Exploring the genetic variability and diversity of pearl millet core collection germplasm for grain nutritional traits improvement

Improving essential nutrient content in staple food crops through biofortification breeding can overcome the micronutrient malnutrition problem. Genetic improvement depends on the availability of genetic variability in the primary gene pool. This study was aimed to ascertain the magnitude of variability in a core germplasm collection of diverse origin and predict pearl millet biofortification prospects for essential micronutrients. Germplasm accessions were evaluated in field trials at ICRISAT, India. The accessions differed significantly for all micronutrients with over two-fold variation for Fe (34–90 mg kg−1), Zn (30–74 mg kg−1), and Ca (85–249 mg kg−1). High estimates of heritability (> 0.81) were observed for Fe, Zn, Ca, P, Mo, and Mg. The lower magnitude of genotype (G) × environment (E) interaction observed for most of the traits implies strong genetic control for grain nutrients. The top-10 accessions for each nutrient and 15 accessions, from five countries for multiple nutrients were identified. For Fe and Zn, 39 accessions, including 15 with multiple nutrients, exceeded the Indian cultivars and 17 of them exceeded the biofortification breeding target for Fe (72 mg kg−1). These 39 accessions were grouped into 5 clusters. Most of these nutrients were positively and significantly associated among themselves and with days to 50% flowering and 1000-grain weight (TGW) indicating the possibility of their simultaneous improvement in superior agronomic background. The identified core collection accessions rich in specific and multiple-nutrients would be useful as the key genetic resources for developing biofortified and agronomically superior cultivars.

. Mean squares for iron (Fe), zinc (Zn), days to 50% flowering and 1000-grain weight of a set of 212 pearl millet core collection accessions evaluated during the 2011 summer and 2011 rainy seasons, Patancheru, India. Values in the parenthesis are degrees of freedom for individual environments. S summer, R rainy, NS non-significant. *,** Significant at the 0.05 and 0.01 probability levels, respectively.
Nutrient-dense germplasm. To identify multiple nutrient germplasm data on 39 core collection accessions (a subset of 212 accessions) that had data on all micronutrients with agronomic traits were considered. The top 10 accessions having superior levels of each nutrient trait were selected except for Ni and Na, for which the accessions having the lowest levels were selected as per desirability (Table 4). All these high nutreint-dense accessions flowered in 45-67 days and had 6-16 g TGW. The low Ni and Na accessions also had a similar range of flowering (45-67 days) and TGW (7-16 g Table S5 and Fig. 1b). The number of accessions in the clusters varied from 6 accessions in cluster 5, 7 each in cluster1 and 3, 8 in cluster 4, and 11 in cluster 2. Cluster1 had higher cluster mean values for Mg, P, S, Cu, and cluster 2 had lower cluster means for most of the traits including Fe. Cluster3 had higher cluster mean values for Fe, Mn, and Ca, and cluster4 had higher cluster mean for TGW and lower value for days to 50% flower, and cluster5 had a higher cluster mean for Zn, Mg, Mo, and K and the lower mean value for Ni. Based on two season pooled data, the correlation analysis was conducted to determine the relationships among grain micronutrients, macronutrients, and with two important agronomic traits (Table 6). Correlation between Fe and Zn was found highly significant and positive (r = 0.43-0.51; p < 0.01). Fe content was positively and significantly (p < 0.05) correlated with Ca (r = 0.33; www.nature.com/scientificreports/ p < 0.05) and Ni content (r = 0.32; p < 0.05). Fe content had non-significant correlations with other micronutrients (r = 0.00 to − 0.18). Zn content was positively and highly significantly (r = 0.36-0.46; p < 0.01) correlated with Mo and S while non-significantly mostly positively correlated with other nutrients. A highly significant and positive correlation of Mg with Na, P, K, S, Cu, and Mo was seen. Ca had a largely positive association with all nutrients but was significantly associated only with Fe and Mn. Although P and K had a positive correlation with other nutrients, significant correlations were observed only for P with K, Cu, Mo, and S whereas K was associated with Mg and Na. Correlation between days to 50% flowering and TGW was negative and significant. All the micronutrients recorded positive correlations with 50% flowering and highly positive and significant correlation of days to 50% flowering with Zn, Mo, Mg, K, and P was seen. Fe content did not correlate with days to 50% flower while Zn content recorded positive and significant correlation. On the other hand, the correlation between Fe density and TGW was low negative and non-significant but a negatively significant correlation was observed between Zn density and TGW. In general, TGW showed an undesirable correlation with almost all the micronutrients, and correlation was highly significant with Mo, Mg, P, and S.

