Rational regional distribution of sugarcane cultivars in China

Knowing yield potential and yield stability of sugarcane cultivars is of significance in guiding sugarcane breeding and rationalising regional distribution of sugarcane cultivars. In the present study, a heritability-adjusted genotype main effect plus genotype × environment (HA-GGE) biplot program was used to analyze the cane and sucrose yields of 44 newly released sugarcane cultivars at eight pilot test sites. The cane and sucrose yields of nine cultivars were higher than those of the control cultivar ROC22. From the perspective of cane yield, cultivars FN 40 and YZ 06–407 were well adapted to a wider range of conditions and produced relatively high cane yields in several pilot sites. From the perspective of sucrose yield, cultivars LC 03–1137, FN 38, FN 41, MT 01–77 and LC 05–136 were well adapted to a wide range of conditions and produced relatively high sucrose yields. Based on these results, three high yielding and widely adapted cultivars, namely, FN 39, LC 05–136, and YZ 05–51 were recommended for production in three major Chinese sugarcane planting areas. The results will provide a theoretical basis for recommending the effective use and rational regional distribution of sugarcane cultivars in China.

Scientific RepoRts | 5:15721 | DOi: 10.1038/srep15721 repeatability) was relatively low. Jackson et al. 21 evaluated the response and association of sugarcane with environmental factors. They found that soil nutrient levels, or other related factors, might be responsible for family × environment interactions in sugarcane. Zhou et al. 22 determined the variance components of G × E and evaluated their relative importance and implications in the breeding programmes. The results showed that the genotype by crop-year variance component was more significant for the rainfed than irrigated programmes, while genotype by location by crop-year interaction was more significant for yield than sucrose content. Besides, breeding and selecting for sugarcane ratooning ability were critical in rainfed regions, highlighting the complexity associated with sugarcane regional breeding. Additive main effects and multiplicative interaction (AMMI) and GGE biplot models are the two most popular statistical methods for data analysis in crop cultivar trials 23,24 . Bissessur et al. 25 used AMMI to depict the sugarcane varieties for both wide and specific adaptations in Mauritius. The GGE biplot method uses plots to analyze diallel data. One key feature of the GGE biplot is to only display the genotype's main effect (G) and G × E interaction. In addition, Yan used an average tester coordinate to simplify the GGE biplot by comprehensive display of information in a two-way data table [26][27][28][29][30] . Yan 24 proposed the use of the GGE biplot method in regional wheat trials. Luo et al. 31 applied the GGE biplot to analyze the data from seven-cultivar trials at seven pilot sites and identified two cultivars FN 39 and YZ 05-51 with relatively high yield and stability. Zhou et al. 32 applied the GGE biplot to analyze experimental data from nine soybean strains at five pilot sites of Heilongjiang Province during 2008-2009 and found that strain Tender 0022111 exhibited high yield with good stability. Zhang et al. 33 analyzed data from national oat regional trials during 2006-2008 and identified three oat strains with high and stable yields. HA-GGE is a heritability adjusted GGE biplot model 30,34 . Xu et al. 34 used the HA-GGE biplot to evaluate the test environments among different cotton planting regions in the Yangtze River basin. They were able to divide the cotton planting regions into different ecological zones based on lint yield 35 .
Pilot demonstrations of sugarcane cultivars provide the basis for regional distribution and are of great importance in accelerating inter-provincial promotion of superior sugarcane cultivars, increasing the income of sugarcane farmers, and promoting the sustainable development of sugarcane industry. Currently, no evaluation has been conducted in China uses the HA-GGE biplot method on yield and yield stability of newly released sugarcane cultivars. In this study, the HA-GGE biplot method was used to the analyze yield trial data involving 44 newly released sugarcane cultivars and eight national pilot experimental sites. The yield trial experiments were designed to screen sugarcane cultivars for both high yield and high yield stability and to distribute top performing cultivars to the three major sugarcane production areas in China. In addition, the discriminating power and representativeness of the test environments were also assessed comprehensively.

