Assessment and phenotypic identification of millet germplasm (Setaria italica L.) in Liaoning, China

Experimental materials

A total of 105 millet germplasm resources preserved in the Germplasm Bank of the Liaoning Academy of Agricultural Sciences (LAAS) were selected for this study. These resources originated from various regions, including Liaoning (52), Beijing (20), Hebei (eight), Jilin (nine), Inner Mongolia (nine), Heilongjiang (two), and Shanxi (five). Approximately half of the materials are from Liaoning, with the remaining materials sourced from area that north of the Yellow River.

Experimental design

The experiment was conducted over three growing seasons, from 2018 to 2020, at the experimental site of the Cash Crops Institute of Liaoning (Liaoyang City, Liaoning Province, China). The soil at the site is sandy loam, with the following composition in the topsoil: 1.97 × 10³ mg/kg organic matter, 800 mg/kg total nitrogen, 73.4 mg/kg alkali-hydrolyzable nitrogen, 23.6 mg/kg available phosphorus, and 247.5 mg/kg available potassium. The materials were arranged sequentially during sowing, with each material planted in four rows with 3–4 cm plant spacing. The length of each row was 5 m, with a spacing of 50 cm, resulting in a plot area of 10 square meters. Sowing took place around May 11^th, and harvesting was performed around September 25^th. A three-compound fertilizer (N:P:K = 15%:15%:15%) was applied as a base fertilizer. Ploughing and weeding were conducted three times during the growth period.

Two points were selected from each plot, and five consecutive plants with uniform growth situation were chosen from each point as samples. Qualitative and quantitative traits were investigated following the Descriptors and Data Standard for Millet (S. italica L.) compiled by Lu (2006). Among the traits, 12 quantitative traits and 19 qualitative traits that shown difference across resources. were selected, ignoring traits without difference.

Data processing

The average and interannual values of the 10 plants were calculated with EXCEL 2007, then analyzed further. Phenotypic diversity analysis and systematic cluster analysis (SCA)were conducted based on all 31 trait, correlation analysis, principal component analysis (PCA), gray relational analysis (GRA), and K-means cluster analysis were performed based on the 12 quantitative traits (Tang, 2010). Shannon diversity index of 19 qualitative traits was calculated using the method described by Wang et al. (2021).

Phenotypic diversity analysis of millet germplasm resources

Table 1 shows the Shannon diversity index (DI) and distribution frequency of phenotypic characters of 19 quality traits. By comparing the frequency, it was found that tiller habit and branch habit showed weak, moderate, and strong characteristics with a uniform distribution, resulting in a greater diversity index (DI). The phenotypic distribution of the other 17 traits showed clear tendencies, with focused on 1 or 2 characteristics, resulting in a lower DI. It is worth noting that traits related to pigmentation show obvious distribution tendencies, with traits related to leaves mainly being green and traits related to flowers primarily being yellow.

The phenotypic diversity analysis of 12 quantitative traits is presented in Table 2. The coefficient of variation (CV), which represents the degree of trait dispersion, ranged from 6.34% to 43.77% for the 12 traits. The highest CV was observed for stem number per plant, followed by main stem node number, seed weight per plant, and ear weight per plant, indicating a high degree of variation and potential for genetic improvement. On the other hand, the subear internode length and growing period had CVs close to or less than 10%, suggesting lower dispersion and relatively stable performance among the varieties. The Shannon DI, which reflects the distribution of trait performance, ranged from 1.0566 to 2.0428. The main stem length, ear weight per plant, seed weight per plant, and grass weight per plant had DI values close to or greater than 2.0, indicating that these traits performance are scattered and susceptible to external conditions. Conversely, Stem number per plant had the lowest DI value, close to 1.0, indicating that these traits performance are stable, reflecting a simple genetic basis.

Correlation analysis of millet germplasm resources

Correlation analysis was performed based on the 12 quantitative traits (Table 3). The ear weight per plant showed a significant positive correlation with seed weight per plant, and both two traits were positve correlatad with the main ear length significantly; Main ear diameter and stem number per plant also showed a highly significant negative correlation with main stem nod number. Ear weight per plant had a significant positive correlation with main ear length and growing period, and seed weight per plant showed significant positive correlations with these three traits. These results indicate that grain yield is positively correlated with ear size (main stem length and main stem diameter) and ear setting potential (growing period and stem number per plant). Additionally, grass weight per plant was positively correlated with main stem nod number and main stem diamete, these two traits represent vegetative growth. Growing period, which represents the growth potential of the plant, showed positive correlations with stem number per plant, main stem length, main ear length, peduncle length, and main ear diameter.

Principal component analysis of quantitative traits

The PCA results for 12 quantitative traits from 105 millet germplasms are presented in Table 4. Seven principal components with eigenvalues greater than or close to one were selected, which collectively contributed to 81.63% of the variance and encapsulated most of the genetic information of the millet germplasms.

