The search for yield predictors for mature field-grown plants from juvenile pot-grown cassava (Manihot esculenta Crantz)

Cassava is the 6th most important source of dietary energy in the world but its root system architecture (RSA) had seldom been quantified. Ability to select superior genotypes at juvenile stages can significantly reduce the cost and time for breeding to bridge the large yield gap. This study adopted a simple approach to phenotyping RSA traits of juvenile and mature cassava plants to identify genotypic differences and the relationships between juvenile traits and harvest index of mature plants. Root classes were categorised and root and shoot traits of eight (8) juvenile pot-grown cassava genotypes, were measured at 30 and 45 days after planting (DAP). The same or related traits were measured at 7 months after planting of the same genotypes grown in the field while yield and yield components were measured in 12-months old field-grown plants. The field experiment was done in 2017 and repeated in 2018. Differences between genotypes for the measured traits were explored using analysis of variance (ANOVA) while traits in juvenile plants were correlated or regressed onto traits measured in 7- and 12-months old plants. The results show significant genotypic variations for most of the traits measured in both juvenile and 7-months old plants. In the 12-months old plants, differences between genotypes were consistent for both 2017 and 2018. Broad-sense heritability was highest for the number of commercial roots (0.87) and shoot fresh weight (0.78) and intermediate for the total number of roots (0.60), harvest index (0.58), fresh weight of roots (0.45). For all the sampling time points or growth stages, there were greater correlations between traits measured at a particular growth stage than between the same traits at different growth stages. However, some juvenile-mature plant trait relationships were significant, positive and consistent for both 2017 and 2018. For example, total root length and the total number of roots in 30 DAP, and branching density of upper nodal roots in 45 DAP, positively correlated with harvest index of 12-months old plants in both 2017 and 2018. Similarly, the diameter of nodal roots, for example, had a negative, significant correlation with fresh shoot biomass of mature plants in both 2017 and 2018. Regression of traits measured in 30 DAP explained up to 22% and 36% of the variation in HI of mature plants in 2017 and 2018, respectively. It is concluded that the simple, rapid, inexpensive phenotyping approach adopted in this study is robust for identifying genotypic variations in juvenile cassava using root system traits. Also, the results provide seminal evidence for the existence of useful relationships between traits of juvenile and mature cassava plants that can be explored to predict yield and yield components.

cassava. Cassava roots, the main economic part of the crop, could be as long as 2m deep in the ground [16], making excavation and quantitative analysis cumbersome, prone to inaccuracies, and ill-suited to screening large genetic populations. The ability to screen for variation in cassava root traits in the early stages of growth could circumvent some of the challenges and limitations outlined earlier and help identify useful root traits in genotypes. Traits related to shoot biomass, root diameter and branching density have the potential for use as predictors of water and nutrient use efficiency in crops [14] and might be useful for the selection of cassava genotypes during the early growth stages. The objectives of the current study were to measure the (i) genotypic variation in root traits of both juvenile and mature cassava genotypes; (ii) explore relationships between the yield of mature field-grown cassava and juvenile root traits.

Plant material
The eight cassava genotypes used in this study have been previously described in [14]. One of the genotypes is released variety called 'Capevars bankye' (herein designated as '8H'). Five of the genotypes (designated 1A, 2B, 5E, 6F and 7G) have recently been recommended for release following field inspection by the National Variety Release and Registration Committee of the Ministry of Food and Agriculture (MoFA), Ghana. Genotypes 1A, 2B and 5E are yellowfleshed varieties with high β-carotene and are suitable for local dishes and processing. Genotypes 6F and 7E are resistant to CMD and are suitable for local dishes and processing. The remaining genotypes included in the study are designated 3C and 4D. All the genotypes have similar time to tuber initiation and mature in 12 months.

Soil and environmental conditions
This study involved both field and pot experiments conducted at the Teaching and Research Farm at the University of Cape Coast (5˚06 N, 1˚15' W). The study site experiences two seasons of rainfall with a peak in May to June and the minor in October. The dry season occurs between November and February [23]. The average precipitation recorded were 438 and 885 mm for the 2017 and 2018 cropping periods, respectively. The average temperatures were 25.8 and 23.9˚C and the average relative humidity was 84.4 and 85.4% for the 2017 and 2018 cropping periods, respectively. Normal day length at the experimental site ranges from approximately 11.30 to 12.40 h [24], while solar radiation ranged from 17-30 MJ m -2 day −1 (mean ca 23 to 25 MJ m -2 day −1 ) day −1 . The properties of the soil for both the field and pot experiments have been described previously [23].

