Morphological diversity and correlation analysis of phenotypes and quality traits of proso millet (Panicum miliaceum L.) core collections

ZHANG Da-zhong , Rabia Begum Panhwar, LlU Jia-jia, GONG Xiang-wei, LlANG Ji-bao , LlU Minxuan, LU Ping , GAO Xiao-li, FENG Bai-li

1 College of Agronomy, Northwest A&F University, Yangling 712100, P.R.China

2 Agricultural Technology Promotion Center of Shenmu City, Shenmu 719000, P.R.China

3 Institute of Crop Sciences, Chinese Academy of Agricultural Sciences, Beijing 100081, P.R.China

Abstract Genetic diversity and comprehensive performance are the basis for the discovery and eff icient use of proso millet (Panicum miliaceum L.) core collections. In this study, 386 proso millet core collections were used as materials to observe inf lorescence color, leaf phase, inf lorescence density, axis shape, branched spike length, panicle type, trichome, measured area of the top3 leaves, and chlorophyll content of the top3 leaves at f illing stage. These core collections were also used to record growth period, plant height, diameter of main stem, plant tiller number, branch number, panicle length, panicle number per plant, and panicle weight per plant at the maturation stage. Starch, fat, protein, and yellow pigment contents in the grain and 1 000-seed weight were also measured after harvest. Then, quantitative traits were used for diversity analysis and comprehensive evaluation of each collection. Correlations between all traits were also analyzed. Results showed that among the 8 quality traits, the Shannon index (H′) of hull color was the highest (1.588) followed by the H′ of inf lorescence density (0.984). However, inf lorescence color and axis shape were lower. The H′ of 16 quantitative traits were signif icantly higher than the quality traits with the following traits having the highest indices: fat content (2.092), 1 000-seed weight (2.073), top3 leaves area (2.070), main stem diameter (2.056), and plant height (2.052). Furthermore, all other traits had a diversity higher than 1.900. After a comprehensive evaluation of phenotypic traits, plant height, diameter of main stem, plant tiller number, leaf area of top3 leaves, and 1 000-seed weight were the biggest contributors to the principal components. Six high-fat and high-protein cultivars, including Nuoshu, A75-2, Zhiduoaosizhi, Panlonghuangmi, Xiaobaishu, and Xiaohongshu were also screened. Correlations between the quantitative traits were signif icant, including the correlation between quality traits and quantitative traits. In conclusion, the core collections can be used as basis for discriminating among proso millet cultivars based on related traits and for further studies on millet with rich genetic diversity, good representation, and signif icant collection between traits.

Keywords: proso millet, phenotypic traits, quality, diversity, correlation, evaluation

1. lntroduction

Proso millet is one of the earliest cultivated crops in China with a short seed-to-seed cycle, and low requirements for water and fertilizer. It plays an important role in the Chinese diet and culture (Magness et al. 1971; Chen 1986). The cultivated acreage of proso millet is about 600 000 ha in China and is mainly distributed in the area along the Great Wall (Lin et al. 2002), indicating that proso millet is an important crop in arid and semiarid areas. Extensive geographical distribution and long-term domestication have led to a rich and varied germplasm resource of proso millet (Wang R Y et al. 2016). Currently, China, Russia, Ukraine, and other countries have collected and conserved more than 24 000 copies of proso millet germplasms with more than 8 700 copies being collected and preserved by China (Chai and Feng 2012). Research and comprehensive evaluation of agronomic and quality traits, richness, and diversity of genes, and correlations between these traits of the germplasms are not only conducive to conservation and further research of high quality germplasms but are also important for its genetic improvement. In the process of core collection and diversity assessment, Hu et al. (2012) evaluated rice core germplasms with genetic diversity index, standard deviation, frequency distribution, and variation coeff icient of phenotype. Liu et al. (2006) analyzed the diversity of mungbean (Vigna radiata) resources by calculating the distribution frequency of quality traits and the maximum, minimum, and mean values, standard deviation, and variation coeff icient of the quantitative traits. Hu et al. (2008) also used a similar method in proso millet.

