Introduction
Foxtail millet (Setaria italica (L.) Beauv.) is one of the ten small-grained cereals (small millets), which is grown as a food crop in Asia and for animal feed in the USA and Europe. The naming of this taxon evolved as the millet having panicles resembling a fox's tail in appearance i.e. a long panicle with soft, long and erect hairs. China, India and Japan are the chief foxtail millet growing countries. The foxtail millet is believed to have first domesticated in the Central China (Chang, Reference Chang1973; Ho, Reference Ho1975). Taxonomically, foxtail millet comprised two subspecies, S. italica subsp. italica and subsp. viridis. The subspecies viridis is considered as the progenitor of the cultivated form.
Based on the comparative morphology of the foxtail millet accessions, Prasada Rao et al. (Reference Prasada Rao, de Wet, Brink and Mengesha1987) have suggested that there are three races of foxtail millet: (i) race Moharia is common in Europe, southeast Russia, Afghanistan and Pakistan; (ii) race Maxima is common in eastern China, Georgia (Eurasia), Japan, Korea, Nepal and northern India (it has also been introduced in the USA); and (iii) race Indica is found in the remaining parts of India and Sri Lanka.
It is difficult to make precise estimates of the area under this crop worldwide, because the area under all the millets is generally reported together. However, the Consultative Group on International Agricultural Research have recently (2006) estimated that foxtail millet is cultivated on about 4.90 m ha annually (available online only at http://www.cgiar.org/research/res_millet.html). China ranks first in the area and production of foxtail millet. The area of 8.09 m ha in 1936 under foxtail millet has subsequently declined to 4.09 m ha in 1983. However, there has been a further upward trend in the area under this crop since 1982 ( Jiaju, Reference Jiaju, Seetharam, Riley and Harinarayana1989). In India, foxtail millet is grown on about 1 m ha. The grain is used both for food and feed. A seed oil content of 5.45%, with a range between 4.0 and 7.3% has been reported in the germplasm (Seetharam et al., Reference Seetharam, Mallikarjunaradhya and Laxminarayana1983). The seeds are also a rich source of protein (12.3%) and minerals (3.3%; Rai, Reference Rai2002). In general, foxtail millet is valued as a crop of short growth duration, which is fairly resistant to insect pests and diseases, and the grains make a nutritious and healthy food. Ramprasad (Reference Ramprasad2005) has highlighted the folk medicine value of the foxtail millet grains.
Foxtail millet has received little research attention in the past years and continues to be a neglected and underutilized crop given its potential. The grain yields are generally low in comparison with other staple cereals; however, yield levels of 1500–2250 kg/ha have been reported from China (Jiaju, Reference Jiaju, Seetharam, Riley and Harinarayana1989). In view of the several merits, this crop deserves increased research attention. The germplasm characterization is now essential in order to breed high yielding varieties. The genebank at International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) is presently holding 1474 accessions of foxtail millet from 23 countries. These accessions need to be evaluated in multilocational trials to identify suitable parental material for use in breeding programmes. However, evaluating the entire collection of this size is a cumbersome and long task. A core collection approach (Frankel, Reference Frankel, Arber, Illmensee, Peacock and Starlinger1984) can be used to overcome this problem efficiently. Though the germplasm collection of 1474 accessions of cultivated foxtail millet is not large when compared with other crops like sorghum in ICRISAT genebank, yet it is large in view of the current low research priority accorded to this crop. The core collection, thus formed, can be evaluated extensively at a relatively low cost and the information derived could be used as a guide towards more efficient utilization of the entire collection (Brown, Reference Brown, Brown, Frankel, Marshall and Williams1989). The core collections of various crops have been developed (see Upadhyaya, Reference Upadhyaya, Dhillon, Tyagi, Lal and Saxena2004) including the other small millets – finger millet (Upadhyaya et al., Reference Upadhyaya, Gowda, Pundir, Gopal Reddy and Singh2006a). In this paper, we describe the strategy in developing the core collection of foxtail millet using passport information and the data on taxonomical and 12 qualitative traits.
