Linkages Between Sorghum bicolor Root System Architectural Traits and Grain Yield Performance Under Combined Drought and Heat Stress Conditions

Abstract

Breeding programs often overlook the use of root traits. Therefore, we investigated the relevance of sorghum root traits in explaining its adaptation to combined drought and heat stress (CDHS). Six (i.e., three pre-release lines + three checks) sorghum genotypes were established at two low-altitude (i.e., <600 masl) locations with a long-term history of averagely very high temperatures in the beginning of the summer season, under two management (i.e., CDHS and well-watered (WW)) regimes. At each location, the genotypes were laid out in the field using a randomized complete block design (RCBD) replicated two times. Root trait data, namely root diameter (RD), number of roots (NR), number of root tips (NRT), total root length (TRL), root depth (RDP), root width (RW), width–depth ratio (WDR), root network area (RNA), root solidity (RS), lower root area (LRA), root perimeter (RP), root volume (RV), surface area (SA), root holes (RH) and root angle (RA) were gathered using the RhizoVision Explorer software during the pre- and post-flowering stage of growth. RSA traits differentially showed significant (p < 0.05) correlations with grain yield (GY) at pre- and post-flowering growth stages and under CDHS and WW conditions also revealing genotypic variation estimates exceeding 50% for all the traits. Regression models varied between pre-flowering (p = 0.013, R² = 47.15%, R² Predicted = 29.32%) and post-flowering (p = 0.000, R² = 85.64%, R² Predicted = 73.30%) growth stages, indicating post-flowering as the optimal stage to relate root traits to yield performance. RD contributed most to the regression model at post-flowering, explaining 51.79% of the 85.64% total variation. The Smith–Hazel index identified ICSV111IN and ASAREACA12-3-1 as superior pre-release lines, suitable for commercialization as new varieties. The study demonstrated that root traits (in particular, RD, RW, and RP) are linked to crop performance under CDHS conditions and should be incorporated in breeding programs. This approach may accelerate genetic gains not only in sorghum breeding programs, but for other crops, while offering a nature-based breeding strategy for stress adaptation in crops.

1. Introduction

Sorghum is an economically important staple crop, providing food for over half a billion people, mostly in arid and semi-arid regions where drought and heat stress significantly impact productivity [1,2]. This is a major problem since climate variability accounts for approximately 33% yield losses for major crops in the world, including maize, soybean, rice, and wheat [3]. Although sorghum is known to be stress-resilient, drought and heat stress threaten its productivity and nutritional quality [1]. This poses dangers to regions that solely depend on this crop for food and nutrition security [4]. Additionally, because plants are sessile, they remain exposed to harsh environmental conditions [5]. Therefore, improving abiotic stress tolerance must be a top priority for breeding programs given the current and predicted climatic scenarios in which the frequency and intensity of climate change-induced stresses are expected to increase [6]. While it is projected that root traits will be implicit in the adaptation and mitigation of climate change-induced abiotic stresses (in particular, CDHS) [7], traditional breeding programs have predominantly focused on above-ground plant parts such as grain production [8,9,10]. This is worrying, given the reports that conventional breeding approaches are becoming increasingly limited in addressing the multigenic nature of abiotic stresses [11,12].

Conventional methods are also reportedly associated with the presence of a high level of genotype via environment interaction (GxE) [13]. This is undesirable since GxE is not heritable and its presence slows genetic gains in selection programs. With all this said, there is an urgent need to integrate new approaches in plant breeding that lead to positive outcomes, encompassing the needs of end users of the breeding products but also preserving the integrity of ecosystem functions and biodiversity [14].

However, the underground component of crops, namely the root system architecture (RSA), has received little attention in the pursuit of crop improvement [15]. This ignores the pivotal roles plant roots play in mobilizing resources, specifically water and nutrients from the soil [16]. The RSA is defined as the physical configuration of the root system that regulates the deployment of roots in the soil in time and space [8,17]. RSA is important as it determines the extent of the region of the soil where water and nutrients can be exploited or accessed [18]. Therefore, enhancing the adaptation of crops to the current and the predicted future climatic conditions may lie in functional root traits [19,20]. For instance, plants are known to adjust their root systems for improved resource mobilization under water-limited conditions, and this buttresses the need for breeding programs to pay attention to root traits [21]. However, specific contributions of RSA in stress tolerance exhibited by some crops (e.g., sorghum) are still poorly established. Therefore, this study sought to investigate the potential role of root traits in conferring sorghum tolerance to CDHS conditions. In doing so we aimed to (i) identify RSA traits of economic importance in breeding; (ii) determine appropriate crop growth stage and growing conditions for sorghum root trait measurements; (iii) establish relationship between RSA traits and grain yield performance in sorghum under CDHS and WW conditions; and (iv) identify superior sorghum genotypes for production under CDHS, selected based on RSA traits and grain yield. We hypothesize that sorghum genotypes adjust their RSA traits for adaptation to CDHS conditions, resulting in variable GY responses. Therefore, unveiling the roles RSA traits play in abiotic stress tolerance could significantly enhance the development and deployment of climate-resilient crop varieties, with the goal of increased food and nutrition in areas prone to climate change-induced abiotic stresses.

