Garden cress (Lepidium sativum L.), belongs to Brassicaceae  Family,   is   a   fast  growing  annual  plant which is cultivated in the temperate and subtropical cold areas  throughout  the  world  for  its  food  and  medicine (Gokavi et al., 2004). It is also an aromatic plant which contains considerable amount of essential oil (Nigist and Sebsebe, 2009) and edible fatty oil rich in its medicinal properties. In addition to its medicinal value, garden cress is used in the form of vegetable (sprout) in North Africa, West and Central Asia and United Kingdom. The breeding program of garden cress relies on improvement of grain yield, oil content of the seeds, resistance to frost, disease and pests (Temesgen et al., 2013a,b; Sabaghnia et al ., 2015). The productivity of garden cress is very low; only 600 to 700 kg/ha in India (Sabaghnia et al., 2015) and 1691 to 3415 kg/ha in Ethiopia (Temesgen et al., 2013a). Hence, it requires the attention of plant breeders to evaluate different genotypes across diverse agro-ecological regions in order to obtain high yielding genotypes. Grain yield is the result of a number of complex morphological and physiological processes that interact with each other and with the environment at different growing stages (Semahegn, 2011). The improvement of landraces for grain yield is not only dependent on the nature and extent of genetic variability, heritability and genetic advance in the base population but also on the association of yield and yield-related traits with desirable biochemical composition (Said, 2012; Temesgen et al., 2013a; Sabaghnia et al., 2015). Since sufficient variability is available in the Ethiopian garden cress (Said, 2012; Temesgen et al., 2013a,b), initiating a robust breeding program with this economically important but under-utilized crop will substantially contribute to its genetic improvement, cultivation and utilization.

Genetic evaluation of large number of garden cress genotypes related to yield and seed quality related traits such as oil and oleoresin contents is of paramount importance. Measurement of simple correlation coefficient helps to identify the relative contribution of component characters towards yield (Panse, 1957). Simple correlation coefficients are not always effective in determining the real relationships among traits. Path coefficients show direct influence of independent variable upon dependent variable. It specifies the cause and effect relationship and measures the relative importance of each variable; therefore, path coefficient would provide a more meaningful interpretation of such association (Dewey and Lu, 1959; Mondal et al., 2011; Malek et al., 2014). It is, therefore, worthwhile to consider a large number of garden cress genotypes to make observations at several locations

So far, the study of interrelationships among yield and yield related components characterized with small number of accessions and/or in non-ideal environment of single location (Temesgen et al., 2013a; Said, 2012) make characterization inefficient demanding a large number of genotypes tested in two suitable apparent locations enable the genotypes to discharge its full potential. Therefore, the present study was undertaken to determine the degree and nature of associations among seed yield and seed quality related characters.

Experimental procedure and data collection

The field experiment was carried out at two locations, namely Haramaya University Agricultural Research Site at Raare (HUARS) and Kulumsa farmer field (KFF) which is 3.4 km East of Kulumsa Agricultural Research Center, Ethiopia, in 2014/2015 growing season. One hundred eight garden cress accessions collected from different agro-ecological regions of Ethiopia were examined. The experiment was conducted in the Randomized Complete Block Design with two replications. Each experimental plot consisted of two 135 cm long rows with inter-row and between plant spacing of 20 and 15 cm, respectively.  Fertilizers were applied at the rate of 18 kg/ha N and 46 kg/ha P₂O₅ and all necessary cultural practices were undertaken as recommended for the crop. On plant basis, data were recorded from five randomly selected competitive plants of each genotype characters namely plant height (PH), number of primary branches per plant (PB), number of secondary branches per plant (SB), grain yield per plant (GYPP), and biomass yield per plant (BYPP). On plot basis, traits such as days to flowering initiation (FI), days to 50% flowering (FL), days to maturity (DM), thousand seed weight (TSW), grain yield per plot (GYPlt), harvest index (HI), and biomass yield per plot (BMPlt) were recorded. The agronomic characters were taken after harvesting the plants. The oil content of the seed was determined at Holetta Highland Oil Crops Laboratory in Ethiopia (Guenther, 2007) while the oleoresin content  was investigated at Wondo Genet Natural Product Laboratory, Ethiopia (Daniel et al., 2008).