Discussion
A large proportion of the global population suffers from micronutrient deficiency. Among the micronutrients, Fe (60-80% population), Zn (30%), and Se (15%) deficiency is widespread 2,3 . The social costs of micronutrient deficiency are devastating to the countries as it results in anaemia (Fe deficiency) and impaired growth (Zn deficiency). Lower intake of other micronutrients such as Mg (for cardiovascular diseases), osteoporosis (for Ca), will lead to the condition of weak bones and teeth (P), hypokalemia (K). The availability of these important micronutrients through the staple diet is a sustainable means to enhance global human capital. This is particularly important in the case of several countries in sub-Saharan Africa and South Asia where pearl millet is a major source of food, fodder and feed. To meet the micronutrient needs, developing nutrient-dense and agronomically desirable cultivars is an important objective in pearl millet breeding at ICRISAT and elsewhere. Pearl millet biofortification breeding program targets breeding pipelines which will have > 50% higher Fe and Zn content (+ 30 mg kg −1 ) over the existing hybrid parents and commercial varieties in Africa and India. To achieve such challenging breeding targets, exploitation of genetic variability present in the gene bank is a fast-track approach for the identification of mineral-dense accessions. Core or mini core collection (10% of core collection or 1% of the entire collection) 16 is a sound scientific strategy to systematically and cost-effectively screen the large germplasm collection conserved in the genebank globally. ICRISAT genebank conserves over 24000 accessions of pearl millet and its wild relatives (http://geneb ank.icris at.org/ accessed on 5 Jan 2020). A core collection consisting of 2094 accessions that representing a diversity of the entire collection was developed by Upadhyaya et al. 15 . This study involved a part (504 accessions) of the core collection that was evaluated in two seasons at the ICRISAT Centre Patancheru, India for grain nutrients and two agronomic traits. Estimation of nutrients in the grains may be influenced by their availability in the soil in experimental fields. The estimated Fe and Zn content in the upper 30 cm soil in the experimental fields were above the critical so as not to limit crop growth 17 . Our results indicated that there was not any Fe and Zn deficiency in the soil and thus no impact on proper phenotyping. Also, balanced nutrition is needed to be emphasized to measure full genetic potential by applying respective micronutrients for creating uniform selection environments. A similar recommendation was reported for pearl millet biofortified hybrids 18 .
ANOVA of the 212 and 39 core collection accessions separately, indicated that the mean square attributable to genotypes was highly significant indicating adequate variability in the accessions of a core collection for micronutrients and agronomic traits (Tables 1 and 2). The mean squares attributable to the environment were also significant for all the micronutrients except Mn and TGW for 39 accessions (Tables 1 and 2) indicating that the two seasons that we used to evaluate the core collection accessions were different and adequate to differentiate the genotypes. The role of G × E interactions was important but much less than that of genotypes as the magnitude of the sum of squares due to genotypes was much greater than the G × E interaction (Tables 1 and 2).
High estimates of heritability were found for most of the micronutrients including Fe and Zn contents. The lower G × E interactions and high heritability for micronutrients suggested the possibility of a good response to selection. Similar high heritability and low G × E interaction for these traits have been reported in pearl millet [19][20][21][22] . Further, the relative performance of accessions did not markedly vary from one season to another which further confirmed that the G × E influences are not affecting ranking and selection of the accessions for micronutrients 23 .
