INTRODUCTION

Sorghum (Sorghum bicolour (L.) Moench) serves both as food for humans and animal feed. Its grain is the main staple food especially for the poverty-stricken and the majority food-in need society, residing primarily in the semi-desert tropics [1,2]. According to [3], report sorghum is 5th in the globe next to wheat, maize, rice and barley.

Moisture stress is a challenge in almost half of the globe’s arable land is a main determinant to the crop productivity worldwide [4]. According to different research studies moisture deficiency at key phases like tillering, seedling setting up and reproductive phases may result in a significant yield reduction and even fatal to the in cereal crops [5,6].

In addition to that, moisture deficient at reproductive stage reduces grain yield more than the moisture stress at another growth phases. Variations in water balance and soil accessible moisture are decisive to crop productivity since they directly interrupt plant physiological growths and responses [7,8]. Although sorghum grows well under unsuitable soil and weather conditions as contrasted to other crops [9]. It can grow in desert and semi-desert regions is affected by moisture deficit at terminal growth phases like flowering and post-flowering that renders the most adverse effect on its yield [10,11]. Gebrekidan [12] reported that it has large genetic diversity in East African countries like Ethiopia, Sudan and Eritrea. It was reported that the presence of significant genetic variability among tested genotypes by [13]. Studies in Ethiopia also revealed that more than 95% of sorghum production area was covered by landraces and with limited use of improved varieties [14]. This implies that the Ethiopian breeding program has not yet efficiently utilized the available genetic variability for genetic improvement. Phenotypic selection based on important characters with bigger heritability along with higher genetic advance (GA) is efficient for the success of breeding to develop best genotypes for sorghum breeding program [15].

Genotypic and phenotypic coefficient of variations, heritability estimates, and genetic advance are important information to an efficient genetic improvement [16]. Hence, searching for sources of extra genetic diversity and largest exploitation of the in-situ genetic variability for selection and development of best and high yielding genotypes can help to increase the yield of sorghum. Godbharle et al. [17] also concluded that quantifying the magnitude and pattern of genetic variability is a precondition to develop best varieties for different objectives of sorghum production. This investigation was conducted to assess the genetic variability in sorghum for yield and yield components under drought conditions.

MATERIALS AND METHODS

The research was implemented at two research sites specifically Sheraro (14°24′N/37°45′E) and Miesso (9°14′N/40°45′E); where sorghum is the major crop (Figure 1). The experiments were conducted during the 2018 growing season. The altitudes for Sheraro and Miesso are 1010 and 1470 meters above sea level, respectively. The soils in both Sheraro and Miesso are dark clay vertisols. The agro-ecology of Sheraro and Miesso is SM1 (SM1-4) sub-moist hot warm lowland and SM1 (SM1-1) hot to warm sub-moist plains, respectively [18]. The range of temperature during the growing season was from 14.38 to 37.76 ℃ and the average of rainfall was 653.9 mm at Sheraro (Supplementary Figure S1). Whereas the temperature range and average rainfall at Miesso during the same period were 16–31 ℃ and 355.3 mm (Supplementary Figure S2).

FIGURE 1

Figure 1. Map of the experimental sites.

Two hundred Ethiopian sorghum landrace collections selected based on adaptation to the moisture stress environments and two hundred genetic lines were used along with known stay-green genotypes such as B-35 (Supplementary Table S1).

Row column experimental design (50 rows × 8 columns) with two replications was used at each site. It was planted based on the onset of rain in the two test sites, which started from end of June until first week of July 2018 crop season. Seeds were planted direct into a three meters plot length with single row with seventy-five centimeters spacing from one row to the next row and fifteen centimeters spacing between plants. Seedlings were thinned to an interplant spacing of 15 cm in two weeks after emergence. Fifty kilograms of urea and 100 kilograms di-ammonium phosphate per hectare of fertilizer was applied as per the national fertilizer recommendation. Half of the urea altogether the Di-ammonium phosphate was applied on planting. Urea was top dressed at the 6 to 8 leaf stage. Each suggested agronomic management (weeding, cultivation, etc.) was used uniformly for both stress and non-stress conditions.

Morphological evaluation of genotypes was done using the sorghum (Sorghum bicolor (L.) Moench) descriptors as perBPGR/ICRISAT descriptor list. The traits used for morphological evaluation are leaf area (cm²), thousand grain weight (g), days to fifty percent flowering, days to seventy five percent maturity, panicle length (cm), plant height (cm) and grain yield (kg/ha). Randomly selected five plants per genotype were used for observations and measurements. Yield was evaluated plot wise.