Pot experiment
Topsoil (0-15 cm) was collected from a land very close to the plots for the field experiments. The setup in [14] was adopted for the pot experiment. In brief, nursery polybags of 45 cm long and 30 cm wide, with drainage holes at the bottom, were used as pots. The pots were filled with air-dried soil to a bulk density of approximately 1.1 g cm -3 and kept under a rain shelter. Before planting, the soil in each pot was watered with tap water to 80% field capacity (FC), determined gravimetrically, and allowed to drain overnight. Stem cuttings of approximately 20 cm of each genotype were randomly planted in the soil at an incline of about 45˚from the centre of the pots, making sure that at least six nodes were within the soil. The cuttings were obtained from disease-free stems. Each genotype had eight replicates for each sampling period. Pots were positioned side by side with the assumption that pots' spacing would not cause mutual shading of the plants at 45 days after planting (DAP). Positions of pots were rotated under the rain shelter every 10 days to reduce the effects of possible environmental gradients [14]. The soils in the pots were maintained at approximately 70% FC during the growth period by watering to weight with tap water every three to four days.
Harvesting or sampling was done at 30 and 45 days after planting (DAP). While we have previously shown that genetic variations in cassava root systems could be evident at 30 and 45 DAP [14], we also considered that screening at these early growth stages was appropriate based on our hypothesis that improving aspects of the juvenile root systems of cassava can accelerate access to soil resources, leading to rapid crop establishment and, consequently, greater yields. At each sampling, a blade was used to carefully cut through the longitudinal stitch lines on the sides of each pot (polybag) to expose the soil and the roots. The exposed soil and roots were then soaked in a basin of water for 3-4 minutes. Roots were removed and gently washed free of residual soil, using water hose at low pressure to minimize damages to roots [14]. Excavation and cleaning of roots were conducted by two individuals and they required an average of 6 minutes to complete the washing of one root system. Visual measurements of the root system were then conducted.

Field experiment
Two independent trials were conducted in 2017 and 2018. Maize and cowpea were normally grown at the site but plots for the experiments had been lying fallow for one and two years, respectively, for the 2017 and 2018 trials. Each of the trials was arranged in a randomized complete block design with two blocks containing the replications of each genotype. In both trials, each plot consisted of five rows, each measuring 10m long, with individual plants separated from one another by 1m in all directions and surrounded by a guard row. Stem cuttings were obtained and planted similarly as in the pot experiment. The experimental field had been slashed, ploughed and harrowed to a depth of about 30 cm, and was not ridged. The cuttings were directly planted into holes dug with a cutlass, typical of cassava planting in the region. In both 2017 and 2018, planting was done in May, typical of the main cassava planting season in Ghana. The plants were grown under rain-fed conditions and managed using best practices typical of the region where no external inputs are applied and weeds are controlled manually when necessary.
Twelve replicate plants within the middle section of the plots were used for analysis in both years. In each year, these plants were harvested 12 months after planting (12 MAP) for final yield parameters. At maturity, plants were excavated by a team of three or four people, using cutlasses, hoes and fork to loosen the soil, taking care not to damage any roots. Once the soil had been loosened, the roots were harvested by pulling the stem along with the roots out of the ground. In most cases, harvesting was done after rainfall to facilitate the process. In the few instances when roots were broken, they were also collected, added to the wholly excavated roots and taken to the laboratory for counting and measuring. In addition to the yield measurements in both years, in 2018, there was additional detailed shoot and root system analysis at 7 MAP, using 4 replicate plants per genotype per plot. In order not to disrupt planting density and the potential effect on yield at 12 MAP, each plot was arbitrarily divided into two in the 2018 trial. Plants from one-half of the plot were harvested at 7 MAP, leaving the plants in the other half of the plot to grow to maturity for the yield measurements at 12 MAP. Roots were excavated similarly as described previously.