In recent years, the study of proso millet core collections has gradually increased, with studies mainly focused on phenotypic traits (Upadhyaya et al. 2011; Dikshit and Sivaraj 2013), quality traits (Wen et al. 2014; Wang et al. 2017), drought and salt resistance (Wang et al. 2007, 2015; Liu et al. 2015; He et al. 2016), and lodging, disease, and insect resistance (Wang et al. 2008; Wang L 2016). Studies at the molecular level increased rapidly (Hu et al. 2009; Hunt et al. 2011). Wang L et al. (2005), Upadhyaya et al. (2011), and Dikshit and Sivaraj (2013) analyzed the phenotypic and quality traits of core collections including 7-8 agronomic traits, such as growth period and plant height, which contribute greatly to variation, and then screened a number of high-quality resources. Wen et al. (2014) and Wang et al. (2017) measured protein, fat, and lysine contents in grains and other quality traits of more than 6 000 resources. Afterwards, they screened out 342 high-quality copies. Wang (2015) and He (2016) identif ied drought resistance of more than 500 resources and screened out 11 Grade I drought-resistant copies. Wang et al. (2007) and Liu et al. (2015) identif ied salt tolerance of over 6 000 proso millet resources and screened 22 salt-tolerant and 3 phosphite salt-tolerant copies. Wang L et al. (2016) screened out 71 resources with lodging-resistance from 1 192 varieties. Wang et al. (2008) identif ied resistance to head smut of more than 6 000 copies by investigating incidence and determining activity of defensive enzymes. Hu et al. (2009), Harriet et al. (2011), and Rajput and Santra (2016) analyzed the genetic diversity of proso millet using SSR markers.

In summary, there are many reports on phenotypic diversity of proso millet core collections, but few studies on quality traits, especially on the relationships among the traits. In this study, 386 core germplasms were used to analyze the genetic diversity and the correlations between the traits by analyzing diversity of phenotypic traits, principal component analysis, and comprehensive evaluation of phenotypes, which is of great importance for the innovation, use, and genetic improvement of the core germplasms.

2. Materials and methods

2.1. Germplasms and experimental condition

A total of 386 core collections (Appendix A) were used as materials, including landraces and cultivars, of which 377 germplasms came from China and 9 germplasms came from foreign countries. All germplasms were provided by the Institute of Crop Sciences, Chinese Academy of Agricultural Sciences who constructed a primary core germplasm using the morphological indicator data recorded in the proso millet database and then used SSR molecular markers to analyze and construct the proso millet core collection. The study was conducted in Zhaojiagou Dryland Agricultural Demonstration Garden (110.53°E, 38.76°N), Shenmu City, Shaanxi Province, China. The plot size was 12 m2(4 m×3 m) and all plants were sown on June 8, 2016.

2.2. Determination of phenotypic traits

During each growth stage, the growing period, inf lorescence color, leaf phase, inf lorescence density, axis shape, branched spike length, panicle type, and trichome were recorded. At the f illing stage, the top3 leaves area (leaf length×leaf width×0.75) was measured with a ruler, and the chlorophyll contents were measured with a chlorophyll meter (SPAD-502Plus, Japan). Five plants were randomly selected from each variety to measure plant height, diameter of the main stem, plant tiller number, branch number, panicle length, panicle number per plant, and panicle weight per plant at maturity stage. After harvesting, seed color was observed, and 1 000-seed weight was measured. The quality traits were described according to the Descriptors and Data Standards for Broomcorn Millet (Panicum miliaceum L.) (Wang and Wang 2006).

2.3. Determination of quality traits

Fat content determination was conducted based on GB/T 5512-2008 (2008). Starch content was determined using the dual wavelength method with a UV-visible spectrophotometer (Blue Star B, LabTech, China). Protein content was assayed by the Kjeldahl method with an automatic Kjeldahl apparatus (K9860, Hanon, China). Flavochrome content determination was conducted based on the ApproVed Methods of the American Association of Cereal Chemists (AACC) (Hentschel et al. 2002).

2.4. Data analysis

Excel 2007 and SPSS 22.0 were used to analyze data and cartography.

