branch angle and leaf angle are also regarded as the potential to improve the productivity of corn (Gong et al. 2015) and sorghum (Truong et al. 2015). This enabled increasing planting density, the capture of more sunlight, and increased photosynthetic efficiency of the crops. Similarly, the below ground root architectural traits can also be used as a target for improvement and optimization according to the necessities based on soil environment. Several studies on crops such as rice (Courtois et al. 2013), wheat (Canè et al. 2014), maize (Pace et al. 2015), cowpea (Burridge et al. 2017), and rapeseed (Wang et al. 2017) phenotyped the root system and mapped associated loci using GWAS for root architectural traits. Thomas et al. (2016) characterized the root morphology of different B. napus growth types with “pouch and wick system” and concluded that root morphology can potentially be used to improve crop yield if suitable markers loci corresponding to agronomic traits can be identified.

The majority of these studies adopted root system phenotyping strategy that is feasible for controlled environment using soil or other growing media and phenotyped the root system of young seedlings. The root systems of young seedlings grown in hydroponics or soil in the controlled environment may not be indicative of the architecture that forms in the actual field condition (Zhu et al. 2011). Several researchers mentioned the importance of monitoring root architectural traits over the whole growth period in the soil condition as the root system architecture changes greatly over time (Hund 2010; Kumar et al. 2012; Passioura 2012; Colombi et al. 2015). Therefore, phenotyping the root system under all stages of growth in the field condition is immensely important, and high-throughput phenotypic techniques should be developed to overcome the difficulties in exploring root system in the field.

Unfortunately, very few high-throughput platforms for phenotyping the root system architecture in the field have been developed to date. Shovelomics is a high-throughput root phenotyping system first described by Trachsel et al. (2011). However, only a few studies have adopted this shovelomics approach to characterize the root system in field conditions in different crops. This includes crops such as maize (Bucksch et al. 2014; Colombi et al. 2015; York and Lynch 2015), in common bean and cowpea (Burridge et al. 2016). Although Trachsel et al. (2011) proposed manual measurement of the data in the original shovelomics approach, most studies used image-based technologies during data collection. Burridge et al. (2017) conducted a GWAS analysis for cowpea root architectural traits with the phenotypic data collected with shovelomics approach. In the current study, we also adopted the field-based shovelomics technique to phenotype 216 B. napus genotypes in multiple years and used that date for a GWAS analysis. This is the first report of shovelomics to characterize the root architectural traits in B. napus.

In the current study, we have phenotyped five root traits and conducted a GWAS analysis. Based on a stringent cutoff at lower 0.01 percentile of bootstrap P values, 31 marker loci were associated with different root traits in 2015, 2016 and in the combined datasets. Markers were distributed in all the B. napus chromosomes except A04, A05, A07, C07 and C08. Five marker loci were detected on chromosome A08 controlling R2Dia and PRB. Four marker loci controlling various root traits were detected on chromosome A09. Multiple co-localized marker loci were detected on chromosome A08 and A09 controlling primary root number under high and low P level (Zhang et al. 2016).

Phenotypic distribution and correlation among the root architectural traits

In the current study, five different root architectural traits R1Dia, R2Dia PRB, RA and RS in 2015 and 2016 were phenotyped. Analysis was performed for both years and the combined dataset. Population means were relatively constant across the years for R1Dia, R2Dia and RS. In contrast, population means for PRB and RA varied across years indicating a genotype x environment effect controlling these traits. Development of lateral root branches is highly responsive to the soil nitrogen (N) and phosphorus (P) condition and water regime. On the basis of range, phenotypic variation within the population was always higher in 2015 to 2016. This also indicates the environmental influences controlling the root traits. Entry mean basis heritability for different traits varied with environments. High heritability values were noted for R1Dia, R2Dia and PRB in 2015. However, the heritability values decreased to moderate levels for all these traits in 2016 suggesting high environmental variation between these two environments.

