Introduction

The human body requires more than 22 mineral elements that can be supplied by an appropriate diet (Philip and Martin 2005). However, diets of human populations subsisting largely on cereals often lack adequate minerals [e.g., copper (Cu), iron (Fe), manganese (Mn) and zinc (Zn)], which are known to play essential roles in physiological processes of the human body. Over three billion people in the world are affected by micronutrient malnutrition resulting in poor health and higher rates of mortality (Cakmak 2002). Many of those afflicted are dependent upon staple crops for their sustenance (Pfeiffer and McClaferty 2007). Among the staple crops, maize (Zea mays L.) is one of the most important crops, accounting for 40 percent of the world’s cereal food production (Bouis and Welch 2010). Together with the fact that maize has the highest average yield per hectare, even a small increase in the nutritive value of maize would be important for human nutrition. Moreover, mineral concentrations (e.g., Cu, Fe, Mn, and Zn) in maize grain are relatively low when compared to animal food products (Wang et al. 2003). The concentration even decreased in the past decades due to breeders selecting exclusively for grain yield but not grain quality (Fan et al. 2008; Šimic et al. 2009; Anandan et al. 2011). Thus, since the concentration of these dietary minerals in maize grain is not sufficient to meet the dietary requirement of humans’ daily intake when these foods are consumed in typical amounts, improving the mineral concentration in maize grain is of great interest.

Traditional efforts on improvement of grain mineral accumulation in crops has been done majorly in three ways: (1) selection of germplasm with greater accumulation of essential minerals (White and Broadley 2009), (2) breeding of mineral efficient crops that accumulate more minerals (Gregorio et al. 2000), and (3) enhancing bio-available minerals in edible portions through genetic engineering (Goto et al. 1999). However, such approaches have not always been successful, mostly because of economic and technical difficulties (e.g., the large amount of time and labor required for breeding new crop varieties with high grain minerals; Mayer et al. 2008). Quantitative trait locus (QTL) mapping is a powerful approach to study and manipulate complex traits important in agriculture (Abiola et al. 2003). A wide range of genetic variation for grain mineral accumulation has been revealed among maize accessions, suggesting that traits of mineral accumulation are complex quantitative traits (Šimic et al. 2012; Qin et al. 2012; Baxter et al. 2013). Therefore, QTL analysis could greatly accelerate genetic modification of mineral amount in maize grain by marker-assisted selection (MAS) and even discovery of underlying candidate genes.

Several QTL studies have been conducted in maize for grain mineral concentration (Garcia-Oliveira et al. 2009; Zhou et al. 2010; Lung’aho et al. 2011; Qin et al. 2012; Šimic et al. 2012; Baxter et al. 2013, 2014). Stable or consistent QTL across different environments could only be identified in some of these above studies. For example, Baxter et al. (2013) identified 27 QTL for 9 mineral concentrations with most expressed in at least two of the six investigated environments. For micro-minerals, Lung’aho et al. (2011) identified 3 QTL for Fe concentration across three environments and Qin et al. (2012) identified 3 QTL for Fe and Zn concentrations across two environments. For macro-minerals, Šimic et al. (2012) identified a major QTL for phosphorus (P) concentration across four environments. However, none of common QTL located in the same genomic region was detected across populations, suggesting that QTL for maize grain mineral concentration were mainly determined by the specific genetic background. Therefore, analysis of QTL using different genetic resources is necessary to enrich number of QTL for controlling mineral accumulation in maize grain.

QTL for mineral accumulation in maize grain are strongly influenced by environment (Šimic et al. 2012; Baxter and Dilkes 2012; Baxter et al. 2013). Soil conditions (location) and climatic factors (year) are considered two important factors for crop production (Pacini et al. 2003). The impact of soil conditions on mineral accumulation has been investigated in Arabidopsis by correlating changes in elemental concentrations to different soil types (Buescher et al. 2010; Baxter et al. 2012; McDowell et al. 2013). Climate across different seasons could affect plant growth and grain yield, which might directly or indirectly affect mineral accumulation in grain. Thus, it is worthy to evaluate the environmental effects as location × years on maize mineral accumulation. Concentration was widely applied for evaluating mineral accumulation in crop grain. Content per kernel (g/kernel; calculated as concentration × kernel weight) was also considered to be important for evaluating mineral amount in a kernel. Despite few studies, analysis of phenotypic and genetic relationship between both concentration and content in a population allows to clarify a dilution effect due to the size of kernel (Cakmak et al. 2010; Imtiaz et al. 2003). Moreover, mineral yield [mineral harvest yield per hectare (g/ha): calculated as concentration × grain yield] was important for evaluating the economic value for global micronutrient malnutrition.