3. Results

3.1. Analysis of genetic diversity

Core germplasm resources of proso millet showed abundant genetic diversity in phenotypic (Fig. 1) and grain traits (Fig. 2). From the perspective of ear type, the f irst 6 of the 16 genotypes shown in Fig. 1 are the scattered type, the 5 in the middle are the compact type, and the last 5 are the lateral type. The resources have numerous, clear differences in plant height, spike type, stem diameter, tiller number, and branch number. The grains also show rich diversity in terms of color, size, and shape (Fig. 2).

Table 1 lists the Shannon indices (H′) of 24 major traits of proso millet resources, including 8 quality traits and 16 quantitative traits. The highest H′ of quality traits was for hull color (1.588), followed by inf lorescence density (0.984); those of inf lorescence color (0.590) and axis shape (0.580) were lower. H′ of 16 quantitative traits were signif icantly higher than quality traits, of which the highest were fat content (2.092), 1 000-seed weight (2.073), top3 leaves area (2.070), main stem diameter (2.056), plant height (2.052), and starch content (2.053).

Diversity of quality traitsThere are signif icant differences in the frequency distribution of different phenotypic and quality traits (Fig. 3). For the 386 genotypes, hull color is mostly yellow (30.3%) and red (24.9%). In terms of inf lorescence density, the osculant type amounts to 62.2% of the total amounts, while the compact type is only 1.8%. As for the leaf phase, the proportion of drooping type, osculant type, and up turned type is 68.4, 24.9, and 6.7%, respectively. Majority of the corn collections have a bent spike principal axis (78.2%) and few are erect (1.6%). The branched spike length is mainly long type (51.6%) and osculant type (40.7%), while the short type was only 7.8%. Osculant type (48.7%) and dense type (47.7%) are main types of trichome. The lateral panicle accounts for a vast majority (76.9%), but the scattered and compact type has a proportion of only 19.4 and 3.6%, respectively. About 3/4 of the corn collections are green inf lorescence, thus 1/4 are purple ones.

Diversity of quantitative traitsAs shown in Table 2,variation amplitude of growing period was 51 to 107 days with an average of 85.7 days; variation coeff icient was 13.51%, and the shortest was only 47% of the longest growing period. Variation amplitude of plant height was 63.6 to 174.9 cm, and 80% of the height among the corn collection was between 102.5 to 154.8 cm. Variation coeff icient of stem diameter was 14.62%, and the variation amplitude was 2.5 to 8.8 mm with an average of 6.3 mm. Stem node numbers were 3.7 to 9.1 with an average of 7.5, and most germplasms are 6-8 knots. As the variation amplitude was from 0 to 6.8, variation coeff icient of tiller number is higher (43.35%) with an average of 2.33. Variation coeff icient of branch number is the highest (69.23%) with an average of 16.3, while some corn collections had no branch. The variation coeff icient of panicle number per plant was also high (46.12%), where the spike length range was from 12.8 to 48.5 cm with an average of 34.78 cm. Thus, the longest ones are nearly 4 times the shortest ones. The variation coeff icient of panicle weight per plant was 45.97%, and the variation amplitude was 4.43 to 96.38 g with a difference of nearly 22 times between the extremums. 1 000-seed weight was between 3.96 to 9.71 g, and the most was between 5.7 and 8.6 g. In terms of the top3 leaf area, the biggest (317.76 cm2) was nearly 8 times the smallest (39.94 cm2). The top3 leaf chlorophyll content variation was only 7.41%. As for quality traits, variation coeff icient of starch content was the smallest (only 1.83%), but the variation coeff icient was relatively high (24.28%). Content of fat ranged from 1.10 to 4.70%, and the variation amplitude was between 10.20 to 16.10%. Flavochrome content ranged from 5.05 to 18.86 mg kg-1with an average of 13.78 mg kg-1.

Fig. 1 The rich agronomic characters of proso millet germplasm resources. Scattered type: (1) Maimizi, (2) Huami, (3) Bendimizi, (4) Minlehongmizi, (5) 6, (6) Guanghehuangmi. Compact type: (7) Bamenghuangshuzi, (8) 2048, (9) Gaoliangshu, (10) 2096, (11) Mi. Lateral type: (12) Bairuanshu, (13) Xiaomaimizi, (14) Heiruanmi, (15) 790051 (India), (16) Yumi 2.