Analysis of variance (ANOVA): Row column design was used to estimate analysis of variance for seven traits. Fit linear mixed-effects models lmer function in lmerTest R package was used for the analysis of variance. Genotypes as fixed effect and replications, row and column as random effect [19,20]. Random effect ANOVA was evaluated using ANOVA-like table for random-effects chi-square test rand function in lmerTest R package [21]. Before computing the analysis of variance homogeneity test for error variance was done.

Estimation of variance components: META-R software [22] was used to calculate genotypic and phenotypic variances and their coefficients of variations as per the formula recommended by [23] as cited by [24] as follows:

$\text{Environmental variance }\left(({\text{σ}}_e^2)\right)=Mse$ (1)

$\text{Genotypic variance }\left(({\text{σ}}_g^2)\right)=\frac{Msg-Mse}{\text{r}}$ (2)

Where $Msg$ is for mean square due to genotypes, $Mse$ is for mean square of error (Environmental variance), $r$ = number of replication.

$\text{Phenotypic variance }\left(({\text{σ}}_p^2)\right)={\text{ σ}}_g^2+{\text{ σ}}_e^2$ (3)

Where, ${\text{σ}}_g^2$ is for genotypic variance and ${\text{σ}}_e^2$ is for environmental variance.

$\text{Phenotypic coefficient of variation }\left((\text{PVC})\right)=\frac{\sqrt{{\text{σ}}_p^2}}{\overline{X}}\times 100$ (4)

$\text{Genotypic coefficient of variation }\left((\text{GCV})\right)=\frac{\sqrt{{\text{σ}}_g^2}}{\overline{X}}\times 100$ (5)

Where, ${\text{σ}}_p^2$ is for phenotypic variation, ${\text{σ}}_g^2$ is for genotypic variation and $\overline{X}$ is for grand mean of the character studied.

Estimation of heritability in broad sense: The ratio of the genotypic variance (${\text{σ}}_g^2$ ) to the phenotypic variance (${\text{σ}}_p^2$ ) was calculated to give broad sense heritability (${\text{h}}^2$ ) expressed on genotype mean as recommended by [25] and as cited in [24] as follows:

${\text{h}}^2=\frac{{\sigma }_g^2}{{\sigma }_p^2}\times 100$ (6)

Estimation of correlation coefficients: As suggested by [26], phenotypic and genotypic correlation coefficients were estimated out of the consequent variance and covariance components as:

$\text{Phenotypic correlation coefficient }\left((r_p)\right)=\frac{PCovxy}{\sqrt{{\sigma }_p^{2}x\times {\sigma }_p^{2}y}}$ (7)

$\text{Genotypic correlation coefficient }\left((r_g)\right)=\frac{gCovxy}{\sqrt{{\sigma }_g^{2}x\times {\sigma }_g^{2}y}}$ (8)

Where, $PCovxy$ is for phenotypic covariance between variables $x$ and $y$ . Similarly $gCovxy$ is genotypic covariance between variables $x$ and $y$ ; ${\text{σ}}_p^{2}x$ and ${\text{σ}}_g^{2}x$ are phenotypic and genotypic, variances for variable $x$ ; and ${\text{σ}}_p^{2}y$ and ${\text{σ}}_g^{2}y$ are phenotypic and genotypic variances for the variable $y$ respectively. The coefficients of correlation were tested using ‘r’ tabulated value at n−2 degrees of freedom, at 5% and 1% probability level, where n is the number of treatments (accessions) as cited by [24].

Path coefficient analysis: was done considering grain yield per hectare as dependent variable and the test characters as independent (causal) variables. Based on [27] path coefficient analysis was evaluated using the phenotypic and genotypic correlation coefficients to determine the direct and indirect effects of yield components on grain yield based on the following relationship:

$r_{ij}=p_{ij}+\sum r_{ik}\times p_k$ (9)

Where $r_{ij}$ is for mutual correlation between the independent character ($i$) and dependent character ($j$) as estimated by the genotypic correlation coefficients. Whereas $p_{ij}$ is for components of direct effects of the independent character ($i$) on the dependent character ($j$) as calculated by the genotypic path coefficients. $\sum r_{ik}\times p_{kj}$ is for summation of components of indirect effect of a given independent character ($i$) on a given dependent character ($j$) via all other characters ($k$). The contribution of the remaining unknown factor was measured as the residual factor (${\text{P}}_{\text{R}}$), which was calculated as:

${\text{P}}_{\text{R}}=(\sqrt{\Sigma r_{ij}{p}_{ij}}$ (10)

The level of ${\text{P}}_{\text{R}}$ indicates how best the causal factors account for the variability of the dependent factor [23]. That is, if ${\text{P}}_{\text{R}}$ result is small (for instance, nearly zero), the dependent character considered (grain yield) is fully clarified by the variability in the independent characters, whereas higher ${\text{P}}_{\text{R}}$ value implies that some other factors which have not been considered, need to be included in the analysis to account fully the variation in the dependent character (grain yield) [24].