Results
Testing for significant differences among the yields of different cultivars at different pilot sites. Analysis of variance showed that the pilot site environmental effect (E), genotype effect (G), GE interaction and the effect of test block within the pilot site were all significant (Table 1). Pilot site (E) was the most important factor affecting yield, accounting for 43.58% (cane yield) and 44.56% (sucrose yield) of the total variations. The next is GE interaction, accounting for 33.42% (cane yield) and 30.74% (sucrose yield) of total variations. Genotype (G) only accounts for a relatively small portion of the total variations: 13.35% (cane yield) and 13.80% (sucrose yield). The G/GGE effect for cane and sucrose yields are 0.29 and 0.31, respectively (Table 1). Therefore, the effect of GE was far greater than the G effect. Based on the two year yield data collected from the eight pilot sites and 45 genotypes ( Optimal regions of sugarcane cultivars. High cane yielding cultivars were identified at different pilot sites based on the diagram of optimal regions for sugarcane cultivars. This, in turn, was based on cane yield in the HA-GGE biplot, in the sectors of FN 40 (35) and YZ 06-407 (3) (Fig. 1A). A relatively large number of pilot sites were fairly widely distributed, suggesting that these two cultivars adapted to a wide range, and their cane yields were high at four test sites, Guangdong Zhanjiang (E1), Guangxi Baise (E2), Guangxi Chongzuo (E3) and Hainan Lingao (E6). The cane yield of MT 01-77 (23) was good in Guangxi Liuzhou (E5) and also relatively high in Yunnan Lincang (E8), suggesting that MT 01-77 had a relatively high adaptability. The cane yields of FN 36 (38), YZ 03-422 (7) and FN 28 (40) were relatively low in all pilot sites. Four cultivars, namely, ROC22 (45), FN 30 (39), GT 97-69 (28) and YT 55 (15), were located inside the polygon close to the origin and were not sensitive to the environmental changes. A diagram of optimal regions for cultivars based on sucrose yield is shown in Fig. 1B (38) and ROC22 (45), were placed inside the polygon close to the origin and were insensitive to the environmental changes, which was in accordance with the claim of breeders that sucrose yield is a relatively stable trait 2 .
The stability of sugarcane cultivars. To select sugarcane cultivars suitable for rational distribution, both high yield and yield stability are important. In Fig. 2A    The representativeness of the test environments. In a biplot diagram, the length of the environment vector represents the discriminating power of the test environment, which is proportional to the vector length. The distance from the origin of the biplot to each icon of the test environment is the length of the corresponding environment vector. In terms of cane yield, the discriminating powers were high for Guangxi Baise (E2), Guangxi Chongzuo (E3) and Yunnan Dehong (E7); medium-high for Guangdong Zhanjiang (E1), Guangxi Liuzhou (E5), Hainan Lingao (E6) and Yunnan Lincang (E8); but low for Guangxi Laibin (E4) (Fig. 3A). In terms of sucrose yield, the discriminating powers were high for Guangdong Zhanjiang (E1), Guangxi Baise (E2), Guangxi Liuzhou (E5) and Yunnan Lincang (E8); medium-high for Guangxi Chongzuo (E3), Hainan Lingao (E6) and Yunnan Dehong (E7); and low for Guangxi Laibin (E4) (Fig. 3B). The representativeness of test environment refers to its consistency with the average of all other test environments in the target region or the average of all test environments, and is represented by the angle between the environment vector and the average environment vector (Average environment coordination axis, AEC axis). The smaller the angle, the higher the representativeness. The AEC axis is the ray with origin as the initial point that connects the origin to the average environment icon, which is the average of the coordinates of all test environments 10 . From the perspective of cane yield, four pilot sites, namely, Guangdong Zhanjiang (E1), Guangxi Baise (E2), Guangxi Chongzuo (E3) and Hainan Lingao (E6), were highly representative environments. Three pilot sites [Guangxi Laibin (E4), Guangxi Liuzhou (E5) and Yunnan Lincang (E8)] were intermediately representative environments. Yunnan Dehong (E7) was the least representative of its target environment. From the perspective of sucrose yield, Guangxi Liuzhou (E5), Hainan Lingao (E6) and Yunnan Lincang (E8) were highly representative environments; Guangdong Zhanjiang (E1), Guangxi Baise (E2) and Guangxi Chongzuo (E3) were intermediately representative environments; and Guangxi Laibin (E4) and Yunnan Dehong (E7) were least representative environments. When combining both the cane and sucrose yield data, six pilot test sites, namely, Guangdong Zhanjiang (E1), Guangxi Baise (E2), Guangxi Chongzuo (E3), Guangxi Liuzhou (E5), Hainan Lingao (E6) and Yunnan Lincang (E8), represented the target environments well. The overall ecological homogeneity was satisfactory and the test results were thus well targeted and of good representativeness. Two pilot sites, Guangxi Laibin (E4) and Yunnan Dehong (E7), represented the test environment relatively poorly.

3-D biplot analysis.