Materials and methods
The Rajendra S. Paroda genebank at ICRISAT, Patancheru, India holds 1474 accessions of the cultivated foxtail millet originating from 23 countries. These accessions were assembled in the ICRISAT genebank over the period from 1976 to 2001. As the number of accessions grew in the collection, they were characterized for 12 qualitative (discrete classes) and 11 quantitative (continuous variation) traits in 1980 (400 accessions), 1984 (317 accessions), 1987 (569 accessions), 1989 (102 accessions) and 2002 (86 accessions). The characterization site, Patancheru, is located at 18°N and 78°E, at an altitude of 545 m, and about 600 km from the sea. Annual rainfall is about 750 mm, most of which occurs during June–September. The germplasm accessions were sown on red soils (alfisols), 75 cm apart on ridges, each accession occupying a single row of 4 m length, plant-to-plant spacing being 10 cm. An application of 20 kg N2 and 50 kg P2O5 per ha was made as basal fertilizer and 45 kg N2 was applied as a top dressing. In all years, sowings were done towards the end of July. Irrigation and hand weeding were given when necessary. The crops were essentially free of disease or insect pests, and no chemical sprays were applied.
The data were recorded on 12 qualitative and 11 quantitative traits following the descriptors of S. italica and Setaria pumila (IBPGR, 1985). The data on all the 12 qualitative traits (plant pigmentation, leaf colour, growth habit, culm branching, bristle length, panicle lobing, inflorescence compactness, lobe compactness, grain colour, plant lodging, leaf senescence and overall plant aspect) and days to flowering (quantitative trait) were recorded on a plot basis. The data on basal tiller number (quantitative trait) were recorded on five representative plants in the plot. The remaining nine quantitative traits (plant height, flag leaf blade length and width, flag leaf sheath length, peduncle length, panicle exsertion, inflorescence length and width and weight of five panicles) were recorded on the main culms of the five representative plants in the plots. During field evaluation, the accessions were classified into their taxonomical races and subraces.
The germplasm accessions were stratified into three taxonomical races (Indica: 996 accessions, Maxima: 235 accessions and Moharia: 243 accessions). The data on the 12 qualitative traits were analyzed using principal coordinate analysis on the accessions in each of the three races, separately, and used in constituting the core subset as the expression of these traits is least influenced by genotype × environment interactions. However, for the validation of the core composition, the data on geographic origin (West Asia, East Asia, Russia and the CIS, South and Southeast Asia, Africa, the Americas, Europe and unknown origin), qualitative as well as quantitative traits was used.
A hierarchical cluster analysis (Ward, Reference Ward1963) was conducted on the scores of the first five principal coordinates delineating all the accessions into 29 clusters. From each cluster, about 10% of the accessions were randomly selected to form the core collection. In clusters having less than ten accessions, at least one was included in the core collection. The frequency of the accessions according to the origin by geographical region, country and affiliation to biological races and subraces in the entire and core collections was analyzed by the χ2-test. Yates' (Reference Yates1934) correction was applied when the frequency in a class (entire collection) was less than five. The means of the entire and core collections were compared using the Newman–Keuls procedure (Newman, Reference Newman1939; Keuls, Reference Keuls1952) for all the 11 quantitative traits. The homogeneity of variances of the entire and core collections was tested with the Levene's (Reference Levene and Oklin1960) test. The percentage of significant differences between the entire collection and the core subset were calculated for the mean difference (MD %), variance difference (VD %), variable rate (VR %) and coincident rate (CR %; Hu et al., Reference Hu, Zhu and Xu2000). The diversity index (H ′) of Shannon and Weaver (Reference Shannon and Weaver1949) was estimated and used as a measure of phenotypic diversity in both the entire and core collections for each trait. The phenotypic correlations between traits in the entire and core collections were estimated separately. This was done to assess the extent to which the trait associations (presumed to be under genetic control) were captured in the core collection.