Clustering and principal component analysis: Average linkage was used to analyze clustering of genotypes using SAS software using seven morphological descriptors that were found significantly different among the genotypes at least at one location. To decide the number of clusters pseudo-F statistics and pseudo ${\text{ t}}^2$ statistics were examined [24]. Mahalanobis ${\text{ D}}^2$ statistics was used to calculate distances between clusters:

$${\text{ D}}^2{}_{ij}=\left(({\text{X}}_i-{\text{X}}_j)\right){\text{ S}}^{–1}\left(({\text{X}}_i-{\text{X}}_j)\right)$$ (10)

Where: ${\text{D}}^2{}_{ij}$ = distance between cases $i$ and $j$ , ${\text{X}}_i$ and ${\text{X}}_j$ is the vectors of the values of the variables for cases $i$ and $j$ ${\text{S}}^{–1}$ = inverse of the pooled variance covariance matrix.

Statistical Analysis System (SAS) software [28] was also used to calculate principal components based on correlation matrix. The ${\text{D}}^2$ values analyzed for pairs of clusters were considered as the calculated values of Chi square (${\chi }^2$) and were tested for significance both at 1 and 5% probability levels against tabulated value of ${\chi }^2$ for P degree of freedom, where P is the number of characters considered [23,24].

RESULTS

Analysis of variance showed highly significant differences among the sorghum genotypes both at Sheraro and Miesso for seven traits, indicating the existence of broad genetic variability among them (Table 1).

TABLE 1

Table 1. Variance of fixed and random effects with significant values among sorghum landraces evaluated at Miesso and Sheraro under rain fed conditions (2018/19).

As indicated in Table 2, the range and mean values for leaf area (cm²) were from 241 to 805 and 416.39, respectively at Sheraro whereas these values for the same trait at Miesso were 213.00–2004.95 and 354.17, respectively (Table 3). The range and mean values for thousand grain weight (g) at Sheraro were 7–51 and 24.3, respectively but at Miesso they were 10–60 and 20.54, respectively. The range and mean of days to maturity (DM) at Sheraro were 85–122 and 106.75 whereas these values at Miesso were 90–138 and 112.23, respectively. The range and mean for plant height (cm) were 22–327 and 226.07, respectively at Sheraro and at Miesso, they were 17–296 and 201.73, respectively. For panicle length (cm) the range and mean values were 7.8–75.4 and 21.87 at Sheraro and 6–35 and 20.88, respectively at Miesso. The range and mean grain yield (kg/ha) values were 216.53–7701.33 and 1914.14 at Sheraro whereas at Miesso they were 10.67–3106.67 and 997, respectively. Range and mean for days to 50% flowering at Sheraro were 50–95 and 73.63, respectively whereas at Miesso they were 52–100 and 73.19, respectively. The range for days to 50% flowering among germplasm was almost a month at both research sites.

Analysis results of phenotypic (${\text{σ}}_p^2$ ) and genotypic (${\text{σ}}_g^2$ ) variances and phenotypic (PCV) and genotypic coefficients of variation (GCV) are presented in Table 2 for the Sheraro site and in Table 3 for Miesso. The genetic coefficient of variation at Sheraro ranged from 24.6% for panicle length to 44.49% for grain yield. The range of genetic coefficient of variance at Miesso was from 29.64% for leaf area to 89.58% for thousand-grain weight. Similarly, the range for phenotypic coefficient of variation at Sheraro was from 44.49% for panicle length to 92.88% for grain yield. The phenotypic coefficient of variation at Miesso varied from 38.57% for DM to 91.98% for thousand-grain weight. The GCV values were lower than that of PCV in this study, signifying that the environment had more effect on the expression of these traits. All the seven traits studied at Sheraro and Miesso had high phenotypic and genotypic coefficients of variance. This reveals that selection may be efficient based on these characters and could be a good indicator of genetic potential.

The heritability estimation for traits under research is presented in Tables 2 and 3. Heritability values are useful in forecasting the expected progress to be achieved through the process of selection. Genetic coefficient of variation along with heritability estimation provides a reliable estimate of the amount of genetic advance anticipated through phenotypic selection. Heritability at Sheraro extended from 30.57% for panicle length to 75.83% for grain yield. The heritability estimates at Miesso varied from 47.4% for leaf area to 96.72% for days to maturity. Days to maturity, days to fifty percent flowering, thousand-grain weight, plant height, and panicle length had high heritability at Miesso. This shows that selection using these traits could effectively select high-yielding genotypes. This is because there would be a close correspondence between the genotypes and the phenotype due to the relatively small contribution of the environment to the total variability. Characters, like plant height, leaf area and grain yield at Sheraro had moderately high heritability. Traits like thousand grain weight, DF, and DM at Sheraro had medium heritability. A character that had low heritability was panicle length at Sheraro. Grain yield had moderately high heritability at Miesso whereas leaf area had medium heritability at Miesso.