Traits measured in pot-grown plants.
Summary of all measurements in this study is shown in Table 1. In the pot-grown plants, shoots, including the leaves, were weighed to determine shoot fresh weight (SFW) and oven-dried at 80˚C for three days to determine shoot dry weight (SDW). Following root measurements, roots were weighed and oven-dried at 80˚C Branching density of upper nodal roots (roots cm -1 ) Branching density of lower nodal roots (roots cm -1 ) for three days to determine root fresh weight (RFW) and root dry weight (RDW), respectively. The root-to-shoot ratio was calculated as the quotient of RDW and SDW. The protocol described by [14] was followed for the extraction and classification of root system features into three categories: upper nodal roots (UNRs), lower nodal roots (LNRs), and basal roots (BRs) ( Fig 1A and Table 1). Total root length (TRL), the number of upper and lower nodal roots (NUNR and NLNR, respectively), the diameter of upper and lower nodal roots (DUNR and DLNR, respectively), and branching densities of upper and lower nodal roots (BdUNR and BdLNR, respectively), and the total number of nodal roots (TNR) were manually measured. Similarly, the number, diameter and branching density of basal roots (NBR, DBR, and BdBR, respectively) were measured. The root-to-shoot ratio was then estimated. Three representative roots were randomly selected from each of the three root categories (Fig 1A) per plant for the measurement of root diameter and branching density. Root diameter for each root type was determined at three (3) cm from the stem cutting using digital callipers. Branching density for each category of the root was determined within six (6) cm distance. Total root length was measured by spreading and suspending the roots in water in a rectangular glass dish with a black background, taking care to avoid roots overlying on each other. Images of the total root system were captured with a Canon EOS 70D DSLR camera (https:// www.usa.canon.com/) held stationary on a tripod at 50 cm above roots. Images were converted to a binary image and TRLs were extracted from root images using skeletonization routines [14] in ImageJ (US National Institutes of Health, Bethesda, MD, USA, https://imagej.nih.gov/ ij/). Depending on the experience of the investigator, measurements on each juvenile root system took between 6 and 8 minutes.
2.5.2 Traits measured at 7 MAP. Twenty-two traits were measured at 7 MAP and included branch level number (BN), commercial roots number (CRN), feeder roots diameter (FeRD), feeder roots length (FeRL), feeder roots number (FeRN) and fibrous roots diameter (FiRD) (Fig 1B). Other traits measured were fibrous roots length (FiRL), root fresh weight (fRFW), shoot fresh weight (fSFW), harvest index (HI), leafless stem height (LSH) and peduncle diameter (PD). Additionally, peduncle extent (PE), peduncle length (PL), primary stem diameter (PSD), primary stem length (PSL), primary stem number (PSN), secondary stem diameter (SSD), secondary stem length (SSL), tuberous roots diameter (TRD), tuberous roots length (TRL) and tuberous roots number (TRN) were measured at 7 months. BN, CRN, FeRN, PSN and TRN were counts of the number of divisions of vegetative branches, commercial or marketable roots, feeder (non-storage) roots, primary stems and tuberous (storage) roots, respectively. FeRD was measured with a calliper 3 cm from the insertion junction on main roots. Subsequently, the feeder roots were severed from the root system and their lengths measured with a ruler to obtain FeRL. The TRL was measured from the neck of the storage root to the beginning of the fibrous root at the end of the storage root. TRD was measured in the middle of the storage root. Fibrous roots are non-storage roots at the end of root tubers ( Fig 1B). FiRD was measured in the middle of the fibrous root with a calliper and FiRL was measured with a ruler. LSH was measured with a tape measure from the soil surface to the longest branch next to the leaf-bearing branch. PSL was measured as the vertical height of the longest stem using a tape measure and from the soil surface to the first primary branch on the stem and the PSD was measured with a calliper in the middle of the primary stem. SSL was determined as the length of the first branch and the SSD determined with a calliper in the middle of the secondary stem. The most frequently occurring root peduncle extent (PE) was scored with a three-score system; 0 for sessile, 3 for pedunculated and 5 for mixed. Where applicable, PL was determined with a ruler and PE with a calliper in the middle of the peduncle. Fresh shoot and root biomass (fSFW and fRFW) were determined with measuring scale and used also to calculate HI.

Traits measured in 12 MAP plants.
At maturity, the above-ground material including stems and leaves were bulked together as shoot biomass and weighed. The yield parameters evaluated at maturity were the total number of roots, root fresh weight at maturity, the number of commercial roots at maturity, shoot fresh weight at maturity and root fresh weight at maturity, all calculated on a plant basis. Harvest index at maturity was calculated as the quotient of root fresh weight at maturity and total fresh biomass.

Statistical analyses
Analyses for the 30 and 45 DAP pot-based screening were independently performed using descriptive statistics. General analysis of variance was performed for the main effect of genotype. Similarly, the data for the 7-month field screening was analysed to determine descriptive statistics, including mean (� x) and range, followed by a general analysis of variance of genotype and block main effect and genotype-by-block interactive effects. Multivariate analysis of trait space was carried out on the 7-month field screening data employing principal components analysis (PCA) to identify major traits accounting for most of the variation among the studied field-grown cassava genotypes. The PCA was based on the correlation matrix and the number of significant principal components was determined based on the eigenvalue-one criterion, retaining any component with an eigenvalue greater than one [19,25,26]. Following the PCA, the squared cosine (cos 2 ) was computed which gave the quality of representation of the variables on the factor map and the total contribution of individual traits (contrib). Analysis of variance was also conducted on yield and yield-related data measured on the mature (12 months old) field-grown plants for both 2017 and 2018. Pearson's correlation coefficients for pairs of the shoot and/or root traits were calculated for plants sampled on pot-grown 30 and 45 DAP juvenile plants, as well as on 7-and 12-month field-grown plants at a level of 5%. Subsequently, simple linear regression models were fitted on significantly correlated traits and where multiple traits at the juvenile stage were significantly correlated with a trait at maturity, Generalized Linear Model (GLM) was applied for multiple regression. For the mature plants' data, broad-sense heritability (H 2 ) across years was estimated as the quotient of the estimated variance associated with the genotypic effect and the total phenotypic variance for a given trait (σg2 /σp2) [23,27]. The phenotypic variance was calculated using Eq (1) as applied by [23]: where: r is the number of replicates, n is the number of years and σg 2 × y is the genotype x year variance. All ANOVA were performed using GenStat (GenStat Release 12.1, VSN International, Oxford, UK). The FactoMineR and the 'corrplot packages in the R software, the Language and Environment for Statistical Computing [28,29,30] were used for PCA and correlation graphics. Regression models were fitted in Microsoft excel and subsequently, multiple regression was conducted in GenStat (GenStat Release 12.1, VSN International, Oxford, UK). The package Factoextra was used for the visualization of cos 2 and the contrib results [28].