In the present study, a recombinant inbred line (RIL) population was used to map main-effect QTL for the accumulation traits (concentration, content, and yield) of seven minerals [Cu, Fe, Mn, Zn, potassium (K), magnesium (Mg) and P] in maize grain based on field measurements. The objectives were to (1) evaluate mineral concentration, content and yield in the RIL population; (2) analyze their consistency across environments using two locations across two years; and (3) map QTL for mineral concentration, content, and yield in this population.

Materials and methods

Plant materials and field design

A population of 218 F8RILs derived from the cross Ye478 × Wu312 was used in this study (Liu et al. 2011). RILs and their parents were grown in two locations Beijing (BJ, 40°06′N latitude, 116°11′E longitude) and Gansu (GS, 38°37′N latitude, 102°40′E) for 2 years 2009 and 2010 (Supplementary Table S1). Before planting, six soil samples were taken from 0 to 30 cm soil layer for each environment, mixed, air-dried, and sieved to remove undecomposed plant material, and used to measure the chemical properties (Supplementary Table S1). Organic matter was determined by K2Cr2O7 method (Walkley 1947), and total nitrogen (N) by Kjeldahl procedure (Bremner 1996). Available P (Olsen-P) was extracted by NaHCO3 and determined using a spectrophotometer (Olsen et al. 1954). Potassium (K) was extracted by NH4OAc and determined at flame photometer (Van Reeuwijk 1992). Micro-minerals (Cu, Fe, Mn and Zn) were extracted with DTPA (diethylenetriaminepentaacetic acid) solution (Lindsay and Norvell 1978) and served for concentration measurement by Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES, Perkin-Elmer, USA) (Xue et al. 2014).

The field was arranged as split-plot design. RILs and parents were planted in one-row plots with three replications. The rows were 4 m long containing 13 plants, with a plant density of 60,000 per hectare. The space between rows was 0.5 m. Seeds were sown in April and May in BJ and GS, respectively, and harvested in September for both locations (Supplementary Table S1). All fertilizers were applied before sowing, except N was supplied with two applications: 50 % at sowing and 50 % at the V6 (6 leave) stage. Rainfall during the growing season in BJ ranged from 400 to 600 mm (Supplementary Table S2), which was sufficient to avoid water deficit stress. Annual precipitation in GS was only 100–300 mm, which was insufficient for maize growth. So, 500 mm irrigation water was additionally supplied. Turf machinery and weeding were applied before sowing to ensure seedling establishment. Pest management was applied at V6 (six expanded leaves) and V8 stages in BJ and GS, respectively.

Phenotyping for quality traits

The analysis methods used are almost identical to those described by Xue et al. (2014). At harvest, five plants were chosen from the middle of each row for trait evaluation. All ears were harvested for evaluation of grain yield (GY), which was expressed as grain weight per hectare. Then, 20 kernels were collected from the middle of each ear, rapidly washed with deionized water and oven-dried at 70 °C to determine kernel weight (KW), which was expressed as dry weight per kernel. For mineral analysis, collected kernels from each row were mixed and ground with a stainless steel grinder GENO-2000 (Spex, Pittsburg, CA, USA) and 0.5 g powder was digested with HNO3–H2O2 in a microwave accelerated reaction system (CEM, Matthews, NC, USA). Mineral concentrations were determined by ICP–OES.

Data analysis

Least square mean (LSMEAN) of replicates was used for QTL analysis of measured parameters from each environment. Then, these data were analyzed by analysis of variance (ANOVA) using the SAS statistics system. For each phenotypic value, the following linear model was used:

$$Y_{ijkr} = \mu \, + \, G_{i} + \, L_{j} + \, Y_{k} + \, G \, \times \, L_{ij} + \, G \, \times \, Y_{ik} + \, L \, \times \, Y_{jk} + \, G \, \times \, L \, \times \, Y_{ijk} + \, e_{ijkr}$$

where observation Y ijkl is the plot-based phenotype as sum of the mean (μ), the genetic effect (G) of the Ith line, the effect of the Jth location (L), the effect of the Kth Year (Y), and their respective interactions G × L ij  + G × Y ik  + L × Y jk  + G × L × Y ijk and the error e ijkr . The PROC MIXED procedure (SAS) was used to estimate LSMEAN values, which were used to analyze trait Pearson correlations (Pillen et al. 2003).