TABLE 2

Table 2. Estimates of genetic parameters for traits of sorghum genotypes evaluated at Shararo in 2018.

TABLE 3

Table 3. Estimates of genetic parameters for traits of sorghum genotypes evaluated at Mieso in 2018.

Estimates of phenotypic and genotypic correlation coefficients between each pair of characters are presented in Tables 4 and 5. The amount and pattern of genotypic correlation coefficients for most of the characters were higher than their corresponding phenotypic correlation coefficients, except for a few cases, which show the masking effect of the environment in the total expression of the genotypes.

The phenotypic and genotypic correlations of grain yield with other characters are presented in Tables 4 and 5. The scope of genotypic correlation of grain yield with other characters starts from 0.809 for thousand grains weight to 0.173 for days to maturity at Sheraro. At Miesso, it ranged from 0.843 for a thousand grain weight to 0.166. Traits that correlate significantly with grain yield may be crucial grain yield predictors in improving grain yield in sorghum. Grain yield showed highly significant and positive genotypic correlation with thousand-grain weight, leaf area, DF, DM, plant height, and panicle length at both sites. Grain yield with panicle length, days to maturity with days to 50% flowering, and grain yield with thousand grain weight had significant positive associations. In contrast, the rest of the correlations among morphological descriptors were non-significant.

The genotypic associations among the agronomic traits at Sheraro were as follows (Table 4): Leaf area had a positive significant genotypic correlation with 1000-grain weight, days to physiological maturity, and panicle length. Thousand grain weight had a positive significant genotypic correlation with DM, plant height, and panicle length. There was a positive significant genotypic correlation of days to maturity with DF, plant height, and panicle length. Similarly, the genotypic associations between yield attributing traits at Miesso were as follows (Table 5): Leaf area had a positive significant genotypic association with thousand-grain weight, days to 50% flowering, and days to physiological maturity. Thousand grain weight was a descriptor with significant and positive genotypic correlation with days to fifty percent flowering, days to physiological maturity, plant height, and panicle length. Days to maturity had a positive-significant genotypic correlation with days to fifty percent flowering, plant height, and panicle length. Days to 50% flowering had a positive-significant association with plant height and panicle length. The range of phenotypic correlation started from 0.89 for grain yield/thousand grain weight up to 0.002 for plant height/panicle length at Sherro whereas at Miesso it was from 0.789 for grain yield/panicle length to 0.01 for thousand-grain weight/leaf area.

TABLE 4

Table 4. Genotypic (above diagonal) and phenotypic (below diagonal) correlation coefficients for traits of sorghum genotypes evaluated at Sheraro in 2018.

TABLE 5

Table 5. Genotypic (above diagonal) and phenotypic (below diagonal) correlation coefficients for traits of sorghum genotypes evaluated at Miesso in 2018.

Path co-efficient analysis delivers an efficient method of splitting direct and indirect causes of correlation. It permits a determinant look to recognize the specific forces acting to produce a given association and measures the relative importance of each causal factor. The genotypic direct and indirect effect of different characters on grain yield kg/ha is presented in Tables 6 and 7. thousand grain weight pursued by leaf area, panicle length, and DM exerted a positive prominent direct effect on grain yield at Sheraro whereas days to fifty percent flowering, 1000-grain weight, and panicle length showed direct positive effect on grain yield at Miesso. Thousand grain weight and panicle length exerted a direct positive effect on grain yield at both sites. Thus, according to both sites, selecting sites selecting genotypes having large thousand grain weight and long panicle length could be used to improve grain yield in sorghum genotypes because of their direct effect on yield.

Days to fifty percent (DF) and plant height at Sheraro and DF, panicle length, and thousand grain weight at Miesso showed a high indirect positive effect via days to maturity on the grain yield kg/ha. Days to maturity and panicle length revealed a high positive indirect effect via leaf area on grain yield at Sheraro. Leaf area and panicle length had a high positive indirect effect on grain yield at Miesso. Therefore, yield can be improved by selecting a1000-grain weight, panicle length, leaf area, plant height, and days to maturity as these traits have an indirect effect on grain yield. The residuals (0.33431 and 0.40460) indicate that traits involved in the path analysis expounded 66.56% and 59.54% of the total variation in grain yield Sheraro and Mieso, respectively. The results from the genetic variability, character association, and path coefficient analyses revealed that morphological descriptors such as thousand grain weight and panicle length were useful concerning genetic variability, correlation, and path-coefficient analysis. The larger variability in these traits could give a main scope for the development of high-yielding lines through selection in the segregating generations.