2. Materials and Methods

2.1. Planting Materials, Study Sites and Experimental Design

Six (pre-release lines + three checks) sorghum genotypes (

Table 1. List of sorghum genotypes evaluated for RSA and grain yield performance at Chiredzi and Chisumbanje Research Station during the 2021–22 winter season under CDHS conditions.

Genotype Name	Description	Origin	Status
SV4	Released grain commercial variety	Crop Breeding Institute, (Harare, Zimbabwe)	Released commercial variety [check]
ICSV111IN	Advanced pre-release line	ICRISAT—Hyderabad, India	Pre-release line
CHITICHI	Local variety	Chiredzi Community Seed Bank (Masvingo, Zimbabwe)	Local landrace variety [check]
MACIA	Released grain commercial variety	Seed Company of Zimbabwe (Harare, Zimbabwe)	Released commercial variety [check]
IESV91070DL	Advanced pre-release line	ICRISAT—Hyderabad, India	Pre-release line
ASAREACA12-3-1	Advanced pre-release line	ICRISAT—Hyderabad, India	Pre-release line

) were established at two low-altitude (i.e., <600 masl) locations with a long-term history of averagely very high temperatures in the beginning of the summer season [22], under two management (i.e., CDHS and well-watered (WW)) regimes during the 2012–22 winter season. At each location (i.e., Chiredzi and Chisumbanje) (Figure 1), the genotypes were laid out in the field using a randomized complete block design (RCBD) replicated two times.

2.2. Trial Establishment and Management

Planting was performed by dribbling seeds in furrows of 5 m in four-row plots with 0.75 m inter-row spacing. After three weeks, post-emergence thinning was performed to achieve 0.2 m in-row spacing, thus resulting in 100 plants per plot. Grain weight (GW) was measured from the grain collected from the heads of plants from the middle two rows of each plot; a 0.5 m border on each end of rows was discarded to eliminate border effects. This resulted in a net plot size of 9 m² (i.e., 3 rows × 0.75 m × 4 m row length), which is equivalent to a population of 40 plants per plot.

Genotypes were subjected to CDHS at the early vegetative stage. This was achieved by establishing the trials early in August under irrigation so that the early vegetative stage coincided with the hottest periods in September to October, where average daily temperatures are above 30 degrees Celsius [22], during which supplementary irrigation was withheld for over two weeks. Exposing genotypes to abiotic stress enabled researchers to investigate the existing variation in root system architecture adjustments that possibly conferred adaptiveness, which was manifested through grain yield performance.

2.3. Data Collection and Exploitation

2.3.1. Root System Architecture Analysis

At both the pre- and post-flowering stages, RSA traits, namely NR, NRT, TRL, RDP, RW, WDR, RNA, RS, LRA, RD, RP, RV, SA, RH, and RA, were measured using the RhizoVision Explorer software [23]. The workflow involved uprooting three plant samples per plot, removing core soils, washing roots with water, taking images of roots using a Canon EOS4000D camera (18MP APS-C CMOS sensor, DIGIC 4+ Image Processor) manufactured by Canon in Tokyo Japan, and then uploading the images into the RhizoVision Explorer software version 2.0.3 for root metrics determination. We used the RhizoVision Explorer software because it is an open source-software, user-friendly, fast, generalist, and all-inclusive tool to facilitate the standardization of RSA trait computations [23]. In previous studies, RhizoVision Explorer was found to be the most efficient and accurate method, providing root trait data without requiring extensive pre-processing of images [23,24].

2.3.2. Grain Yield Parameters

Grain yield (GY) from harvested sorghum heads per net plot was measured after the threshing and cleaning of grains. This was computed and expressed in tons per hectare using the following formula:

$$Grain yield (\mathrm{t}/{\mathrm{h}\mathrm{a}}^{-1})=1000 \times {\displaystyle \frac{Net plot grain weight \left(\mathrm{k}\mathrm{g}\right)}{Net plot size \left({\mathrm{m}}^2\right)/\mathrm{10,000}}}$$

For accuracy, GY data were measured at 12% moisture contents, as determined using a KM 36 G cereal grain moisture meter (Corousell, Manila, Philippines). Grain yield was an assumed response variable, indicating the adaptiveness of genotypes under CDHS conditions.

2.4. Characterization of RSA Traits and Their Linkages with GY Performance

To identify the RSA traits of high importance for this study, principal component analysis (PCA) [25] was deployed. These traits of economic importance were identified based on their high variability, as identified through PCA using data collected at pre- and post-flowering growth stages. The choice between using PCA information generated using data collected at the pre- or post-flowering crop growth stage was based on which one optimized the variability of RSA traits. The PCA can extract the most important dimensions to precisely identify which data to focus on detecting patterns. The method, therefore, dealt with the problem of multiple indicators by only identifying the most informative variables for further analysis. PCA is suitable for scenarios that require fast and accurate extraction of key information, such as data compression and pattern recognition [25]. In this study, screen plots were used as graphical tools to help decide on the number of principal components to retain, picking PCs with eigenvalues of at least 1, while also aiming for the selected PCs to explain at least 80% of the variance. The genotypic coefficients of variation for RSA traits of economic importance were estimated using the Multi-Environment Trial Analysis with R (METAR) software v 6.0 [26].