Genotypic variation in traits of juvenile plants
Significant differences between genotypes were found for most of the traits examined (14 of the 18 traits examined at both 30 and 45 DAP; p<0.01; S1 Table). For some traits, the ranking of genotypes was however not consistent for the two sampling dates (S1A-S1H Fig Table).

Variation in 7 months' field-grown genotypes.
Significant phenotypic variation (p<0.05 or p<0.001) was observed for 18 of the 22 shoot or root traits examined in plants sampled at 7 MAP (S2 Table). Largely, variation between blocks was significant for a few traits and block-by-genotype interaction was also rare (S2 Table). The minimum value for some traits was zero and for such traits, there was a greater range of variation. These traits included branch level number (BN), commercial roots number (CRN), feeder roots-, fibrous roots-, and peduncle-, as well as secondary stem-related traits (S2 Table). Among the traits which recorded nonzero minimum values, traits with a 10-fold or greater range of variation included root fresh weight and shoot fresh weight, for which there were approximately 13 and 30-fold range of variation, respectively (S2 Table). One of the more broadly variable root architectural traits was tuberous roots number, for which the maximum value was 7.5 times greater than the minimum value. Phenotypic variation was also observed for primary stem length, tuberous roots diameter, and tuberous roots length, as well as leafless stem height and harvest index (HI) with approximately 8, 2, 3, 4 and 1-fold range of variation, respectively (S2 Table).
Shoot fresh weight varied from 0.95± 0.14g (7G) to 3. Again, while genotype 5E may be partitioning root biomass into larger root mass, genotype 8H produced many roots of various categories ( Fig 2B). Secondary stem length varied from 10.   Fig 2A  shows the quality of representation of the variables (cos 2 or squared coordinates) on the factor map for dimensions, including those considered significant following the PCA. Traits including BN, PSD, SSL, SSD, fSFW and fRFW were well represented on PC1 with cos 2 between 0.46-0.57 (Fig 2A). Peduncle extent and length were well represented on PC2 with cos 2 of 0.46-0.57, whilst FiRD and FeRD were respectively well represented on PC6 and PC7 with cos 2 of approximately 0.55 (Fig 2A). Plot highlighting the most contributing variables for each dimension considered significant after PCA is presented in Fig 2B. The first seven principal components (PCs) with an eigenvalue greater than one explained 78.1% of the total variation for the 22 shoot and root system traits examined for the 7-month field-grown cassava plants (S3 Table and Fig 2B). The first dimension loadings, which accounted for 22% of the variation, largely separated stem-and biomass-related traits (e.g.: primary stem diameter, secondary stem length and diameter, leafless stem height, fresh shoot and root biomass) from other traits (S3 Table and Fig 2B). The relative magnitude of eigenvectors for PC2 was 17.4%, explained mostly by the peduncle-and tuberous root number-related traits (S3 Table and Fig 2B). One (feeder roots number), three (primary stem number, tuberous roots length and fibrous roots diameter), two (fibrous roots length and harvest index), one (feeder roots length) and one (feeder roots diameter) traits were significantly correlated to the third to seventh dimensions, with contributions of 11.5, 8.1, 6.9, 6.5 and 5.8%, respectively (S3 Table and Fig 2B). Ten traits, including SSL, SSD, fRFW, PL, BN, fSFW, CRN and TRN contributed to the variability in the first two dimensions (Fig 2C). A similar observation was made in the analysis of the total number of roots, which comprised of the sum of commercial, non-commercial and feeder roots (Fig 6G and 6H). The effects of genotype and the interaction between genotype x year of trial x block accounted for most of the experimental variation ( Table 2). The effect of genotype alone ranged from 3.2% for root fresh weight to 43.6% for shoot fresh weight ( Table 2). Broad-sense heritability (H 2 ) estimates were generally intermediate to high. The H 2 was highest (0.87) for the number of commercial roots, followed by (0.78) for shoot fresh weight. Root fresh weight had the least H 2 (0.45; Table 2). There were also significant correlations between traits measured in juvenile and matured plants but these correlated traits were generally fewer than the number of correlated traits at a given growth stage. Approximately between 12 and 13% of all potential correlations were statistically significant (p < 0.05) between traits measured in juvenile plants and those at 7 MAP (Fig 3-II). Similarly, approximately between 9 and 31% of all potential correlations were statistically significant (p < 0.05) between traits measured in juvenile and 12-month old-fieldgrown plants (Fig 3-IV). Traits measured in plants at 7 and 12 MAP also showed significant correlations for approximately between 14 and 19% of all potential correlations (p < 0.05; Fig 3-V).