Fig. 2 The rich grain color of proso millet germplasm resources.

Table 1 The Shannon index (H′) of proso millet germplasm resources

Fig. 3 The distribution frequency of each phenotype in quality traits. Hull color: a, gray; b, multiple; c, white; d, brown; e, red; f, yellow. Inf lorescence density: a, dense; b, sparse; c, slightly dense; d, osculant. Leaf phase: a, upturned; b, osculant; c, drooping. Axis shape: a, erect; b, slightly curved; c, bent. Panicle branch length: a, short; b, osculant; c, long. Trichome: a, sparse; b, dense; c, osculant. Panicle type: a, compact type; b, scattered type; c, lateral type. Inf lorescence color: a, purple; b, green.

3.2. Phenotype evaluation

Princ ip al c omp onent analysis and evaluation of phenotypic traitsPrincipal component analysis was used to evaluate the genotypes of phenotypic traits. Results (Table 3) show that the cumulative contribution rate of the f irst 4 principal components is 81.26%, indicating that these principal components contain the most genetic information of the phenotypic traits and can be used for comprehensive evaluation of core collections. The contribution rate of the f irst principal component was 37.52%, and the power vector of plant height, top3 leaf area, and stem diameter were greater than for other characters, indicating that the f irst principal component consisted of these 3 characters. Contribution rate of the second principal component was 18.81%, and the power vectors of stem node number and 1 000-grain weight were the largest. Moreover, the power vector of the panicle number per plant was the highest in the third principal component, showing that the third principal component was a comprehensive response of panicle number. Contribution of the 4th principal component was 10.69%, of which the panicle length was signif icantly higher than the other coeff icients. This means that panicle length had the largest contribution to the 4th principal component.The composite score (F) was calculated for each germplasm based on standardized values of 10 phenotypic traits, normalized scores of the 4 principal components, and the calculated principal component weighting coeff icients (0.461, 0.232, 0.175, 0.132). Higher F values indicate more comprehensive phenotypes. Among the 386 germplasms, the best 10 germplasms came from Shaanxi Province, including Hongruanmi, Bairuanmi, Hongyingmi, Hongmizi, Huangmizi (shu), Huangmizi, Dahongpao, Zigaitouhuangmi, Heiruanmi (shu), and Baiyingmi. The worst 10 were from Heilongjiang and Inner Mongolia, including Balinzuogedashu, Xiaobaishu, Huangmizi, 2096, Anchunwei, Yimengliang 56-2, Baimizi, Jan-55, Eheitou, and Maimizi.

Evaluation of quality traitsAmong 386 germplasms,217 were non-waxy germplasms and 169 were waxy germplasms. The content of amylose was between 1.57 and 45.61%, of which Neimi 5 was the highest and Zhengninghongnian was the lowest. The average content of grain protein was 12.3%, and amplitude was 10.2 to 16.1%, of which 19 germplasms are above 14%, including Noushu, Helanerhuang, 2048, Maimizi, Heimizi, Bamenghuangshuzi, 6, A75-2, Anchunwei, Aoduoaosizhi, Dahuangmi, Linghehongnianmi, 15, Panlonghuangmi, Xiaobaishu, Yemizi, Xiaohongshu, Hongmizi, and Nianfeng 5. The amplitude of fat content was 1.1 to 4.7% with an average content of 3.1%, and 61 genotypes were higher than 4%. Above all, there were 6 germplasms with both high protein (more than 14%) and high fat content (more than 4%), including Nuoshu, A75-2, Aoduoaosizhi, Panlonghuangmi, Xiaobaishu, and Xiaohongshu. Average of f lavochrome content was 13.78 mg kg-1, and the amplitude was 5.05 to 18.68 mg kg-1. Hejianbaishuzi, Liaomi 56, Gumi 21, Hongruanmi, Bairuanmi, and Xiaotoumi had a content of f lavochrome above 18 mg kg-1.

