Vous êtes sur la page 1sur 99

SECTION II

Breeding & Seed Production

105

Variabilité génétique des ecotypes locaux de maïs pour la tolérance à la sécheresse au Sénégal

Abdou Ndiaye,

ISRA–CNRA BP, 53 Bambey, Sénégal

Résumé La variabilité génétique est un élément essentiel pour tout programme de sélection. Ainsi, 36 écotypes locaux de maïs et 5 populations tolérantes à la sécheresse en provenance du CIMMYT ont été évalués en station sous deux régimes: i) sous irrigation complète et ii) sous stress hydrique induit à la floraison pour l’identification de source de tolérance à la sécheresse. Plusieurs caractères agromorphologiques comme la hauteur moyenne du plant et de l’épi, la date de floraison, le rendement et ses composantes ont été mesurés. L’utilisation de l’analyse de variance et de l’analyse en composantes principales a permis d’identifier et de décrire deux groupes répartis en fonction de la nature du régime hydrique. Quelques associations significatives entre caractères ont été trouvées et pourraient être dues à la pléotropie, au linkage chromosomique, au déséquilibre gamétique ou à la sélection. Etant donné le nombre important de paramètres à prendre en considération pour la tolérance à la sécheresse, seuls ceux qui semblent plus importants seront utilisés dans le cadre du criblage au champ ou au laboratoire à la fois sous régime complet ou sous stress afin de combiner rendement élevé et tolérance à la sécheresse.

Abstract Genetic variability is essential for continued genetic improvement of any crop species for stress tolerance. A total of 36 maize (Zea mays L.) local ecotypes and five drought tolerant populations from CIMMYT were evaluated on station under two regimes: i) complete irrigation and ii) water stress induced at flowering stage. The objective was to identify sources of drought tolerance for future breeding work. Many agro-morphological characters, such as plant and ear heights, number of days from planting to 50% silking, grain yield and yield components were measured. Using analysis of variance and principal component analysis, two groups of populations were identified and described according to their response to the water regime. Significant associations between characters were found. These relationships could have been caused by genetic effects (pleiotropy, for example), selection effects, multivariate effects (multiple correlation) and sampling effects. Since a large number of parameters are known to be responsible for drought tolerance, only those of direct relevance may be used

106

for laboratory screening followed by field test under stress as well as under normal conditions for combining high yield and drought tolerance.

Introduction

Tout travail de sélection exige l’examen d’un matériel aussi vaste que possible, d’autant plus que le nombre de problèmes dont on

espère trouver la solution est élevé. Dans quelque domaine que ce soit, les buts principaux à atteindre sont : grande productivité, qualité et régularité de la production : résistance et/ou la tolérance

aux différents aléas climatiques dont la sécheresse et les maladies. L’adaptation de la plante aux conditions du milieu physique tels les

inconvénients climatiques (sécheresse, fortes variations thermiques, vents desséchants, etc.), le faible niveau de fertilité

minérale des sols (teneur faible en azote) et la demande progressivement croissante sur le marché légitiment et commandent la recherche de techniques permettant une

production accrue du maïs.

De nos jours, certaines limites dans les capacités d’amélioration des principales cultures ont été atteintes et de façon générale dans la recherche et le choix de nouvelles sources de variabilité pour la

tolérance aux différents stress biotiques et abiotiques. C’est en particulier le cas du maïs dont la base génétique des principales

variétés cultivées en France est très étroite, avec la prédominance de quelques lignées “élites’’ et de leurs dérivés. Aux Etats-Unis, seuls quelques génotypes issus des meilleures populations

synthétiques ou traditionnelles (Iowa Stiff Stalk Synthétic, Lancaster Sure Crop, Reid’s Yellow dent et Krug) sont utilisées. En

Europe tous les hybrides cultivés sont, pratiquement issus d’un parent denté américain et d’un parent corné européen, qui est en général apparenté aux deux (2) lignées F2 et F7 issues de la

population Lacaune. Au CIMMYT et en Afrique occidentale et centrale, le développement de populations tolérantes à la

sécheresse s’est limité au criblage d’un nombre plus ou moins restreint de variétés performantes et d’écotypes. C’est ainsi que, suivant différentes approches de sélection pour la tolérance à la

sécheresse, des populations sources ont été développées telles Pool 16 sequia, Tuxpéno sequia, DTP1, et DTP2 à partir d’un schéma de

sélection récurrente intrapopulation d’une part, et d’autre part par le cribl age et la combinaison de variétés (lignées, populations,

hybrides, écotypes etc…) tolérantes et performantes en conditions de sécheresse (Edmeades et al. 1997). En Afrique occidentale et

107

centrale, deux populations DRT-EW (blanc denté précoce) et DRT-EY

(jaune corné précoce) ont été obtenues à partir d’écotypes locaux (Thé et al. 1997).

Cette approche est limitée, entre autres, par les nombreuses interactions génotype x environnement et le déterminisme

plurigénique de la tolérance au déficit hydrique. Cette limite d’ordre génétique, qui peut présenter des risques agronomiques certains,

est liée ou sinon due à la non-utilisation de toute la variabilité actuellement disponible, soit par souci d’efficacité à court terme, soit par manque de méthodologie pour pouvoir utiliser toute cette

variabilité. C’est pour répondre à ces différents aspects et pour préparer le progrès génétique à moyen et long terme que le

programme s’est fixé comme objectifs de :

??

rassembler un grand nombre de d’écotypes locaux adaptées aux conditions climatiques en particulier de sécheresse ;

??

se donner les moyens d’étude et de conservation de cette variabilité ;

??

préciser les méthodologies de gestion et d’utilisation de cette variabilité en sélection pour la tolérance à la sécheresse.

En définitive, le but de ce travail consiste à effectuer l’évaluation de

la variabilité des populations locales de maïs pour la tolérance à la sécheresse afin d’élargir la base génétique existante d’une part et à proposer d’autre part des méthodes d’intégration de ce germplasme

dans le programme de recherche d’une nouvelle variabilité.

Matériel et Méthodes

Matériel végétal

Le matériel végétal étudié est composé de 36 écotypes locaux issus

de la prospection de 1990 à travers toutes les zones maïsicoles du Sénégal et 5 populations tolérantes à la sécheresse en provenance du CIMMYT. La liste des écotypes locaux figure au niveau du

tableau.

Tableau 1. Liste des écotypes locaux.

N o

Code

N o

Code

N o

Code

N o

Code

N o

Code

1

TB1

9

VG9

17

KD19

25

KD43

33

NR50

2

TB2

10

VG10

18

KD20

26

KD46

34

NR54

3

TB4

11

VG11

19

KD22

27

SD28

35

NR55

4

TB5

12

VG12

20

KD23

28

SD29

36

NR56

5

TB57

13

KD13

21

KD26

29

SD30

37

DTPW

6

TB58

14

KD14

22

KD27

30

SD31

38

DTPY

7

TB60

15

KD16

23

KD40

31

SD32

39

TS6

8

VG6

16

KD18

24

KD41

32

SD34

40

PSC

 

41

PSS

108

Méthodes

Le dispositif expérimental utilisé est un essai en blocs complets randomisés avec 2 répétitions. Les lignes sont séparées par des

allées de 0,80 m, l’écartement entre les poquets est de 0,50 m soit une densité optimale de 50 000 pieds/ha. Deux régimes hydriques

ont été appliqués : un régime hydrique « sans stress » du semis à la maturité (irrigation complète) et un régime hydrique avec un stress simulé pendant la phase de préfloraison —floraison par arrêt de

l’irrigation jusqu’à la maturité. Tous les essais ont été semés sur la même bande durant la contre-saison de 1999 et 2000 et ont reçu

respectivement 300 kg/ha de N.P.K. (8.18.27) avant le semis, 150 kg d'urée à la montaison et 150 kg/ha d'urée à la floraison.

L’essai a été mis en place à la station de recherche de Nioro du Rip situé à 14° 8 latitude et à 16° 4 longitude au niveau de la zone de

culture par excellence du maïs. Elle se caractérise par un climat sahélo - soudanien, une pluviométrie variant entre 400 et 600 mm,

des températures minimale de 15 à 18 °C, maximale de 35 à 40 °C et moyenne de 25 à 28° C. Le choix des caractères mesurés a été effectué de façon à ce qu’ils apportent le maximum d’informations

sur la plante. Il s’agit des caractères agronomiques (rendement, précocité), de caractères quantitatifs associés au rendement

(mesures morphologiques) et des facteurs de régularité du rendement telle la tolérance à la sécheresse et autres aléas.

L’étude de chacun des aspects de la variabilité génétique nécessite l’emploi de techniques de description et d’analyse différentes

suivant que l’on désire étudier chaque caractère indépendamment ou leur ensemble, de façon synthétique. L’analyse de variance a été utilisée pour quantifier les variations inter et intrapopulations qui

constituent la variation totale. La structure générale de la variabilité a été ensuite étudiée par des analyses multivariées

comme l’analyse en composantes principales. Cette analyse a été utilisée par différents auteurs : Hugues de Cherisey (1983) sur le

millet et Ndiaye (1985) sur le maïs.

