Multivariate analysis of phenotypic variability in Tef [ Eragrostis tef ( Zucc . ) Trotter ] genotypes from Ethiopia

Tef [Eragrostis tef (Zucc.) Trotter] is an important food crop in Ethiopia. The present research was conducted to characterize the phenotypic variability of 68 tef genotypes collected from Ethiopia. Where a Bi-replicated 7×10 alpha lattice design was used to evaluate the 70 tef genotypes at Holetta and Debre Zeit Research Centers during 2015. Based on the results of cluster analysis (CA), genotypes were grouped into twelve clusters and twenty nine genotypes formed a single cluster; whereas, nine clusters comprised of five or few genotypes. The first five principal components (PC) with eigenvalue greater than one accounted for 80% of the total genetic variation, height related traits, the diameters of the two basal culm internodes, and number of spikletes and primary branches per main panicle were traits that chiefly contribute for the total variance accounted for by the first PC. The second PC gross variation originated due mainly to variations in yield and yield related traits like grain yield, total biomass, straw yield and harvest index. In addition, genetic distances (D 2 ) which ranged from 326.22 to 25.07 were measured among the 12 clusters. Thus, indicates their chance of giving better genetic recombination and segregation of progenies.


INTRODUCTION
Tef (Eragrostis tef (Zucc.)Trotter) is traditionally grown as a staple cereal crop in Ethiopia and it is produced by more than 6.5 million small scale farmers (CSA, 2015).The grain is ground into flour, which is used to make a pancake-like local bread called "injera" (Ketema, 1997).
The grain is also used to make a local drink.In addition, tef has been used as a forage or pasture crop for cattle in some parts of the world (Assefa et al., 2009).The straw also serves as bedding material, mulch and domestic fuel source (Assefa et al., 2001b).Tef is better adapted to excessive or low soil moisture conditions than other cereals and often sown as a rescue crop (Tefera and Ketema, 2001).
Therefore, tef is considered an important food security crop.In Ethiopia, tef shows low productivity, because of the lack of lodging resistant varieties, low yielding varieties under a wide range of cultivation, pest problem, drought and labor intensive nature of cultivation (Assefa *Corresponding author.E-mail: thomas.tsige@yahoo.com.et al., 2013).The development of improved tef varieties had been successful (Assefa et al., 2013;MoARD, 2016).The existence of genetic variability is an important factor in the development and selection of improved varieties.Therefore, estimating the genetic variation among landraces will enhance breeding activities (Assefa et al., 2015;Kefyalew et al., 2000).Tef is an ancient crop in Ethiopia and cultivated across a wide range of environments, which can contribute to greater genetic variation.The Ethiopian Biodiversity Institute (EBI) currently holds 6000 tef landraces mostly from altitudes ranging from 800 to 3200 m.a.s.l.(Tesema, 2013).Hence, multivariate analysis is a useful tool for characterization and classification of plant genetic resources evaluated for several pheno-morphic and agronomic traits (Assefa et al., 2003).The present study was conducted to study variability of newly collected local tef genotypes.

MATERIALS AND METHODS
The experiment was planted during 2015 growing season at Holetta (9°.03'N and 38°30'E) and Debre Zeit (8°44'N and 38°58' E).Sixtyeight locally collected genotypes along with two checks genotypes were evaluated in a 7×10 alpha lattice designs with two replications (Table 1).Tef accessions were initially (each contain 50-100 panicles selected from individual plants) collected from farmers' field within 15 km interval and also sown in separate rows for purification at Debre Zite Agricultural Research Center during the 2013 and 2014 main-cropping seasons and the 2015 off-season.
Genotypes were planted in a plot area of 1 m 2 (1 m × 1 m).A spacing of 0.2, 0.5 and 1.5 m were used between rows, plots and replications, respectively.For the seven month (from June to December) growing season, average rainfall, minimum and maximum temperatures of Holetta and Debre Zeit were: 710 mm, 5. 6°C, 19.5°C and 73 mm, 18.8°C, 24.9°C, respectively.Data collections were made on eighteen traits.Days to heading, days to maturity, days to grain filling period, lodging index, total biomass (g), grain yield (g), straw yield (g), thousand seed weight (g), and harvest index (HI) where taken on plot base.In contrast, plant height (cm), panicle length (cm), culm length (cm), number of total tillers per plant, number of fertile tillers per plant, number of spikelets per panicle, number of primary branches per main panicle, first basal culm internode diameter (mm), and second basal culm internode diameter (mm) were recorded on five random sample individual plants.
For multivariate analysis, the mean data of the 70 test genotypes for each of the traits were the first pre-standardized to mean zero and variance unity to avoid bias due to differences in measurement scales.
Multivariate statistical analysis methods included cluster analysis (CA) and principal component analysis (PCA) using the MINITAB statistical computer package, version 14.00 (MINITAB, 2003).Points where local peaks of the pseudo F statistic join with small values of the pseudo t 2 statistic followed by a larger pseudo t 2 for the next cluster fusion were observed to decide the number of clusters (SAS Institute, 2002).
Genetic distance between clusters was computed using the generalized Mahalanobis's D 2 statistics formula as suggested in Singh and Chaudhary (1996) and distance analysis was computed using the SAS computer software (SAS Institute, 2002).It was also made based on the mean values for the 18 quantitative traits and 70 tef genotypes over the two locations.where D 2 p= total generalized distance based on p characters, Xi and Xj are the p mean vectors of 70 test genotypes I and j, respectively, S -1 =pooled error variance and co-variance matrix.
The D 2 value obtained for pairs of clusters was considered as the calculated value of Chi-square and was tested for significance at 5 and 1% levels of probability against the tabulated values of 2 at 'q' degrees of freedom, where q represents the number of traits studied (Fikreselassie, 2012).