Results
The entire collection of foxtail millet in the ICRISAT genebank (1474 accessions) was grouped into 29 clusters. Race wise, the number of clusters was 12 (1–12) in Indica, 11 (13–23) in Maxima and 6 (24–29) in Moharia. The number of accessions in clusters ranged between 7 and 134. From each cluster, 10%, or at least one accession, was included in the core collection resulting in a selection of 155 accessions, which is 10.51% of the entire collection. The number of core accessions in individual clusters ranged from 1 to 14. The list of 155 core collection accessions with the data on race, subrace and the country of origin is given in Supplementary Table S1, available online only at http://journals.cambridge.org.
The χ2-values of frequency distribution for the 12 qualitative traits (plant pigmentation, leaf colour, growth habit, culm branching, bristle length, panicle lobing, inflorescence compactness, lobe compactness, grain colour, lodging, senescence and over all plant aspect) and trait classes (three to six classes) were non-significant, suggesting that the representation of diversity from the entire to the core collections was adequate (Supplementary Table S2, available online only at http://journals.cambridge.org). The entire collection of foxtail millet is represented by 23 countries spread over Asia, Africa, Europe and North America. Seven countries (Ethiopia, Iran, Malawi, South Africa, Spain, Switzerland and UK) and the group of accessions with unknown origin, had less than five accessions in the entire collection. A χ2-test was carried out to determine whether the sampling of accessions in the core collection represented the entire collection satisfactorily. The χ2-value was non-significant for all regions except Africa. The accessions from Africa were over represented in the core collection. There were 13 accessions from Africa, and the clustering based on 12 qualitative traits delineated into four clusters, having 1, 8, 1 and 3 accessions, individually. Including at least one accession from each of these clusters resulted in higher (significant) representation in the core collection from this region. The heterogeneity for all the regions was non-significant (P = 0.328–0.999). The countries within the regions and the unknown group had non-significant χ2-values, indicating that all the countries were optimally represented in the core collection (Table 1).
Table 1 Number and percentage of accessions from different regions and countries (within region) in entire and core collections of the foxtail millet germplasm
In foxtail millet, the race Indica is characterized by tall plants (average 122 cm) with three to four tillers and 18 cm long panicles. On a panicle, spikelets are loosely arranged and have three bristles below each spikelet (Prasada Rao et al., Reference Prasada Rao, de Wet, Brink and Mengesha1987). The germplasm collection in ICRISAT genebank is dominated by this race with 996 accessions (67.57%). The race Moharia is phenotypically similar to subsp. viridis, except that the latter has the grain-shattering feature. The plants are relatively short (about 60 cm), producing about nine tillers and also having a tendency to produce culm branches. The panicles are about 8 cm long and compact in appearance (Prasada Rao et al., Reference Prasada Rao, de Wet, Brink and Mengesha1987). This race is represented by 243 accessions (16.49%) in the ICRISAT genebank. The race Maxima is highly variable. It is identifiable by closely arranged spikelets on long panicles, giving the panicles a tightly lobed appearance. On average, plants grow mid-tall (85 cm) with one to two unbranched tillers and with panicles of 15 cm length. The race Maxima is represented by 235 accessions (15.94%) of the foxtail millet germplasm in the ICRISAT genebank. These races were adequately represented in the core collection having 102 accessions of Indica (65.81%), 29 accessions of Moharia (18.71%) and 24 accessions of Maxima (15.48%). The χ2-values for the races for frequency distribution between the entire and the core collections were non-significant (χ2 = 0.071, 0.020 and 0.465, and P = 0.789, 0.886 and 0.495 for Indica, Maxima and Moharia, respectively), revealing that the sampling technique to constitute the core collection was effective. Considering the representation of the foxtail millet germplasm at subrace level, race Indica comprised four subraces, namely, Erecta, Glabra, Nana and Profusa. Of these subraces, Nana is predominant (79.41%) followed by Glabra (15.69%). The race Moharia comprised three subraces, of which Glabra is more common (58.62%) followed by Aristata (37.93%) and Fusiformis (3.45%). The race Maxima also comprised three subraces, of which Compacta is more common (83.33%) and the other two subraces are Spongiosa (12.50%) and Assamense (4.17%). The frequency distribution of the accessions by subraces within races in the entire and core collections indicated a good fit, again suggesting that the core collection represented the entire collection at the subrace level also (Table 2).