Il est à noter que pour les analyses suivantes, les écotypes sont identifiés par trois chiffres dont le premier indique le régime

hydrique (un nombre pa ir correspond au régime sous irrigation complète ; un nombre impair au régime stress hydrique) et les deux

109

derniers chiffres correspondent au numéro du génotype. Dans le

cadre de l’analyse de variance combinée, les essais sont identifiés par le régime complet(IR pour irrigation complète, ST pour le stress hydrique, ces lettres sont suivis par les deux derniers chiffres de

l’année d’expérimentation.

Resultats et Discussion

Analyse de variance

Les résultats de l’analyse de variance individuelle des essais

montre que tous les caractères présentent un effet génotypique hautement significatif. Les résultats de l’analyse de variance combinée suite au regroupement des essais deux à deux (Tableaux

2a et 2b) montrent un effet “régime d’irrigation’’ hautement significatif excepté les essais sous irrigation complète (IR9900) pour

les caractères comme la durée semis-floraison femelle à 50%, la hauteur du plant (HMP). Pour le cas de l’essai “Irrigué + Stressé’’ (IRST99), les effets “régime d’irrigation’’ et “génotype ’’ sont

significatifs pour la hauteur moyenne du plant (HMP) et pour le rendement (RDT) au seuil de 5%. Cet effet “régime’’ dénote de

l’efficacité du stress hydrique appliqué et qui se traduit par un indice de sévérité de la sécheresse variant entre 52 et 85% en fonction du génotype.

Tableau 2a. Analyse de variance selon le modèle interactif complet sur IRST99, sur IR9900 et sur ST9900 (Somme des Carrés des Ecarts).

Régimes

IRST99

 

IR9900

 

ST9900

Source

ddl

HMP

RDT (10 5 )

HMP

RDT (10 5 )

HMP

RDT (10 5 )

Génotype

40

32012.2*

3508

17817.9ns

5686**

17329.8*

290*

Régime

1

29317.2*

4438

7994.0ns

498ns

9225.0ns

932ns

BlRégime

2

3301.5

2257

3301.5

174

4222.5

150

G x R Résid.

40

9514.0ns

2117

15499ns

332ns

8875.0ns

198ns

80

36885.9

6776

32985

1000

29552.4

332

Tableau 2b. Analyse de variance selon le modèle interactif complet sur IRST00 et IRST9900 (Somme des Carrés des Ecarts).

Régimes

IRST00

IRST9900

Source

ddl

HMP

RDT(10 5 )

HMP

RDT(10 5 )

Génotype

40

12598.7*

391*

25023.6*

445*

Régime

1

27127.4*

246*

73641.6*

6744*

110

Bl(Régime)

2

4222.5

99

7524.1

325

G x R

40

14397.5ns

336ns

26498.9ns

844ns

Résid.

80

25652.4

655

62538.4

1332

Pour le regroupement des essais de l’an 2000 (IRST00), en plus de l’effet “génotype’’, l’effet “régime hydrique’’ est significatif pour les

variables comme la hauteur du plant et le rendement. Cette variabilité inter- régime hydrique montre à nouveau l’efficience du stress appliqué comme cela a été le cas en 1999.

Pour le regroupement des essais stressés (ST9900), les effets

“génotype’’ et “régime hydrique’’ sont significatif pour les caractères tels le rendement et la hauteur moyenne du plant. Comme dans le groupe précédent, les génotypes semblent avoir

réagi de la même manière face au stress appliqué pendant les deux années d’expérimentation.

Pour l’analyse globale, l’effet “régime’’ est significatif pour le rendement et la hauteur moyenne du plant, ce qui confirme les

résultats des analyses précédentes des essais regroupés par année et par régime. Il faut cependant noter l’absence d’une interaction

génotype x régime hydrique sur les groupes IRST99, IRST00 et sur l’ensemble des quatre essais regroupés IRST9900 (cf. Tableau 2b).

Association des caractères

Seuls les caractères quantitatifs et d’intérêt agronomique ont dû

présenter des corrélations hautement significatives. Ainsi observe- t-on :

?? de fortes corrélations positives existent entre la hauteur totale

(HMP), la hauteur de l’épi (HEPI), la date de floraison (FLOR) et la verse. Il semble donc qu’une floraison tardive s’accompagne d’un fort développement végétatif et d’une sensibilité à la

??

verse ; la vigueur au départ et la floraison (r = -0,62). On en déduit que

??

les populations précoces paraissent donc mieux adaptées aux conditions en début de végétation la hauteur moyenne de l’é pi, la date de floraison et la pourriture

des tiges. Un bon développement végétatif est lié à une diminution de la pourriture des tiges mais cependant, une

floraison précoce est statistiquement associée à une

111

augmentation de la pourriture des tiges car les populations

précoces ont des tiges plus vieilles à la récolte.

On peut ainsi suggérer les causes suivantes pour expliquer ces

associations de caractères. Il s’agit de la pléiotropie génétique assimilable à une corrélation physiologique, du linkage

chromosomique, du déséquilibre gamétique, de la sélection, de l’association fortuite par dérive génétique ou plus simplement par effet d’échantillonnage et enfin du jeu des quatre premiers facteurs

entre chacun des deux caractères et un troisième caractère.

An alyse en composantes principales

L’analyse en composantes principales (A.C.P) a été effectuée à la

fois sur les caractères quantitatifs et qualitatifs dans le but de mieux cerner la variabilité existante. Pour ce faire, les essais ont

été regroupés par année et par régime. Il s’agit de IRST9900 (essai irrigué + essai stressé en 1999 et en 2000) soit 4 essais ayant fait

l’objet d’une analyse globale, de ST9900 (regroupement des essais stressés de 1999 et de 2000), et de IR9900 (regroupement des essais sous irrigation complète en 1999 et en 2000).

Pour l’essai IRST9900, les quatre premiers axes représentent

74,5% de la variabilité totale, le cinquième axe avec 8,2% d’inertie contribue moins qu’une variable mesurée à la description de la variabilité totale (Fig. 1). Pour l’essai ST9900, les quatre premiers

axes expliquent environ 79,6% de la variabilité totale (Fig. 2) et enfin pour le groupe IR9900, les quatre axes expliquent 74,2% de la

variabilité totale (Fig. 3). Un résumé de l’analyse en composantes est donné par le Tableau 3. Ainsi :

??

La première composante principale est un axe de rendement et d’architecture. Elle oppose principalement les deux régimes

hydriques c’est-à-dire le régime sous irrigation complète et celui à régime stressé induit à partir de la floraison. On trouve les génotypes à haut rendement, de grande taille du côté du

régime sous irrigation complète et à l’opposé des génotypes ayant des rendements faibles, une verse et une casse

importantes et une incidence à la sécheresse très élevée.

112

n

Valeur

Pourcent

Cumul

0

3.0499

1

3.0499

33.89

33.89

 

2

1.4328

15.92

49.81

   

3

1.2431

13.81

63.62

 

4

0.9763

10.85

74.47

 

5

0.7370

8.19

82.66

 

6

0.5856

6.51

89.16

 

7

0.4498

5.00

94.16

 

8

0.4016

4.46

98.62

 

Variance totale = 9.0

Figure 1. Valeurs propres du groupe d’essais IRST99.

n

Valeur

Pourcent

Cumul

0

2.7347

1

2.7347

34.18

34.18

 

2

1.5816

19.77

53.95

   

3

1.2385

15.48

69.44

 

4

0.8128

10.16

79.60

 

5

0.5968

7.46

87.06

 

6

0.4767

5.96

93.01

 

7

0.4037

5.05

98.06

 

Variance totale = 8.0

Figure 2. Valeurs propres du groupe d’essais ST9900.

n

Valeur

Pourcent

Cumul

0

2.2159

1 2.2159

27.70

27.70

2 1.5719

19.65

47.35

3

1.1837

14.80

62.14

4

0.9632

12.04

74.18

5

0.7654

9.57

83.75

6

0.6851

8.56

92.32

7 0.4356

5.45

97.76

8 0.1791

2.24 100.00

Variancetotale = 8.0

Figure 3. Valeurs propres du groupe d’essais IR9900

?? La deuxième composante est un axe de caractéristiques de la précocité. Elle oppose les populations de grande taille plus précoces sous irrigation complète à celles de taille souvent

réduite et de durée semis-floraison femelle plus longue en raison du stress appliqué.

?? La troisième composante est un axe caractéristique de la sensibilité à la verse et à la sécheresse. Elle oppose les écotypes

??

113

trop sensibles au stress hydrique et qui ont une verse

importante à ceux plus ou plus tolérants aux conditions de sécheresse et ayant une incidence à la sécheresse relativement faible.

La quatrième composante principale est un axe caractéristique de l’architecture avec des génotypes ayant une bonne densité à

la levée et qui manifestent une casse relativement importante vers la récolte.

Tableau 3. Résumé de l’analyse en composantes principales sur les moyennes.