The eigenvector value of each trait indicates its contribution to the principal component (PC). By comparing the eigenvector values, the following observations were made: the 1^st PC was primarily associated with main ear length (−), main stem length (+), seed weight per plant (+), grass weight per plant (−), and growing period (−); the 2^nd PC was mainly influenced by main ear diamete (+), ear weight per plant (+), and main stem nod number (+); the 3^rd PC was primarily influenced by stem number per plant (+), 1,000-seed weight (−), and Ear weight per plant (+); the 4^{th PC} was mainly influenced by MED (−), GP (+), GWP (+) and EWP (−); the 5^th PC was mainly influenced by main stem diameter (+), peduncle length (−), main stem nod number (+), main ear length (+) and seed weight per plant (+); the 6^th PC was primarily influenced by main stem length (−), peduncle length (+), seed weight per plant (+), and ear weight per plant (+); and the 7^th PC was mainly influenced by main ear length (+) and main stem nod number (−). Compared to correlation analysis, PCA provided an interpretation of the correlation between quantitative traits and specifically emphasized the importance of the GP and TSW.

Gray relational analysis of quantitative traits and yield

After standardized processing, the seed weight per plant was taken as the reference column, and the correlation between the main quantitative traits and the seed weight per plant was analysed (Table 5). Among the tested quantitative traits, ear weight per plant had the greatest impact on the seed weight per plant, followed by stem number per plant, main stem length, and main ear diameter, these four traits contribute most to the ear-bearing capacity of the plant. The coefficients of growing period, main stem length, and peduncle length were very close, these three traits are mainly related to maturity and main stem growth. The main stem diameter, grass weight per plant, and main stem nod number represent the vegetative growth status of the plant, and their contributions to the seed weight per plant decrease in turn.

System cluster analysis of all traits

Based on the performance of the 31 studied traits (12 quantitative traits + 19 quality traits) of 105 millet resources, system cluster analysis (SCA) analysis was conducted via standardized data transformation-Euclidean distance-deviation square sum method. All materials were divided into six groups at Euclidean distance = 17.09 (Fig. 1).

The 19 quantitative traits for each group were analyzed and compared (Table 6). The 1^st group consisted of 11 samples, which exhibited mid-earlier maturity, fewer tillers, taller and slender main stem with multiple nodes, smaller main ears, small seeds with medium grain yield, and higher grass yield. The 2^nd group included 30 materials, showing earlier maturation, fewer tillers, lower and slenderer main stem with multiple nodes, smaller main ears, larger seeds but lower grain yield, and higher grass yield. The 3^rd group consisted of 33 materials displaying mid-later maturation, medium tiller ability, taller main stem with multiple nodes, medium-sized main ears, larger seeds with higher grain yield, and medium grass yield. The 4^th group comprised six materials, showing later maturation, multiple tillering, taller and stronger main stems with multiple nodes, larger main ears, smaller seeds but higher grain yield, and greater grass yield. The 5^th group contained five materials, exhibiting mid-later maturation, multiple tillering, lower and stronger main stem with fewer nodes, shorter and thicker main ears, medium-sized grains, lower grain yield and lower grass yield. The 6^th group consisted of 20 materials that exhibited mid-later maturation, multiple tillering, taller and slenderer main stems, smaller main ears, smaller seeds but higher grain yields, and lower grass yields.

The 12 quantitative traits were analyzed and compared too (Table 6). The qualities of each group were also compared. The 1^st group had unique characteristics in terms of leaf sheath color and bristle length that were not found in the other groups. The 2^nd group displayed common characteristics for all traits. The 3^rd group exhibited abundant traits characteristics, including red seed color, which was not present in the other groups. The 4^th group, with its six resources, had relatively simple and concentrated trait phenotypes, but also had a rare occurrence of purple bristle color. The 5^th group, consisting of only five resources, had scattered trait phenotypes, with kernel color, seed coor, and shattering habit showing unique characteristics. The 6^th group showed abundant trait characteristics, with lodging resistance and ear shape displaying characteristics not found in the other groups.

Further examination of the geographical distribution of resources within each group revealed that resources from Liaoning had certain advantages, particularly in Group 4, which exclusively consisted of materials from Liaoning. Resources from Group 1 were also concentrated in Liaoning. Group 2 resources were relatively dispersed and covered almost all geographical origins. Groups 3 and 6 mainly sourced their resources from Liaoning, Beijing, and Inner Mongolia. Group 5 resources were evenly distributed from Liaoning and Jilin.

K-mean clustering analysis

K-means clustering analysis was conducted to analyse the 12 quantitative traits, and the theoretical distribution intervals and central values of these traits are shown in Table 7. The performance of each trait for the studied resources was found to be concentrated near the central value, with more distribution around the minimum value and less around the maximum value.

Based on these results, specific resource materials with desirable performance in terms of maturity, plant height, tiller habit, ear size, and grain weight were identified. After a comprehensive evaluation, four specific resources were selected: Dungu No. 1, an early-maturing and draft small-ear and small-seed resource from Taiyuan, Shanxi Province; Xiao-li-xiang, a small-seed and early-maturing resource from Shijiazhuang, Hebei Province; Basen Shengu, a later-maturing and taller large-ear resource from Fuxin, Liaoning Province; and Yuhuanggu No. 1, a later-maturing and low-yield large-seed resource from Chifeng, Inner Mongolia.