Table 2 Number (percentage) of accessions belonging to different races and subraces (within race) in entire and core collections of the foxtail millet germplasm
Differences between the means of the entire and core collections were non-significant for all the 11 quantitative traits (Table 3). The homogeneity test of the variances between the entire and core collections revealed homogeneity of the variances for all quantitative traits except for the inflorescence width. This indicated a VD % of only 9% ( < 20%; Hu et al., Reference Hu, Zhu and Xu2000) with the variance in the core being greater than the entire collection. The CR value of 87.7% (>80%) and VR of 104% in the core (Table 4) indicated a fairly high representation of diversity from the entire collection to the core subset.
Table 3 Range, means, and variances in entire collection and core collection of the foxtail millet germplasm
1 Means were tested using Newman–Keuls test and were non-significant for all characters.
2 Variances were tested by Levene's test.
Table 4 Coincident rate percentage and variable rate percentage of 11 quantitative traits of the core collection of the foxtail millet germplasm
The estimates of Shannon–Weaver diversity index (H ′) reveal the allelic richness and evenness in a population. Overall, the H ′ estimates for all the 23 traits, and the qualitative (12) traits, separately corresponded well between the entire and core collections. The mean H ′ for quantitative traits was 0.588 ± 0.016 and 0.575 ± 0.018 for the entire and core collections, respectively. The values are nearly similar, revealing that the diversity of the entire collection has been effectively sampled in constituting the core collection (Supplementary Table S3, available online only at http://journals.cambridge.org).
Phenotypic correlations were calculated between the 11 quantitative traits in the entire and core collections separately (Supplementary Table S4, available online only at http://journals.cambridge.org). In the entire and core collections, 53 and 41 correlations, respectively, were significant at the 1% level of significance. The significance levels in the entire and the core collections were different due to the different numbers of degree of freedom, 1472 for the entire collection and 153 for the core collection, and hence, the correlations between the two sets could be considered having a similar trend.
Discussion
Foxtail millet is an important crop in some niches, namely, northern hills, Assam and Andhra Pradesh in India, and Northeast provinces of China; however, the precise estimates of the area under cultivation are not available. For productivity also, except from China, where grain yield as high as 2250 kg/ha has been reported (Jiaju, Reference Jiaju, Seetharam, Riley and Harinarayana1989), the information is not available. The grains of foxtail millet are nutritious and they also possess high dietary values, e.g. high oil contents (5.45%; Seetharam et al., Reference Seetharam, Mallikarjunaradhya and Laxminarayana1983), protein (12.3%), minerals (3.3%; Rai, Reference Rai2002), crude fibre (10.8%) and iron (49 mg/kg of grain; Beghel et al., Reference Beghel, Netke and Bajpai1985). The grains also make an ideal food for people suffering from diabetes and gastric problem as well used to treat chickenpox (Ramprasad, Reference Ramprasad2005). In general, this crop is valued for its short duration, cultivated with limited farm inputs and remaining free from any disease or insect pest. In view of the several merits of this crop and a very limited research undergone, foxtail millet deserves greater attention. There is a need to identify the germplasm accessions with economic traits that can be used in breeding cultivars with high and stable yields. The core collection, which has reduced number of accessions, though containing almost full diversity, can be used for such purposes cost effectively.
Our results indicated that the germplasm collection at ICRISAT possess good amount of diversity for important agronomic traits (Table 3 and Supplementary Table S3, available online only at http://journals.cambridge.org) and serves as a good basis for developing a core collection. The present core collection of 155 accessions is a good representative of the entire collection as indicated by the non-significant differences in the means and homogeneity of variances of the entire and core collections. The negligible estimates of MD (0%), VD (9%) and low discrepancy of CR (87.7%) and VR (104%) suggest that the variability on the 11 quantitative traits have been captured in the core collection and corresponds well with the hypothesis of Hu et al. (Reference Hu, Zhu and Xu2000). They considered a core collection to be the representative of the entire collection when no more than 20% of the traits have different means and variances (significant at 0.05) and the CR and VR % retained by the core collection are no < 80%.