Axe

Pourcentage d’inertie expliquée

Variables contribuant à la définition de l’axe

Axe 1

34.1

RDT (+0.161)

 

HMP (+0.188)

HEPI (+0.200)

Axe 2

19.8

FLOR(+0.164)

 

HUM (+0.126)

Axe 3

15.5

VERS(+0.148)

 

INCI (0.195)

Axe 4

10.9

NPL (+0.600)

 

CAS (+0.100)

L’observation du Tableau 3 montre que les deux premiers axes sont caractérisés par des caractères quantitatifs contrairement aux troisième et quatrième axes qui le sont par les caractères

essentiellement qualitatifs. La Figure 4 donne la projection des caractères ou les cercles de corrélation sur les plans engendrés par

les axes principaux 1 et 2.

1,0000 2 VER CAS HMP RDT NPL HE -1,6400 1,6400 1 INCI HUM FF50
1,0000
2
VER
CAS
HMP RDT
NPL
HE
-1,6400
1,6400
1
INCI
HUM
FF50

-1,0000

Figure 4. Projection des caractères sur le plan engendré par les axes 1 et 2.

114

La projection des écotypes sur le plan engendré par les axes 1 et 2

(Fig. 5 et Fig. 6) montre deux groupes principaux :

?? le groupe qui s’identifie au régime sous irrigation complète matérialisé par les numéros dont le premier chiffre est impair.

Il est composé des écotypes locaux à potentiel de rendement élevé, une hauteur moyenne du plant et de l’épi importantes et,

le groupe qui caractérise essentiellement le régime stressé matérialisé par les numéros dont le premier chiffre est pair. Il est essentiellement composé d’individus à rendement très

réduit en raison de leur sensibilité la sécheresse. Ces derniers sont de taille relativement réduite en raison du régime auquel

ils ont été soumis.

??

3,4000 2 433 423 324 305 333 235 302 436 214 209 331 220 325
3,4000
2
433 423
324
305
333
235
302
436
214
209
331
220
325
315
313
222
HMP RDT
211
216
-5,3000
229
5,3000
1
219
327
326
338
221
426
230
308
135
407
212
317
113
239
241

-3,4000

Figure 5. Dispersion des écotypes dans le plan engendré par les axes 1 et 2.

3,4000 3 133 108 114 214 134 HMP -5,3000 327 220 5,3000 326 1 219
3,4000
3
133
108
114
214
134
HMP
-5,3000
327
220
5,3000
326
1
219
315
212
216
211
313
236
222
229
335
308
235
303
341
204
224
306
234

-3,4000

115

Figure 6. Dispersion des écotypes dans le plan engendré par les axes 1 et 3.

En reprenant l’analyse en composantes principales avec comme variables de choix le rendement ou la hauteur moyenne du plant,

on obtient une nouvelle distribution des individus et qui semblent être les plus représentatifs compte tenu de leur contribution à la

définition des axes. Ainsi, avec la variable rendement (Fig. 7), l’axe 1 oppose les individus KD14, KD41, NR55, KD13, NR54, SD31, VG12, et KD22 par exemple aux individus tels KD16, DTPY, VG6, SD28,

KD19, TB60, et VG10. Concernant la hauteur moyenne (HMP) (Fig. 8), deux groupes bien distincts apparaissent. Il s’agit du groupe

comprenant les génotypes TB1, KD22, KD26, et VG11 et le groupe concernant les écotypes tels que KD26, TB57, VG10, KD16, DTPY, KD13 et La Posta séquia.

4,5990 2 110 133 108 138 109 430 305 333 310 214 421 141 104
4,5990
2
110
133
108
138 109
430 305
333
310
214 421
141
104
437
315
423 209
410 436
401
HMP HE
121
235
220
134
414
303
413
341
411
-5,6650
6,0460
211
222
216
338
234
327
326
1
215 335
407
219
132
415
221
135
230
224
426
203 408
308
113
123
440
317 417
419
212
241
2 39
-4,3000
215 335 407 219 132 415 221 135 230 224 426 203 408 308 113 123

Figure 7. Dipersion des écotype dan le plan engendré par les axes 1 et 2 sur la base du rendement corrigé.

3,5000 HMP 2 N_Rep +113,0-+166,5 +166,5-+220,0 305 333 310 HMP 315 121 -5,0000 313 5,0000
3,5000
HMP
2
N_Rep
+113,0-+166,5
+166,5-+220,0
305
333
310
HMP
315
121
-5,0000
313
5,0000
101
1
119
338
126
111
340

116

Figure 8. Dispersion des écotypes sur le plan engendré par les axes 1 et 2 sur la base de la hauteur du plant.

Il faut cependant noter la quasi-simulitude des résultats entre

l’analyse de variance et l’analyse en composantes principales quant à la séparation des écotypes en fonction du régime. Cela s’est traduit par l’absence d’une interaction génotype x environnement,

le régime hydrique pouvant être considéré comme un environnement. La séparation des individus sur le premier axe en

deux groupes pose en fait le problème de l’efficience du criblage sous régimes hydriques différenciés et/ou de l’utilisation de ces

écotypes pour la tolérance à la sécheresse. Cela revient à dire que la sélection doit s’effectuer au niveau intragroupe dans un premier temps dans le souci d’identifier les sources de tolérance d’une part

et de potentialité de rendement d’autre part avant de passer à la phase de sélection intergroupe.

Au vu des résultats, la nécessité d’une compréhension profonde des sources de variabilité devient impérative en vue de leur

introduction dans le programme de sélection. En d’autres termes, la connaissance des relations entre les écotypes d’une part, la

meilleure prise en compte des réactions spécifiques aux environnements génétiques et écologiques (avec ou sans sécheresse) d’autre part, et enfin une plus large compréhension de

leur potentiel de combinaison spécifique peuvent, non seulement révéler le taux de variabilité disponible mais aussi les méthodes de

sélection adéquates. Parmi les méthodes préconisées, on peut citer celles qui semblent pouvoir permettre de surmonter les problèmes d’adaptation aux conditions pédoclimatiques (sécheresse, faible

niveau de fertilité des sols etc.) et ceux liés à l’amélioration du niveau moyen de performance. Toutefois, des méthodes sont

courantes et ont déjà été utilisées (Hallauer and Sears 1972). Il s’agit de la sélection massale, de la sélection récurrente avec descendances et du back-cross.

117

En pratique, toute sélection est nécessairement multicaractère et les améliorateurs de plantes doivent choisir la stratégie la plus efficace compte tenu de leurs contraintes et de leurs moyens.

Gallais (1990) a montré l’intérêt et la supériorité d’une sélection multicaractère par index lorsque le nombre de caractères à

sélectionner est grand, et qu’ils sont d’égale importance, ce qui est une condition d’efficacité de la sélection à long terme. Ceci est d’autant vrai que les caractères sont liés, et ce d’autant que la

corrélation est faible ou négative. Mais qu’elle que soit la liaison entre les deux caractères, un progrès important sur un caractère

limitera le progrès s ur l’autre (Gallais 1990). Donc, plus la structure des corrélations est forte, plus le progrès simultané sur plusieurs caractères sera difficile. La connaissance de l‘intensité de la

structure des corrélations dans une population, en plus de ses performances moyennes et de sa variabilité pour les différents

caractères, pourrait donner des indications sur le progrès potentiel que l’on peut espérer obtenir en sélectionnant cette population.

References

Edmeades, G.O., M. Bänziger, and D. Beck. 1997. Development and per se performance of CIMMYT maize populations as drought

tolerant sources. Pages 254–262 in G.O. Edmeades, M. Bänziger, H.R. Mickelson, and C.B. Pena-Valdivia (eds.) Developing drought and low-N tolerant maize. Proceedings of a

Symposium, March 25–29, 1996, CIMMYT, Mexico.

Gallais, A. 1990. Théorie de la sélection en amélioration des plantes. Editions Masson, Paris.

Hallauer, A. R., and J.H. Sears. 1972. Integrating exotic gerpmplasm into corn-belt maize breeeding programs. Crop Sci. 12: 203–206

Hugues

de

Chérisey.

1983.

Contribution

à

l’évaluation

des

ressources

génétiques

du

millet

Setaria

italica

L.

P.B.

Variabilité des caractères quantitatifs. Thèse de Docteur Ingénieur. Univ. Paris XI Orsay.

Lavergne, V. 1988. Variabilité entre populations de maïs (Zea mays

L.). Approche biométrique et approche enzymatique. Thèse de Docteur de l’INA-PG.

Ndiaye, A. 1985. Etude de l’organisation de la variabilité dans les

populations de maïs (Zea mays L.) Mémoire de DEA, Univ. Paris

Sud Orsay.

118

Thé, C., C. Zonkeng, and S.K. Kim. 1997. Identification and

development of drought tolerant maize cultivars in Cameroon. In G.O. Edmeades, M. Bänziger, H.R. Mickelson, and C.B. Pena- Valdivia (eds.) Developing drought and low-N tolerant maize.

Proceedings of a Symposium, 25–29 March 1996, CIMMYT, Mexico.

119

Molecular marker analysis: Population size, type of trait, and method of analysis for QTL detection

Mashark S. Abdulai

Savanna Agricultural Research Institute, PO Box 52, Tamale, Ghana. E-mail:sari@africaonline.com.gh.