Heritability for each trait was calculated from an ANOVA fitting effect of genotype (G), environment (E; two locations × two years were combined into 4 environments), and G × E interactions, as

$$H^{2} = \sigma_{G}^{2} / \, (\sigma_{G}^{2} + \sigma_{GE}^{2} / \, n + \sigma_{E}^{2} /nr) \times 100$$

where H 2 is broad sense heritability, σ 2 G is genotypic variance, σ 2 GE is genotype × environment variance, σ 2 E is error variance, n is number of environments and r is the number of replications (Nyquist 1991).

Using 184 polymorphic markers, a genetic linkage map was constructed for this RIL population with a total length of 2084 cM and an average interval length of 11.3 cM in our previous work (Liu et al. 2011). QTL were detected by composite interval mapping (Zeng 1994) using Windows QTL Cartographer version 2.5 (Model 6) (Wang et al. 2012). The threshold LOD value was determined with 1000 permutations at P = 0.05 level (Churchill and Doerge 1994). QTL positions were assigned underneath maximal LOD scores. Loci detected within the average interval (11.3 cM) in different environments were considered to be the same locus for each trait.

Results

Grain mineral concentration, content and yield in parents and RILs

All RILs and parents were grown to maturity in four environments: GS09, GS10, BJ09 and BJ10 (Supplementary Table S1). Grain mineral accumulation traits including concentration, content, and yield were investigated for four different micro-minerals (Cu, Fe, Mn and Zn), and three different macro-minerals (K, Mg and P). GY and KW were also evaluated.

Between the two parents, line Ye478 had significantly higher GY and KW compared to Wu312 across all environments (Table 1). For most minerals, the concentration showed similar levels in both parents, while the content and yield were higher in Ye478 with a higher KW and GY. However, Fe concentration was significantly higher in Ye478 across all environments, except in BJ09, and the P and Zn concentrations were significantly higher in Wu312 in BJ09.

Table 1 Parent performance of mineral concentration, content and yield as well as kernel weight (KW) and grain yield (GY) in maize inbred lines Ye478 and Wu312 under four environments
Full size table

Among RILs the values of all investigated traits segregated continuously and approximately fit normal distributions with absolute values of both skewness and kurtosis being less than 1.0. All traits showed transgressive segregation in both directions (lines with lower values than the lowest parent or higher values than the highest parent) (Supplementary Fig. S1). These results suggest presence of multiple genes controlling the investigated traits. Average KW was similar across environments, while GY and some minerals markedly varied among different environments (Table 2). Grain yield in GS was similar in both years and higher than that in BJ. At BJ, the lowest GY was found in 2010, likely a result from drought stress at maize silking stage (approx. 25 % less than average monthly rainfall in July 2010: Supplementary Table S2). Concentration and content of most minerals (except Cu and Zn) were similar across environments with the exception of BJ10 where higher value was found likely due to the concentrating effect by its lower GY (Table 2). Mineral yields were higher for all elements in GS than BJ.

Table 2 Grain mineral concentration, content and yield as well as kernel weight (KW) and grain yield (GY) in the RIL population grown in four environments
Full size table

Phenotypic variation and heritability for mineral concentration, content, and yield

The effect of genotype was significant for all traits (α = 0.05) (Table 3). Effects of environment (year and location) and genotype × environment were also significant, suggesting the presence of strong environmental effects on mineral accumulation in maize grain across these two locations and years. Within environmental effects, location had stronger effects than year for GY, KW, and all mineral yields. Cu and Zn had higher location effects for both concentrations and contents, while Mn, K and P showed stronger year effects. Fe and Mg had higher year effects for their concentrations, but lower for contents.

Table 3 Analysis of variances (ANOVA) and heritability (H 2) on grain mineral concentration, content and yield as well as grain yield (GY) and kernel weight (KW) in the RIL population grown in four environments
Full size table

Despite the lower level of accumulation, Cu and Mn had higher coefficient of variation (CV %) values (18.2–28.8, 22.3–31.9, 33.4–55.3 %) for concentration, content and yield, respectively (Table 2). The heritability (H 2) of different traits varied from 48 to 85 % with an average of 70 % (Table 3). Mineral concentration and content had similar H 2 at higher levels of 72–85 %, while mineral yield had lower levels of H 2 ranging from 48 to 63 %.