The trend of similar phenotypic correlations in the core and entire collections in the present study supported the hypothesis put forward by Ortiz et al. (Reference Ortiz, Ruiz-Tapia and Mujica-Sanchez1998) that the adequate and proper sampling is essential in developing a representative core collection, requires consideration of the sampling of phenotypic associations arising out of coadapted gene complexes. The information from phenotypic correlations can also be used to identify traits, which need not be recorded in the initial characterization, when information is revealed by an easily measurable associated trait. The proportion of variance in one trait that can be attributed to its linear relationship with a second trait is indicated by the square of the correlation coefficient (Snedecor and Cochran, Reference Snedecor and Cochran1980). Furthermore, the correlation coefficients with an absolute value >0.71 or < − 0.71 have been suggested to be meaningful (Skinner et al., Reference Skinner, Bauchan, Auricht and Hughes1999), so that 50% of the variation in one trait is predicted by the other. In the present study, three meaningful correlations (r>0.71) were: (i) plant height vs. inflorescence length, (ii) flag leaf blade length vs. inflorescence length, and (iii) peduncle length vs. panicle exsertion in the entire and core collections. The examination of these combinations revealed that the inflorescence length corresponds well with the plant height and the flag leaf blade length and peduncle length corresponds well with the panicle exsertion. Hence, to be cost effective and reduce some redundancy in the foxtail millet research in the future, of the five traits involved in the three trait pairs, panicle exsertion and inflorescence length need not be recorded. This corresponds well with a study on finger millet germplasm (Upadhyaya et al., Reference Upadhyaya, Gowda, Pundir, Gopal Reddy and Singh2006a), where the peduncle length was correlated with the panicle exsertion and the inflorescence length with the length of the longest finger and concluded that the two traits, panicle exsertion and longest finger length could be ignored in initial characterization studies.
The present core collection has merits as this has been developed based on germplasm characterization and clustering and expected to be more efficient than the core developed using the passport data and random selection of accessions (Diwan et al., Reference Diwan, McIntosh and Bauchan1995). Furthermore, this core has high level of representativeness of the entire collection and is likely to provide easy access to the entire collection.
In view of the limited resources available for agricultural research in developing countries, especially for a low priority crop such as foxtail millet, the core collection will provide a good working collection that can be extensively characterized for all economically important traits including the reactions to various stress factors and for nutritional traits, such as protein, β-carotene, iron and zinc contents. The molecular characterization of the core collection would help in revealing the population structure and in assessing the genetic diversity at DNA level with reduced resources. In fact, the characterization data on morphological, agronomical, biochemical and molecular traits should be integrated because the different types of traits provide complementary information (Singh et al., Reference Singh, Gutierrez, Molina, Urrea and Gepts1991). The tolerance to high and low temperatures is going to be an important factor in view of global climate change. The core collection due to its drastically reduced size can be conveniently evaluated to identify tolerant/resistant sources using lesser resources. With the use of the core and mini core collections of chickpea (Upadhyaya and Ortiz, Reference Upadhyaya and Ortiz2001), sources of high grain yield (Upadhyaya et al., Reference Upadhyaya, Dwivedi, Gowda and Singh2007), tolerance to drought (Kashiwagi et al., Reference Kashiwagi, Krishnamurthy, Upadhyaya, Krishna, Chandra, Vincent and Serraj2005) and diseases (Pande et al., Reference Pande, Kishore, Upadhyaya and Rao2006) have been identified. Similarly, new sources of tolerance to drought (Upadhyaya, Reference Upadhyaya2005) and low temperature at germination (Upadhyaya et al., Reference Upadhyaya, Gowda, Reddy and Singh2009b), and for early maturity (Upadhyaya et al., Reference Upadhyaya, Reddy, Gowda and Singh2006b), were identified in the groundnut core and mini core collections. For additional sources of useful traits, the accessions from the reserve collections (remaining part of the entire collection) can also be examined selectively from the same cluster from which the accession in the core collection has been identified. The core collection is expected to facilitate efficient access to the germplasm by the user community. This core collection is not static and should be revised periodically when new accessions and information becomes available as in the case of pearl millet (Reference Upadhyaya, Reddy, Dwivedi, Gowda and SinghUpadhyaya et al., in press).