Abstract The detection and location of genes that contribute to the expression of quantitative characters is a problem to plant breeders. Different breeding designs and statistical packages have been developed aimed at identifying the best combinations that will solve the problem. The objectives of the study were to use simulations to compare three statistical methods for the detection of quantitative trait loci (QTLs) in three different population sizes for four traits with heritabilities 1.00, 0.75, 0.50, and 0.25, respectively. The power to detect a QTL was related to the heritability of QTLs, the number of genotypes analyzed, and the method of detection. The single analysis of variance (F-test) using SAS, the LOD score which is the likelihood ratio test of absence of linkage between the maker and the QTL using Mapmaker, and LOD score using regression employed by Plabqtl were the methods used. Power was measured as the mean number of QTLs detected for each situation. When heritability of the trait was very low (<25%), more than 300 genotypes may be required to obtain the same level of power which will be detected by 200 genotypes when heritability of the trait was high (100%). Mapmaker was able to detect the maximum number of QTLs in all cases, whereas QTLs detected by Plabqtl and SAS detected were comparable at only higher population levels. The ANOVA method of SAS was able to detect QTLs but the number of pseudo-QTLs rose as the number of genotypes analyzed was increased beyond 200. The optimum number of genotypes to be used in this type of study should be between 200 and 300 for all methods. The ANOVA method of SAS should be the first option before the other two methods are used.

Résumé La détection et la localisation des gènes qui contribuent à l'expression de caractères quantitatifs est un problème pour les sélectionneurs de plantes. Différents modèles de sélection et paquets statistiques ont été développés afin d’identifier les meilleures combinaisons qui résoudront le problème. Les objectifs de l'étude étaient d’utiliser des simulations pour comparer trois

120

méthodes statistiques pour détecter les loci du caractère quantitatif (QTLS) dans trois différentes tailles de population pour quatre caractères avec pour héritabilités 1.00 ; 0,75 ; 0,50 et 0,25, respectivement. L'analyse de variance simple (F- test) utilisant SAS, le score LOD qui est le test de la proportion probable d'absence de liens entre le marqueur et le QTL utilisant le MAPMAKER et le score LOD utilisant la régression employée par PLABQTL étaient les méthodes utilisées. La puissance était mesurée comme étant le nombre moyen de QTLS détecté pour chaque situation. Le pouvoir de détecter un QTL était relié à l’héritabilité de QTLS, le nombre de génotypes analysés et la méthode de détection. Quand l’héritabilité du caractère était très bas (< 25%), plus que 300 génotypes étaient nécessaires pour obtenir le même niveau de pouvoir détecté par 200 génotypes lorsque l’héritabilité du caractère était élevé (100%). Le MAPMAKER était capable de détecter le nombre maximum de QTLS dans tous les cas, tandis que les QTLS détectés par PLABQTL et ANOVA de SAS étaient comparables seulement aux tailles de population élevées. La méthode ANOVA de SAS était capable de détecter les QTLS mais le nombre de pseudo - QTLS a augmenté quand le nombre de génotypes analysés s’est accru au- delà de 200. Le nombre optimum de génotypes employé dans ce type d'étude devrait être entre 200 et 300 pour toutes les méthodes. La méthode ANOVA de SAS devrait être la première option avant que les deux autres méthodes ne soient utilisées.

Introduction

The genetic improvement of a species through artificial selection depends on the ability to capitalize on genetic effects that can be distinguished from environmental effects (Stau b and Sequen 1996).

The effects can be identified more easily if they can be linked to genetic markers. King and Standfield (1997) defined genetic

markers as genes with known locations on a chromosome and with clear-cut phenotypes used as points of reference. Breeders may use

genetic markers as selection tools (Beckmann and Soller 1983; Darvasi and Soller 1994). Hospital et al. (1992) and Openshaw et al. (1994) worked on marker-assisted selection and concluded that the

procedure can increase the effectivene ss of backcrossing by increasing the probability of obtaining a suitable conversion while

decreasing the time required to achieve an acceptable recovery.

Most traits of agronomic importance, such as yield, nutritional

quality, maturity, and stress tolerance are quantitatively inherited (Allard 1960; Hallauer and Miranda 1988). Several or many genes

with small, additive effects usually control quantitatively inherited traits. They also interact with each other and with the environment

121

to determine the ultimate phenotype of the individual. Classical

biometrical methods provide information on the inheritance of the traits but usually, plant breeders know little or nothing about the number of genetic factors (genes or loci) involved in the expression

of the trait. The ability to manipulate the genes may, therefore, be low.

One of the fundamental problems of quantitative genetics is that genotypes with quantitative trait loci (QTL) cannot be determined by

inspecting the distributions of trait phenotypes alone (Knapp 1994). Genetic markers have therefore been used effectively in the

analysis of genetic diversity and germplasm organization in a number of crop species.

There is a minimum variability required to detect the presence of QTL. Therefore most QTL studies have utilized populations in which

the alleles of both parents occur at a relatively high frequency. Tanksley and Nelson (1996) referred to such populations as balanced populations. These populations include F 2 families,

backcross-1 (BC1) families, doubled haploid (DH) families, and recombinant inbred lines (RIL). Whereas these populations are good

for mapping, Tanksley and Nelson (1996), using simulated data, pointed out that because undesirable QTL alleles from unadapted parents occur at high frequency thus preventing meaningful data

collection, epistatic interactions are difficult to detect and subtle but often negative pleiotropic effects may go unnoticed. Balanced

populations are therefore not very suitable populations for detecting and transferring useful QTL from unadapted germplasm to elite

breeding lines. Instead, Tanksley and Nelson (1996) proposed the application of advanced backcross QTL analysis in which QTL mapping is delayed to either the BC2 or BC3 generations.

Carbonell et al. (1992; 1993) reported that DH experimental design

was more powerful in detecting QTL than BC experimental design while keeping actual Type I errors very low. For the DHs, the authors noticed that the power of detecting a QTL was related to the

heritability; the higher the QTL contribution to the total heritability of the trait, the higher the probability of being detected. A highly

heritable trait is usually detected with minimum effort. Nevertheless, most of the traits of economic importance have low heritability, thus, special skills are needed to be able to detect their

presence.

122

For backcrosses, the ranking of power was determined by the

proportion of the square of the sum of the additive and dominance variations. Tanksley and Nelson (1996) pointed out that a single selfing generation (i.e. F 2 ) was approximately twice as powerful as

BC1 generation in detecting a QTL with additive gene action and no epistatic interactions; however, some QTL might remain

undetected.

The statistical detection of a QTL is likely to depend not only on the

type of population, the population size (Jeon 1988; Darvasi et al. 1993) and marker density (Darvasi et al. 1993), but also on the

intra-locus and inter-loci interactions of the segregating QTL. A recessive QTL will go undetected in any backcross generation since no genotypes homozygous for the donor allele would occur in such

generations.

Marker-based techniques may be able to detect the number and relative contribution of each locus to the expression of the trait. Using simulation data, Touzet et al. (1995) and Le-Roy and Elsen

(1995) compared four statistics for the detection of QTL in daughter and grand daughter designs, previously defined by Soller and Genizi

(1978) and Weller et al. (1990). In all cases, the power to detect QTL was higher for segregation analysis than for linear method or logarithm of odds favoring linkage (LOD) score. Touzet et al. (1995)

and Le-Roy and Elsen (1995) noted that the differences could be significant when the QTL had small effects and also not closely

linked to the marker. Muranty (1996) reported that the power of the test for QTL detection based on linear models showed that for a

given number of parents in given conditions, the mating design had no effect on power. The power for testing hypothesis means of QTL genotypes is increased by increasing replication of genotypes

(Knapp and Bridges 1990), the total number of individuals genotyped (Knapp and Bridges 1990; Muranty 1996), and mapping marker

density of the genome (Moreno-Gonzalez 1992a). The power of the test is also higher when the tests are on additive effects only assuming that dominance is absent when it could exist (Mangin et

al. 1993).

The objectives of this research were to estimate the power to:

i. detect QTL affecting four traits with he ritability of 1.00, 0.75, 0.50,

and 0.25, respectively using populations of different sizes, and

123

ii. compare the power for the test of hypothesis using different

statistical packages.

Materials and Methods

Simulations for QTL detection

Data were simulated using a program written in Pascal

programming language on IBM compatible computer. The assumption was that two diploid maize inbred lines P 1 and P 2 were crossed to produce the F 1 progeny. Plants from the F 1 progeny were

self pollinated to produce F 2 lines. The data were simulated to have four traits having heritability of 1.00, 0.75, 0.50 and 0.25,

respectively. Each of the ten chromosomes had one quantitative trait locus placed in the center, five markers were then placed on each side of the QTL and each marker was 20 cM away from the

next marker. Three different populations were simulated each of size 100, 200 and 300 families. Each population was replicated five

times.

Method of QTL detection

Different strategies have been suggested for the identification of

single QTL (Edwards et al. 1987; Jian and Zeng 1995). In this study, three different statistical packages, SAS (1996), PLABQTL (Utz and

Melchinger 1996), and MapmakerQTL (Lincoln et al. 1990) were used to analyze and identify the presence of the QTL. Power of the test was calculated as the proportion of the number of QTL identified

by using each of the statistical packages.