Correlations of mineral concentration, content, and yield

Significant positive correlations were found for grain concentration between several minerals (Table 4). Some of the minerals were highly correlated while many had weak positive correlations. Among these minerals, P had the closest correlation to Mg concentration (r = 0.65, P < 0.01), and medium correlation to Fe, Mn, Zn and K concentrations (r = 0.39–0.48, P < 0.01). All mineral concentrations showed no significant correlation to GY or KW. Close correlations (significantly positive) were observed for content and yield between each two minerals (r = 0.33–0.96, P < 0.01). Kernel weight and GY showed positive correlations to all mineral contents (r = 0.29–0.86, P < 0.01) and yields (0.69–0.89, P < 0.01), respectively.

Table 4 Correlation coefficients (r) between each two traits for grain yield (GY), kernel weight (KW) and grain mineral concentration, content and yield using LSMEAN values across four environments
Full size table

Significant positive correlations (r = 0.22–0.80, P < 0.01) were observed for a mineral between each two of concentration, content and yield, with an exception of a lower correlation between Zn concentration and Zn yield (r = 0.17, P = 0.013) (Supplementary Table S3). Additionally, the correlations between concentration and content were higher (r = 0.42–0.80, P < 0.01) than those either between concentration and yield (r = 0.17–0.50, P < 0.05) or between content and yield (r = 0.27–0.54, P < 0.01).

Identification of grain yield, kernel weight and grain mineral accumulation QTL

LOD threshold values were 2.9–3.3 for different traits (Supplementary Table S4). Among the 23 investigated traits, we identified 74 QTL in total for 21 traits across four environments (Fig. 1; Supplementary Table S5). No QTL were found for Cu concentration and Fe yield. The percentage of explained variance for QTL varied from 5.84 to 38.14 %. Two QTL, qZnCT1 and qMgCC10, explained more than 30 % of the phenotypic variance with 31.4 and 38.1 %, respectively. Another four QTL explained between 20 and 30 % phenotypic variance (Supplementary Table S5). Five and six QTL were identified for GY and KW, respectively. For the mineral traits, 24, 27, and 12 QTL were identified for concentration, content and yield, respectively.

Fig. 1
figure 1

Chromosome localization of QTL for grain yield, kernel weight, and grain mineral concentration, content and yield in maize detected in Ye478 × Wu312 RIL populations. Each QTL to the left of the linkage map are above the LOD threshold for composite interval mapping (CIM) analysis. QTL are marked by its environment name with different sharp for different traits. Reference mark, hash, filled triangle, filled star, filled square, filled lower right triangle, filled circle, filled inverted triangle and filled diamond represented grain yield (GY), kernel weight (KW) and iron (Fe), manganese (Mn), zinc (Zn), copper (Cu), potassium (K), phosphorus (P) and magnesium (Mg) traits, respectively. Red, blue and yellow letters represent grain mineral concentration, content and yield, respectively. Chromosome Bins with yellow background represented important regions for controlling grain mineral traits with several QTLs co-localized here

Full size image

QTL repeatedly detected across the different environments were considered as stable QTL (sQTL). About 77 % QTL (57 out of the 74 QTL) were only detected in a single environment, while other 17 QTL were detected in at least two environments. As a consequence, 8 sQTL were identified based on these 17 QTL with one (sQTL3.1) detected in three environment and 7 in two environments (Table 5). sQTL3.1 was detected for KW located at chromosome Bin 3.08/3.09 from three environments (GS09, GS10, and BJ09), explaining phenotypic variance of 20.8, 16.8, and 27.3 %, respectively (Table 5; Supplementary S5). Another sQTL(sQTL3.3) for K content was also detected at Bin 3.08/3.09, with explained phenotypic variance of 6.2 and 27.8 % in GS10 and BJ09 environments, respectively (Table 5; Supplementary Table S5). Moreover, three sQTL were identified for Mn concentration at Bin 1.04/1.05, 2.03/2.04 and 4.05/4.06 (Table 5). sQTL for Mg and Zn concentrations, Zn content were also detected.