In this study, it was critical to determine the optimal crop growth stage at which RSA trait measurements provided meaningful and accurate breeding information. A Paired T-test analysis was then deployed to compare mean RSA trait values between the pre- and post-flowering crop growth stages. The Paired T-test is reportedly the common method for comparing mean values between two samples that show a “before and after” scenario or in cases where objects in one sample are all measured twice [27]. The test aimed at ascertaining if the sampling time (i.e., pre- and post-flowering) influenced RSA measurements. To investigate correlations between individual RSA traits and GY performance, we deployed the Pairwise Pearson correlation analysis. This analysis enabled us to understand the strength and direction of their relationships, also testing for statistical significance.

The multiple linear regression analysis (MLR) was used establish the relationships between RSA traits and GY performance. This was on the premise that RSA traits do not operate in isolation but instead have a collective influence on grain yield performance. The MLR is an important statistical method that uses several independent variables to predict the outcome of a response variable [28]. Regression analysis also made it possible to quantify the amount of variation in GY that the RSA traits were explaining, thus providing a metric to gauge the power of the existing relationship. To demonstrate the most significant RSA traits and their relative importance in explaining the variance of the dependent variable (GY). The assumptions of MLR (i.e., existence of a linear relationship between independent and dependent variables, no multicollinearity, multivariate normality, and homoscedasticity) were checked to ensure the model’s validity and reliability.

2.5. Regression Model Validation and Selection

The validation and selection of appropriate MLR models between pre- and post-flowering and between WW and CDHS was based on the Akaike information criterion (AIC), Bayesian information criterion (BIC), Durbin–Watson Statistic, Predicted Residual Error Sum of Squares (PRESS), R²_Predicted, and the linear determination index denoted by R² dealing with regression analysis. Tolerance accounts for how much the variance of a regression coefficient is inflated due to the multicollinearity of predictor variables.

2.6. Identifying Superior Sorghum Genotypes for Production Under CDHS

Superior sorghum genotypes were selected based on analysis of variance (ANOVA) and the Smith–Hazel Multi-Trait Stability Index (MTSI) analysis.

Combined analysis of variance (ANOVA) as an inferential method was used to test the equality of means, including Turkey’s multiple comparisons of means at a 5% probability level using GenStat Software 21st Edition [29]. Results from GY ANOVA were used as a basis to compare and validate the feasibility and practicality of using RSA traits in plant selection. To quantify and visualize the changes in the GY performance of genotypes between WW and CDHS conditions within and across sites, we generated dumbbell plots developed using the ‘geom_dumbbell’ function in the ggalt v4.2.3 R package.

We deployed the Smith–Hazel Multi-Trait Stability Analysis (MTSI) [30] to identify genotypes with superior genetic worth based on the joint effects of RSA traits. The method was selected because it was anticipated that individual RSA traits would not operate in isolation in influencing crop performance but would instead have a combined effect. The Smith–Hazel analysis was applied after building an appropriate multiple linear regression (MLR) model, which was used to identify the RSA traits of economic importance in breeding. The identified RSA traits of importance were then subjected to the Smith–Hazel MTSI analysis to identify superior sorghum genotypes. The basic concept of using the Smith–Hazel index is to define the genetic worth of an individual based on a linear function of the genetic values of multiple traits, each weighted to a pre-assigned relative economic value [31].

3. Results

3.1. Characterization of RSA Traitsand Their Linkages with GY Performance

3.1.1. Traits of Economic Importance at the Individual Level

Principal component analysis conducted using the RSA traits data collected at the post-flowering stage identified NRT, RDP, TRL, and NR as the traits of economic importance in breeding. For instance, from PC1, it was observed that NRT had the highest score of 0.346, RD had the second largest score of 0.336, and TRL had the third highest score of 0.333. On the other hand, RV (PCA score = 0.461), RD (PCA score = 0.421), and SA (PCA score = 0.398) recorded the highest contributions to PC2 (

Table 2. Principal component analysis results for RSA traits and grain yield measured at two sites under combined drought and heat stress conditions at the post-flowering crop growth stage during the 2021–22 winter season in Zimbabwe.

	Principal Component
	1	2	3	4
Eigenvalues	7.8041	3.0351	2.4212	1.0611
Proportion variance (%)	0.459	0.179	0.142	0.062
Cumulative variance (%)	0.459	0.638	0.780	0.842
Number of roots	0.314 *	−0.133	−0.134	0.145
Number of root tips	0.346 *	−0.057	−0.064	−0.011
Total root length	0.333 *	−0.011	−0.193	−0.019
Root depth	0.336 *	0.018	−0.009	−0.182
Root width	0.301	−0.064	0.251	0.233
Width–depth ratio	−0.075	−0.141	0.310	0.761
Root network area	0.299	0.022	0.305	−0.055
Root solidity	−0.117	0.372	0.199	−0.273
Lower root area	0.064	−0.330	0.401	−0.112
Root diameter	0.163	0.421	−0.193	0.101
Root perimeter	0.303	−0.057	−0.310	0.084
Root volume	0.023	0.461	0.284	0.122
Surface area	0.153	0.398	0.338	0.034
Root holes	0.254	−0.258	0.250	−0.278
Root angle	0.098	0.237	−0.032	−0.079
Grain yield	0.243	0.177	−0.249	0.292

Key: RSA with high values in principal component 1 demonstrated high variability, hence being of economic importance in plant selection. The mark * indicate RSA traits that exhibited high variability.