Due to the large number of cultivars tested, the information ratios (IR) of the third principle components for cane and sucrose yields were 1.024 and 0.928, respectively (Table 3). A 2-D biplot is not sufficient to show all the characteristics of the data. Nevertheless, the goodness of fit of the first three principle components in the 3-D biplot for cane and sucrose yields reached 72% and 67.9%, respectively, to allow a characterisation of the data (Fig. 4)

Discussion
Forty-five sugarcane cultivars were evaluated for yield potential and yield stability in this study. These included 12 YZ-and one DZ-series cultivars from Yunnan, four LC-and six GT-series cultivars from Guangxi, nine FN-and one MT-series cultivars from Fujian, ten YT-series cultivars from Guangdong, one GN-series cultivar from Jiangxi, and one check cultivar ROC 22. These cultivars were developed recently by several Chinese sugarcane breeding institutes and are highly representative products of the sugarcane breeding efforts in China. In addition, eight pilot test sites were chosen from the major sugarcane production areas for evaluation and demonstration purposes. Hence, the results of this study are of significance in guiding the selection and recommendation of superior sugarcane cultivars in China.
Nine cultivars, namely, YZ 06-407, LC 03-1137, FN 38, FN 41, DZ 03-83, MT 01-77, FN 39, LC 05-136 and YZ 05-51, produced both high cane and high sucrose yields as were predicted by the HA-GGE biplots. Five cultivars, namely, YG 40, YZ 06-189, YZ 05-49, YZ 06-80 and GN 02-70, produced relatively high sucrose yields, but lower cane yields compared to the check ROC 22. Cultivars GT 29, YT 60, FN 40 and YT 96-86 produced relatively high cane yields, but lower sucrose yields. These 18 cultivars had either cane or sucrose yield exceeding that of ROC22, and therefore were recommended for planting using a reasonable distribution of sugarcane cultivars based on local conditions. Because of local conditions in China, an ideal sugarcane cultivar would have high and stable yield and would adapt to a wide range of environments [36][37][38] . In the present study, only four cultivars, namely, YZ 06-407, FN 39, YZ 05-51 and LC 05-136, fit that definition by producing higher and more stable cane and sucrose yields than the check cultivar ROC22. Two other cultivars, FN 38 and FN 41, produced higher cane and sucrose yields than ROC22 with a relatively high stability on cane yield but a relatively low stability on sucrose yield. Still, two other cultivars, LC 03-1137 and DZ 03-83, produced higher cane and sucrose yields than ROC22 with a relatively low stability on cane yield but a relatively high stability on sucrose yield. These result are consistent with a previous report by Luo et al. 31 , who also concluded that cultivar FN 39 was of high cane and sucrose yields as well as being generally stable. The fore-mentioned eight sugarcane cultivars are of high production value and are being recommended as widely adaptable cultivars for different trial stations.
Assessment of test environments in a target region is an important step during sugarcane cultivar trials 28 . HA-GGE biplot-based assessments can directly reveal the yield and stability properties of the cultivar as well as the discriminating power of test sites 24,[26][27][28][29][30][31]34 . This helps sugarcane breeders identify test sites of good discriminating power and improve the accuracy of varietal trials [29][30][31]33,34,[36][37][38][39][40] . Test sites that were selected based on a good discriminating power and a good representativeness of all test sites in the target region were proven to be ideal sites 39 . In the current study, the HA-GGE biplot was applied in preliminary assessments of eight pilot test sites. When taking both the cane and the sucrose yields into account, six pilot sites, namely, Guangdong Zhanjiang (E1), Guangxi Baise (E2), Guangxi Chongzuo (E3), Guangxi Liuzhou (E5), Hainan Lingao (E6) and Yunnan Lincang (E8) were shown to be highly representative of their target environments. The overall ecological homogeneity of these sites was satisfactory; hence, the test results were representative of these six pilot test sites. In contrast, two pilot test sites, Guangxi Laibin (E4) and Yunnan Dehong (E7), were not representative of their test environments.
How to assess both yield and yield stability of newly released sugarcane cultivars is a difficult research topic. Nevertheless, this information is very critical for guiding the rational distribution of sugarcane cultivars, the purpose of sugarcane breeding research. By using the HA-GGE biplot, ten newly released sugarcane cultivars, namely, DZ 03-83, FN 38, FN 39, FN 40, FN 41, LC 03-1137, LC 05-136, MT 01-77, YZ 05-51 and YZ 06-407, produced both higher cane and sucrose yields than the check cultivar ROC22. Eight cultivars are being recommended for rational distribution based on local conditions. Three cultivars were recommended for commercial production in the three dominant sugarcane production areas in China, namely, the Central and Southern Guangxi (LC 05-136), the Southwestern Yunnan (YZ 05-51), and the Western Guangdong and Northern Hainan (FN 39). Cultivars DZ 03-83 and YZ 06-407 were recommended for commercial production in all sugarcane plant areas except for Guangxi Chongzuo and Hainan Lingao. Cultivars MT 01-77, FN 38 and FN 41 were recommended for commercial production in Guangdong Zhanjiang, Guangxi Baise and Guangxi Chongzuo. LC 03-1137 was recommended for production in Guangxi Liuzhou, Hainan Lingao, and Yunnan Lincang. LC 03-182 was recommended for production in Guangxi Laibin and Yunnan Dehong. These results are of significance in guiding sugarcane breeding and rational regional distribution of sugarcane cultivars in China.