Single factor analysis of variance using SAS

Dudley (1993) noted that the methods used for identifying QTL can

be divided into those which consider one marker locus at a time and those which consider all marker loci at once and attempt to

estimate effects for each locus while adjusting for the other loci. To examine the relationship between marker locus genotypes and quantitative trait expression, some theoretical consideration is

inevitable.

Consider an F 1 derived from two inbreds, for which a chromosome arm is heterozygous at a co-dominant marker locus, designated

M/m. The F1 may also be heterozygous at QTL, Q/q which is li nked to the marker locus with some recombination frequency,

124

designated r. The F 2 progeny from this theoretical situation will

consist of nine genotypic classes with respect to the two loci. The relative frequencies of these genotypes are functions of r, and the expressions of these genotypes for a given quantitative trait are

assigned based on the genotype at the QTL. However, since the genotype at the QTL cannot be discriminated, their effect must be

inferred via association with the genotypes at the marker locus.

The sums of frequency times expression, across each of the three

marker locus genotypes in terms of both r and d genotypic effects due to the QTL are presented below:

Marker class

Mean expression

MM

(1-2r)a + 2r(1-r)d

Mm

[(1-r) 2 + r 2 ]d

mm

(1-2r)(r-a) + 2r(1-r)d

The expressions for MM, Mm and mm simplify to the assigned values of QTL genotypes, a, d and -a, respectively. When r = 0, in

this situation, the marker locus is itself, responsible for the detected quantitative effects. Single factor analysis of variance

employed by SAS involves comparison among the phenotypic means of appropriate marker classes of the progeny. In this study the design is a completely randomized design in one environment. The

linear model is:

Y ij = µ + g i + e ij

where Y ij is an observation of the j th line of the i th QTL genotype,

µ is the population mean,

g i is the QTL genotype, e ij is the ij th residual effect, i = 1,2,….,q; j = 1,2,…

q is the number of QTL genotypes.

, n,

Doerge et al. (1994) and Knapp (1994) have given a review of how this method was developed. The three marker class means may then be subjected to analysis of variance to test the hypothesis of no

linkage between marker and trait.

Interval mapping using PLABQTL and MAPMAKER/QTL

125

PLABQTL is a software that localizes and characterizes QTL in

mapping populations derived from a biparental cross by self pollination, similar to the simulated data of this study. Simple and composite interval mapping are performed using multiple

regression procedure according to the approach described by Haley and Knott (1992). The model assumes a QTL (Q) lying between two

co-dominant flanking markers (M1 and M2). The F 2 generation of a cross between two inbred lines carry different alleles for all three loci. The F 1 cross between the two inbred lines has the following

chromosome array:

 

M1

r1

Q

r2

M2

o

|

|

|

o

|

|

|

m1

q

m2

where r1 and r2 are the recombination frequencies between M1 and Q, and between Q and M2, respectively; r is the recombination

frequency between the marker loci. The genotypic effects of the

three QTL genotypes possible in the F2 are set at (m + a), (m + d) and (m - a) for QQ, Qq or qQ and qq, respectively, where m is the

mean of homozygotes and a and d are additive and dominance

deviations, respectively. The expected means in terms of the putative QTL for each F 2 marker genotype is derived and then used

to fit a and d by multiple regression. For PLABQTL, regression is

maximum likelihood when errors are independent and normally

distributed. It can be written in terms of the residual sum of squares (RSS) of the full model, the reduced model and the number of observations as;

Likelihood ratio test = nLog e (RSS reduced /RSS full ).

When the likelihood ratio is above the threshold, it is assumed that a QTL is located in that region.

Mapmaker uses the same values but the test is maximum

likeli hood analysis. In this procedure, a combined test for the presence of p parameters, can be obtained from the maximized likelihood of the model in which the p parameters are estimated

compared with the maximized likelihood from which the p parameters are omitted or set at some value. Nine marker

classes are distinguishable for F2 intercrosses. Mapmaker then estimates the parameters as the means and variances among the

126

marker classes. Calculations are then performed by stepping along

the marker interval and assigning appropriate recombination values r1, r2, for a case where Q is within the marker interval. The likelihood for Q being unlinked to both markers is compared to the

likelihoods that it is at specific interior points in the interval. The LOD is estimated as:

LOD = - log 10 (L parameters /L no parameters )

The LOD score does not provide a test for the presence of a QTL between the markers but it compares the likelihood of the QTL

being at the position characterized by the recombination fractions r1, r2 against the likelihood that it is at some position unlinked to the interval. The map position at which the LOD is greatest is

likely to be close to the location of the QTL.

Results

Power was measured as the mean number of QTLs detected across the replications for each population and for each statistical method.

The mean number of QTLs, correctly identified, or falsely identified for all the traits using the various statistical methods are presented in Figures 1 and 2, and Table 1, respectively. The data were

simulated, and conscious efforts were done to put QTLs at specific locations. Therefore any QTLs identified at any location other than

those specified were declared pseudo-QTLs. The power to detect QTLs increased as the number of genotypes was increased from 100

to 300 across heritabilities for all the methods of detection (Figures 1 and 2). The results show that when the heritability of a trait is as high as 100% (Figure 1a), the power to detect QTL increased as the

number of genotypes was increased from 100 to 200. A further increase of the number of genotypes beyond 250 did not increase

the power. However, when the heritability of the trait was low (25% as in Figure 1d), the power to detect QTLs was low even when 300 genotypes were analyzed with the most e fficient statistical package.

In general the power to detect QTLs was dependent both on the number of genotypes analyzed and the heritability of the trait the

QTLs are contributing to the expression of the trait.

When a QTL was identified at a position known to have none, then that QTL was declared a pseudo-QTL. Also when more than one QTL was identified on a chromosome it was evident that the additional

QTLs were pseudo-QTLs. The ability to falsely identify QTLs was related to the heritability of the trait and the number of genotypes

127

analyzed. When the number of genotypes analyzed was low, raising

the heritability of the trait increase the power to detect pseudo- QTLs (Table 1). However, when the number of genotypes analyzed was high, the power to falsely detect QTLs was very low.

The power to detect QTLs depended on the methodology (Figures 1

and 2). There were differences among the methods of detection based on the number of QTLs detected across the traits. When the

number of genotypes analyzed was 100 (Figure 2a) not all the QTLs could be detected
number of genotypes analyzed was 100 (Figure 2a) not all the QTLs
could be detected even when the heritability of the trait was 100%.
In Figure 2c when the number of genotypes was 300 it appeared
12
12
that the maximum number of QTLs was detected by all the methods
a
10
even at lower than 100% heritability.
10
For any of the number of
b
genotypes analyzed Mapmaker was always the best in terms of the
8
8
number of QTLs detected. At lower heritabilities, Mapmaker was
still better than both PLABQTL and single analysis of variance using
6
6
SAS, but as the number of
genotypes
increased to 300 and the
heritability also
increased to 100% (Figs 1a and 2c) the
difference
4
4
among the methods of detection was negligible.
2
2
0
0
100
150
200
250
300
350
number of QTL
100 150 200 250 300 350 8 7 c 6 5 4 3 2 1
100
150
200
250
300
350
8
7
c
6
5
4
3
2
1
0
100
150
200
250
300
350
number of QTL

population size

6 5 d sas plabqtl mapmaker 4 3 2 1 0 100 150 200 250
6
5
d
sas
plabqtl
mapmaker
4
3
2
1
0
100
150
200
250
300
350

Population size

Figure 1. Power of test of difference between number of QTL identified. Power was estimated using MAPMAKER, PLABQTL, and SAS for heritabilities, a = 1.00, b = 0.75, c = 0.50 and d = 0.25.

128

Table 1. Average number of falsely identified QTL using the different statistical packages and different popul ation sizes for traits 1 to 4 for heritabilities 1.00, 0.75, 0.50, and 0.25, respectively for QTLs.

 

Trait

Method

Population size

1

2

3

4

SAS

100

2.4

0.8

0.4

0.0

200

5.8

3.4

1.6

0.8

300

10.4

7.6

3.4

1.4

PLABQTL

100

1.4

1.0

0.2

0.2

200

1.4

1.6

1.0

0.2

300

0.0

0.8

0.6

0.6

MAOMAKER

100

0.8

0.8

0.4

0.0

6 5 a 4 3 2 1 0 0 0.5 1 1.5 number of QTL
6
5
a
4
3
2
1
0
0
0.5
1
1.5
number of QTL
12 10 c 8 6 4 2 0 0 1 2 number of QTL
12
10
c
8
6
4
2
0
0
1
2
number of QTL

heritability

sas plabqtl mapmaker Poly. (sas)
sas
plabqtl
mapmaker
Poly. (sas)
12 10 b 8 6 4 2 0 0 0.5 1 1.5
12
10
b
8
6
4
2
0
0
0.5
1
1.5

heritability

Figure 2: Power of test for number of QTL detected for a) population size 100, b) population size 200, and c) population size 300.