Significant and high correlation was observed among the traits within the same year and in the combined datasets. R1Dia and R2Dia are highly correlated in both environment and combined datasets. A high correlation was also observed between PRB and R1Dia. This is expected as the taproot having high diameter will have more surface area to grow more primary and lateral root branches and the root will be more vigorous. This statement is supported further as a high positive correlation was observed between R1Dia and RS and between PRB and RS. Visual root scoring was conducted based on the overall appearance of the root system on top soil considering diameter and root branches. Low but significant positive correlation was detected between R2Dia and PRB indicating lesser role of bottom tap root diameter on root branches. Interestingly, we did not observe any significant correlation between RA and other root traits in the current study. A lack of correlation between other traits and RA suggests that RA might be an independent trait which depends on soil environment rather than other root characteristics in B. napus.

Root diameters can be a proxy for other root traits

In the current study, multiple marker loci associated with R1Dia were also found to be associated with RS and PRB. Two marker loci on chromosome A01 (position about 2.04 Mbp) and chromosome C05 (position about 0.68 Mbp) were found to be associated with both R1Dia and RS and detected in multiple datasets (Fig. 6). Candidate gene CYC2; 3 and PGP11 were identified close to the marker loci on chromosome C05 (position 0.68 Mbp). CYC2 is involved in lateral root formation and triple CYC2 mutant gives rise to defective lateral roots compared to the wild type (Vanneste et al. 2011). The PGP family of proteins consists of members of ABC transporter superfamily and few of them directly act as auxin efflux carrier in basipetal auxin transport and affect lateral root and root hair formation (Santelia et al. 2005). Two other significant marker loci, chrA09_32034411 and chrA09_32034416, are physically close and were found to be associated with RS and R1Dia, respectively (Fig. 6). Candidate gene STRUBBELIG-receptor family 3 (SUB) was found in the same region on chromosome A09. SUB is expressed in the root throughout the root development process and controls root meristem specification and epidermal root hair specification in Arabidopsis (Savage et al. 2013; Kwak and Schiefelbein 2014). Marker locus chrC01_rand_171065 was associated with PRB and R1Dia. In addition to these, two other marker loci on chromosome C05 are located within 9 kbp and are associated with both R2Dia and PRB. Candidate gene Aminotransferase-like, plant mobile domain family protein or MAIN or MAINTENANCE OF MERISTEM was found close to this marker locus. Experimental evidences suggest that mutants of maintenance of meristem have a defective stem cell niche in Arabidopsis roots (Wenig et al. 2013). MAINTENANCE OF MERISTEM homolog MAIN-like 1 mutant has short primary root branches (Ühlken et al. 2014).

Fig. 6
figure 6

Manhattan plot and QQ plot showing the common significant markers associated with R1Dia and RS. a R1Dia_2015, b R1Dia_2016, c R1Dia_combine, d RS_2016 and e RS_combine

Full size image

Common marker loci or a region associated with R1Dia, RS, PRB and R2Dia is expected because we took PRB data from the top 10 cm of the root system and RS scoring was based on visual appearance of the root system consisting of both PRB and taproot diameters. In our phenotypic study, R1Dia, PRB and RS were highly correlated with each other in both years. Higher soil level tap root diameter has higher root surface area to accommodate higher root lateral branches and root hairs. A root system with high tap root diameter is correlated with higher xylem area in cowpea (Burridge et al. 2017) which may transport water and nutrients from the root to the shoots more efficiently. A high phenotypic correlation between R1Dia and R2Dia also indicates that root diameters may have a positive role in defining root length. Burridge et al. (2017) suggested that taproot diameter at 10 cm may be related to the plants ability to explore a deeper soil horizon in drought conditions for cowpea. Considering all these aspects, we propose that tap root diameter in B. napus can be a good indicator for other root architectural traits and may serve as a proxy trait for the whole root system. Importantly, the soil level taproot diameter R1Dia can be measured without destroying the plant.