Table 2 The variation distribution of quantitative traits

Table 3 Power vector (PV), eigenvalues (E), contribution rate (CR), and cumulative contribution rate (CCR) of the f irst 4 principal components based on 10 phenotypic traits

3.3. Correlations between traits

Table 4 shows the correlations between the 5 major and 13 representative quantitative traits. The ways to describe the correlation of quality and quantitative traits are different, so 3 categories were created: correlation between quantitative traits, correlation between quality traits, and correlation between quality and quantitative traits. Correlation between the quality traits was not signif icant, except that leaf phase and spike branch length were signif icantly correlated. Therefore, the following sections only describe the correlations between quantitative traits and correlations between quality and quantitative traits.

Correlation between quantitative traitsAs seen in Table 4, growing period and plant height, stem node number, panicle length, panicle weight per plant, 1 000-seed weight, top3 leaves area, grain protein, and starch content had highly signif icant correlations. Plant height, stem node number, panicle length, branch length, panicle weight, 1 000-seed weight, and top3 leaves area were all positively correlated with growing period, while grain protein content and number of panicle per plant were negatively correlated with growing period. Correlations between top3 leaves chlorophyll, fat, and f lavochrome contents were not signif icant. Plant height had signif icantly positive correlations with stem node number, panicle length, panicle weight, top3 leaves area, and 1 000-seed weight. Remarkably, plant height had a negative correlation with panicle number and grain protein content, and no signif icant correlation with top3 leaves chlorophyll, fat, starch, or f lavochrome contents. There were clear, negative correlations between stem node and panicle number per plant and top3 leaves chlorophyll and protein content. A signif icant positive correlation between panicle weight and top3 leaves area was also seen. There was a positive correlation between 1 000-seed weight and starch content, but it was not signif icant. Spike length had signif icantly positive correlations with panicle weight, top3 leaves area, and 1 000-seed weight but was not correlated with panicle number, top3 leaves chlorophyll content, or quality traits. In terms of panicle number per plant, remarkable positive correlations were seen with panicle weight, top3 leaves chlorophyll, and protein content, all of which were signif icant, and there was a clear and negative correlation with top3 leaves area and 1 000-seed weight. Panicle weight per plant was signif icantly and positively correlated with top3 leaves chlorophyll content, top3 leaves area, starch content, and 1 000-seed weight, but negatively correlated with grain protein content. Chlorophyll content was signif icantly and positively correlated with top3 leaves area and grain protein content, and signif icantly negatively correlated with starch content. Compared to protein content, starch content and 1 000-seed weight were signif icantly positively correlated with top3 leaves area. A signif icant positive correlation between grain fat and protein content was also seen, and both had obvious negative correlations with other quality traits. Starch content was signif icantly and positively correlated with 1 000-seed weight and f lavochrome content. Flavochrome content and 1 000-seed weight also had a signif icant positive correlation.

Correlation between quality traits and quantitative traitsAs shown in Table 4, 4 quality traits, including leaf phase, hull color, inf lorescence color, and panicle type, had highly signif icant correlations with many quantitative traits. The following is the detailed analysis of correlations between the quality and quantitative traits.

Different frequency distributions of the quantitative traits were seen for 3 leaf phase genotypes (Fig. 4). As shown in Fig. 4-A, the growth period of the drooping type was longer than that of the osculant and upturned types with the upturned type having the shortest growth period with an average of 76.5 days, shorter than the dropping type by 12.2 days. Top3 leaves area of the dropping type was 30% higher than that of the upturned type with an average of 205 cm2(Fig. 4-B). Grain protein and top3 leaves chlorophyll content of drooping and osculant genotypes were higher than that of the upturned type (Fig. 4-C and D). The dropping germplasms have the highest 1 000-seed weight and f lavochrome content with 1 000-seed weight being 20% higher than in the upturned type (Fig. 4-E and F).

The red genotypes had the highest 1 000-seed weight (Fig. 5) with an average of 7.61 g and amplitude of 4.64 to 9.50 g, followed by the gray type between 4.75 and 8.83 g with an average of 7.42 g. Other genotypes are as follows: brown (amplitude of 4.80 to 8.94 g with an average of 7.11 g), yellow, and multi colored (amplitude of 4.31 to 9.52 g with an average of 6.88 g). As shown in Fig. 5-B, f lavochrome content was also the highest in red seeds with an average of 14.5 mg kg-1, followed by gray, brown, yellow, and multi colored seed, while in white seed the f lavochrome content was the lowest (13.1 mg kg-1).