200

0.8

1.0

0.8

0.4

300

0.0

0.6

0.4

0.2

For these materials, no matter the population size or the heritability of the trait, MAPMAKER always identified the lowest number of pseudo-QTLs. When 100 genotypes were analyzed

PLABQTL was better than SAS. Increasing the number of genotypes

129

analyzed raised the number of falsely identified QTLs for SAS while

there was no difference between PLABQTL and MAPMAKER.

Discussion

The total sample size of an experiment equals the number of QTL

genotypes times the number of replications of the genotypes or lines. We simulated five replications per population. Power is efficiently increased by increasing the number of replications of

QTL genotypes or the size of the population independent of the value of the variance for lines within marker (s 2 n:q ). Knapp and Bridges

(1990) have shown that when s 2 n:q is equal to zero, then increasing the number of replications will result in the increase in the power to detect QTLs. However, when s 2 n:q is not equal to zero analyzing

more than one replication would not efficiently improve the power. Under this situation, it is important to use larger number of

genotypes rather than replicating the lines within the experiment, hence the power was measured as the mean number of QTLs identified for the five replications for each population. These results

show that an experiment of the size greater than 100 genotypes, Mapmaker is capable of detecting relatively small quantitative

effects under high heritability of the trait (Fig 2a). It is however, clear that as soon as the heritability of the trait begins to decrease

the power of the test begins to decrease even with the same number of genotypes. To be able to obtain the optimum power without increasing the number of genotypes, a compromise can be

reached by examining the curves. When heritability of the trait is high, 200 genotypes are adequate. When heritability of the trait is

very low, more than 300 genotypes may be required to obtain the same level of power, for this reason, the size of the experiment can be controlled if the heritability of the trait is known.

For a given nominal type I error, Carbonell et al. (1993) and Jansen (1994) have shown that tests using F values are more powerful than with LOD score. However, Dudley (1993) cautioned that with very

large numbers of markers and a single factor analysis of variance for each marker locus, a certain proportion of the effects are

declared significant when in fact there is no association between the marker and the QTL. We therefore followed the suggestion of Lander and Botstein (1989) to use a probability level of 0.001 to be

equivalent to LOD score of 3.0 in order to increase the power and obtain Type I errors similar to nominal ones. However, this could

130

lead to an increase in Type II errors. From our results we detected

that using single analysis of variance of SAS increased the Type I error rate. Since the method assumes the marker to be the QTL, with a high density of markers as this, those markers close to the

QTLs were probably falsely identified as being associated with the QTLs.

MAPMAKER and PLABQTL used interval and composite interval mapping, respectively, to identify the QTLs. This procedure is a

positional identification. Le Roy and Elsen (1994) therefore believe that it should be used only when a QTL is known to exist. In this

case we know the position of the QTLs so the two methods were appropriately applied. Nevertheless, when the p-values of SAS and LOD scores of both MAPMAKER and PLABQTL are standardized, the

power to detect QTLs become similar. Even though MAPMAKER was the best in identifying QTLs and lowest in type II error rates, it was

cumbersome and not user friendly. Several steps were required before the final results were obtained. The program is therefore limited to computer literates. Single analysis of variance methods

on one hand were simple to compute once the correct model was identified, and could be used even when high-density marker data

was not possible. We recommend them as the first approach for QTL detection, however, caution should be taken to watch out for deviations from normality. PLABQTL was as impressive as

MAPMAKER in the power to detect QTLs. Therefore, it is the program for composite interval mapping in its simplicity in

computing. It does require the least programming and computing time. It also has the advantage of handling data for multiple

environments that are not easily done using MAPMAKER.

Acknowledgment

The author expresses sincere appreciation to Dr Shawn M. Kaeppler

of the Agronomy Department, University of Wisconsin-Madison, USA, for helping to simulate the data and for all the encouragement given during the preparation of this manuscript. Many thanks to my

colleagues at the Savanna Agricultural Research Institute for proofing the manuscript.

References

Allard, R.W. 1960. Principles of plant breeding. John Wiley and Sons, New York.

131

Beckmann, J. S., and M. Soller. 1983. Restriction fragment length

polymorphisms in genetic improvement: methodologies, mapping and cost. Theoretical and Applied Genetics 67:35–43.

Carbonell, E.A., M.J. Asins, M. Baselga, E. Balansard, and T.M.

Gerig. 1993. Power Studies in the estimation of genetic parameters and the localization of quantitative trait loci for backcross and doubled haploid populations. Theoretical and

Applied Genetics 86:411–416.

Carbonell, E.A., T.M. Gerig, E. Balansard, and M.J. Asins. 1992. Interval mapping in the analysis of non-additive quantitative

trait loci. Biometrics 48:305–315.

Darvasi, A., A. Weinreb, V. Minke, J.I. Weller, and M. Soller. 1993. Detecting marker-QTL linkage and estimating QTL gene effect

and map location using a saturated genetic map.

134:943–951.

Genetics

Darvasi, A. and M. Soller. 1994. Optimum spacing of genetic

markers for determining linkage between marker loci and quantitative trait loci. Theoretical and Applied Genetics

89:351–357.

Doerge, R.W., Z.B. Zeng, and B.S. Weir. 1994. Statistical issues in the search for genes affecting quantitative traits in populations. Pages 15–26 In Proc. Analysis of Molecular marker data.

Dudley, J. W. 1993. Molecular markers in plant improvement:

Manipulation of genes affecting quantitative traits. CropScience

33:660–668.

Edwards, M.D., C.W. Stuber, and J.G. Wendel. 1987. Molecular marker facilitated investigations of quantitative trait loci in maize: I Numbers, genetic distribution and types of gene action.

Genetics 116:113–125.

Haley, C.S. and S.A. Knot. 1992. A simple regression method for mapping quantitative trait loci in line crosses using flanking

markers. Heredity 69:315–324.

Hallauer, A.R. and J.B. Miranda. 1988. Quantitative genetics in maize breeding. Iowa State University Press, Ames, Iowa.

Hospital, F., C. Chevalet, and P. Mulsant. 1992. Using markers in

ge ne introgression breeding programs. Genetics 132:1199–1210.

Jansen, R.C. 1994. High resolution of quantitative traits into multiple loci via interval mapping. Genetics 136:1447–1455.

132

Jeon, G.J. 1988. The effects of population size and dominance of

quantitative trait loci (QTL) on the detection of linkage between markers and QTL for livestock. Asian-Australian Journal of Animal Science 8:651–655.

Jian, C. J. and Z.B. Zeng. 1995. Multiple trait analysis of genetic mapping for quantitative trait loci. Genetics 140:1111–1127.

King, R.C. and W.D. Stanfield. 1997. A dictionary of genetics. Oxford University Press, New York and Oxford.

Knapp, S.J. 1994. Mapping quantitative trait loci. Pp 58-96 in P. L. Phillips and I. K. Vasil (eds.) DNA-based markers in plants,

Kluwer Academic Publishers.

Knapp, S.J. and W.C. Bridges. 1990. Using molecular markers to estimate quantitative trait locus parameters: Power and genetic variances for unreplicated and replicated progeny. Genetics

126:769–777.

Lander, E. S., and D. Botstein. 1989. Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps.

Genetics 121:185–199.

Le-Roy, P. and J.M. Elsen. 1994. Numerical comparison between powers of maximum likelihood and analysis of variance methods

for QTL detection in progeny test designs: the case of monogenic inheritance. Theoretical and Applied Genetics 90:65–72.

Lincoln, S.E., M.J. Daly, and E.S. Lander. 1990. Constructing genetic linkage maps with MAPMAKER: a tutorial and reference manual. A Whitehead Institute for Biomedical

Research Technical Report. 2nd edition.

Mangin, B, B. Goffinet, and A. Rebai. 1993. Constructing confidence intervals for QTL location. Genetics 138:1301–1308.

Mansur, L.M., K.G. Lark, Kross, and A. Oliveira, 1993. Interval

mapping of quantitative trait loci for reproductive, morphological, and seed traits of soybean, Glycine max L. Theoretical and

Applied Genetics 86:907–913.

Moreno-Gonzalez, J. 1992a. Estimates of marker-associated QTL effects in Monte Carlo backcross generations using multi ple regression. Theoretical and Applied Genetics 90:65–72.

Moreno-Gonzalez, J. 1992b. Genetic models to estimate additive and non-additive effects of marker-associated QTL using multiple

133

regression techniques. Theoretical and Applied Genetics 85:435–

444.

Muranty, H., 1996. Power of tests for quantitative trait loci detection using full-sib families in different schemes. Heredity 76:156–

165.

Openshaw, S.J., S.G. Jarboe, and W.D. Beavis. 1994. Marker- assisted selection in backcross breeding. Pages 41–43 in Proc. Analysis of Molecular marker data.

SAS Institute Inc. 1996. SAS users Guide: Statistics. SAS Institute, Cary, North Carolina, USA.

Soller, M. and A. Genizi. 1978. The efficiency of experimental

designs for the detection of linkage between a marker locus and locus affecting a quantitative trait in segregating populations. Biometrics 34:47–55.