TABLE 6

Table 6. Estimates of direct (bold diagonal) and indirect (off diagonal) genotypic effects of traits on grain yield in sorghum genotypes evaluated at Sheraro in2018.

TABLE 7

Table 7. Estimates of direct (bold diagonal) and indirect (off diagonal) genotypic effects of traits on grain yield in sorghum genotypes evaluated at Miesso in 2018.

The first six PCAs from Sheraro data are presented in Table 8. Sixty-three and 69/100 percent (63.69%) of the total variability among the four hundred genotypes evaluated for different morphological traits was scored by the first three principal components with Eigenvalue > 1. The remaining PCAs provided only 36.31% of the total morphological diversity among the genotypes studied. PCA 1 provided the greatest variability of 25.03% pursued by PCA 2, which donated a total phenotypic variability of 23.53%, and PCA 3 supplied 15.13% of the total variation. The useful morphological descriptors in PCA 1 that were the main contributors to the variations among the accessions were days to maturity, 1000-grain weight, plant height, days to fifty percent flowering, and grain yield which had positive factor loading value. PCA 2 was related to diversity among genotypes due to days to fifty percent flowering, days to maturity, and panicle length. Similarly, PCA 3 expressed positive loading values for variations among genotypes due to panicle length, and plant height.

The principal component analysis on data from Miesso (Table 9) showed that out of eight components, three principal components scored most the genetic variance. Sixty-one and 6/10 percent of the total variability among the four hundred genotypes evaluated for seven morphological traits was scored by the first three principal components with Eigenvalue > 1. Only 38.4% of the total morphological diversity among the genotypes studied was scored by the remaining PCAs. The main morphological traits in PCA 1 that contributed to the variations among the accessions were days to maturity (DM), days to 50% flowering (DF), leaf area, panicle length, and plant height which had positive factor loading value whereas grain yield and thousand grain weight had contributed negatively. PCA 2 was related to diversity among sorghum genotypes due to thousand-grain weight, plant height, grain yield, DM, DF, and leaf area. Panicle length had a negative factor loading value.

TABLE 8

Table 8. Percentage, cumulative variance and Eigen vectors on the first six principal components for characters of sorghum genotypes evaluated at Sheraro in 2018.

TABLE 9

Table 9. Percentage, cumulative variance and Eigen vectors on the first six principal components for characters of sorghum accessions evaluated at Miesso in 2018.

At Sheraro, the $D^2$ analysis of the 400 genotypes grouped them into six clusters with different numbers of genotypes in each cluster (Table 10). Cluster I had the largest number of entries, which is 181 (45.25%), followed by Cluster IV with 102 (25.5%) genotypes, cluster II with 67 (16.75%) genotypes, and cluster III with 41 (10.25%) genotypes. The remaining clusters (cluster V and cluster VI) had eight (2%) and one (0.25%) genotype, respectively. The $D^2$ analysis for the data from Miesso showed that the 400 germplasms were grouped into six clusters (Table 11). Clusters II and I had the highest number of genotypes with 156 (39%) genotypes each, and Cluster III is next with 44 (11%) genotypes, cluster IV with 38 (9/5%) genotypes, cluster V with 4 (1%) genotypes and cluster VI with 2 (0.5%) genotypes. A cluster with a single genotype may be due to total isolation preventing the gene flow or intensive natural/human selection for diverse adaptive complexes. From the breeding point of view, these genotypes are very important. They can have very important genotypes with desirable characteristics.

TABLE 10

Table 10. Grouping of sorghum genotypes into different diversity classes based on data from the Sheraro trial (2018).

TABLE 11

Table 11. Grouping of sorghum genotypes into different diversity classes based on data from the Miesso trial (2018).