Fig. 6 shows that inf lorescence color was signif icantly correlated with the growing period and f lavochrome content. The purple germplasms had a long growing period with an average of 89.1 days and the green germplasms had an average of 84.4 days. The green germplasms had an average f lavochrome content of 13.95 mg kg-1, which was higher than the purple germplasms (13.34 mg kg-1).

As shown in Fig. 7-A, lateral germplasms had a longer growth period with an average of 87 days and amplitude of 60 to 107 days. Compact and scattered types had the same average growing period of 81 days but with different amplitudes, ranging from 68 to 100 days and 51 to 101 days, respectively. The lateral type had the largest top3 leaves area (Fig. 7-B) with an average of 200 and amplitude of 102 to 317 cm2, followed by the compact and scattered types with average values of 175 and 167 cm2and amplitudes of 40 to 295 cm2and 118 to 236 cm2, respectively. The 1 000-seed weight of lateral type germplasms had an amplitude range of 3.96 to 9.71 g with an average of 7.28 g, that was 14.3 and 30.7% higher than scattered and compact germplasms, respectively (Fig. 7-D). Grain protein content of compact and scattered germplasms were both about 12.7%, which was signif icantly higher than that of lateral germplasms at 12.2% (Fig. 7-D).

Table 4 Correlation coefficients of traits

Fig. 4 Frequency distribution of quantitative traits with different leaf phase genotypes. a, drooping type; b, osculant type; c, upturned type.

4. Discussion

4.1. Diversity and evaluation of proso millet core collections

Fig. 5 1 000-seed weight and f lavochrome content in different hull color genotypes. a, red; b, yellow; c, white; d, brown; e, gray; f, multiple.

Fig. 6 Growing period and f lavochrome content of different inf lorescence color genotypes. a, green; b, purple.

Results suggest that H′ of hull color in quality traits was the highest (1.588), and the predominant color was yellow and red (55.2%), which is similar to the statistical results of 8 515 genotypes of proso millet resources (52.6%). Moreover, the distributions of each type of panicle and inf lorescence color are also similar to a previous study (Wang et al. 2005). H′ of the quantitative traits of core collections were generally higher compared to the study of Hu et al. (2008). The H′ of 1 000-seed weight and panicle weight per plant were 2.073 and 1.978 with variation coeff icients of 17.90 and 45.97, respectively. H′ of grain protein and fat content were 2.028 and 2.092 with ranges of 10.2 to 16.1% and 1.1 to 4.7%, respectively. Variation coeff icients and amplitudes of other traits were also high, indicating that the core collections had abundant variation and can accurately represent the genetic diversity of the proso millet resources as required by core germplasm construction (Upadhyaya et al. 2011) and genetic improvement of proso millet. At present, F-value analysis based on normalization of phenotypic data and principal component scores are mostly applied to the comprehensive evaluation of phenotypic traits of core collections (Dong et al. 2015). This method was also used in this study, and the 10 germplasms with the highest F-values came from Shaanxi Province, including Hongruanmi, Bairuanmi, Hongyingmi, Hongmizi, Huangmizi (shu), Huangmizi, Dahongpao, Zigaitouhuangmi, Heiruanmi (shu), and Baiyingmi. In addition, other resources with high F should also be attractive, for example, Zhiduoaosizhi from Russia that had good phenotypic and quality traits. Although the protein content of Hongruanmi and Hongyingmi is generally normal, fat content and comprehensive phenotypic traits were ranked in the top3. Nuoshu, A75-2, Panlonghuangmi, Xiaobaishu, and Xiaohongshu have excellent quality characteristics, which can be applied to the breeding of high quality varieties.

4.2. Proso millet genetic improvement

Fig. 7 Growing period, top3 leaves area, 1 000-seed weight, and protein content of different panicle genotypes. a, scattered type; b, lateral type; c, compact type.