Staub, J. E., and Sequen. 1996. Genetic markers, map construction, and their application in plant breeding. Horticultural Science

31:729–741.

Tanksley, S.D. and J.C. Nelson. 1996. Advanced backcross QTL analysis: a method for the simultaneous discovery and transfer of valuable QTLs from unadapted germplasm into elite breeding

lines. Theoretical and Applied Genetics 92:92–203.

Touzet, P., R.G. Winkler, and T. Helentjaris. 1995. Combined genetic and physiological analysis of a locus contributing to

quantitative variation. Theoretical and Applied Genetics

91:200–205.

Utz, H.F. and A.E. Melchinger. 1996. PLABQTL: A program for

composite interval mapping of QTL. J. Quantative Trait Loci,

Madison, Wisconsin, USA. Vol. 2.

Weller, J.I., Y. Kashi, and M. Soller. 1990. Power of daughter and

grand daughter designs for determining linkage between loci and quantitative trait loci in dairy cattle. Journal of Dairy Science

73:2525–2537.

Weller, J.I. 1986. Maximum likelihood techniques for the mapping and analysis of quantitative trait loci with the aid of genetic markers. Biometrics 42:627–640.

134

135

AMMI analysis of genotype x environment interaction of open pollinated maize varieties evaluated in the major agro-ecologies of Nigeria

S.R. Ajibade 1 , B.A Ogunbodede 1 , and B.A. Oyejola 2

1 Institute of Agricultural Research and Training, Obafemi Awolowo University, PMB 5029, Ibadan, Nigeria. 2 Department of Statistics, University of Ilorin, PMB 1515, Ilorin, Nigeria.

Abstract

An Additive main effect and multiplicative interaction (AMMI) model was used to analyze data from two sets of open pollinated (OP) maize (Zea mays L.) varieties (early- and late-maturing). Each set was separately evaluated for grain yield in 18 representative locations of the main agroecological zones of Nigeria. Mean grain yields were 2.3 t/ha for the early- and 2.7 t/ha for the late- maturing varieties. Location effects accounted for 76 and 79% of the total variation in the early- and late-maturing varieties,

(GL)

interaction occurred within each set of trial. GL interaction accounted for only 7.4 and 5.6% of the total variation for the two sets of trials, respectively. The first three interaction principal component axes (IPCAs) were significant in the two sets accounting for 82% and 79% of the sum of squares (SS) due to GL interaction for the early and late varieties, respectively. Ten sites within the mangrove and forest agro-environments consistently formed a cluster in each of the two sets of data indicating that these sites induced similar adaptation patterns on the maize varieties. To minimize costs, four testing sites in the mangrove-forest agro- environments would be adequate for conducting yield trials of maize varieties similar to those evaluated in this study. Identification of sites for maize testing in other ecologies awaits further studies.

respectively.

Significant

variety

and

variety

x

location

Résumé

Le model principal effet additif et l`interaction multiplicative (AMMI)

a été utilisé pour analyser les données de deux groupes de

pollinisation ouverte (op) du mais (Zea mays L.) (variétés précoces et tardives). Chaque groupe a été évalué séparément pour le rendement en grains dans 18 localités représentatives des principales zones agro/ écologiques du Nigéria. Le rendement moyen des grains a été de 2,3 t/ha pour la variété précoce et 2,7 t/ha pour les variétés tardives. Les effets des localités sont de 76 et 79% de la variation totale des variétés précoces et tardives

136

respectivement. Une interaction significative des variétés et des variétés x location (GL) a été noté dans chaque groupe d`essais. L’interaction GL compte pour 7,4 et 5,6% de la variation totale pour les deux groupes d’essais respectivement. Les trois premières composantes principales d’interaction (IPCAs) sont significatives dans les deux groupes et comptent pour 82% et 79% de la somme des carrés (SS) dûes à l’interaction GL pour les variétés précoces et tardives respectivement. Dix sites dans la mangrove et les environnements agro forestiers ont formé un cluster dans chaque groupe de données montrant ainsi que ces sites offrent les conditions d’adaptations de ces variétés de maïs. Pour minimiser les coûts, quatre sites de test dans les environnements de la mangrove – forêts sont suffisants pour conduire des essais de rendement des variétés de maïs semblables à celles utilisées dans cette étude. L’identification des sites pour les essais maïs dans d’autres écologies sera faite par des études futures.

Introduction

The primary aim of multilocational trials in plant breeding is to estimate yield of genotypes across diverse environments.

Differential genotypic response to variable environmental conditions associated with changes in the ranking of genotypes

may limit accurate yield estimates and identification of high yielding, stable genotypes. The conventional method of partitioning total variation into components due to varieties, environment and

variety x environment interaction conveys little information on the individual patterns of response (Kempton 1984).

Other methods, such as regression analysis, have been used extensively to partition genotype x environment (GE) interaction SS

as discussed by Gauch (1988). Multivariate analysis techniques such as principal component analysis are often used to simplify interpretation of GE structure by representing complex

relationships among locations or genotypes in a scatter plot (Westcott 1987). Cluster analysis is also used to group locations that

discriminate among genotypes in a similar manner or to summarize the pattern of genotypic performance across environments (Crossa et al. 1991).

An additive main effects and multiplicative interaction (AMMI)

model combines analysis of variance for the genotype and environment main effects with principal component analysis for the

GE interaction (Gauch 1992). With AMMI analysis, the total SS is partitioned into a pattern-rich model and a discarded noise-rich

137

residual thereby adjusting the yields and gaining accuracy (Gauch

and Zobel 1996). In a biplot from AMMI analysis, points for varieties and environments are displayed on the same scattergram so that the expected responses of varieties in a particular environment

may be derived from visual inspection of their relative positions on the plot (Kempton 1984). When subregions are not clearly definable

on geographical basis or breeding is targeted to wide adaptation, the description of genotype x location (GL) effects can contribute to identification of crucial test sites within a given region

(Annicchiarico 1992). The AMMI model may provide a powerful insight into environmental and varietal factors influencing GL

interaction (Wallace et al. 1993; Badu -Apraku et al. 1997).

AMMI analysis has been used to guide and modify the selection of

test sites to increase research efficiency. For instance, M’ Benga (1989) used AMMI model and found that fewer sites could be used for

maize trials in the Gambia. Other researchers have reported similar results for maize (Badu-Apraku et al. 1997; Ndiaye et al. 2001) and other crops ( see inter alia Saindon and Schaalje 1993).

Maize is cultivated throughout Nigeria for domestic consumption

and for industrial uses. The country is endowed with diverse climatic conditions ranging from the high rainforest in the south to the Sahel savanna in the northern parts of the country. The

ultimate objective of the Nationally Coordinated Maize Research Project (NCMRP) in Nigeria is to develop and identify stable high

yielding maize varieties adapted to the specific ecologies in the country. For this reason, the NCMRP annually conducts performance evaluation of candidate maize varieties in the

different ecologies of the country. Although the genotype x environment interaction analysis of the earlier yield trials had

been done, the AMMI model has never been used to analyze the trials. Therefore, the results of those earlier analyzes may not have

properly identified varieties adapted to specific ecologies.

The objectives of this study were to (i) identify stable, high yielding

OP maize varieties, (ii) evaluate the yield potential of the various testing sites, and (iii) use the results as a guide to modify and

select maize test sites in Nigeria.

Materials and Methods

138

Two sets of trials involving open pollinated maize varieties were

conducted in Nigeria during the major growing season (June– November) in 1996 under the auspices of the NCMRP. The trials were conducted at 18 locations across the different agroecological

regions of the country (Table 1), with 7 and 8 early and late varieties, respectively (Table 2). Seed of the varieties were obtained

from the International Institute of Tropical Agriculture (IITA), Ibadan.

A randomized complete block design with four replications was used at each location for each set of varieties. Plots consisted of four

rows, 5 m long, spaced 0.75 m apart with 0.25 m spacing between hills within the row. Two seeds wee planted per hole and the seedlings were later thinned to one plant per hill resulting in a

population density of about 53 333 plants/ha. Fertilizer was applied at the rate of 80 kg N, 40 kg P and 40 kg K per ha at each location

for optimum plant growth.

Data were taken only from the two central rows of each plot. Ears

were shelled mechanically and grain yield (t/ha) was calculated at 15% moisture content. Altitude and rainfall data were obtained from

the records of the Nigerian Department of Meteorological Services, Oshodi, Lagos.

Table 1. Codes, ecology, mean annual rainfall, altitude and mean grain yield of the 18 locations used for evaluating the performance of the early and late maturing maize varieties in Nigeria, 1996.

 

Rainfall

Altitude

Grain yield, t/ha

 

Code

Site

Ecology

(mm)

(m)

Early OPs

Late

 

OPs

A Ibadan

Forest

1365.00

224.00

3.36 b

5.15

b

B Ife

Forest

1610.60

-*

3.06 bc

2.23

h

C Orin-Ekiti

Forest

1653.00

-

1.22 fg

2.63 g

D Ikenne

Forest

-

60.00

1.19 fg

2.94 fg

E Ilorin

SGS

945.30

344.40

3.29 b

4.63

c

F Heipang

Mid-alt.

