雙親本雜交與輪迴選種結果皆顯示非選拔性狀 (抽穗期或株高) 較 327 品系 早和矮，但最顯著的遺傳增進發生在早世代選拔，之後的變化幅度較小，僅輪迴 選種 BLUE 選拔初始親本族群隨著循環數增加株高不斷降低，從雙親本雜交選出 之親本校正外表型判斷應是此族群產量高之品系同時具備早抽穗、株高矮特性之 故 (圖九)。或許因此族群材料為進入區域產量試驗之品系，已為育種選出之優良 品系，故具高產又早抽穗之違背一般生理原則特性。Yamamoto et al. (2016) 不只 模擬輪迴選種，在每個 Cycle 另外模擬產生自交系 (inbred lines, ILs)，結果顯示 自交系的性狀表現皆在其衍生之族群附近，隨著輪迴選種循環數增加才有明顯增 進，否則只雜交一次與選拔一次的自交系表現無法達到育種目標，顯然如輪迴選 種般多次雜交重組與選拔是必要的。本研究輪迴選種以 GEBV 選拔親本之 Cycle 5 平均產量和最大值分別為 5948 kg/ha 和 6195 kg/ha (表十三)，高於雙親本雜交 10 個組合各自 F6族群產量平均值 5641-5926 kg/ha 和 9 個組合之最大值 5936-6232 kg/ha (表十)，兩者同為經過 5 次選拔。雖然輪迴選種可能因族群多樣性下降
導致產量增長幅度不大，但平均值仍略高於 F6族群，且優良個體表現較佳，顯然
多次雜交重組仍有效改良族群表現並產生優良個體。玉米等異交作物常以輪迴選 種改良族群再選出優良個體做雙親本雜交，水稻雖為自交作物但仍可透過輪迴選 種獲得親本材料，若族群具有雄不稔基因可便利雜交工作的進行 (Grenier et al.
2015)，再藉由 GS 能加速世代推進、縮小種植規模並節省大量調查外表型工作。
doi:10.6342/NTU201701806
30
五、結論
本研究以模擬方式探討在世代推進的情形下 GS 在水稻育種計畫中之應用，
證實在理想狀況下，基因體選種於雙親本雜交之產量性狀選拔效率優於外表型選 種，且基因體選種在輪迴選種中能有效改良族群抽穗期與產量性狀表現、快速累 積有利對偶基因頻度。未來若欲將 GS 應用在實際育種計畫中，須首先了解訓練 族群與育種族群關係及目標性狀最適統計模型以提高預測準確度，可參考本研究 之流程，先以 PCA 分析族群結構，接著以 GWAS 分析目標性狀之遺傳結構，再 透過交叉驗證選出各性狀最佳統計模型。接著或可先以短期輪迴選種快速累積有 利對偶基因頻度改良族群，從中選拔優良品系進行純化或選出優良親本進行雜 交，於 F2早世代及 F6以 GS 方式節省大量人力且較精確選拔出優良品系。
doi:10.6342/NTU201701806
31
表一、327 品系在各年份、季節的品系數、區集數和各性狀缺值數
Table 1. The No. of lines, the No. of blocks, and the No. of missing data for 327 lines in each year and each season
Year Season lines with data No. Blocks No. FL missing No. PH missing No. Yield missing
2011 dry 326 3 0 0 4
2011 wet 326 3 1 0 1
2012 dry 325 2 0 0 5
2012 wet 323 3 2 0 3
doi:10.6342/NTU201701806
32
表二、GEBV 預測產量前十名品系及校正外表型產量 BLUE 前十名品系 Table 2. The top ten lines in yield of GEBV and BLUE
GEBV BLUE
Entry Yield Entry Yield B1024 6371.054 B1024 6247.8 B1027 6370.615 B1027 6091.709 B1026 6160.164 A1271 5989.436 A1271 6115.114 B1026 5914.709 B1019 6018.69 B1118 5894.527 M1476 6003.656 A1274 5816.527 A1338 5993.299 M1470 5812.963 M1470 5984.14 M1476 5808.073 B1025 5983.807 B1147 5752.982 M1405 5961.804 B1070 5685.436 灰底為兩者重疊之品系。
Lines overlapped in both estimation are highlighted in gray.