The present study revealed the several-folds variability for most of the essential nutrients as well as agronomic traits such as TGW and days to 50% flowering (Table 3). Simultaneous improvement of different micronutrients and agronomic traits is possible when they are not negatively correlated to each other and with agronomic traits. Our study revealed that the Fe was positively correlated with Zn in both the seasons (r = 0.43-0.51, p < 0.01) and in the pooled (r = 0.46, p < 0.01; Table 6) among highly selected core collection 39 accessions. A similar positive and highly significant correlation between Fe and Zn observed among 212 accessions in both the seasons (r = 0.57-0.69, p < 0.01) and in the pooled (r = 0.64, p < 0.01), displaying that simultaneous selection for high-Fe and Zn densities could be very successful in pearl millet. Such a positive direction correlation between Fe and Zn has been reported in several pearl millet studies 14,19 21,24-30 . This tight linkage among Fe and Zn densities could be an indication of some physiological functions responsible for Fe and Zn uptake and translocation are governed by similar genes or do not have an antagonistic effect. For instance, QTLs identified for Fe and Zn contents are co-localized in pearl millet, thus hypothesizing a common transporting pathway 31 . The biofortification initiative led by HarvestPlus (a CGIAR program) recommended pearl millet biofortification breeding target is for Fe while Zn can be improved as associated trait 32 . The Zn content was positively correlated with Mo and S indicated that these nutrients could be combined in biofortified lines. Similar reports on the increase in Zn leading to an increase in the S in wheat grains 33,34 . The witness of highly significant and positive correlation of Mg with Na, P, K, S, Cu, and Mo imply Mg can be improved with these nutrients and Mg could be playing an important symbiotic role in assimilating these nutrients to grains). Bashir et al. 25 reported a positive correlation between Mg with P and a non-significant correlation with Na in pearl millet, a study in rice reported a positive correlation between Mg with P and Cu 35 . Given the observed positive association of Ca with Fe and Mn, this will consistently increase these nutrient densities.
Flowering always had a positive correlation with all the nutrients, which shows the late-flowering accessions had a longer time to accumulate nutrients than early flowering accessions. This could be of an artifact that influence of reduced grain weight in late-flowering accessions. Thus, higher nutrient content is observed in small shrunken grains. Earlier studies in pearl millet reported a positive correlation between Fe and Zn with TGW 20,24 . In another study non-significant negative and positive correlation was observed between Fe and Zn with TGW 26 , while Bashir et al. 25 reported none of the micronutrients showed significant correlation with TGW. In sorghum 36,37 and wheat 38 reported Fe and Zn had a negative correlation with TGW. These negative linkages can be broken in breeding populations using selective crosses involving contrasting genotypes and selection in larger segregating populations.
Scientific Reports | (2020) 10:21177 | https://doi.org/10.1038/s41598-020-77818-0 www.nature.com/scientificreports/ Ten top accessions of core collection were identified based on per se performance for each nutrient. Some of the accessions were promising for more than one nutrient. Considering the importance of Ni in nitrogen metabolism and fixation in crops, its adequacy critical for germination and crop growth 39 . However, an elevated level of Ni causes stunted growth, chlorosis, nutrient imbalance in crops, and health burdens to humans, thus lower level is desirable. The top 15 accessions with multiple nutrients (Table 5) were identified. These 15 accessions originated from five countries in Africa (Ghana, Burkina Faso, Niger, Zimbabwe) and Asia (India) indicating their geographical diversity. The clustering of 39 accessions (comes from 9 countries), indicating that these 15 accessions occurred in four of the five clusters. The seven accessions from India were clustered into two clusters; cluster 1 had four accessions from India and one from Ghana whereas cluster 2 had three accessions from India and one from Niger (Fig. 1b). These 15 multiple nutrients accession including six accessions with higher levels for eight or more nutrients, also showed wide variability for days to 50% flowering (42-64 days) and