Data analysis

Correlation coefficients

The correlation coefficients were calculated to determine the degree of association of characters with yield and among themselves. Phenotypic and genotypic correlations were computed by using the standard procedure suggested by Johnson et al. (1955), Singh and Chaudhary (1985) and Dabholkar (1992) using the following formula:

r_gxy = [Cov_gxy/( (s²_gxx s²_gy )]

r_pxy=[Cov_pxy/( (s²_pxxs²_py )].

where r_pxy = phenotypic correlation coefficient between characters x and y, Cov_pxy = phenotypic covariance between characters x and y, s²_px = phenotypic variance for character x and s²_py = phenotypic variance for character y, r_gxy = genotypic correlation coefficient between characters x and y, Cov_gxy = genotypic covariance between characters x and y, s²_gx= genotypic variance for character x and s²_gy = genotypic variance for character y. The coefficients of correlation were tested using ‘r’ tabulated value at n - 2 degrees of freedom, at 5 and 1% probability levels, where n is the number of treatments (genotypes) as described by Robertson (1959).

Path analysis

Path analysis was used for exhibiting the direct and indirect effects on seed yield according to the method suggested by Dewey and Lu (1959) using phenotypic as well as genotypic correlation coefficients as:

rij = pij + Æ©rikpkj

where rij = mutual association between the independent character i (yield-related trait) and dependent character, j (grain yield) as measured by the genotypic correlation coefficients, Pij = components of direct effects of the independent character (i) on the dependent character (j) as measured by the path coefficients, and ∑rikpkj =  summation of components of indirect effects of a given independent character (i) on a given dependent character (j) via all other independent characters (k), where: I = any trait in the model, and r = correlation coefficient between any trait i and the dependent variable.

Data processing was done by MS Excel 2007. The contribution of the remaining unknown characters was measured as the residual effect as demonstrated by Singh and Chaudhary (1985). SPAR2.0 (IASRI, 2005) and META-R version 5 (Alvarado et al., 2015) statistical packages were used for analysis.

Phenotypic and genotypic correlations

Estimates of genotypic and phenotypic correlation coefficients at Haramaya and Kulumsa environments are shown in Tables 1 and 2, respectively. The analysis of correlations in each location showed that genotypic correlation coefficients were generally higher than their corresponding phenotypic correlations indicating strong hereditary association among the traits due to genetic factors such as linkage and/or pleiotropic effect enabling consistent performance across wide range of environments as described by Waitt and Levin (1998). Similar findings were observed by Tahira et al. (2011) and Hasan et al. (2015) on the study of Brassica. Grain yield is a very complex character which is governed by polygenes and the result of numerous simple characters presumably showing considerable variations from one environment to another (Singh et al., 2014). The correlation between the traits may be due to linkage or pleiotropy (Allard, 1999) or environment (Aastveit and Aastveit, 1993). Similarly, a narrow difference between phenotypic and genotypic correlation coefficient was obtained for almost all the pairs of characters studied showing that masking or modifying effects of the environment was little demonstrating that the presence of an inherent association among these characters. Similar results were reported in barley (Azeb et al., 2016).

Genotypic path analysis

In the present study, grain yield  per  plot  (GYPLt) was the dependent variable while other evaluated traits were considered as independent variables. Direct and indirect effects of these components were determined on grain yield and their contributions in each of the two environments are shown in Table 3. The path coefficient analysis at genotypic level revealed at HRS and KFF, respectively that biomass per plant (1.143 and 0.996) had the highest positive direct effect on yield followed by harvest index (0.694 and 0.274) and grain yield per plant (0.171 and 0.134). Similarly, harvest index and grain yield per plant had highly significant and positive correlations with grain yield in each location at genotypic level, indicating the need for direct selection for these traits in order to improve grain yield of garden cress.  At   HRS,   the   highly    significant    direct correlation of grain yield with days to flowering initiation (0.126), with plant height (0.274) and with number of secondary branches (0.182) (Table 3) shows the usefulness of selecting of these traits for grain yield improvement. Similar findings were reported by Temesgen et al. (2013a) on their study with the Ethiopian garden cress. However, Hasan et al. (2015) indicated that plant height had negative direct effect while grain yield per plant had strong positive and direct effect on grain yield of Brassica napus at genotypic level indicating correlation of traits varied depending on the environment and nature of the crop.