doi:10.6342/NTU201701806
33
表三、產量 GEBV 前十名品系及校正外表型產量 BLUE 前十名品系其產量及抽穗期平均值於各季原始排名前二十名之整理 Table 3. Summary for the GEBV-top 10 and the BLUE-top 10 which were among the top-20 in each trial
GEBV BLUE
Entry Yield top 20th season FL top 20th season Entry Yield top 20th season FL top 20th season B1024* 2011D, 2011W, 2012D, 2012W 2011W B1024* 2011D, 2011W, 2012D, 2012W 2011W
B1027* 2011D, 2011W, 2012D, 2012W 2011D, 2011W, 2012W B1027* 2011D, 2011W, 2012D, 2012W 2011D, 2011W, 2012W B1026* 2011W, 2012D, 2012W 2011D, 2011W, 2012W A1271* 2011D, 2011W, 2012D, 2012W 2012D
A1271* 2011D, 2011W, 2012D, 2012W 2012D B1026* 2011W, 2012D, 2012W 2011D, 2011W, 2012W B1019 2011W, 2012W 2011W, 2012W B1118 2011W, 2012D, 2012W 2012D, 2012W M1476 2011D, 2011W, 2012D 2011D, 2012D A1274* 2011W, 2012D, 2012W 2011W, 2012D, 2012W A1338 2011D, 2012D M1470 2012D & 2012W NA 2012D & 2012W NA M1470 2012D & 2012W NA 2012D & 2012W NA M1476 2011D, 2011W, 2012D 2011D, 2012D
B1025 2011D, 2011W 2011W, 2012W B1147 2011D, 2012D
M1405 2011W 2012W B1070 2012D, 2012W
灰底為兩者重疊之品系。2011 與 2012 為年分，D 為乾季，W 為濕季。M1470 品系在 2012 年兩季缺值故不列入考慮。具*之品系為選出之親本。
Lines overlapped in both estimation are highlighted in gray. 2011 and 2012 are year of trial, D stands for dry season, W stands for wet season. M1470 line is not taken into consideration for its value was missing in both dry and wet seasons in 2012. The lines with the symbol * are the selected parental lines.
doi:10.6342/NTU201701806
34
表四、雙親本雜交之親本
Table 4. The parents of bi-parental crosses
Parent 1 Parent 2
Cross 1 A1271 A1274
Cross 2 A1271 B1024
Cross 3 A1271 B1026
Cross 4 A1271 B1027
Cross 5 A1274 B1024
Cross 6 A1274 B1026
Cross 7 A1274 B1027
Cross 8 B1024 B1026
Cross 9 B1024 B1027
Cross 10 B1026 B1027
doi:10.6342/NTU201701806
35
表五、輪迴選種起始親本
Table 5. The initial parents (Cycle 0) for recurrent selection
GEBV BLUE
Yield FL Yield FL
B1024 B1007 B1024 A1311
B1027 B1035 B1027 A1259
B1026 B1025 A1271 M1463
A1271 B1008 B1026 B1007
doi:10.6342/NTU201701806
36
表六、外表型變方分析表
Table 6. The ANOVA table of phenotype
Source d.f. Type III SS MS F P FL
Line 326 24652.5 75.6211 24.4444 < 0.001 Environment 3 46.0663 15.3554 4.96362 0.00195 Block (E) 7 189.668 27.0954 8.75855 < 0.001 Line × Environment 970 16011 16.5061 5.33558 < 0.001
Residuals 2265 7007 3.0936 PH
Line 326 53821.4 165.096 5.04052 < 0.001 Environment 3 673.986 224.662 6.85911 < 0.001 Block (E) 7 1257.81 179.687 5.48597 < 0.001 Line × Environment 970 55770.9 57.4958 1.75539 < 0.001
Residuals 2265 74285.7 32.7972 Yield
Line 326 4.8×108 1.5×106 4.5049 < 0.001 Environment 3 2.1×106 714760 2.20257 0.08582 Block (E) 7 9.8×106 1.4×106 4.32246 < 0.001 Line × Environment 969 6.9×108 711583 2.19278 < 0.001
Residuals 2256 7.3×108 324512
doi:10.6342/NTU201701806
37
表七、外表型資料性狀間相關係數
Table 7. The coefficients of correlation between phenotypes
2011 Dry 2011 Wet 2012 Dry 2012 Wet BLUE
r(FL, PH) 0.21 0.32 0.43 0.28 0.33
r(FL, Yield) -0.31 -0.56 -0.25 -0.15 -0.42 r(PH, Yield) -0.088 -0.26 -0.12 -0.088 -0.19 FL 為抽穗期、PH 為株高、Yield 為產量，BLUE 為校正外表型。樣本數介於 323 至 327 之間。
FL, flowering time; PH, plant height; BLUE, the adjusted phenotypic values. The sample size is between 323 and 327.