Comparison of juvenile and mature cassava traits
Simple linear regression models of selected significant correlations between juvenile and 7 MAP, and between juvenile and 12 MAP are presented in Figs 4-7. A simple linear model of the number of roots (Fig 4A), plant biomass ( Fig 4B) and TRL (Fig 4F) measured in 30-day old juvenile plants, with primary stem length from the 7 MAP as the dependent variable, produced a significant positive relationship. Other positive relationships between 30 DAP and 7 MAP traits included DLNR and RFW (Fig 4C), branching density of basal roots and number of feeder roots (Fig 4D and 4E). Biomass (Fig 4G) and root number (Fig 4F) of 30 DAP plants produced a significant negative relationship with the primary stem length of 7 MAP.
A linear model of branching density of different root categories in 45 DAP showed a significant positive relationship with primary stem length ( Fig 5A) and leafless stem length (Fig 5C) of plants at 7 MAP. Other positive relationships between 45 DAP and 7 MAP traits included the number of roots and primary stem length (Fig 5B), number of basal roots and biomass ( Fig  5D) and number of feeder roots (Fig 5F), as well as TRL and tuberous root number (Fig 6E). The diameter of roots of plants at 45 DAP produced a significant negative relationship with the primary stem length of plants at 7 MAP (Fig 5G). Total root length at 45 DAP also produced a significant negative relationship with the primary stem length at 7 MAP (Fig 5H).
Pearson's correlation coefficient analysis of juvenile and mature plant traits revealed intermediate positive and negative significant correlations between some traits measured in 30  were between -0.44 to -0.30 and 0.29 to 0.55 (Fig 3A and 3B). Most 30 DAP traits which correlated with mature-plant traits measured in 2017, also correlated with same mature-plant traits

PLOS ONE
Predicting cassava harvest index from young plants  (Figs 3A, 3B and 6). For example, biomass (Fig 6A), total root length ( Fig  6B) and the total number of roots (Fig 6D) positively correlated with both 2017 and 2018 harvest index at maturity. Similarly, the diameter of nodal roots consistently had a negative significant correlation with both 2017 and 2018 fresh shoot biomass (Fig 6F). There were some traits measured in 30 DAP plants that correlated with mature plants traits only in one of the trial years. These included branching density at 30 DAP and number of commercial roots at maturity in 2018 (Fig 6C), root diameter at 30 DAP and HI at maturity in 2018 (Fig 6E), total root length at 30 DAP and shoot fresh biomass at maturity in 2018 ( Fig 6G) and diameter of upper nodal roots at 30 DAP and the total number of roots at maturity in 2017 (Fig 6H).  7F). Fig 7G-7I show examples of significant negative correlations between 45 DAP and mature plant traits. Although many traits measured from 45 DAP juvenile plants significantly correlated with traits of mature field-grown cassava plants, the coefficient of determination from the simple linear regression models, which ranged from 0.083 to 0.22 (Fig 7A-7I

Genetic variation and robustness of phenotyping approach
The identification of superior genotypes based on multiple traits is a key objective of cassava improvement trials. In root crops, the ability to identify these genotypic differences, especially in root system architecture, at the juvenile stage could substantially reduce the cost and time of selection and breeding efforts. This could be aided by a robust, inexpensive and rapid phenotyping approach. In the current study, multiple traits, including those said to be important for resource capture in cassava [6,16,22], were measured. At the juvenile stage, we adopted the root categorisation presented in [14] and [16] to divide the cassava roots into three classes, namely, upper nodal, lower nodal and basal roots. Previously, evidence was provided that welldeveloped branching pattern, including the number and lengths of nodal and basal roots, were associated with resource capture in cassava [16]. The results in the current study showed significant genotypic variations in many of the traits measured from both the juvenile plants (S1 Fig) and plants at 7 months after planting (MAP) (S2 Fig) and 12 MAP (S3 Fig). At maturity, the differences found were consistent across the two years of screening. Evidence has been provided of similar genetic variation for juvenile traits among cassava lines [31]. Indeed, [14] reported of similar root genotypic variation in total root length, root numbers, root diameters and root branching density in the same panel of cassava cultivars used in the present study. This suggests that our simple, inexpensive phenotyping approach to characterize cassava RSA at the juvenile stage is robust and could provide a time-saving and less laborious option for a multi-trait selection in breeding for improved genotypes in cassava. The simple phenotyping approach adopted in this study helps obviate several constraints to root phenotyping in general and cassava in particular.
Genotype H8 is a white-flesh cultivar called 'Capevars bankye'. This cultivar, which is highly preferred by farmers, has been released and cultivated since 2005. Consistent with the data in [14], it was ranked among the top-performing lines for traits such as branching density and total root length (S1 Fig). The remaining seven genotypes used in the present study were bred for high carotenoid content and resistance to cassava mosaic disease and five genotypes, namely 1A, 2B, 5E, 6F and 7G, have just been recommended for release in Ghana. Generally, these genotypes were ranked above genotypes 3C and 4D for many traits that were measured at 30 and 45 DAP and, therefore, it is interesting that 3C and 4D have not been endorsed for release. Indeed, poor yield potential of 3C and 4D was more evident in the 12 MAP field experiment when they obtained the least harvest index and shoot fresh weight, respectively (S3 Fig). The significant differences found between the genotypes in the juvenile experiment thus emphasise the importance of screening cassava at the juvenile stage for desirable characteristics, which include both farmer-preferred traits such as yield potential and breeder-preferred traits such as soil resource capture and use efficiency. Certain general trends were evident from the RSA measurements at the juvenile stage. There was a significant difference in most traits between the two sampling dates (S1 Table). While branching density, the number of roots and specific root length decreased from 30 to 45 DAP, total root length and root diameters increased between the two sampling time points (S1 Fig). In [16], similar observations were made where the total number of axile roots increased until 60 DAP, thereafter decreased, except that in the present study, decrease in axile started only after 30 DAP. Since root diameters increased from 30 to 45 DAP, we hypothesize that the abscission of axile roots from the cutting leading to a reduction in root number and density might be a cue for the initiation of storage roots thickening [16]. The initiation of storage root formation in cassava could start as early as 45 DAP [32]. Moreover, the soil used in this study was unamended with any additional nutrients and so it is conceivable that after 30 DAP soil nutrients would have decreased. Reduced root number and branching, in tandem with increasing root length, might be adaptive strategies to reduce the metabolic costs of soil exploration and improve the acquisition of limiting soil resources [33,34].