).

In addition, all the identified RSA traits of economic importance revealed high genotypic variation estimates (

Table 3. The genotypic variation estimates of the RSA traits of plant selection economic importance.

RSA Trait	Genotypic Variation Estimate	Category
Number root tips	0.96	Traits of economic importance at the individual level
Root depth	0.89
Total root length	0.96
Number of roots	0.89
Root holes	0.97	Traits of economic importance in combination
Root diameter	0.95
Root width	0.97
Lower root area	0.82
Root perimeter	0.93

Key: Genotypic coefficient of variation estimate of an RSA trait represents likelihood to be controlled by additive gene action, classified as low if (h < 0.15), medium (0.15 < h < 0.50), and high magnitude (h > 0.50).

), which were determined based on Multi-Environment Trial Analysis aimed at accounting for the genotypic and environmental variance.

3.1.2. Association of RSA Traits with GY Peformance

RSA traits were significantly (p = 0.000) associated with GY performance, collectively explaining 85.64% of the variation in GY (

Table 4. Analysis of variance of the multiple linear regression model built to represent the relationship between RSA traits measured at post-flowering and grain yield performance during the 2021/22 season under combined drought and heat stress conditions at Chiredzi and Chisumbanje Research Station. The modeled relationship helped identify RSA traits of economic importance in breeding.

Source	DF	Seq SS	Contribution (%)	Adj SS	Adj MS	F Value	p Value
Regression	5	19.1427	85.64	19.1427	3.8285	21.47	0.000
Root holes	1	0.1027	0.46	2.6238	2.6238	14.72	0.001
Root diameter	1	11.5769	51.79	2.0534	2.0534	11.52	0.003
Root width	1	1.9519	8.73	0.6859	0.6859	3.85	0.065
Lower root Area	1	0.1105	0.49	1.5517	1.5517	8.70	0.009
Root perimeter	1	5.4006	24.16	5.4006	5.4006	30.29	0.000
Error	18	3.2094	14.36	3.2094	0.1783
Total	23	22.3521	100.00

Key: % contribution is the amount of variation a predictor variable (RSA trait) is contributing to the total variance (R² = 85.64%) being explained in the MLR model. The model represents the relationship between RSA traits and GY performance of sorghum genotypes.

). In this relationship RD was identified as the RSA trait of highest economic importance, contributing 51.79% of the 85.64% of the model’s total explained variance, followed by RP (24.16%), whilst RH was ranked the least (Table 4).

All the four traits had calculated VIF values of less than 10, as well as recording tolerance values of less than 0.2 (

Table 5. The multiple linear regression model parameters of the model developed to represent the relationship between RSA traits and grain yield performance at post-flowering under CDHS conditions during the 2021/22 season at Chiredzi and Chisumbanje Research Station.

Term	Coef	SE Coef	95% CI	T Value	p Value	VIF	Tolerance
Constant	−1.412	0.358	(−2.165, −0.660)	−3.94	0.001
Root holes	−0.002742	0.000715	(−0.004244, −0.001240)	−3.84	0.001	4.19	0.24
Root diameter	0.000116	0.000034	(0.000044, 0.000187)	3.39	0.003	1.93	0.52
Root width	0.000063	0.000032	(−0.000004, 0.000131)	1.96	0.065	3.52	0.28
Lower root Area	0.000000	0.000000	(0.000000, 0.000000)	2.95	0.009	2.56	0.39
Root perimeter	0.000003	0.000000	(0.000002, 0.000004)	5.50	0.000	2.17	0.46

Key: The variance inflation factor (VIF) is a model diagnostic tool used to detect multicollinearity. VIF > 5 indicate high multicollinearity and VIF > 10 indicates serious multicollinearity problems. Tolerance (1 − R²) measures the influence of one predictor variable on all other predictor variables, where tolerance < 0.1 might indicate presence of multicollinearity.

).

All the RSA traits showed significant positive correlations with grain yield performance at the post-flowering growth stage but not at the pre-flowering stage, except for RD (

Table 6. Across-site Pairwise Pearson correlation results indicating the individual association of RSA traits with GY performance at pre- and post-flowering growth stages under CDHS conditions during the 2021–22 season.

RSA Trait	Pre-Flowering (r)	Post-Flowering (r)
Number of root tips	0.312 ns	0.652 *
Root depth	0.639 *	0.611 *
Total root length	−0.039 ns	0.708 *
Number of roots	0.271 ns	0.614 *

Key: * indicates significant (5% probability level) correlation between an RSA trait and GY performance, ns indicates no significant correlation.

).

All the RSA traits showed significant positive correlations with GY performance under WW and CDHS conditions (

Table 7. Pairwise Pearson correlation results indicating the individual association of RSA traits with GY performance under WW and CDHS conditions during the 2021–22 season.