doi:10.6342/NTU201701806
38
表八、外表型資料校正前後之相關係數與廣義遺傳率 (ℎ2)
Table 8. The coefficients of correlation between trait values before and after adjustment, and the broad-sense heritability
2011 Dry 2011 Wet 2012 Dry 2012 Wet ℎ2
FL 0.88 0.92 0.91 0.88 0.528
PH 0.86 0.87 0.84 0.91 0.382
Yield 0.78 0.73 0.79 0.76 0.353
doi:10.6342/NTU201701806
39
表九、8 種統計方法於各性狀之預測準確度
Table 9. The prediction accuracy of eight statistical methods for each trait
FL PH Yield
RR-BLUP 0.595±0.015 e (0.544-0.648) BayesA 0.690±0.013 b
(0.636-0.731)
0.485±0.022 a (0.406-0.546)
0.474±0.021 ab (0.385-0.537) BayesB 0.700±0.010 a
(0.662-0.741)
0.486±0.022 a (0.405-0.544)
0.473±0.021 bc (0.395-0.54) BayesC 0.636±0.015 c
(0.584-0.68)
Mean ± standard deviation; minimum and maximum are provided between the brackets;
the letters show Tukey’s test results (α = 0.05), different letters signify significant differences. FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
Mean ± standard deviation; minimum and maximum are provided between the brackets; different letters signify significant differences (𝑝 < 0.05/21).
doi:10.6342/NTU201701806
41
表十一、GS 及 PS 之抽穗期 GEBV 結果
Table 11. The flowering time GEBV for GS and PS
FL mid-parent F2 F2 F6 F6 best line
Mean ± standard deviation; minimum and maximum are provided between the brackets; different letters signify significant differences (𝑝 < 0.05/21).
doi:10.6342/NTU201701806
42
表十二、GS 及 PS 之株高 GEBV 結果
Table 12. The plant height GEBV for GS and PS
PH mid-parent F2 F2 F6 F6 best line
Mean ± standard deviation; minimum and maximum are provided between the brackets; different letters signify significant differences (𝑝 < 0.05/21).
doi:10.6342/NTU201701806
43
表十三、輪迴選種各循環 F2族群性狀 GEBV
Table 13. GEBV for F2 population at each cycle of recurrent selection
G0 G1 G2 G3 G4 G5
Mean ± standard deviation; minimum and maximum are provided between the brackets; different letters signify significant differences (𝑝 < 0.05/21).
doi:10.6342/NTU201701806
44
圖一、外表型資料與基因型資料篩選
虛線上方為 Spindel et al. (2015) 篩選流程；下方為本研究再次篩選條件。
Figure 1. The filtering of the phenotypic data and the genotypic data
Above the dash line is the filtering process of Spindel et al. (2015); below are the additional filtering steps conducted in this study.
doi:10.6342/NTU201701806
45
圖二、雙親本雜交流程
GS 為基因體選種，PS 為外表型選種，於 F2及 F6世代以不同方法選拔產量性狀 而分成兩種路徑，其餘世代皆以 PS 進行選拔，藉此比較 GS 與 PS 兩種選拔方法 的遺傳增進效果。
Figure 2 The bi-parental cross scheme
GS, genomic selection; PS, phenotypic selection. Two paths are divided from F2
generation in order to compare the genetic improvement between GS and PS.
doi:10.6342/NTU201701806
46
圖三、雙親本雜交流程之 GS 路徑示意圖，以譜系法推進世代
Figure 3. The bi-parental cross scheme for GS by modified pedigree method
doi:10.6342/NTU201701806
47
圖四、輪迴選種流程
首先從 327 品系中選出產量前四名與抽穗期前四名之品系相互雜交，共 16 個雜 交組合，每個組合產生 1 株 F1再自交產生 24 株 F2，共 384 株 F2組成 Cycle 1 族 群。接著同樣從 Cycle 1 族群中選出產量前四名與抽穗期前四名之個體相互雜 交，每個組合產生 2 株 F1再自交產生 12 株 F2，維持相同族群大小，共 384 株 F2
組成 Cycle 2 族群，重複此步驟至產生 Cycle 5 族群為止。
Figure 4. The recurrent selection scheme
From the initial 327 lines, we selected the top four lines in yield and the top four lines in flowering time, and then cross them to produce 16 crossing combinations. Each
combination produces one F1 and then self to produce 24 F2 to form a total of 384 F2 in Cycle 1. From Cycle 1, we selected the top four individuals in yield and the top four individuals in flowering time, and then cross them to produce 16 crossing combinations.
Each combination produces two F1 and then self to produce 12 F2. A total of 384 F2
forms Cycle 2. The process was iterate till cycle 5.
doi:10.6342/NTU201701806
48
圖五、327 品系各性狀於四個環境下之表現 FL 為抽穗期、PH 為株高、Yield 為產量。
Figure 5. The performance of the 327 lines in four trials FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
49
圖六、以 PCA 針對 327 品系結合 38,639 個 SNP 所得之族群結構分析圖 左為各成分解釋變異比例，右為第一、二主成分所畫散佈圖。
Figure 6. Applying PCA to 327 lines × 38,639 SNPs for a population structure summary analysis
The left part is the variance proportion explained by each component, and the right part is the distribution of the 327 lines based on the first and the second components.
doi:10.6342/NTU201701806
50
圖七、327 品系和 38,639 個 SNP 之 GWAS 結果 QQ plot 在左、Manhattan plot 在右；
虛線為 FDR = 0.05，無虛線表示無顯著 SNP。FL 為抽穗期、PH 為株高、Yield 為產量。
Figure 7. GWAS results for 327 lines × 38,639 SNPs
QQ plot is at the left-hand side and Manhattan plot at the right-hand side.