Trait contribution to variation
The results of the PCA (S3 Table) suggest that in the panel of cassava genotypes used in the present study, shoot and root-related traits are contributing almost equally to the genetic variability of near-maturity field-grown cassava, and therefore both could be exploited in the breeding of genotypes for targeted environmental conditions and higher harvest index. If the first four PCs, which accounted for approximately 60% of the total variation are considered, then primary stem-related traits (diameter, number, length), secondary stem length and diameter, leafless stem height, fresh shoot and root biomass, peduncle-related traits (length, diameter and extent), tuberous rootsrelated traits (number and length), feeder roots number, and fibrous roots diameter, were responsible for most of the phenotypic variation in the 7-months old field-grown plants. As was observed in the PC loadings on PC1 and PC2, the results of quality representation showed that BN, PSD, SSL, SSD, fSFW and fRFW were well represented on PC1 with cos 2 between 0.46-0.57. Traits that are correlated with PC1 and PC2 could be the most important in explaining the variability in the dataset and may be considered in efforts to breed for improved genotypes [14,23]. Here, 10 traits, including SSL, SSD, fRFW, PL, BN, fSFW, CRN and TRN contributed to the variability in the first two dimensions (S3C Fig), suggesting that these traits are sufficient to differentiate cassava genotypes. Screening for these physiological traits at 7 MAP, could be time-and resource-saving, given that main agronomic evaluation of cassava are commonly conducted at the end of the crop cycle, making the selection process protracted and costly [6].

Components of variance and heritability in 12-months field-grown cassava
Conventional physiological traits measured in studies of cassava at 12 MAP include above and below-ground biomass and number of storage roots [6]. Similarly, these traits were measured at 12 MAP and subsequently computed HI for two years of field evaluation. Reliably estimating variance components and heritability are vital requisites for selection gain and improvement in quantitative traits [23,35]. Over the two years, the effects of genotype, block and year of trial accounted for most of the experimental variation but various interactions contributed relatively little (Table 2). It was evident that some vagaries in experimental conditions between the two trials might be contributing to the variation in some of the traits examined, in which case, the residual proportion (s 2 ε ) was larger than the genotypic (s 2 g ), the block (s 2 b ) or the trial (s 2 y ) effect (S2 Table). It has been suggested that under conditions of high residual variance, the within-genotype variation might be high and/or there may be the need for a more parsimonious model [23]. Broad-sense heritability was highest for the number of commercial roots (0.87) and shoot fresh weight (0.78) and intermediate for the total number of roots (0.60), harvest index (0.58), fresh weight of roots (0.45) ( Table 2). The H 2 estimates here are comparable to that reported by Oliveira et al., (2015) for the number of roots (0.51±17) and root and shoot biomass (0.80±0.21) of 12 MAP field-grown cassava under irrigated conditions. Shoot-related traits have been reported to have larger H 2 than root-related traits [27,36,37], so it is interesting to note that the number of commercial roots obtained the highest H 2 in the present study. The relatively higher heritability estimates for root number and fresh weight of shoot might be indicative of fewer genes controlling these parameters and low environmental influence on the expression of these phenotypes [38,39], and pointing to the possibility of using simple selection methods for the improvement of these traits in response to stress conditions [39,40]. This hypothesis is also supported by the fact that the heritability estimate for HI was relatively lower, given that HI is a complex trait computed from multiple traits and therefore is expected to be under the control of several genes.