RSA Trait	WW Condition (r)	CDHS Condition (r)
Number of root tips	0.674 *	0.652 *
Root depth	0.788 *	0.611 *
Total root length	0.825 *	0.708 *
Number of roots	0.768 *	0.614 *

Key: * indicates significant (5% probability level) correlation between an RSA trait and GY performance.

).

3.1.3. Influence of Time of Sampling and Management Regime on RSA Trait Measurements

There were no significant changes in RSA measurements at the pre- and post-flowering crop growth stages across all RSA traits of economic importance at the individual level (

Table 8. Paired T-test results comparing mean RSA trait values between the pre- and post-flowering crop growth stages using the RSA traits of economic importance at the individual level and in combination as determined through PCA and MLR analysis, respectively.

RSA Trait	Mean Pre-Flowering	Mean Post-Flowering	Category
ns Number of root tips	603.9	492.3	Traits of economic importance at the individual level
ns Root depth	14,645	13,328
ns Total root length	692,723	599,049
ns Number of roots	28.38	24.50
* Root holes	564.3	343.5	Traits of economic importance in combination
* Root diameter	4396	8151
* Root width	20,729	16,410
ns Lower Root Area	93,846,371	70,910,745
* Root perimeter	612,743	431,615

Key: * indicates significant (5% probability level) difference between RSA trait measurements at pre-and post flowering crop growth stages, ns indicates no significant difference.

). All the RSA traits identified to be of economic importance in combination (Table 4) significantly varied in their measurements at the pre- and post-flowering growth stages, except for LRA (Table 8).

RSA trait measurements did not significantly (p > 0.05) differ between WW and CDHS regimes, except for NR, RH, and RW (

Table 9. Paired T-test results comparing mean RSA trait values between the well watered (WW) and combined drought and heat stress (CDHS) regimes based on the performance of RSA traits of economic importance at the individual level and in combination as determined through PCA and MLR analysis, respectively.

RSA Trait	WW Conditions	CDHS Conditions	Category
ns Number of root tips	440.2	492.3	Traits of economic importance at the individual level
ns Root depth	12,632	13,328
ns Total root length	640,968	599,049
* Number of roots	20.33	24.50
* Root holes	528.1	343.5	Traits of economic importance in combination
ns Root diameter	6931	8151
* Root width	12,329	16,410
ns Lower Root Area	70,276,230	70,910,745
ns Root perimeter	450,930	431,615

Key: * indicates significant (5% probability level) difference between RSA trait measurements under WW and CDHS conditions, ns indicates no significant difference.

).

3.1.4. Model Validation Selecting Optimal Crop Growth Stage and Management Condition for RSA Trait Measurements

Under CDHS conditions, the relationship between RSA traits and GY performance at the post-flowering crop growth stage revealed a higher coefficient of determination (R²), R² Predicted and a higher Durbin–Watson Statistic, but lower PRESS, AIC, and BIC values compared to pre-flowering (

Table 10. Model validation information used to select appropriate MLR models between pre and post-flowering and between under WW and CDHS conditions during the 2021/22 season.

Management Condition	Crop Growth Stage	R2	PRESS	R2_Predicted	AIC	BIC	Durbin–Watson Statistic
CDHS	Post Flowering	85.64%	5.96812	73.30%	40.82	42.07	1.93320
	Pre Flowering	47.15%	15.7983	29.32%	68.04	70.16	1.91583
WW	Post-flowering	71.24%	4.94478	66.35%	33.62	35.96	1.75446

Key: Durbin–Watson Statistic tests for autocorrelation and by rule of thumb values in the range 1.5 < d < 2.5 show no autocorrelation. Akaike information criterion (AIC) is used in model selection; the smaller the AIC value, the better the model fit. Bayesian Information Criterion (BIC) is used in model selection and the model with the lowest BIC is preferred or selected.

). The relationship between RSA traits and GY performance under CDHS conditions also showed higher values than under WW across all validation metrics (Table 10).

3.2. Identification of Superior Sorghum Genotypes for Production Under CDHS

3.2.1. Superior Genotypes Based on the Smith–Hazel Multi-Trait Selection Index Analysis

The Smith–Hazel MTSI analysis, at a 50% selection intensity, ranked genotypes based on their individual genetic worth (

Table 11. Genetic worth (index) of genotypes based on the Smith–Hazel MTSI analysis. The genetic worth determined based on RSA traits, namely root diameter, width, and perimeter including grain yield.

Genotype	Description	Genetic Worth (V1)
SV4	Semi dwarf open-pollinated variety, 113 to 127 days to maturity, grain yield potential is 3.4 to 9.0 t/ha	401.2685
ICSV111IN	An elite sorghum breeding line at advanced testing by ICRISAT in Zimbabwe, preferred by farmers for high grain yield, white color, and drought tolerance	362.6735
ASAREACA12-3-1	An elite sorghum breeding line at advanced testing by ICRISAT in Zimbabwe, high-yielding and stable variety preferred by farmers in Zimbabwe	260.9641
IESV91070DL	An elite sorghum breeding line at advanced testing by ICRISAT in Zimbabwe, preferred by farmers for high grain yield, white color, and drought tolerance	249.3526
Chitichi	Local landrace variety commonly grown in the south-eastern lowveld communal areas of Zimbabwe, white, and is drought-tolerant	212.5624
Macia	A white open-pollinated sorghum variety, yield potential of up to 5 tons per hectare, physiological maturity is 115–120 days	156.4162

Key: The genetic worth (index) represents breeding value and the higher the value the more the breeding gain likely to be achieved after selection.

