2.1 Plant materials

Seeds from 46 diverse industrial hemp genotypes were provided by New West Genetics, Inc. (NWG). These include European dual-purpose varieties, Canadian oilseed varieties, Chinese fiber varieties, US high-CBD (cannabidiol) varieties and breeding populations created by NWG derived from crosses among them (S1 Table).

2.2 Raised-bed construction and preparation

The experiment was conducted in a large raised-bed structure constructed inside a greenhouse at the Donald Danforth Plant Science Center (38.6745° N, 90.3971° W). The unit measured 6 m × 3 m × 1 m (length × width × height), resulting in an 18 m³ growth volume, and was built using dimensional treated lumber. The walls consisted of 5 cm × 30 cm × 304 cm boards (nominal size: 2 in × 12 in × 10 ft) attached horizontally to vertical 5 cm × 10 cm boards (nominal size: 2 in × 4 in) at each end using construction-grade lag screws. A 5 cm × 10 cm × 3 m (nominal size: 2 in × 4 in × 10 ft) cross brace was placed across the bottom to connect the two long sides at the midpoint, and 5 cm × 30.5 cm × 3 m (nominal size: 2 in × 12 in × 10 ft) boards were installed diagonally across each 3 m wall to increase rigidity. At the midpoint of each long wall, a 10 cm × 10 cm vertical post (nominal size: 4 in × 4 in) was positioned externally for additional structural support. A 5 cm × 15 cm × 3 m (nominal size: 2 in × 6 in × 10 ft) board was mounted across the top edges to prevent the walls from spreading as the bed was filled with growth media. To reinforce the structure, a polyester tension band (rated to 40,000 N) was wrapped around the perimeter to maintain the structural integrity under load. Steel angle iron caps were placed at each corner to protect the band from abrasion and to allow tightening without fraying. This design ensured long-term stability of the raised-bed unit under the weight of the growth media.

2.3. Growth conditions and experiment design

Plants were grown in a greenhouse between January 5, 2022 and March 8, 2022. Seeds of 46 genotypes were sown in 36-well plug trays containing soil and peat planting medium (50% Jolly Gardener/ 50% peat). Germination rates for all genotypes used were above 70% with 15 evenly germinated plants used for the experiment. The seed source was unfeminized seeds with the males thinned later as detailed below [35]. The trays were kept in a greenhouse with a light sensor connected supplemental lighting set to a day length of 12 hours, temperature 30–35°C both day and night, and a humidity of 50–70% for 16 days. After 16 days of growth, the seedlings were transplanted in a randomized block design with three replicate blocks in a large format grow bed containing turface. Five seedlings per genotype per block were spaced 11.5 cm apart with 20 cm row spacing with an equivalent planting density of low-density grain production (43.5 pl m⁻²). Each block served as an independent replicate containing all genotypes arranged in randomized order. Irrigation was applied using 450 drip stakes (0.5 gallon h⁻¹ each), providing a total delivery rate of approximately 47 L m^-2 h^-1 across the raised bed. A 15-16-17 fertilizer (Jack’s Professional Water-soluble fertilizer, J.R. Peters Inc., Allentown, PA) at 200 ppm Nitrogen was applied in 15-minute pulses every 45 minutes, three days per week (Monday, Wednesday, and Friday). Tempered water was applied on the remaining four days per week at the same drip rate (Tuesday, Thursday, Saturday, and Sunday). With the drip irrigation and rate used there were no signs of drought or nutrient stress.

The sex of the hemp plants was determined after flowering by visually inspecting each plant and identifying female or male phenotypic markers. Female inflorescences were identified by their abundance of stigma in multiple colas or floral clusters, while male hemp plants did not have stigma protruding from the flower clusters and had pollen sacks. In this study, all identified male plants were removed and discarded, with an average culling of two plants per genotype and block, while female and monoecious plants were kept through to the harvest.