The mean inter-cluster D² values for Sheraro are given in Table 12. The χ²-test for the 6 clusters showed that there was a statistically highly significant difference (P = 0.1%) in all traits except for cluster I with II (7.825), I with III (29.27), I with IV (12.84), II with III (6.922) and III with V (14.05). The topmost mean inter-cluster D² was recorded between cluster IV with cluster VI (D² = 301.5) followed by cluster I with cluster IV (D² = 197.6) and cluster IV with cluster V (D² = 159) which had shown the largest inter-cluster distance. Likewise, the average inter-cluster D² values for Miesso are presented in Table 13. The χ²-test for the six clusters indicated that there was a statistically highly significant difference (P = 0.1%) probability level in all characters except cluster I with II (7.645), I with III (5.884), I with IV (27.43), II with III (26.04), III with IV (8.999) and IV with V (8.523). The highest average inter-cluster D² was recorded between cluster V with cluster VI (D² = 202.9) followed by cluster IV with cluster VI (D² = 158.5) and cluster III with cluster VI (D² = 126) which had shown the largest inter-cluster distance. These clusters were genetically more divergent from each other than any other clusters.

In the case of Sheraro, the mean value of the quantitative morphological descriptors in each cluster is shown in Table 14. Cluster I had the description of the shortest days to maturity and shortest plant height. The remaining traits were reasonable in magnitude. Cluster II was described by moderate magnitudes of all the characters. The smallest panicle length characterized cluster III and the remaining traits were modest in magnitude. The smallest grain yield, smallest thousand-grain weight, and the longest DF illustrated cluster IV. Cluster V was characterized by the largest thousand-grain weight; the following features characterized Cluster VI: largest grain yield, shortest DF, longest DM, tallest plant height, and largest panicle length. The average value of the quantitative characters at Miesso is given in Table 15. All the traits in cluster I showed moderate results in magnitude. The longest DF, longest DM, largest panicle length, smallest grain yield, and smallest thousand grain weight characterized cluster II. The shortest DF, shortest DM, and largest thousand grain weight characterized cluster III. Cluster IV was characterized by the shortest plant height whereas. Cluster V was characterized by grain yield and tallest plant height; the smallest panicle length characterized Cluster VI.

TABLE 12

Table 12. Pairwise generalized squared distance (D²) among sorghum accessions in six clusters at Sheraro (2018).

TABLE 13

Table 13. Pairwise generalized squared distance (D²) among sorghum accessions in six clusters at Miesso (2018).

TABLE 14

Table 14. Cluster means for six characters of sorghum accessions evaluated at Sheraro in 2018.

TABLE 15

Table 15. Cluster means for six characters of sorghum accessions evaluated at Miesso in 2018.

DISCUSSION

The analysis of variance for different traits including yield and yield components was highly significant at both research sites, indicating the presence of a significant amount of variability for these traits among the genotypes, which provides an opportunity for selection of desirable genotypes for further genetic improvement. Similarly, [29] reported significant differences for days to flowering, plant height, grain yield per panicle, panicle length, number of tillers per plant, panicle weight, and panicle exertion at Kobo; and days to flowering, days to maturity, plant height, grain-filling period, grain yield per panicle, panicle length, panicle harvest index at Miesso. Likewise, [30] reported that analysis of variance showed significant differences among fifty-two sweet sorghum genotypes for days to 50% flowering, days to physiological maturity, plant height, stem girth, internode length, number of leaves, stay green trait, brix content, fresh cane weight, mill-able cane weight, juice weight, juice volume, sucrose percentage, juice extractability percentage, cane yield, juice yield, and sucrose yield. The range of days to flowering period among genotypes was almost a month at both research sites and this might be attributed to the diversity of genotypes with different genetic backgrounds and adaptations.

In this study, the GCV values were lower than that of PCV, indicating that the environment had an important role in the expression of these characters. Generally, quantitative or agronomic traits are highly influenced by the environment. Adane et al. [31] reported a similar result that the phenotypic variances were slightly higher than the genotypic variance for days to 50% flowering, plant height, hundred seed weight, and overall agronomic aspect, signifying the influence of environment on these traits. All the seven traits studied at the Sheraro and Miesso sites had high phenotypic and genotypic coefficients of variance. This indicates that selection may be effective based on these characters and their phenotypic expression would be a good indication of genetic potential. There is a large scope for selection based on these traits and the diversity in the genotypes provides a huge potential for future breeding programs. Yaqoob M et al. [32] reported similar higher estimates of PCV and GCV for plant height, leaf area index, and grain yield, suggesting that environmental influence was moderate for these traits as the experiment was managed under uniform input level to all the genotypes. Elangovan and Babu [33] also reported higher PCV and GCV values for days to 50% flowering, plant height, and 100-seed weight. On the contrary, [34] reported a low coefficient of variation for plant height, days to flowering, thousand seed weight, and medium value for grain yield.