). The Smith–Hazel Multi-Trait Stability Index identified the check genotype SV4 and two pre-release lines (i.e., ICSV111IN and ASAREACA12-3-1) as the most productive and stable sorghum genotypes under CDHS conditions (Figure 2). Following the Smith–Hazel MTSI analysis, the check genotype SV4 showed the most genetic worth, followed by the pre-release line ICSV111IN and the check genotype Macia. In addition, the RSA traits of economic importance also revealed unequal importance based on their index coefficients (

Table 12. The calculated Traits’ individual index coefficients (b) based on the Smith–Hazel MTSI analysis.

Trait	Root Trait Measurement Description	Individual Index Coefficient (b)
Root diameter	The distance transform value at each skeletal pixel is the radius at that pixel and is doubled to give the diameter.	5.0391258
Root width	The maximum horizontal distance the root crown grew at the time of imaging.	1.4706810
Root perimeter	The sum of the Euclidean distances between the connected contour pixels in the entire segmented image of the plant root.	0.6103781
GYD	Grain weight from the harvested and threshed sorghum heads per net plot after cleaning the grain.	48.5630463

The description of the RSA trait measurements adopted the outlined guidelines in [23]. The index coefficient (b) or weight value of an RSA trait is its contribution to the overall selection index, and the higher, the better and more important in selection.

) and selection differential (

Table 13. Selection differential (SD) (per RSA trait) based on the Smith–Hazel MTSI analysis.

Trait	Xo	Xs	SD	SDperc	Sense	Goal
Root diameter	8.150958	9.50875	1.357792	16.65806	increase	100
Root width	16.409667	18.53550	2.125833	12.95476	increase	100
Root perimeter	431.614500	607.86925	176.254750	40.83615	increase	100
GYD	1.179250	1.95500	0.775750	65.78334	increase	100

Key: Xo is the mean of original population, Xs is the mean of selected genotypes, and SD is selection differential.

).

3.2.2. Superior Genotypes Based on GY ANOVA

Genotypes revealed a statistically significant difference (p < 0.001) in grain yield performance across sites (

Table 14. Grain yield combined analysis of variance results for sorghum genotypes evaluated during the 2021/22 season under combined drought and heat stress conditions.

Genotype Name	Grain Yield (t/ha)	Status
SV4	1.955 a	Released commercial variety [check]
ICSV111 IN	1.800 a	Pre-release breeding line
MACIA	1.060 b	Released commercial variety [check]
ASARECA 12-3-1	0.910 bc	Pre-release breeding line
IESV91070DL	0.763 bc	Pre-release breeding line
CHITICHI	0.588 c	Local landrace variety [check]
p value	<0.001
LSD	0.3905
CV%	15.0

Key: GY combined analysis of variance results after testing the equality of means, including Turkey’s multiple comparisons of means at a 5% probability level. Genotypes sharing the same letter are not statistically different in their grain yield performance.

). The pre-release line ICSV111IN demonstrated the same high grain yield superiority as the winning candidate, check genotype SV4; however, this was still significantly higher than the other check genotypes MACIA and CHITICHI (Table 14).

3.2.3. Stability in Grain Yield Performance

Furthermore, the same pre-release line ICSV111IN consistently recorded a higher grain yield performance under CDHS than in WW conditions across the two study sites (Figure 3 and Figure 4).

4. Discussion

Underground traits in the form of RSA traits remain an underexploited target for crop improvement in the quest to develop climate-resilient crops [8]. On the contrary, it is projected that the bulk of abiotic stresses associated with climate change will be most acutely expressed by the plant at the root–soil interface, making plant roots an important future breeding strategy [7]. However, there is a lack of knowledge about variety-specific root traits for adaptation to drought [19], hence highlighting the need to determine how these traits contribute to species adaptation in local environments [32]. Here, a study was conducted to investigate the linkages between root traits and the agronomic performance sorghum genotypes under CDHS conditions, centered on the hypothesis that depicting the roles of RSA in abiotic stress tolerance could significantly enhance the development and deployment of climate-resilient crop varieties for use as increased food and nutrition in areas prone to climate change-induced abiotic stresses.

Based on the evidence generated from this study, we can confirm that a linkage exists between RSA traits and plant (Sorghum bicolor) response to combined drought and heat stressors. To assess this, we matched the RSA traits of sorghum genotypes to their GY performance under CDHS conditions. The study identified RD, RW, and RP as multiple RSA traits of economic importance in sorghum breeding for tolerance against combined drought and heat stress. In addition, the study found the post-flowering crop growth stage appropriate and more informative for RSA measurements compared to pre-flowering, also identifying ICSV111IN and ASAREACA12-3-1 as superior pre-release lines for production under CDHS. Previous studies revealed that traits such as root diameter, specific root length, root area, and density contribute to crop adaptiveness to drought stress conditions [33]. RSA traits response to temperature changes is reportedly crop-specific as increases in temperature can promote or inhibit plant growth [34].