TGW (7-16 g) ( Table 5). These germplasm resources can be exploited to improve the multiple nutrients in different maturity and seed size that are missing in elite pearl millet lines and hybrid parents. Using the core/mini core approach multiple nutrient germplasm accessions have been identified in peanut 40 , sorghum 41 , finger millet 42 , foxtail millet 43 , and kodo millet 44 , These core collection accessions offer diverse sources for traits introgression and wider scope for selection for the improvement of target micronutrients in breeding populations. Recently, the baseline for Fe and Zn has been established in the India national pearl millet cultivar release policy based on micronutrient levels in commercial cultivars. This is mandatory for both summer and rainy season cultivars. It is very important to note that all 39 accessions (of which, 13 were from India) were found to exceed the Indian pearl millet cultivars baseline of 42 mg kg −1 Fe and 32 mg kg −1 Zn. While, 17 accessions including eight accessions from India, exceeded the biofortification breeding Fe target (72 mg kg −1 ). Previous studies in pearl millet 14,24 reported that high-Fe/Zn sources are largely or entirely from iniadi germplasm (early-maturing, larger grain-size collections from Togo regions) and no reports on the availability of Indian origin germplasm for higher micronutrients than iniadi. The present study revealed that there are 7 accessions from India that had 72-88 mg kg −1 Fe content and 42-62 mg kg −1 Zn content. This opens new opportunities for utilization in the initial crossing program for developing parental pipelines to meet the national minimum standards for these micronutrients 45 . Application genomic tools would help assess the genetic constitution and identify genetically dissimilar accessions among these multiple nutrient accessions. Their use in the breeding programs will help in the development of and dissemination of high Fe and Zn containing genotypes within high yielding breeding pipelines to address both food and nutritional insecurity in a target millet region. A limited quantity of seed of these lines is available from the ICRISAT Genebank under Standard Material Transfer Agreement.

Materials and methods
Plant materials and field evaluation. A set of 504 pearl millet accessions, that were part of the 2094 accessions of the core collection, originating from more than 25 countries representing major pearl millet growing regions of the world 15 were used in this study (Supplementary Table S1). While developing the pearl millet core collection, duplicates were excluded and only distinct accessions were included 15 . These accessions were grouped into four groups based on similar phenology using days to 50% flowering from the Genebank database. Thus group 1 consisted of accessions that flowered in 40-60 days, group 2 in 61-70 days, group 3 in 71-80 days, and group 4 in 81-120 days. Quantity of seed received from gene bank was limited, thus, trials were evaluated in two sets following an augmented design 46 in the 2009 summer season (February-June) with two controls, ICMB 88004 and ICTP 8203 at every 10th accessions at the ICRISAT, Patancheru, India (17.53°N latitude, 78.27°E longitude, and 545msl), in the alfisols. ICMB 88004 is a potential breeding line derived directly from germplasm and ICTP 8203 a popular variety derived from landraces 47,48 . Each accession was planted in one row of 4 m in length with the row-to-row spacing of 60 cm. The distance between plants within a row was maintained at 10 cm by thinning at 10-12 days after planting. A basal dose of diammonium phosphate at 100 kg ha −1 (i.e., 18 kg N and 46 kg P) was applied before planting with 100 kg ha −1 of urea (i.e., 46 kg N) side-dressed 15 days after planting. The field was manually weeded just after thinning and irrigated at every 10 days interval to protect from any moisture stress. At the time of panicle emergence, the main panicles of 6-8 plants in each plot were covered with parchment paper bags for selfing. All the selfed panicles were harvested at/or after physiological maturity and sundried for 10-15 days before threshing. Due to very late flowering (> 75 days), we could not harvest the grains from some accessions as the field was maintained till 110 days. Few accessions also had a poor self-seed set (< 10%). Seventy-six accessions that matured late and had poor seed set were not included and selfed seed from 428 accessions was used for grain micronutrient analysis as described below.