2.3. Sample collection and harvest measures

The root and shoot harvest were conducted seven-weeks after transplanting. All genotypes had already flowered at seven-weeks after transplanting and were harvested on the same day. All irrigation was stopped two days before the shoot harvest to allow for the turface to partially dry. The stems of all female plants were tagged with a wrap tag and cut 8 cm above the turface line. The shoots were imaged within one hour of harvest using a Sony ILCE-7 camera equipped with a Sony FE 28 mm f/2.0 lens and a ring light (LS Photography PPH67 ring light) for illumination. A single image of the whole intact shoot was taken per plant on a black background with a poker chip in view for scale at a 1.8 m distance from the lens. The shoot images were analyzed using the program RhizoVision Explorer [36]. Images were converted to grayscale and threshold applied to separate shoot from the background, then the shoots were segmented and skeletonized. After imaging, the samples were dried at 50°C for five days and then weighed to obtain their dry biomass weights (Controlled Environment Room, Darwin Chambers, MO, USA). Eleven imaged-based shoot traits were extracted, as well as shoot biomass and root to shoot mass fraction measures.

For the root harvest, three panels on the side of raised bed were removed for direct access to the roots. Carefully utilizing a vacuum while holding the plant stem, the turface was removed to reveal the roots, minimizing any mechanical damage. Once excavated, the harvested roots were wrapped in wet germination paper before same-day imaging. A single image of the whole root system was taken per plant on a black background with a poker chip in view for scale at a 90 cm distance from the lens. After image capture, images were processed using the whole-root mode in RhizoVision Explorer software to extract measures of root system shape. The images of the intact root were converted to grayscale and a threshold applied to separate the root from the background, then the roots were segmented and skeletonized to obtain 13 whole root image traits [37]. The roots were then cut, spread out on a water-filled clear Perspex tray, and imaged on a flatbed scanner with a transparency unit at 600 dpi (Epson Expression 12000XL, Epson America Inc., CA, USA) to obtain root images suitable for root length extraction. These flatbed scanner root images were analyzed using the broken-root mode in RhizoVision Explorer to extract root length related traits by root class, resulting in a further 23 broken root image traits [36]. After flatbed scanner imaging, the roots were transferred into paper bags and dried at 50°C (Fischer Scientific Isotemp Incubator, PA, USA) for up to 5 days to obtain dry biomass weights, specific root length, and root to shoot mass fraction measures. Root and shoot images and RhizoVision Explorer configuration files are accessible at 10.5281/zenodo.17783092.

2.4. In silico analysis

The construction of a phylogenetic tree for hemp root genes and the identification of core root system architecture genes was based on previously published reports on root system architecture genes. The primary transcript protein sequences of maize, rice, and Arabidopsis root system architecture genes predicted by [38] were downloaded from the Phytozome database (https://data.jgi.doe.gov/). The protein sequences of Cannabis sativa (GenBank assembly - GCA_900626175.2) were used for BLAST against the (B73 RefGen v4), rice (MSU RGAP Release 7), and Arabidopsis (Araport11) databases using the blastp command. The primary goal was to identify high-confidence, functionally plausible candidate rather than a complete orthogroup assessment. The BLAST results were therefore filtered for a sequence similarity of 95% to prioritize high-confidence orthologs with a stringent identity cutoff to reduce false positives. These results were then combined into a single FASTA file for further analysis. A FASTA result file was used to align the protein sequence in MEGA11 using the MUSCLE option in the MEGA11 GUI [39]. Phylogenetic analysis was conducted using the Maximum Likelihood method with 1000 bootstrap replications. To visualize the orthologous genes among Cannabis sativa, Zea mays, Arabidopsis thaliana, and Oryza sativa, a phylogenetic tree was created using FigTree v1.4.4. [40]. BLAST, alignment, and phylogenetic project files for the in silico analysis are accessible at 10.5281/zenodo.17870483.

2.5. Statistical analysis

Statistical analyses were conducted using R version 4.4.3 [41]. The raw data is available in S1 Data. The following measurements were derived from image analysis and biomass data: Specific Root Length (m^-1 g^-1) was calculated as total root length divided by root dry biomass. Root Mass Fraction was calculated as root dry biomass divided by total dry biomass. Lateral Root Length (mm) was determined as the sum of root length within diameter ranges 2 and 3. Lateral Root Fraction was calculated as the proportion of lateral roots to total root length.