Although sorghum grown in arid and semi-arid environments is typically deep-rooted for increased uptake of resources and grain production [9,35], this was not entirely the observed scenario in this study. The study identified NRT, RDP, TRL, and NR as RSA traits of economic importance at the individual level and RH, RD, RW, LRA, and RP as traits of importance in breeding operating in combination or multiple traits. The observed genotypic variation in RSA traits was expected, as previous studies have reported the presence of variation in RSA traits in response to drought [1,36]. Generally, genotypes with deeper roots are expected to exhibit higher crop productivity due to improved access to stored water and nutrients even in the deeper soil layers [37]. In the context of this study, root depth (RDP) was not of practical importance in conferring CDHS tolerance since it was not modeled in the relationship between RSA traits and GYD performance (Table 4). This could suggest that resources were not limited vertically in the soil profile but horizontally. This might be the reason why RD, RW, RP, and LRA are the main RSA traits of importance (Table 4) for increased root surface area with soil moisture and soil volume under CDHS conditions. However, RDP remains an important RSA trait in plant selection under CDHS conditions based on its high variability (Table 2) and high genotypic variation estimate (Table 3). The presence of genotypic variation in RSA traits presents opportunities for plant breeders to utilize them for crop improvement.

The root system architecture plays a critical role in conferring tolerance to abiotic stressors. Previous studies identified root length and depth as critical traits for optimizing uptake of water and nutrients, leading to up to 0.5 t/ha yield increases in wheat [16]. In this study, TRL and RDP, including other related RSA traits, were also identified as traits of economic importance (Table 2 and Table 4). Reaching out for resources in deeper soils requires that plants arrest lateral root growth and instead prioritize axile root development [10]. During drought, plants invest more in root development to achieve maximum resource capture by reducing the number of lateral roots and promoting deeper rooting [9]. Increased lateral root branching density in the topsoil facilitates the capture of immobile resources such as phosphorus, whereas reduced lateral root branching density in the subsoil enhances the efficient uptake of mobile resources, e.g., water and nitrogen [17]. To develop climate-resilient varieties, breeders must therefore aim to identify RSA traits that facilitate early establishment of an extensive root system [15].

RD, which is the RSA trait and predictor variable with the highest contribution (51.79%) to the regression model developed to represent the relationship between RSA traits and GY performance (Table 4), could be a target for plant selection to develop climate-resilient sorghum varieties. It enables plants to overcome mechanical resistance as they penetrate hard soils for more exploration into new soil volumes for more nitrogen sources [38]. Thicker roots are reported to have greater penetration ability in hard soils as they provide structural support to withstand buckling and deflection [39].

Although the linkages between individual RSA traits (i.e., NRT, RDP, TRL, NR) and the GY performance of genotypes are insightful (see Table 6 and Table 7), they may not be of practical importance. It is assumed that these RSA traits operate singularly, yet it is expected that RSA traits have a combined influence on plant adaptability to abiotic stressors. To circumvent this, a multiple linear regression (MLR) analysis was used to better represent the relationship between RSA traits and GY performance. Previous studies have demonstrated that RSA traits interact to optimize resource uptake and plant survival under environmental stress [10].

Future breeding would obviously require guidance on the appropriate crop growth stage and management conditions where Sorghum bicolor RSA trait measurements can be recorded to effectively inform plant selection decisions. The MLR model, built to represent the relationship between RSA traits and GY performance across crop growth stages, shows post-flowering as more appropriate and informative (R² = 85.64%) with a higher predictive power (73.30%) compared to that at pre-flowering, which accounted for less variation in GY and showed lower (29.32%) predictive power (see Table 10). In addition, the results generally show significant and stronger correlations between RSA traits and GY performance at post- than at pre-flowering (Table 6). At pre-flowering, the RSA traits did not show significant correlations with GY, except for RDP (Table 6). Breeders will therefore need to collect RSA trait information at post-flowering. This recommendation is further supported by the fact that the MLR model at post-flowering revealed a lower PRESS, AIC, and BIC, including a higher Durbin–Watson Statistic value, compared to pre-flowering (Table 10). A Durbin–Watson Statistic value closer to 2 indicates no significant autocorrelation [40].

The selected model is identified by the minimum value of the Bayesian information criterion (BIC), which tends to go for models that are more parsimonious than those favored by AIC [41]. The modeled relationship between RSA traits and GY performance under WW conditions showed lower AIC and BIC values compared to under CDHS (Table 10). This implies that the model under WW is more optimal in balancing trade-offs between model fitness and complexity compared to under CDHS [40]. However, the RSA traits and GY-modeled relationship is more accurate and informative since it accounted for more variation (R² = 85.64%) in GY compared to WW (Table 10). Coefficients of determination (R²) greater than 50% generally indicate a strong model for regression analysis [28].

The developed MLR model had predictor variables with variance inflation factors (VIF) of less than 5 (Table 5), thereby showing no presence of multicollinearity. Inflation variance factors provide insights into the validity of multiple regression models. Generally, VlF values above 5 or 10 indicate the presence of multicollinearity [42], leading to inaccurate estimation of the associated coefficients due to the correlations among the predictor variables. In this study, we intentionally used a model validation criterion that excluded predictor variables with VIFs greater than 5 to circumvent the possible problem of multicollinearity. Therefore, the established relationships between RSA traits and GY performance across crop growth stages (pre- and post-flowering) and management conditions (WW and CDHS) were rigorously tested and can be relied upon.