Table 5 Summary of stable QTL (sQTL) for all investigated traits across four environments GS09, GS10, BJ09 and BJ10
Full size table

Discussion

Genetic and phenotypic relationship between mineral concentration, content, and yield in maize grain

Enhancement of mineral accumulation (biofortification) in crop grain is one of the efficient ways to solve global nutrient malnutrition. One key step is to detect favorable QTL/genes/alleles. Several QTL mapping studies for maize grain minerals have detected QTL distributed across the ten chromosomes, and focused primarily on the most important minerals, Fe and Zn (Zhou et al. 2010; Lung’aho et al. 2011; Šimic et al. 2012). Beside Fe and Zn, QTL for Mg and P concentration were analyzed to detect three P-related QTL co-localizing on chromosome 3 (Šimic et al. 2012). A comprehensive study on 9 minerals in maize grain was also recently reported through single-kernel ionomic profile (Baxter et al. 2013). This work demonstrated the positive correlations among grain mineral concentrations with high correlation coefficients between Fe and Zn, P and Mg with no correlation found between mineral concentration and KW. In rice grain, high phenotypic correlation and co-localization of QTL for Mg and P concentration were also found (Zhang et al. 2014). Similarly, in this work, we detected significant correlations between each pair of mineral concentrations (except Cu) with the closest correlations between P and other minerals (Table 4). We also failed to show any significant correlation between mineral concentration and either KW or GY (Table 4). Moreover, we could not find any QTL co-location between mineral concentration (except Mn and Mg) and either KW or GY (Supplementary Table S6). These results indicated that the correlations of mineral concentration exist in crop grain, but correlations existing between mineral concentration and kernel mass traits were low or nonexistent. Thus, it is possible to increase several mineral concentrations as a whole without affecting grain yield in crop.

In contrast to the previous studies focused on grain concentration, we analyzed three mineral accumulation traits: concentration, content, and yield. QTL analysis for both concentration and content helped to identify co-localized concentration and content loci for each mineral. Four chromosomal regions were identified containing both concentration and content QTL (Supplementary Table S6). QTL for Mg concentration (qMgCC3.1 and qMgCC6) and content (qMgCT3 and qMgCT6) were found at Bin 3.04 and Bin 6.07/6.08, respectively. Co-localization of concentration and content QTL were also found for Mn and Zn at Bin 1.04/1.05 and Bin 5.05, respectively. Together, the fact that the favorable alleles for these four pairs QTL came from the same parent (Supplementary Table S5), suggests genetic correlations between mineral content and concentration in maize grain.

Environmental influences to mineral accumulation in maize grain

Mineral accumulation in maize grain is a complex trait affected by a number of factors, including genotype, environmental conditions and their interactions (House 1999; Baxter 2009; Baxter and Dilkes 2012). Plant breeders usually evaluate genetic materials in several environments to minimize the environmental effects and identify stable alleles (and genotypes). In this work, four environments with two locations × two years were applied to test the environmental effects on mineral accumulation in maize grain. The ANOVA showed significant (P < 0.01) effects for both year and location (Table 3) for all investigated mineral concentrations, contents and yields, in agreement with previous studies (Qin et al. 2012; Baxter et al. 2013).

Soil conditions and regional climatic factors considerably impact the environmental performance for crop production (Pacini et al. 2003). Climatic conditions during maize-growing season differed in these four experimental environments (Supplementary Table S2). BJ10 was extremely dry, which resulted in approximately 70 % yield loss compared to the average yield from the other three environments (Table 2). Higher values for all tested mineral traits were found in BJ10, which may be attributed to a concentration effect, i.e., reverse dilution effect due to lower grain yield and lower carbohydrate accumulation. However, the genetic features for the tested mineral traits in BJ10 did not differ from those in other environments because of the similar variance coefficients (Table 2), number of QTL detected (Supplementary Table S5), number of sQTL detected (Table 5), and number of co-localized QTL identified (Supplementary Table S6). Thus, BJ10 environment was able to be included for QTL analysis as a comprehensive experiment despite the presence of drought stress.