The estimate of GCV and PCV alone is not very helpful in determining the heritability of a trite portion. The amount of genetic advance to be expected from the selection can be achieved by estimating heritability along with the coefficient of variability [35]. Heritability values help predict the expected progress to be achieved through the process of selection. The genetic coefficient of variation along with the heritability estimate provides a reliable estimate of the amount of genetic advance to be expected through phenotypic selection. Days to maturity, days to 50% flowering, thousand seed weight, plant height, and panicle length had high heritability at the Miesso site. This indicates that selection using these traits could be effective. This is because there would be a close correspondence between the genotypes and the phenotype due to the relatively small contribution of the environment to the total variability. Characters, like plant height, leaf area and grain yield at the Sheraro site had moderately high heritability. Traits like thousand seed weight, days to flowering, and days to maturity at the Sheraro site had medium heritability. The character that had low heritability was panicle length at the Sheraro site. Heritability for grain yield was moderately high at Miesso. However, leaf area had medium heritability at Miesso.

Although variability estimates provide information on the extent of improvement, they do not throw much light on the extent and nature of the relationship, which exists between the characters [36]. This could be obtained from simple association analysis. Knowledge of the association of component characters with single plant yield may greatly help in making selection precise and accurate. The greater the magnitude of correlation coefficients, the stronger the association. The magnitudes of genotypic correlation coefficients for most of the characters were higher than their corresponding phenotypic correlation coefficients, except in few cases, which indicates the influence of the environment in the total expression of the genotypes was low. Such results are in concurrence with the results of [37].

Traits significantly correlated with grain yield may be important yield predictors in sorghum breeding. Grain yield showed positive and highly significant genotypic association with thousand seed weight, leaf area, days to 50% flowering, days to physiological maturity, plant height, and panicle length at both sites. Grain yield with panicle length, days to maturity with days to flowering, had significant positive associations while the rest of the associations were non-significant. Similarly, [37–40] reported that seed yield exhibited a positive and significant correlation with thousand seed weight. In contrary to the present study, [29] reported that thousand grain weight had a non-significant association with grain yield. Khandelwal et al. [37] also reported that grain yield had a non-significant positive genotypic correlation with the thousand grain weight of sorghum.

Path co-efficient analysis provides an effective means of partitioning direct and indirect causes of association. It permits a critical look to recognize the specific forces acting to produce a given correlation and measures the relative importance of each causal factor. In the present study, thousand grain weight followed by leaf area, panicle length, and days to maturity exerted a positive prominent direct effect on grain yield at Sheraro. Days to flowering, thousand seed weight, and panicle length also showed a positive direct effect on seed yield at the Miesso site In line with the present study, [41] reported panicle length having a direct effect on grain yield. Patel et al. [42] reported that thousand grain weight had a direct effect on sorghum grain yield. Similarly, [43] also reported that panicle length had a direct effect on sorghum grain yield. This information would be of greater value in selecting the useful traits and thus optimize the data recording by taking observations on a few related traits in the preliminary trials involving a large number of genotypes [44]. Therefore, according to the data from both sites, selecting genotypes having high seed weight and long panicle length could be used to improve grain yield in sorghum. Similar to this study, [45] reported that days to maturity, panicle length, and thousand grain weight had shown a positive direct effect on grain yield. It was reported that panicle length had a positive direct effect on seed yield [37,43]. Iyanar et al. [41] also reported a direct effect of thousand grain weight on grain yield. However, days to 50% flowering and plant height at Sheraro; leaf area, days to maturity, and plant height at Mieso showed a negative direct effect on grain yield. These traits only contributed to seed yield mainly via their positive indirect effect with other characters. Plant height had a negative direct effect on seed yield at both sites. A similar result, that plant height had a negative direct effect on seed yield, was reported by [45] and [46]. Days to flowering and plant height at Sheraro and days to flowering, panicle length, and thousand grain weight at Miesso showed a high positive indirect effect via days to maturity on the grain yield/ha. Days to maturity and panicle length showed a high positive indirect effect via leaf area on seed yield at the Sheraro site. Leaf area and panicle length revealed a high positive indirect effect on seed yield at the Miesso site Therefore, yield can be improved by selecting for thousand seed weight, panicle length, leaf area, plant height, and days to maturity due to their indirect effect on yield. The residuals (0.3343)1 for Sheraro and (0.40460) for Miesso indicate that characters included in the genotypic path analysis explained 66.56% at Sheraro and 59.54% at Mieso of the total variation in seed yield which indicates that there may be some more components that are contributing towards seed yield.