The tidyverse packages including dplyr and broom were used for data processing, and ggplot2 and patchwork were used for visualization [42]. Analysis of variance (ANOVA) was conducted using linear mixed-effects models fitted with lme4::lmer(), with genotype as a fixed effect and block as a random effect. ANOVA tables for testing the effect of genotype were obtained from the fitted models using the anova() function. The agricolae package was used to perform post-hoc multiple comparison tests [43]. Broad-sense heritability was using the equation . The variables , , and r represent the variances of the genotypic effect, environmental effect, and the number of replicates blocks, respectively. Correlations among the root and shoot traits extracted from the 46 hemp genotypes were explored. Pairwise pearson correlation coefficients were computed using the stats::cor() function with pairwise complete observations to handle missing data. Corresponding significance p-values were calculated using the ggcorrplot::cor_pmat() function [44].

To classify genotypes based on coordinated root and shoot development, percentile ranks were calculated for six key traits: root dry biomass, total root length, root crown convex area, shoot dry biomass, shoot height, and shoot convex area. Percentile ranks were computed using the dplyr::ntile() function, dividing each trait distribution into 100 quantiles. Genotypes were then assigned to one of four quadrants representing major root–shoot allometric patterns using case dplyr::case_when() function. Trait data were mean-centered and scaled prior for visualization and follow up analysis. Principal component analysis (PCA) with K-means clustering was performed using the FactoMineR and factoextra packages to identify natural groupings among genotypes [45]. Bootstrap Jaccard similarity values were calculated using fpc::clusterboot(). Clustering was tested across three and four cluster solutions, and the resulting group assignments were compared to predefined quadrant-based groupings to evaluate concordance between unsupervised clustering and trait-based classifications.

3.1. Phenotypic variation in hemp genotypes for root and shoot traits

In a raised-bed growth system, 46 hemp genotypes were grown and harvested after seven weeks. Individual root and shoot system photos were taken for each plant, resulting in 637 images, and an additional 375 flatbed scanner images were taken to accurately determine root length. Root and shoot phenotypes were extracted from the images, and significant phenotypic differences were observed for multiple root and shoot traits (S2 Table).

For root length traits, significant genotypic differences were observed including total root length, lateral and second-order lateral root lengths, average root diameter, and specific root length (p < 0.05, S2 Table). Heritable variation for root length and biomass traits was found in hemp with broad-sense heritability (H²) scores of 0.63 for total root length, 0.5 for average root diameter, and 0.47 for specific root length (S3 Table). Broad-sense heritability represents the proportion of phenotypic variance explained by genetic differences under the specific environment tested in the study; values above 0.4 indicate a moderate to high genetic contribution. Genotype 1583 had the greatest mean root length, which was 175% larger than the smallest genotype, Earlina08FC (Fig 1A, S4 Table). For root system convex area, genotype 1583 had the greatest area which was 160% larger than the smallest genotype Earlina08FC (Fig 1A, S4 Table). Genotype 1583 and Earlina08FC also had the largest and smallest lateral root fractions, with Genotype 1583 having a 42% greater fraction than Earlina08FC (S4 Table). For specific root length, genotype 4603 had the greatest value at 153 m g^-1, whereas genotype Han NW had the smallest specific root length at 20 mg^-1 (S4 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Large diversity observed in root and shoot size traits among 46 hemp genotypes.

(A) Boxplots for the following extracted root traits. Root dry biomass, convex hull area, and total root length. (B) Boxplots for the following extracted shoot traits. Shoot dry weight, convex hull area, and height. (C, D) Representative greyscale root and shoot images for a small and large genotype, genotype 2729 and genotype 1583, respectively. Scale bar = 20 cm. Data from S3 Table.

https://doi.org/10.1371/journal.pone.0339929.g001

For the root system shape and distribution traits, significant genotypic differences were observed including for root convex network area, root crown maximum width, and lower root area (S2 Table). Root system shape and distribution traits were found to be heritable with broad-sense heritability scores of 0.64 for root convex area, 0.51 for lower root area, and 0.45 for root crown maximum width (S3 Table). Genotype 1583 had the largest convex hull area which was 160% larger than the smallest genotype Earlina08FC (Fig 1A). Genotype 1583 also had the largest root crown maximum width, which was 105% wider compared to the smallest width of genotype 4479 (S4 Table). For lower root area, genotype 1583 had the largest area with a 172% increased area compared to the smallest genotype 4603 (S4 Table).