The analysis of more than 40 proso millet cultivars bred in recent years showed that the growing period is between 93.9 and 111.0 days, which is basically consistent with the growth period of high-yield genotypes in this study. The slight difference may be due to the differences of planting date and environment (Yang et al. 2017). The reason why plant height of these variants (120.4 to 175.1 cm) was generally higher than core collections (63.6 to 174.9 cm) may be because higher plant height was selected indirectly at high yield selections. However, if plant height of cultivars was higher than the range of 118 to 150 cm, it would be bad for resistance to lodging (Wang L et al. 2016), which may lead to diff iculties in mechanized harvesting. Spike length of varieties was 24.3 to 40.3 cm, and most of them were lateral type, which was consistent with this study. The average grain weight of cultivated varieties was 6.2 to 9.9 g, which was lower than that of high yield core collections (7.4 to 9.7 g). Fat and protein contents of these cultivars ranged from 1.7 to 5.6% and 9.64 to 15.72%, respectively, and content of some collections was high, but the average value was low. Compared with amylose content of landraces (1.6 to 45.6%), amylose content of the varieties was 2.24 to 38.67%, indicating that the selection of strong non-waxy and waxy varieties was not prominent. From the above results, we knew that though trait amplitudes of the varieties were small, the diversity within the cultivars was still high. Therefore, rich variations can still be seen. Thus, the selection pressure of proso millet breeding was not enough to reduce variation (Wang H G et al. 2016), and there still is great potential for optimizing plant shape traits and yield, and for improving quality.

4.3. Correlation between the traits

The correlation analysis of traits in this study was based on large samples. Although some correlation coeff icients were small, correlations were still signif icant or highly signif icant. The correlation between most quantitative traits was either signif icant or very signif icant. The growing period is clearly and positively correlated with plant height, diameter of main stem, top3 leaves area, and 1 000-seed weight, indicating that the genotypes with long growing periods are strong and have high yields. However, the yield will decrease because of the extremely long growing period, for example, the average 1 000-seed weight of genotypes with a growth period of more than 100 days is 6.5 g, which is lower than the 7.4 g of 80 to 90 days. The result is consistent with results of Lang et al. (2012) on rice and Yuan et al. (2016) on foxtail millet. A possible reason for this is the matter accumulation of the germplasms with short growing periods is less than medium germplasms. Also, germplasms with a long growth period are not conducive to grain f illing and maturity due to environmental factors, such as lower light and decreasing temperature during the latter growth period, which results in decreased yield (Yuan et al. 2016). Therefore, breeders can increase yields by modestly extending the growing days.

Effects of growth period on quality traits were different. The content of grain fat and protein f irst decreased and then increased with the extension of the growing period. On the other hand, the content of starch and f lavochrome had an opposite trend, which may be related to the characteristics of germplasm accumulation in proso millet (Han 2017). There was a signif icant and positive correlation between characteristics of vegetative organs such as plant height, stem diameter, stem node number, panicle length, and top3 leaves area. Results of this study are consistent with the f inding of Upadhyaya et al. (2011) showing that in vegetative organs, growing period is consistently correlated with yield and quality traits. There is a marked positive correlation between grain fat and protein content, and they both have signif icantly negative correlations with 1 000-seed weight, indicating that high protein and fat are not contradictory and could be achieved in some varieties. In contrast, obtaining both high yield and high quality is diff icult. These f indings are consistent with Hou et al. (2001).

The correlation between leaf phase and quantitative traits was the highest among quality traits. The correlation coeff icients between leaf phase and plant height, growing days, and 1 000-grain weight were -0.476, -0.378, and -0.318, respectively. Since leaf phase also is the key to optimizing plant shape (Jiao 2010), it is necessary to study further. Other quality traits and quantitative traits are also signif icantly related, but the correlation coeff icient was low, and the relationships need to be further studied.

5. Conclusion

The core collections had abundant variation. Therefore, they can accurately represent the genetic diversity of the proso millet resources as required by germplasm construction and can be used as basis for discriminating among proso millet cultivars based on related traits and for further studies on millet with rich genetic diversity, good representation, and signif icant correlation between traits.

Acknowledgements

Work in my lab was supported by the National Natural Science Foundation of China (31371529), the earmarked fund for China Agriculture Research System (CARS-06-A26), and the National Key Technologies R&D Program of China during the 12th Five-Year Plan period (2014BAD07B03).

Appendixassociated with this paper can be available on http://www.ChinaAgriSci.com/V2/En/appendix.htm