Significant marker loci associated with same trait in multiple datasets

The marker locus chrA01_rand_2039614 was significantly associated with R1Dia in both 2015, 2016 and combined datasets. Marker locus chrC01_rand_171065 was also significantly associated with R1Dia in 2015 and the combined dataset. Marker locus chrC05_rand_678307 was identified in the combined dataset for R1Dia at 0.01 P value cutoff. However, with a less stringent 0.1 percentile P value cutoff, these marker loci were found to be significant in both 2015 and 2016. For R2Dia, three closely located marker loci on chromosome A08 (position about 18.4 Mbp) were found to be significantly associated in both 2015 and combined datasets. Two candidate genes, AXR3 and CRY2, were detected in the same region of chromosome A08 (Fig. 7). Auxin Resistant 3 or AXR3/IAA17 overexpression caused defective roots in Arabidopsis (Kim et al. 2006). Knox (2003) reported that AXR3/IAA17 inhibits root hair initiation and elongation. The relative abundance of AXR3/IAA17 and SHY2/IAA3 (gene which induce root hair formation and elongation) in a cell is the determinant of root hair formation in Arabidopsis. Overexpression of CRY2 in blue light results in reduced primary root elongation (Canamero et al. 2006).

Fig. 7
figure 7

Manhattan plot and QQ plot showing the results of marker/trait association of R2Dia. a Manhattan plot of R2Dia in 2015, b QQ plot of R2Dia in 2015, c Manhattan plot of R2 in combined dataset, d QQ plot of R2 in combined dataset

Full size image

For PRB, one marker locus chrC01_rand_171065 was repeatedly significant in 2015 and the combined dataset. Marker locus chrA06_2653673 was significantly associated with PRB in combined dataset was also detected in 2015 and 2016 at the less stringent 0.1 percentile P value cutoff. Lateral Organ Boundary domain 1 (LBD1) was detected close to this marker locus. Constitutive expression of LBD1 in the Medicago truncatula root system subjected to salt stress regulates the overall root architecture (Ariel et al. 2010). Primary lateral roots appear from the taproot near the soil surface and are necessary for nutrient acquisition. Lateral roots in plants are highly responsive to the soil microenvironment. It has been reported that the length of lateral root was increased under low N condition in Arabidopsis, and lateral root density was increased in P-deficient soil conditions in rice (Desnos 2008; Gruber et al. 2013; Kong et al. 2014). Therefore, the significant markers identified for PRB in this study may be useful in future marker-assisted breeding program to select for improved nutrient acquisition.

Root angle is an important trait for root architecture where a narrower root angle allows the root to penetrate deeper into the soil horizon which reduces the competition between the neighboring plants. In addition, the plant will have an excess of moisture from the deeper soil horizon too. Seven marker loci associated with root angle (RA) were identified in 2015, 2016 and the combined dataset. Among them, the marker locus chrA02_18466347 was found to be significant in both 2015 and the combined dataset. Root score was measured visually on the basis of overall appearances of the root system and eight marker loci were identified related to root score (RS). Marker locus chrA01_rand_2039614 was associated with RS in both 2016 and the combined datasets.

In summary, we followed the shovelomics approach to phenotype five root architectural traits and use the data for GWAS of these root traits with SNP markers. In this study, a total of 31 significant marker loci associated with different root traits were identified. Several marker loci were found to be associated with multiple root traits and appeared in multiple environments and potential for future marker-assisted selection for root traits in rapeseed/canola. It appears that tap root diameter in rapeseed/canola may be a determinant of the total root system in canola and can be a proxy trait for other root architectural traits. Tap root diameter at soil level (R1Dia) can be measured without uprooting the plant from soil and, therefore, can potentially be a non-destructive measurement for root system characterization in rapeseed/canola. Fifteen candidate genes related to different root traits were detected within 100 kbp downstream and upstream of different significant markers.