Cluster analysis
Using a 73% similarity level, the genotypes formed 12 clusters (C).The number of genotypes in each cluster ranged from 1 to 29 (Figure 1 and Table 2).The largest cluster (C-3) contains different tef germplasm collected from all zones, while C-6 was the second largest cluster and it comprised 14 germplasm accessions of which 13 were from Jimma and Horo Gudru Zones of Oromya, while the remaining tef germplasm lines were from North Wello Zone of Amhara Region.The third big cluster (C-5) constituted the improved variety Quncho (DZ-Cr-387) and other 8 local germplasm accessions of which two (Oro-ACC#8-L13 and Oro-ACC#9-L45) were from Jimma zone and the remaining 6 were equally distributed between North Wello and West Shewa zones of Amhara and Oromya, respectively.Beside those major clusters, each of clusters 7 and 2 comprised 5 and 3 tef germplasm lines, respectively.Regarding their origin, cluster 7 comprised of tef germplasm entirely collected from North Wello and North Shewa Zones of Amhara Region, while those in cluster 2 originated from North Wello Zone of Amhara.In addition, clusters 1, 4 and 9 each comprised two germpalsm accessions, with the former two containing types from North Wello of Amhara and West Shewa of Oromya Region, while the latter one contained accessions collected from Jimma Zone of Oromya.Four of the twelve clusters comprising single genotype including the germplasm accessions Oro-ACC#8-L30 (C-8), Oro-ACC#4-L18(C-10), Oro-ACC#4-L25) (C-11), and the released variety "Tseday" (C-12).Those unclear patterns of genotypes grouping in respect to their origin could be a result of free exchange of genotypes and the expansion of improved tef varieties.
In line with the present results, Assefa et al. (1999) categorized 320 tef lines into 14 major complexes consisting of 1 to 183 tef lines.Previous cluster analyses with different sets of tef materials have also demonstrated variable groupings of tef genotypes based on similarity (Assefa et al., 2000(Assefa et al., , 2001a(Assefa et al., , 2003)).
The cluster mean comparison for the 18 traits evaluated depicted that the first cluster consisted of tef germplasm lines with early panicle emergence and maturity, short grain filling period, thin first and second basal culm internodes.In contrast, this cluster is characterized by tef materials having high harvest index, lodging index and grain yield.On the other hand, except relatively high values of tiller number (total and fertile) and harvest index, the remaining characters of the genotypes included in the second cluster scored small values (Tables 3 and 4).However, most quantitative traits of tef germplasm lines measured within C-3, 5, 6 and 10 showed relatively high values.Unlike their common characters, C-3, 5 and 6 contained the largest number tef germplasm lines, while C-10 contained a locally collected single tef germplasm line (Oro-ACC#4-L18).In addition, both C-5 and C-6 showed lower tiller numbers (total and fertile) and lodging index values.On top of this, relatively lower values of harvest index were exhibited by cluster 10.Days to maturity and grain filling period showed the highest mean value in C-7, but the lowest cluster mean values of total biomass and grain yield, lower value of straw yield, lodging index, number of fertile and total tillers were noted for this cluster.Cluster 4 is characterized by tef germplasm lines which have relatively small number of primary panicle branches, high total tiller number, longer grain filling period, late maturity, and high values of total biomass, grain yield and straw yield.The lowest cluster mean values of most traits were noted for C-8, which contained the single tef germplasm line (Oro-ACC#8-L30)
Cluster 11 which comprised the solitary tef germplasm line Oro-ACC#4-L25, scored the lowest number of primary branches per main panicle and harvest index (Tables 3 and 4).Additionally, this cluster is characterized by tef genotype which had high mean value for height related traits, longer grain filling period and low grain yield.The last cluster (C-12) which contained only the single released variety "Tseday", scored the lowest values in phenological traits (days to maturity and grain filling period) and the highest value in total and fertile tiller numbers.Similarly, higher harvest index value, earliness in panicle emergence, lower total biomass, and straw yield were the characteristics features of this cluster.In addition, this specific character of the tef variety "Tseday" is in agreement with the inherent nature of the variety, because this variety is manly released for use in low moisture stress areas, and it possesses characteristics of earliness in phenological traits to escape terminal drought.Finally, most of tef germplasm lines which included in the two big clusters (3 and 5) had best performance with respect to most important traits under consideration.Those genotypes, therefore, can be recommended for further evaluation.