Methods
Ethics Statement. We confirm that no specific permits were required for the described locations/ activities. We also confirm that the field studies reported did not involve endangered or protected species. Pilot sites. The field evaluation trials were conducted at the field pilot sites of eight sugarcane integrative experiment stations located in eight counties (or cities) of four provinces (synonymous with autonomous regions for this paper) in China, namely, Guangxi, Yunnan, Guangdong, and Hainan. These pilot sites were designated as GDZJ (E1) in Guangdong Zhanjiang, GXBS (E2) in Guangxi Baise, GXCZ (E3) in Guangxi Chongzuo, GXLB (E4) in Guangxi Laibin, GXLZ (E5) in Guangxi Liuzhou, HNLG (E6) in Hainan Lingao, YNDH (E7) in Yunnan Dehong, and YNLC (E8) in Yunnan Lincang. The elevation, latitude, longitude, precipitation, soil type, and other environmental parameters of these pilot sites 36 are described in Table 4 .   Experimental methods. Test plot sites were selected that had flat terrain with uniform but representative local fertility levels. All had a convenient irrigation facility and grew the same crop prior to the test. All were at a considerable distance from tall buildings. A randomized block design was used with three replications. Each block had four 8-meter long rows with 1.1-meter spacing. Planting started from late February to early March at a rate of 105,000 cutting per hectare. Field management levels were the same as those for the local commercial fields. Cultivation, plowing, fertilization, irrigation and pest control were carried out on time. For field operations, the same technical measures were completed at the same pilot sites on the same day. Trials were conducted for two consecutive years to include data from two plant-cane and the first ratoon crops. The sugarcane crops were harvested in March of the following year. Before harvest, yield estimates were obtained as follows: the stalks in the two middle rows, the experimental rows, central row were manually harvested and mechanically weighed; the actual harvest area was measured, and the number of stalks harvested was counted. The cane yield per unit area was then calculated based on the area, number of stalks, and stalk weight. In the middle of each month from November to March of the following year, six millable stalks were collected and sucrose content was measured using the polarimetry method 37 . Sucrose yield was then calculated based on average sucrose content and cane yield using Equation (1): Sucrose yield (t ha -1 ) = Cane yield (t ha -1 ) × Sucrose content (%) Data analysis. The GGE-Biplot software was used for HA-GGE biplot analysis 26,40 . The cane yield data obtained from all pilot sites were summarized into a cultivar-plot site two-way table, and the average trait value of the corresponding cultivar at the corresponding trial site was taken as each value. The general model for GGE biplots is 26,40 : The response value (G) observed in target environment j' due to indirect selection in test environment j is calculated 26,40 : In the HA-GGE biplot model, the correlation between indices was evaluated by the cosine of the angle between adjacent indices. Starting from one index vector the cosine of the angle between the other index vector and the starting index vector is the correlation coefficient of two indices. The position of projection of the sugarcane cultivar or pilot site on the average-tester (AT) axis is used to assess the average performance of the cultivar or the representativeness of the site. The length of the projection of the sugarcane cultivar or plot site on the AT axis is used to determine the stability of the cultivar or the discriminating power of the plot site 30 . In the "appropriate combination of genotype and environment" in a functional diagram of the HA-GGE biplot, if the icons of the outmost cultivars are connected to form a polygon, the icons of all cultivars will be enclosed within the polygon. If a line is perpendicular to each side of the icons, all cultivars will be enclosed within the polygon. If a line perpendicular to each side of the polygon is drawn from the origin of the biplot, the polygon will be divided into different sectors, and all pilot sites within the same sector form a test site combination. Within each sector, the cultivar located at the polygon vertex is the cultivar with the best performance in all test sites of that sector, i.e., the best cultivar of that test site combination 30 .
To assess the test environment with a high accuracy, in the HA-GGE biplot the parameter used to assess the discriminating power of the test environment is the square root of the heritability, which approximately equals the length of the test environment vector. The parameter used to assess representativeness is the genetic correlation coefficient (r), which approximately equals the cosine of the angle between the test environment vector and the average environment axis. The ideal index of the test environment is the product of the square root of the heritability of the length of the vertical projection of test environment vector on the average environment axis and the genetic correlation coefficient r [26][27][28][29][30] .
Analysis of variance was conducted based on the general linear model for randomized complete blocks and the significance of various effects was based on F test against the experimental errors within environments 40 .