Relationships between juvenile RSA traits and yield components of mature plants
There is a pressing need to use root phenotyping to assess the productivity of crop plants at an early stage [6]. To this end, we explored the relationships between phenotypic traits measured at the juvenile stage (30 and 45 DAP) and root or yield component traits measured in plants at 7 or 12 MAP. Comparing juvenile plant RSA traits with field traits or field agronomical performance has been reported in several studies, for example in alfalfa [41]; bread or durum wheat [42][43][44][45]; common bean [46]; maize [47][48][49]; oilseed rape [50]; potato [11]; red clover [51] and tomato [52]. The phenotyping technique adopted in the current study allows costeffective and rapid measurement of cassava root traits, but the roots of juvenile plants growing in pots may not develop exactly as in nature [11,42,53]. Moreover, there is no evidence for predicting performance in heterogeneous environments based on a single RSA trait [54]. It has also been suggested that plastic responses of growth linked to the different experimental systems (screen-house versus field), might lead to overly large variations to identify trends in a dataset [12]. Notwithstanding, a systematic root phenotyping system was adopted (Fig 1), over two sampling times (30 and 45 DAP), to provide a useful array of root and shoot traits in juvenile cassava that were related to desirable traits in the mature cassava plants.
Significant correlations (p<0.05) between the juvenile RSA traits, and between field-measured traits were observed at 7 MAP (Figs 4 and 5) and between field-measured yield and yield-related traits at 12 MAP (p<0.05) (Figs 6 and 7), suggesting that the RSA of 30-or 45 day-old young cassava plants could be explored for predicting field performance of agronomic traits. Moreover, there were several significant, positive inter-traits correlations for a given growth stage than between growth stages in both juvenile and mature plants (Fig 3), suggesting that these correlated traits could be improved simultaneously or some, selected for, as proxies for others [23].  (Fig 4). However, there were negative relationships between some traits. For example, the diameter of upper nodal roots consistently had a negative relationship with the root-to-shoot ratio at both 30 DAP (r = -0.32) and 45 DAP (r = -0.55), pointing to a positive allometric relationship between the diameter of upper nodal roots and root biomass in cassava.
The root system size of juvenile cassava plants, as represented by the number of roots, diameters, fresh biomass of roots, branching density of different root classes measured in both 30 DAP (Fig 4) and 45 DAP (Fig 5) plants, was positively associated with various traits measured in 7 MAP plants, including primary stem length (Figs 4B and 5A), leafless stem length (Fig 5C), number of feeder roots (Fig 4D and 4E), biomass ( Fig 5D) and tuberous root number (Fig 5E). A few inconsistencies found in the correlations between traits measured at 30 and 45 DAP with the traits measured at 7 MAP could be due to some form of stress which was possibly initiated in the plants grown up to 45 days as a result of a decline in soil nutrients in the pots. The present results suggest that an increase in shoot and root biomass in 7-month fieldgrown plants could be linked to the root system size of younger plants. Evidence has been provided that bigger and vigorous root systems of younger plants are beneficial to acquire more soil resources for early plant growth [42,55,56,57]. Some cassava genotypes, especially early maturing cultivars are harvestable at 7 MAP. [6] suggested that evaluating cassava genotypes at 7 MAP can be used to select good varieties of potentially good yield at 12 MAP. The results of the present study suggest that at the juvenile growth stage, it would be possible to identify genotypes with longer primary stems, more tuberous and feeder roots, as well as increased biomass, which is some of the main characteristics associated with an improved final yield of cassava in the field.
The best GLM for HI for measurements taken at 30 DAP included some consistent traits including fresh shoot and root biomass, as well as traits related to root system size such as total root length and the total number of roots, for both the 2017 and 2018 datasets and explained up to 36% of the variation in HI of 12 MAP field-grown plants. The best GLM for HI for measurements taken at 45 DAP included some consistent traits including branching density of upper nodal roots and number of basal roots, for both the 2017 and 2018 datasets and explained up to 26% of the variation in HI of 12 MAP field-grown plants. Similarly, the best GLM for the total number of roots at 12 MAP for both 2017 and 2018 consistently included the total number of roots from measurements taken at 45 DAP, suggesting that many root traits at the early growth stage are associated with the number of tubers and yield at maturity. Overall, GLMs for HI for measurements taken at the juvenile stage of growth included fresh shoot and root biomass, branching densities and numbers of upper, lower and basal roots and diameter of lower nodal roots, suggesting that root production and size at the juvenile stage are important determinants of economic yield at maturity. Basal roots, for example, might be vital for water uptake and anchorage while nodal roots originating from the nodes of the propagated cutting and spreading horizontally might be more important in nutrient acquisition and tuberization [12].