It was observed that plant height, days to maturity and oil content exhibited positive and relatively low direct effect on yield at KFF (Table 3).  These  traits  except  for oil content had highly significant positive correlation with grain yield in an earlier study (Temesgen et al., 2013a). On the contrary, the number of primary branches, biomass per plot, thousand seed weight, and oleoresin content were negative and low direct effect for grain yield in both  locations.  Days to 50% flowering and days to maturity at HRS and days to flowering initiation and number of secondary branches at KFF had negative and low direct effect to grain yield but positive and significant correlation with grain yield except for  oleoresin content which was  non-significant (Table 3). These implied that almost all of these traits might probably contribute for grain yield through indirect effects requiring indirect selection. Halder et al. (2016) demonstrated similar high positive direct effect of plant height, thousand seed weight, and days to 50% flowering on grain yield/ha. These results are also in agreement with Marjanovic Jeromela et al. (2008) and Halder et al. (2016) who reported that plant height exerted the highest positive significant direct effect on grain yield. In the study of rapeseed, Uddin et al. (2013) and Ara et al. (2013) also reported significant direct effect from primary branches per plant and number of secondary branches per plant on grain yield.

Traits such as days to flowering initiation, days to 50% flowering, plant height, number of primary and secondary branches, days to maturity, biomass per plot, grain yield per plant and thousand seed weight exerted indirect effects on grain yield via biomass per plant in both environments (Table 3). This demonstrated that the relationship between these traits and grain yield in the two environments were predominantly due to indirect effects. Harvest index showed relatively higher negative indirect effect to grain yield through biomass per plant. According to Temesgen et al. (2013a), all investigated traits showed direct effects to grain yield except for days to 50% flowering and plant height which revealed indirect effect. This was probably due to the type of genotypes used in the study and environmental factors that could also substantially contributed for the deviation of the result.

Generally, the direct and indirect impact of traits in path analysis in each location depicted considerable variation. This was presumably due to significant differential performance across location which resulted in substantial changes in the direction and magnitude of genotypic correlations (Table 3). Accordingly, the grain yield per plot of garden cress can be increased by selecting genotypes having higher biomass per plant, harvest index and grain yield per plant at both environments and higher plant height and number of secondary branches at HRS. This is in agreement with earlier work on lentil by Hegazy et al. (2012). As shown in Table 3, most traits at both locations exerted their indirect effect to increased grain yield through biomass per plant. Similar findings were reported from B. juncea (Hasan et al., 2015).

Path coefficient analysis revealed that biomass per plant, harvest index, and grain yield per plant had high positive direct effects on yield at both environments and plant height, flowering initiation and number of secondary branches showed direct and significant contributions to grain yield at HRS unlike at KFF. A higher positive indirect contribution by most of the yield components via biomass per plant was responsible for highly significant genotypic and phenotypic correlation coefficients. It could be said that biomass per plant, harvest index, and grain yield per plant were consistent and the most important traits affecting the yield  per  plot  of  garden cress.  Thus, due attention must be given to these traits in the selection program. Along with these, days to flowering initiation, days to 50% flowering, plant height, number of primary and secondary branches, days to maturity, biomass per plot, and thousand seed weight contributed indirectly to grain yield at both locations. Therefore, including these traits directly or indirectly in the selection criteria is important towards developing high grain yielding garden cress genotypes.

In the present study, residual effect at HRS and KFF were 0.098 and 0.072, respectively (Table 3). This means, characters in the path analysis expressed in grain yield contributing traits by 90.2% at HRS and 93.8% at KFF, while the remaining needs additional characterization for the future breeding program.

The genotypic correlation coefficient was generally higher than the corresponding phenotypic correlation coefficient. Genotypic and phenotypic correlations between grain yield and majority of investigated traits were highly significant at each location except oil and oleoresin contents. Plant height, days to flowering initiation and number of secondary branches had high direct contribution to the grain yield at HRS unlike at KFF. Genotypic path coefficient analysis revealed that biomass per plant, grain yield per plant, and harvest index had relatively higher positive direct effect on grain yield in each location. In general, this study has clearly indicated the need for focusing on biomass per plant, grain yield per plant, and harvest index traits towards improving grain yield of garden cress.

The authors have not declared any conflict of interests.