For the measured shoot traits, large differences among the 46 hemp genotypes were observed (Fig 1B). Significant genotypic differences were observed for shoot size and distribution traits, including biomass, height, network area, and ratio of shoot width to height (p < 0.05, S2 Table). Shoot traits were highly heritable with broad-sense heritability scores of 0.77 for shoot dry biomass, 0.66 for shoot convex area, and 0.50 for shoot height (S3 Table). Genotype 2729 exhibited the smallest shoot biomass, while genotype 1583 had the largest shoot biomass with a 175% increase (Fig 1B). Genotype X-59 had the smallest shoot convex hull area, whereas genotype 1583 again was the largest with a 153% greater convex hull area (Fig 1B, S4 Table). For shoot distribution traits, genotype 4000 had the smallest width-to-depth ratio at 0.24, while genotype X-59 had the largest ratio, which was 71% greater at 0.54 (S4 Table). In addition, there was a significant genotypic effect for root to shoot mass fraction with genotype 4479 having the smallest mass fraction of 0.045 and genotype 4646 with the greatest root to shoot mass fraction of 0.14 (p < 0.05, S3 Table, S4 Table).

Post-culling counts by genotype and block were used to conduct a limited sensitivity check excluding plants from plots with the most extreme culling (S2 Data). Approximately 12% of plots showed substantial culling (≥4 removed plants) and were excluded from this analysis. The primary genotype rankings, heritability estimates, and major root–shoot correlations were unaffected. Three traits were no longer significant for genotypic effects once heavily culled plots were excluded (root mass fraction, root lateral fraction, and root median diameter, ns). Conversely, two traits that were previously non-significant became significant (root crown average hole size and shoot lower area; p < 0.05).

3.2. Correlation among traits

Analysis of the whole dataset revealed a major cluster of positively correlated traits encompassing biomass traits, plant size traits, and root and shoot shape traits (Fig 2A, S5 Table). In contrast, the majority of plant traits were negatively associated with specific root length. For shoot width to depth ratio, little to no correlation with other plant traits was observed. A strong positive correlation was observed between root and shoot biomass (R = 0.93, p < 0.01, Fig 2B), and a similar moderate relationship was found between shoot biomass and root system maximum depth (R = 0.50, p < 0.01, Fig 2C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Phenotype correlations among 46 hemp genotypes.

(A) Correlation matrix with positive correlations between traits depicted in red color, negative correlations in blue color, and X symbol marks all bivariate correlations that were not significant with a P value set at 0.05. (B) Linear regression between shoot dry biomass and root dry biomass. (C) Regression between shoot dry biomass and root system maximum depth. Data from S5 Table.

https://doi.org/10.1371/journal.pone.0339929.g002

3.3. Hemp genotype quadrant grouping by root and shoot size traits

Hemp genotypes were classified into four groupings by root and shoot size traits (S6 Table). Four groupings were devised: (I) big root and big shoot, (II) big root and small shoot, (III) small root and small shoot, (IV) small root and big shoot (Fig 3A). The phenotypes used for the classification were root biomass, total root length, root convex area, shoot biomass, and shoot convex area. Percentile rankings were assigned to each hemp genotype per phenotype and the phenotype data scaled. Quadrant groupings were assigned when a genotype fulfilled the upper or lower 50% percentile of each respective phenotype (Fig 3B). The majority of the hemp genotypes fell into group I and group III, with 16 and 15 genotypes, respectively. One genotype was assigned into group II and one genotype was assigned to group IV. The remaining 13 genotypes did not fulfil all requirements to be assigned a quadrant group, one genotype resembled closely to group I, four genotypes resembled group II, five genotypes resembled group III, and three genotypes to group IV. K-means clustering of the PCA results largely supported the initial genotype groupings (S1 Fig). When clustered into three groups, most genotypes in group I and group III corresponded to clusters 1 and 3, respectively (S6 Table). Increasing the number of clusters to four introduced more fine-scale genotype differences, however groups I and III remained in general distinct, and the genotypes that had not fulfilled a quadrant grouping were also in a separate and distinct cluster (S6 Table). Bootstrap Jaccard similarity values indicated low–moderate stability for the three-cluster solution (0.43–0.52). With four-clusters it showed one highly stable cluster (0.81) and three clusters of moderate stability (0.43–0.60).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Hemp genotype quadrant grouping by root and shoot size traits.