The dotted line is signifies FDR = 0.05, and no dotted line indicates no significant SNP.
FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
51
圖七、327 品系和 38,639 個 SNP 之 GWAS 結果 (續)
Figure 7. GWAS results for 327 lines × 38,639 SNPs (continued)
doi:10.6342/NTU201701806
52
圖八、八種統計方法在各性狀的預測準確度比較圖
每個方法以十折交叉驗證重複 1000 次，再以 Tukey’s test 多重比較檢定各統計方 法間是否具有顯著差異。FL 為抽穗期、PH 為株高、Yield 為產量。
Figure 8. The prediction accuracy comparison of eight statistical methods for each trait Each method was 1000 times cross-validated (10-fold cross-validation), and then Tukey’s test was used to determine whether there was a significant difference between the statistical methods. FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
53
圖九、327 品系校正外表型分布圖
五個親本之表現亦標於圖上；FL 為抽穗期、PH 為株高、Yield 為產量。
Figure 9. The distribution of the adjusted phenotypes of 327 lines
The performances of the five selected parents are marked on the figure. FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
54
圖十、10 個雜交組合之產量 GEBV 與抽穗期 GEBV 散佈圖
黑點為 327 品系，紫三角為親本，灰十字為 F2族群，紅十字為 GS 選出 F6優良品 系，藍十字為 PS 選出 F6優良品系。
Figure 10. Yield and days to flowering GEBV in 10 bi-parental crosses
Black spot represent the 327 initial lines, purple triangles for the parents, gray crosses for the F2 group, red crosses for F6 best line selected from GS, and blue crosses for F6
best line selected from PS.
doi:10.6342/NTU201701806
55
圖十、10 個雜交組合之產量 GEBV 與抽穗期 GEBV 散佈圖 (續)
Figure 10. Yield and days to flowering GEBV in 10 bi-parental crosses (continued)
doi:10.6342/NTU201701806
56
圖十一、10 個雜交組合選出 F6優良品系之表現最佳個體之產量 GEBV 與抽穗期 GEBV 散佈圖
Figure 11. Yield and flowering time GEBV of the best individual of the F6 best line selected from the ten crossing combinations
doi:10.6342/NTU201701806
57
圖十二、327 品系與 10 個雜交組合 GS 與 PS 選出 F6優良品系性狀之 GEBV 分布 Figure 12. Distribution for the 327 lines and the F6 best lines selected from 10 crossing combinations
doi:10.6342/NTU201701806
58
圖十三、輪迴選種五次循環 F2族群之產量 GEBV 與抽穗期 GEBV 散佈圖 左方為 GEBV 選出初始親本，右方為校正外表型 BLUE 選出初始親本。黑點為 327 品系，綠三角為初始親本，紅至紫十字依序為第一至第五次循環之 F2世代。
Figure 13. Yield and flowering time GEBV of the F2 population at each cycle of the recurrent selection
Results from GEBV-selected parents are on the left panel, results from the BLUE-selected parents are on the right panel. Black spots represent the 327 initial lines, the green triangles for the initial parents, red to purple crosses in order for the first to the fifth cycle of F2 generation.
doi:10.6342/NTU201701806
59
圖十四、輪迴選種第五次循環 F2族群性狀 GEBV 盒鬚圖
由左至右依序為 327 品系、GEBV 選拔初始親本之 Cycle 5 數值、以 BLUE 選拔 初始親本之 Cycle 5 數值。經 Welch’s t-test 兩兩比較，皆具顯著差異 (𝑝 <
0.0001)。Yield 為產量、FL 為抽穗期、PH 為株高。
Figure 14. Boxplot of F2 populations at Cycle 5 of the recurrent selection
From left to right: the initial 327 lines,Cycle 5 of GEBV-selected initial parents, and Cycle 5 of BLUE-selected initial parents. Welch’s t-tests show significant differences between any two groups (𝑝 < 0.0001). FL, flowering time; PH, plant height.
doi:10.6342/NTU201701806
60
圖十五、輪迴選種各循環 F2族群相對 327 品系之標準化遺傳增進變化 抽穗期與株高為取絕對值結果。Yield 為產量、FL 為抽穗期、PH 為株高。
圖十五、輪迴選種各循環 F2族群相對 327 品系之標準化遺傳增進變化 抽穗期與株高為取絕對值結果。Yield 為產量、FL 為抽穗期、PH 為株高。