Soil properties for the two locations GS and BJ showed differences in mineral availabilities. Available Zn in BJ was higher than that in GS. Consistently, maize grain Zn accumulation in BJ was also found to be higher (Table 2). In addition, ANOVA analysis found higher location effects for both Zn concentration and content, but not for other elements (except Cu) (Table 3), further suggesting that soil Zn availability plays a critical role for its accumulation in maize grain. These results were consistent with other studies (Sharma and Bapat 2000; Chahal et al. 2005), which suggested that Zn accumulation in maize grain can be significantly improved by increasing soil Zn availability, e.g., through Zn fertilizer. Indeed, a recent work has found that Zn deficiency could be corrected by the application of Zn fertilizer to soil (Kumar and Bohra 2014).

For other elements, e.g., Cu, Fe, and Mn, the effects of soil element availability for grain mineral accumulations have not been resolved. Kumar et al. (2009) reported that Cu concentrations in leaves, grain, and straw increased significantly with an increase in the level of applied Cu in soil. Antunović et al. (2003) found that the mineral concentrations of Mn and Fe in maize grain were not in connection to the status of the corresponding elements in the experimental environments. It was presumed that interaction with other elements (e.g., Ca or Mg) could be responsible for this inconsistence for Cu, Fe, and Mn. We found that soil available Cu, Fe, and Mn were higher in GS location (Table 1). However, grain mineral accumulation in GS was similar to that in BJ (Table 3), except that BJ10 had higher mineral concentrations due to drought stress. Thus, our results suggested that soil availability of Cu, Fe, and Mn has little effect on their accumulations in maize grain.

Important QTL clusters for mineral accumulation in maize grain

Limited information is available about the genetic control and molecular mechanism contributing to high accumulation of minerals in maize grain. QTL analysis provides a powerful tool to identify chromosomal locations of genes suitable for breeding programs. In this work, we conducted a QTL analysis in four environments and found several chromosomal regions which contained stable QTL (across different environments) within a grain mineral trait or contained co-localized QTL across different traits (Fig. 1; Supplementary Table S6). In general, 52 QTL (70 %) out of the 74 detected QTL fell into 12 QTL clusters in which 3–6 QTL were grouped for different mineral traits or from different environments within a trait.

At chromosome Bin 6.07/6.08, five QTLs were detected from GS10, with one for KW and four for Zn, K, Mg and P contents (Supplementary Table S6). High phenotypic correlations between mineral contents and KW suggested the presence of corresponding QTLs for both traits in this QTL cluster. Moreover, in this region, we also detected a QTL for Mg concentration from BJ09 besides that for Mg content (Fig. 1; Supplementary Table S5, S6). Zhou et al. (2010) identified two QTL located in this region for Zn concentration using different populations. The fact of QTL co-localization for mineral concentration and content, and KW, suggested an important QTL at Bin6.07/6.08 for controlling mineral accumulation in maize grain. It also implicated that improvement with this QTL region could enhance grain mineral content via concentration and KW simultaneously.

Identified QTL for maize grain mineral accumulation traits have been always determined by specific genetic populations (Garcia-Oliveira et al. 2009; Zhou et al. 2010; Lung’aho et al. 2011; Šimic et al. 2012; Baxter et al. 2013, 2014). In this work, we identified 3 sQTL for Mn concentration at Bin1.04/1.05, 2.03/2.04 and 4.05/4.06. Among them, a sQTL (sQTL4.1) at Bin 4.05/4.06 contained 2 QTL for Mn concentration from environments GS10 and BJ10 explaining 12.3 % and 15.3 % of the phenotypic variance, respectively (Supplementary Table S5). The existence of QTL affecting Mn concentration was also reported in the same region from another maize population (Zhou et al. 2010). Moreover, a QTL for Mn concentration in rice grain has been reported on chromosome 2, which is a syntenic region of maize Bin 4.05/4.06 (Schnable et al. 2009; Zhang et al. 2014). These results suggest that Bin4.05/4.06 may contain common QTL for Mn concentration across different genetic background and even different crop species.

Author contribution statement

Conceived and designed the experiments: Lixing Yuan, Fanjun Chen and Riliang Gu. Analyzed the date: Riliang Gu, Fanjun Chen, Xin Wang and Jordon Pace. Performed field experiments: Fanjun Chen, Jianchao Liu, Pengcheng Li and Qingchun Pan. Performed lab measurement: Bingran Liu. Wrote the paper: Riliang Gu. Made the final revision: Lixing Yuan, Guohua Mi, Thomas Lübberstedt, Jordon Pace and Ayaz-Ali Soomro.