)
Most inter cluster distances showed highly significant (P<0.01)differences, while there were no significant intercluster distances between C-3 and C-5 and C-3 and C-6 (Table 5).In addition, the shortest (D 2= 25.07) inter-cluster D 2 values were estimated between C-3 and C-5, while the largest (D  2= 326.22) was estimated between C-8 and C-10, each of which contain one local tef germplasm line Oro-ACC#8-L3 and Oro-ACC#4-L18, respectively.Similarly, C-8 and C-11 comprised the second most divergent (D 2= 275.22) groups and in this case Oro-ACC#8-L30 formed far inter-cluster distance with Oro-ACC#4-L25.In addition, the other clusters (C 8 and C 12) which in that order contain the solitary local tef germplam  Overall, the released variety "Tseday" and the locally collected tef germplasm line (Oro-ACC#8-L30) had large genetic distance with most of the other clusters in this experiment.On top of this, the high inter-cluster distances noted among different clusters may result from locations in which those tef germplasms were collected and different genetic background of those tef materials (released vs. local tef germplasm lines).Generally, a wide generalized squared distance (D 2 ) serves as a better indicator for selecting crossing materials.Consequently, most divergent clusters noted in this study are expected to give maximum genetic recombination and genetic variation in the subsequent segregating generations.

Principal components analysis
The first five principal components (PCs) having a minimum eigenvalue of one accounting for 80% of the total variability observed among the 70 tef test genotypes (Table 6).Of these, the first PC alone explained about 40% of the total variance mainly due to the variations in height related traits (that is, plant height, panicle length, and culm length), first and second basal culm internode diameters, and number of spiklets and primary branches per main panicle.On the other hand, even if relatively lower percent variation was explained by PC 1 in the studies of Assefa et al. (1999Assefa et al. ( , 2000Assefa et al. ( , 2001a, b), b), most of the traits responsible for variation in PC 1 showed similarity with the current study.In addition, another experiment of Assefa et al. (2003) with seventeen traits of 60 tef germplasm population showed similarity in both percent variation explained, and the traits contributing to the variation in PC 1.However, the first PC in the studies of Adnew et al. (2005) and Jifar et al. (2015) explained relatively high proportion of the variation than that in this study.
Unlike, the first PC, most yield related traits like grain yield, total biomass, straw yield, harvest index and lodging index contributed to about 16% of the gross variation accounted for by the second PC (Table 6).This is line with results of the second PC of Assefa et al. (2000).However, slightly larger variability was reported by Assefa (1999Assefa ( , 2001bAssefa ( , 2003) ) in other studies, whereas Assefa et al. (2001a) and Adnew et al. (2005) reported that the second PC, respectively explained 7.1% more and 5.6% less variability than that in the current study.Furthermore, about 9, 8 and 6% of the total genotype variance was explained on the basis of the third, fourth and fifth PCs, respectively (Table 6).The former was largely due to the variations in phenological traits (that is, days to maturity and grain filling period), lodging index and number of primary panicle branches, whereas, number of total and fertile tillers were the primary

Conclusion
The grouping of tef genotypes into twelve clusters at 73% similarity level confirmed the existence of important trait variability among tef genotypes that could be recommended for further evaluation and regarding conservation of the indigenous tef genetic resources in Ethiopia, unclear patterns of genotypes grouping in respect to their origin in this experiment showed the importance to address each tef growing zones of the country.Height related traits (that is plant height, panicle length and culm length), first and second basal culm internode diameters and number of spiklets and primary branches per main panicle contributed more for the 40% variation explained by the first PC.In addition, most of tef germplasm lines which were included in the two big clusters (3 and 5) had best performance with regard to most important traits under consideration.Moreover, the higher mean values of most yield related traits of Oro-ACC#8-L30 (C-8) and earliness in maturity, higher tiller number and harvest index of "Tseday"(C-12), in line with their large genetic distance with most of the other clusters could make them source of elite materials for future use.
Author(s) agree that this article remain permanently open access under the terms of the Creative Commons Attribution License 4.0 International License

Figure 1 .
Figure 1.Dendrogram of seventy genotypes based on average linkage and Euclidean distance of 18 traits evaluated at two locations.

Table 1 .
Tef genotypes used in the study and area of collection.
*Year of release.

Table 2 .
List of tef genotypes grouped in 12 clusters (average linkage Euclidean distance cluster analyses) using 18 traits evaluated at two locations.

Table 3 .
Means for the 12 clustered for the 18 quantitative traits of 70 Tef genotypes clustered into 12 groups.

Table 4 .
The Generalized Squared inter cluster distance of 18 quantitative traits of 70 tef genotypes evaluated over two locations

Table 5 .
Eigenvectors and eigenvalues of the first five principal components for 18 traits of 70 tef genotypes evaluated at Debre Zeit and Holetta during the 2015 main cropping season

Table 6 .
Eigenvectors and eigenvalues of the first five principal components for 18 traits of 70 tef genotypes evaluated at Debre Zeit and Holetta during the 2015 main cropping season.