Utility in cassava improvement programmes
There are effectively four main factors of the genetic gain equation (ΔG) that influence breeding progress [58]. These include the phenotypic variation in the population (σ p ), the heritability (h 2 ), and selection intensity (i). By increasing σ p , h 2 or i, ΔG can be improved but this must be concomitant with a decrease in the last factor, L, which describes the length of time necessary to complete a cycle of selection [58]. The present study could be instrumental in the efforts to decrease L because selection based on juvenile plant traits could determine how quickly generations can be completed and how many generations can be completed per any given time. [59] noted that selection from juvenile traits can be an efficient method for acquiring ΔG while minimizing rotation time. The present work could also have utility in indirect selection for storage root yield, where selection is (i) based on the correlation between traits of the juvenile plants and same traits of mature plants or (ii) applied on another trait as that which shall be improved [60]. It is, however, worth noting that certain prerequisites are important to achieve this benefit. Based on the mathematical rule for the standardized selection response, the success of the indirect selection is dependent on the presence of a genetic correlation between two different traits (correlated response) [60]. Indirect selection can only be advantageous if an indirect character has a greater heritability than primary character, and the genetic correlation between primary and indirect characters is high [60]. Also, for indirect selection to be worthwhile in breeding programmes, it should be easier to measure the indirect character more accurately than it is to measure the primary character and the i of the indirect character must also be much greater than that of the primary character [58,60].
The present protocol can be applied to cassava populations to identify markers which segregate with specific rooting characteristics and ultimately reduce the lengthy time involved in breeding for improved genotypes in cassava. Going forward, it would be ideal to use genetic analysis, which shows closely linked genes or pleiotropy, to confirm the phenotypic correlations of fresh biomass, branching density and diameter-related traits, root length and numbers with yield of mature field-grown plants. This could facilitate the identification of traits which are beneficial for improved resource acquisition and use-efficiency under field and stressed conditions and also enhance the potential of introgressing coincident quantitative trait loci (QTLs) to improve RSA, yield and yield-related traits in cassava.
It must, however, be said that inferences from this paper may be somewhat limited because the genotypes used may not cover the complete natural range of variation in cassava and may not provide full insight into the genetic properties of the species. Thus, the genetic component in this study is more fixed, than random. Ideally, attempts to exploit indirect selection in an applied cassava breeding programme should be based on a random set of genotypes similar to the populations of a conventional breeding programme. It would, therefore, be appropriate to validate our results so that our approach can effectively be applied to actual breeding programmes. There is an ongoing project of using nuclear applications, including gamma rays for mutation induction in cassava, at the Nuclear Agricultural Research, Biotechnology and Nuclear Agriculture Research Institute in Ghana. It is possible to validate our approach using inbred populations of various mutant families, derived from irradiated propagation materials.
While it is imperative to validate our results, indirect selection based on the protocol proposed here could have some improvements over conventional breeding. Conventional cassava breeding is protracted and complex because it involves several stages and many of the traits under selection have a quantitative inheritance with many active genes and strong environmental influence. The accuracy of selection depends on the experimental differences in all of the stages of the breeding programme [61]. We hypothesise that the protocol proposed could be exploited to reduce the duration required in clonal evaluation trials (CET) and even preliminary yield trials (PYT), provided high heritability and genetically correlated traits could be identified in the early growth stage. As part of the ongoing cassava improvement programme, we intend to exploit juvenile traits in CET and/or PYT, alongside the conventional approaches. Hopefully, the genetic value of the clones in the CET phase obtained indirectly through selection from juvenile plants will be a good predictor of the performance of the genotypes in the PYT phase, thereby simplifying the process of variety development.
In a typical breeding programme, thousands of genotypes are involved. The benefit of an indirect selection will be realised if the system is of high throughput, allowing the testing of a much larger number of progenies than otherwise possible so that the selection pressure can be significantly increased. Depending on the experience of the technician, excavation, soaking and cleaning of root systems and measurements of traits required 15 to 18 minutes in the present study. [62] has reported that the soil type in which plants are grown can influence the time needed to uproot and evaluate the roots, suggesting that the time recorded here could be reduced if a more friable soil is used for this work. In the present study, based on the results of the PCA, it may also be feasible to reduce the number of traits to be evaluated to increase the speed of the proposed method. To enable high-throughput phenotyping of the root systems, automated or semi-automated image analysis and low-cost computer vision technologies could be adapted or developed and coupled to the present approach.

Conclusions
There is an urgent need for rapid, inexpensive and robust phenotyping of root system traits of juvenile plants as a basis for identifying genotypic variations and predicting growth and yield performance of mature plants. This can substantially reduce the time, cost and effort for selection and breeding for crop improvement in response to environmental stresses and higher productivity. Despite its global socio-economic importance as a key food security crop, cassava is not easily amenable to such approaches. The results in the current study demonstrate the possibility of applying a simple, rapid, and robust phenotyping approach to identifying important traits in juvenile cassava that can be a basis for predicting field performance of mature plants. Significant genotypic variations were observed in both juvenile and mature plants, and traits with broad-sense heritability were identified. The results provide insights into the dynamics of cassava root and shoot traits, at 30 and 45 DAP and their relationships with traits in 7 and 12 months old (mature) plants. The explanatory power of fresh shoot and root biomass, total root length and the total number of roots measured at 30 DAP as predictors of HI index in mature plants was consistent for both 2017 and 2018. Genotypic differences in some important traits, such as shoot and root fresh weight, in mature plants were consistent for both 2017 and 2018 experiments or screening. The results provide seminal evidence for potentially useful relationships between traits in juvenile and mature cassava plants that can be further explored for predicting harvest index and supporting efforts for improving the crop.