The association between the characters and their contribution to the diversity can also be confirmed by PCA analysis. The principal component analysis (PCA) using the mean values of the genotypes provides a reduced dimension model that would indicate measured differences among the germplasm. In this study, only PCAs with eigenvalues greater than one, which determines as a minimum of ten percent of the variation, was considered as recommended by [47]. The largest eigenvalues have the largest attributes in principal components. The results at Sheraro showed the importance of the first three principal components (PC) with eigenvalue > 1 in discriminating the entire germplasm. The percentage of variation explained by these PCs was more than 63.69% of the total variability among the four hundred sorghum genotypes evaluated for different morphological traits. The remaining components contributed only 36.31% towards the total morphological diversity among the genotypes studied. Similarly, [48] reported that the first three principal components with eigenvalue > 1 contributed about 66.35% of the total variability among the 100 sorghum germplasm genotypes evaluated for different morphological traits. For the Mieso site, the principal component analysis revealed that out of eight components, three principal components scored most the genetic variance. The first three principal components with eigenvalue > 1 contributed about 61.6% of the total variability among the four hundred sorghum genotypes evaluated for seven morphological traits. Jain and Patel [49] also reported in line with this report.

Genetic divergence in a population, especially in relation to important traits, is necessary and an indispensable pre-requisite for successful plant selection work. Different members within a cluster are assumed to be more closely related in terms of the traits under consideration with each other than those members in different clusters. Similarly, members in clusters with no significant distance are assumed to have more close relationships with each other than they are with those in significantly distant clusters. A better understanding of the genetics of morphological characteristics is required by the breeder to increase the efficiency of the selection of more diverse and adapted parents for crop improvement [50]. The success of any crop-breeding program is based on the knowledge and availability of genetic variability for efficient selection [2]. At both research sites, the cluster analysis of the 400 genotypes grouped them into six clusters with a variable number of entries in each cluster, indicating the presence of a wide range of genetic diversity among the genotypes under investigation. The formation of solitary clusters in the present study may be due to total isolation preventing the gene flow or intensive natural/human selection for diverse adaptive complexes. These genotypes may be unique and useful from a breeding point of view. This genetic diversity can be due to the geographic diversity of the collection sites of the sorghum landraces used in this study. In line with this study, [51] reported six clusters based on D² statistics among one hundred fifty sorghum genotypes. Vijaylaxmi et al. [52] also reported six hierarchical clusters containing one to sixty-one sorghum genotypes. Similar results were observed by [53–60] who had reported 14, 13, 23, 10 and 14 clusters of sorghum genotypes, respectively.

Maximum inter-cluster distances were observed between cluster 4 and cluster 5 at Sheraro and between clusters 5 and 6 at Miesso, indicating that genotypes belonging to these groups were genetically most divergent. Such genetically diverse sorghum genotypes could be effectively utilized as parents in the hybridization program. This type of hybridization would be useful for obtaining the highest number of valuable segregates along with maximized vigor. Kadam et al. [61], Shridher et al. [55], and Yirgalem et al. [24] reported similar instances of inter-cluster distances.

In conclusion, the overall genetic diversity analysis and phenotypic evaluation have identified four common genotypes that are consistently superior in all experiments, having unique alleles and drought tolerance traits. The genotypes that showed promising results for drought tolerance include Acc#239130, Acc#220255, Acc#235810 and Acc#220253. It is therefore recommended that these four genotypes be used as breeding material for further improvement of the crop.

SUPPLEMENTARY MATERIALS

The following supplementary materials are available online at https://doi.org/10.20900/cbgg20240008. Supplementary Table S1: Sorghum (Sorghum bicolor (L.) Moench) genotypes (accessions) used in this study (2018); Supplementary Figure S1: Monthly rainfall amount at Sheraro (2018); Supplementary Figure S2. Monthly rainfall amount at Miesso (2018).

DATA AVAILABILITY

All data supporting the findings of this study will be available on paper and its supplementary materials published online. The original breeding data is uploaded to the Journal through online uploading system. The data is uploaded to the publisher consumption only. The original data can’t be shared to any third party. Seed of the breeding lines can be obtained from Melkassa Agricultural Research Center and it can be shared.

AUTHOR CONTRIBUTIONS

Conceptualization, YG; Methodology Formulation, YG; Software Analysis, YG; Writing—Original Draft Preparation, YG, KB, FA, TT and AT; Writing—Review & Editing, YG, KB, FA, TT and AT; Supervision, KB, FA, TT and AT.

CONFLICTS OF INTEREST

The authors declare that they have no conflicts of interest.

FUNDING

This work was supported by TERRA project coordinated in by Ethiopian Institute of Agricultural Research Institution. TERRA project is among the projects financed by Bill & Melinda Gates Foundation. The project has terminated after the fieldwork.

ACKNOWLEDGEMENTS

We thank to Melkassa Agricultural Research Center and Shire Maitsebri Resesarch center for providing experimental site. We especially thank to the sorghum research national coordination center staff based at Melkassa Agricultural Research Center, Adama, Ethiopia for providing planting material, technical support during data collection.