With the RSA traits accounting for a higher variation in GY under CDHS conditions (85.64%), compared to 71.24% under WW (Table 10), this could suggest that RSA traits express themselves both more and better under abiotic stressors. This may further imply that when breeders aim for developing resilient varieties but select based on GY, neglecting the underground root traits, they are missing an important plant breeding point. The PCA did not identify GY as a trait of economic importance (Table 2), meaning that its variability was not high. The low variability in GY is undesirable in plant selection as it may translate to low breeding information, which may slow breeding gains.

Although GY as a trait for plant selection exhibited the highest individual index coefficients (Table 12), it recorded the lowest selection differential compared to RSA traits (Table 13). The selection differential is a key parameter in the prediction of response to selection [43]. Given that plant selection has traditionally focused on the above-ground plant parts such as GY performance [9], these findings then suggest that vast amounts of crop improvement information and potential breeding gains remain untapped by not utilizing RSA traits.

To buttress the point that RSA traits could be of value in breeding, they demonstrated high genotypic variation estimates (Table 3). The high genotypic coefficient of variation estimates for RSA traits imply that they are controlled by additive gene action [44,45]. According to [46], heritability can be classified as low (h < 0.15), medium (0.15 < h < 0.50), and high magnitude (h > 0.50). Thus, these RSA parameters are highly heritable and can be relied upon in breeding programs as they can successfully be perpetuated to future generations of crop plants. In addition, RSA traits collectively revealed significant association with GY performance (Table 4) but differentially explained the variation in GY performance across crop growth stages (pre-and post-flowering) and between CDHS and WW management regimes (Table 10). RSA traits also showed significant and positive pairwise correlations with GY performance (Table 6 and Table 7). The findings demonstrate the close linkages between RSA traits and crop performance, providing compelling evidence that RSA traits could be relied upon in plant selection.

Study findings on RSA trait measurements across crop growth stages (Table 8) and management conditions (Table 9) can inform breeders on the appropriate time and conditions to collect plant selection information.

The selection of the pre-release lines ICSV111IN and ASAREACA12-3-1 through the Smith–Hazel MTSI index, together with the best check genotype SV4 (Figure 2), suggests that they are potential candidates for release as climate-resilient varieties. Their selection (Figure 2) means that they possess high breeding value or genetic gain across the traits used in the selection index. Interestingly, the grain yield performance of the pre-release line ICSV111IN did not differ from that of SV4 but significantly outperformed Macia, another check genotype (Table 14). Therefore, the genotype ICSV111IN can be targeted as the most promising pre-release line, followed by ASAREACA12-3-1. Using the Smith–Hazel index, breeders aim for multiple-trait improvement simultaneously [47]. The selections based on the Smith–Hazel MTSI index (Figure 2) suggest how the multiple RSA traits, namely RD, RW, and RP, including GY performance, combine to differentially determine the ultimate individual genetic worth of sorghum genotypes to influence their selection (Table 11 and Figure 2). The red circle (Figure 2) represents a cut off or threshold line which is set based on the desired selection intensity and serves to mark the minimum index value needed for a genotype to be selected [48]. It follows that only genotypes with index values above the cut off point are selected. In previous studies, the application of the Smith–Hazel index improved selection efficiency by an average of 14% compared to single-trait selection [31].

The fact that the RSA traits are of unequal breeding value (Table 12 and Table 13) based on their index coefficients (b) and selection differential (SD), and that they differentially contribute to the genetic worth of genotypes demonstrates the presence of genotypic variation that can be exploited in the existing plant breeding. The information generated from this study can surely guide breeders on what RSA traits to target for crop improvement.

The pre-release line ICSV111IN consistently recorded a higher grain yield under CDHS than under WW conditions across the two study sites (Figure 3 and Figure 4), which could qualify it as a novel finding and innovation. This contradictory pattern emerging within the data challenges the general assumptions that the agronomic performance of crops is higher under favorable or optimum conditions. This study, therefore, proposes ICSV111IN as an elite, climate-resilient, and unique line that could be targeted for further research and variety release suitable for the arid and hot areas of Zimbabwe.

5. Conclusions

The study successfully demonstrated the linkage between RSA traits and sorghum GY performance under CDHS conditions. Sorghum genotypes exhibited genotypic variation in RSA traits and GY performance. The RSA traits significantly associated with GY performance with RD, RW, and RP promise to be of economic importance in plant breeding based on their high genotypic coefficient of variation estimates and strong positive correlations with GY performance under CDHS. Generating evidence to inform plant selection decisions, the modeled relationships between RSA traits and GY performance across crop growth stages, and management conditions show post-flowering and CDHS as more appropriate and informative methods. Sorghum genotypes, namely ICSV111IN and ASAREACA12-3-1, are the superior pre-release lines that can be targeted for further evaluation and possible release for production under CDHS. The study findings present opportunities to integrate RSA traits in crop improvement programs which may improve plant selection efficiency, also enhancing accelerated breeding gains.