Selection and re-evaluation in two seasons. Based on agronomic performance and high grain Fe content (> 70 mg kg −1 ), 212 accessions (Supplementary Table S2 Table S3) having high-Fe, 1000-grain weight (TGW) were analyzed for allgrain nutrients using ICP-OES at Waite Analytical laboratory, Adelaide University, Australia. The alfisol fields at the ICRISAT are precision fields with a gentle slope of 0.5%. The planting, plant spacing, and crop-management practices were similar as described above except the row to row spacing of 75 cm in the 2011 rainy season. Days to 50% flowering were recorded on a plot basis when at least 50% of plants in an accession had full exsertion of stigmas. Considering the high cost of self-seed production and good reliability of open-pollinated grain samples in pearl millet for estimation of nutrients 49  www.nature.com/scientificreports/ were harvested in a cloth bag at after physiological maturity. These were hand threshed and approximately 20 g of grains were collected from each plot for determining 1000-grain weight (TGW) and for use in grain micronutrient analysis (described below). For estimating TGW a random sample of 200 grains was used from each plot, weighed, and multiplied with a factor of five.

Micronutrient analysis.
Selfed-seed samples from 428 accessions in the 2009 rainy season were analyzed for grain micronutrients using Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES) following the method described by Wheal et al. 50 at Waite Analytical laboratory, Adelaide University, Australia. Briefly, grain samples were oven-dried overnight at 85 °C prior to digestion, grounded enough to pass through a 1 mm stainless steel sieve using Christie and Norris hammer mill, and stored in screw-top polycarbonate vials. Grain samples were digested with di-acid (nitric and perchloric acid) mixture. After digestion, the volume of the digest was made to 25 mL using distilled water, and the content was agitated for 1 min by a vortex mixer. These digests were used for determination using ICP-OES. All 212 replicated accessions samples from the 2011 summer and rainy seasons were analyzed using X-ray Fluorescence (XRF) Spectrophotometry at ICRISAT, Patancheru, India. Destructive (wet lab) micronutrient analytical cost is very high, adding additional cost burdens to the breeding program when dealing with larger and replicated germplasm samples. Therefore, the present study explored the potential use of X-ray Fluorescence (XRF) Spectrophotometry for Fe and Zn content. XRF analysis, a scanning based method 51 was used to estimate Fe and Zn. Briefly, XRF uses aluminium sample cups of 30 mm diameter, 36 mm depth and > 20 g of grain weight capacity, combined with polypropylene inner cups sealed at one end with 4 μm Poly-4 XRF sample film. Cups in a batch of 10 were filled with 8-12 g of grain and were shaken to distribute grain evenly within the cup, which was then loaded in the XRF instrument holder. The XRF method is highly reliable in pearl millet and the readings were positively and significantly correlated (r ≥ 0.90 for Fe and Zn) with the ICP method 51,52 . Experiment field representative soil samples collected from the top 30 cm layer were analyzed for extractable Fe and Zn content using atomic absorption spectroscopy (AAS) as described by Lindsay and Norvell 53 .
Statistical analyses. Analysis of variance (ANOVA) of all the trials was performed for both individual environments and pooled data using the PROC GLM procedure in SAS 14.1 software 54 . Genotypes (accessions) were considered as random and environment (season) as fixed. Variance attributable to genotypes (σ 2 g) in individual seasons and pooled over two seasons were estimated for all traits. In the pooled analysis, σ 2 ge) variance components were estimated for all traits. The homogeneity of error variance for different seasons was tested by Bartlett's test 55 , which is sensitive to departures from normality 56 . Broad-sense heritability (H 2 b) was estimated across environments using the following formula where σ 2 g is the genotypic variance, σ 2 ge is the genotype × environment interaction variance, and σ 2 error is the residual variance, r is the number of replications and e is the number of environments. The cluster analysis was carried out using R v3.5.1 57 and the clustering of genotypes was performed using Ward's 58 methods. The significance of differences between cluster mean values were tested following the Newman-Keuls procedure 59,60 . Phenotypic correlations were estimated among all the traits in each environment and pooled data and between XRF and ICP for 39 accessions and tested for significance 56 . Promising core collection accessions for Fe, Zn, and other micronutrients were identified based on the per se performance.