(A) Quadrant designations assigned based on 50th percentiles for multiple root and shoot size traits (Root biomass, total root length, root convex hull area, shoot biomass, shoot convex hull area, shoot height). (B) Normalized root and shoot convex hull data with the multiple trait quadrant color assignment. If a genotype did not fulfill the requirements of 50th percentiles for each trait then they were grouped as NA.

https://doi.org/10.1371/journal.pone.0339929.g003

3.4. Identification of root candidate genes in Cannabis in hemp

Despite the growing economic and ecological interest in Cannabis sativa, little is known about the genetic regulation of its root traits due to a historical lack of genetic resources and breeding efforts. To address this, we have employed root phenotyping and in silico comparative genomics to provide a framework for dissecting root system architecture (RSA) in industrial hemp. The phenotyping effort revealed substantial and heritable variation in root traits such as total root length, convex hull area, and biomass among 46 diverse hemp genotypes, highlighting the potential for genetic improvement. Linking this phenotypic diversity with underlying genetic determinants, an in silico analysis identified a total of 301 root system architecture-related orthologues from maize, rice, and Arabidopsis, including 74 high-confidence hemp candidate genes with functional relevance (S7 Table). Duplicate gene entries represent genes that are orthologous to multiple loci in maize, rice, or Arabidopsis. These RSA genes were primarily grouped into four clusters and further divided into nine clusters based on phylogenetic analysis using a 95% sequence similarity threshold for these genes (Fig 4). Among these 301 genes, 74 root genes identified as homologs of known RSA genes which have potential control of root system architecture in hemp (S7 Table). The detailed list of homologous hemp genes with known root system architecture genes from maize, rice, and Arabidopsis is provided in S7 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Dissection of priority root candidate genes for hemp.

(A) Maximum likelihood phylogenetic tree of 301 Cannabis sativa genes identified as orthologs of known RSA regulators from Arabidopsis thaliana, Oryza sativa (rice), and Zea mays (maize). Protein sequences were aligned using MUSCLE, the tree was constructed in MEGA11 with 1,000 bootstrap replicates, visualized in FigTree. Gene clusters were grouped based on ≥95% sequence similarity and are color-coded into nine distinct functional clusters. (B) Key root architecture genes identified through phylogenetic analysis, along with their known functions and representative homologs from model species. (For additional genes, refer to S7 Table).

https://doi.org/10.1371/journal.pone.0339929.g004

Key root genes were primarily highlighted in grey (7 genes), yellow (8 genes), and dark grey (5 genes). Among them, seven genes that have been functionally characterized in maize (ZmDRO1, ZmRt1), rice (OsNRT1, OsDRO1), and Arabidopsis (AtDro1, AtNRT2.1, AtUTR7) were identified. The XP_030478835.1, XP_030482378.1, XP_030490379.1, and XP_030490380 genes are homologs of DEEPER ROOTING 1 (DRO1), an auxin-response gene that is directly regulated by Auxin Response Factors that control root growth angle and promote deep rooting [17]. The maize rootless1 (ZmRt1) encodes a bHLH transcription factor that governs nodal root development in maize and shares phylogenetic similarity with three hemp genes (XP_030484730.1, XP_030481898.1, and XP_030484738.1). Seven genes homologous to the Arabidopsis and rice genes (AtNRT2.1 and OsNRT1), known nitrogen transporters, were identified, which might be potentially responsible for the nitrogen transport pathway in hemp [46,47]. A homolog (XP_030502130.1) to AT4G31600 (AtUTR7) encodes a Golgi-localized UDP–glucose/UDP–galactose transporter that regulates lateral root emergence in Arabidopsis was also identified [48]. This genome-wide analysis underscores the conservation of RSA regulatory networks across plant species and provides a curated gene set for downstream validation in hemp. These genes are prime candidates for genetic manipulation via marker-assisted selection or gene-editing technologies to improve RSA-related traits in breeding programs. A stringent identity cutoff was employed to reduce false positives and to focus on a conservative set of orthologs with the strongest likelihood of functional equivalence and therefore this approach prioritizes confidence over completeness and may under-represent the total number of true functional homologs. Further development of breeding populations and validation with experimental evidence will be necessary to confirm the functions of candidate genes related to root traits.