1382.00

1290.00

1.50 f

1.50

I

G P. Hacourt

MG

2339.70

29.00

1.41 fg

0.61

j

H Zaria

NGS

1043.20

655.00

4.47 a

5.72 a

I Saminaka

Mid-alt.

-

-

2.16 e

2.64 g

J Benin City

Forest

2549.20

78.00

2.27 de

2.64

g

K Abeokuta

Forest

1471.60

106.00

1.25 fg

1.33

I

L Mokwa

SGS

-

302.00

3.06 bc

3.27 ef

M Asaba

MG

-

97.60

0.32 h

0.91 j

139

N Gembu

Mid-alt.

-

-

2.96 c

2.14 h

O Makurdi

NGS

1323.90

106.00

2.32 de

2.63

g

P Samaru

NGS

1043.20

675.00

4.23 a

3.98 d

Q Yola

SS

935.60

190.00

2.57 d

3.41 e

R Nsukka

Forest**

2213.60

390.00

1.12 g

0.66 j

Mean

2.32

2.72

SE

0.11

0.13

Means followed by the same letter are not significantly different at 0.05

probability level. SGS = southern Guinea savanna; NGS = northern Guinea savanna; Mid-alt = mid-altitude; MG = Mangrove forest. * Not available; ** Forest with acidic soil, pH = 4.2

Analysis of variance was carried out separately for each set of trial

and partitioning of GL interaction was done using the AMMI model as described by Zobel et al. (1988). The AMMI model first fits additive

effects for the main factors; that is, ge notypes (G) and environments (E), using the additive analysis of variance procedure. Subsequently,

the program fits multiplicative effects for the interaction term; that is, genotype x environment interaction, using the principal component analysis (PCA). The AMMI model is:

Y ger = µ + ? g +

where:

ß e + S n ? n ? gn d en + ? ge + ? ger

Y

ger

=

the observed yield of the g th genotype in e th location and r th replicate;

µ

=

the grand mean;

?

g

=

the deviation of the mean of the g th genotype from µ;

ße

=

the deviation of mean of the e th environment from µ;

?

n

=

singular value for the n th interaction principal component axis (IPCA);

?

gn

=

genotype eigenvector values for the n th PCA axis;

d en

=

environment eigenvector values for the n th PCA axis;

?ge

=

the residual effect;

? ger

=

the error term (Gauch and Zobel 1996).

Means separation was also carried out by using Duncan multiple range test (DMRT).

Table 2. Code, name, source population, grain type, and mean grain yield of seven early- and eight late-maturing o pen pollinated maize varieties evaluated in 18 locations in Nigeria, 1996.

Code

Grain

Yield

 

Name

Source population

type

(t/ha)

Early-maturing varieties

140

1.

TZE Comp. 4 DMR BC 1

E8430-SR, 1K8149-SR

D*

2.52

2.

TZE Comp. 3 C1

TZE SR-W, DMR-ESRW

D

2.29

3.

DMR-ESR-Y

DMR x TZSR

F

2.29

4.

SUWAN-2-SSR

SUWAN 2

F

2.41

5.

EV8730-SR

POP 30

F/D

2.30

6.

Acr 92 TZE Comp. 5-W

TZE Comp.5

D

2.39

7.

Acr 90 Pool 16-DT

Pool 16

D

2.04

Mean

2.32

SE Late-maturing varieties

 

0.07

1.

TZL Comp. - DMR

TZL Comp. 4

D

2.93

2.

TZL Comp. 3 x 4 C 0

TZL Comp. 3 x 4

F/D

2.85

3.

Akure 93 DMR-LSR-W

DMR-LSR-W

D

2.56

4.

Akure 9322- DMR-SR

Pop. 22-SR

D

2.67

5.

TZB-SR

TZB

F/D

2.53

6.

TZPB-SR-C2

TZPB

D

2.80

7.

Ikenne 9129-SR

Pop.

29-SR

D

2.76

8.

TZ 9043 DMR-SR

Pop. 43-SR

D

2.68

Mean

2.72

SE

0.09

*D = dent, F = flint

Results and Discussion

Mean grain yield across locations was 2.32 t/ha for early- and 2.72 t/ha for the late-maturing maize varieties (Table 2). For both sets, location H (Zaria) had the highest grain yield of 4.47 t/ha for early

and 5.72 t/ha for late varieties while the lowest yields were obtained at Asaba and Nsukka (Table 2). Differences among

locations accounted for most of the variation observed in both sets of trials, about 76% in the early and 79% in the late maturing varieties, respectively (Table 3).

Romagosa and Fox (1993) reported that the proportion of sums of

squares due to differences among sites in most yield trials range from 80–90% and variation due to GE is usually larger than genotypic variation. Similar results have been consistently reported

for maize in West and Central Africa and our results corroborate the earlier studies (Badu-Apraku et al. 1997; Fakorede et al. 1989).

Table 3. AMMI analysis of variance for yield of early and late- maturing open pollinated maize varieties evaluated in 18 locations in Nigeria, 1996.

S.V

DF

SS

% total SS

MS

Early-maturing maize

141

Total

503

822.73

Location

17

628.72

76.42

36.98

**

Genotype

6

9.53

1.16

1.59

**

G

x L

102

60.81

7.39

0.60

**

IPCA1

22

23.25

38.23

1.06

**

IPCA2

20

15.74

25.88

0.79

**

IPCA3

18

10.73

17.65

0.60

**

Residual

42

11.10

18.25

0.26

Error

378

123.66

15.03

0.33

Late-maturing maize Total

575

1502.62

Location

17

1182.63

78.71

69.57

**

Genotype

7

9.85

0.66

1.41

**

G

x L

119

84.31

5.61

0.71

**

IPCA1

23

30.27

35.90

1.32

**

IPCA2

21

20.24

24.10

0.96

**

IPCA3

19

15.89

18.87

0.84

**

Residual

56

17.92

21.22

0.35

Error

432

225.83

15.03

0.53

**Significant at 0.01 level of probability.

Although statistically significant, variation among the maize varieties within each set of trials was small (1.16% for early and 0.66% for the late OPs). Variations due to GL interaction effects ( 7%

and 6% for the early and late maturing varieties, respectively) were relatively larger and also statistically significant.

The first three IPCAs were significant in the two sets and together

accounted for 82 and 79% of the SS due to GL interaction for the early and late varieties, respectively. The biplot (Figs.1a and b) displayed main effects means (across locations and varieties) on the

abscissa and IPCA1, as the ordinate.

All of the early-maturing varieties appeared to be stable and similar in grain yield except variety 7 (Across 90 Pool 16-DT) that had a

high interaction effect (Fig. 1a). All the late maturing maize varieties also appeared to be similar in yield potential. The very

small SS due to genotypes indicated the similarity in grain yields among the maize varieties in each set of trial (Table 3).

The plot of IPCA against mean grain yield clearly showed that six of the seven early maturing varieties loaded on the center of IPCA 1

(Fig. 1a). Variety 7 (Across 90 Pool 16-DT) was completely separated from the others and must have, therefore, been the primary cause

142

of the significant variety mean squares associated with this set of

material. From the biplot of the late-maturing varieties, no variety could be said to be distinctly isolated from all the others (Fig. 1b). Therefore, the varieties may have contributed nearly equally to the

significant variety source of variation in this set.

1.5- 7 I P O C 6 NB E A 0.0- M RCK GF IJ+
1.5-
7
I
P
O
C
6
NB
E
A
0.0-
M
RCK GF
IJ+ 1Q
L
A
H
1
D
34
5
2
-1.5-
P
-+---------+---------+---------+---------+---------+---------+
0.0
0.8
1.6
2.4
3.2
4.0
4.8
Mean grain yield (t/ha)
1.2-
B
2
I
3
1
Q
P
C
K
4
D
A
0.0-
GR M
N
?+
E
A
1
F
6
L
H
5O
7
8
-1.2-
P
-+---------+---------+---------+---------+---------+---------+
0.0
1.0
2.0
3.0
4.0
5.0
6.0

Mean grain yield (t/ha)

Figure 1. Biplot of the AMMI analysis for (a) early-maturing and (b)

143

late-maturing open pollinated maize varieties ( ? = Location points C, I and J superimposed) evaluated in 18 locations in Nigeria. Environment and genotype codes are as in Tables 1 and 2.

The locations showed patterns similar to the corresponding varieties in the biplot of IPCA 1 with mean grain yield (Figs 1a and

b). For the early varieties, clusters were formed as follows:

i. M alone (Asaba: Mangrove),

ii. P alone (Samaru: NGS),

iii. H alone (Zaria: NGS),

iv. RCKGD together (Nsukka, Orin Ekiti, Abeokuta, Ikenne —all in the Forest Zone and Port Harcourt- Mangrove),

v. all others together (all in the savannas, except A and B in

the Forest Zone).

Clustering of the locations for the late-maturing varieties was much more diffused than that of the early varieties. That

notwithstanding, the following groups could be identified:

(i)

P alone (Samaru: NGS),

(ii)

EAH together (Ilorin: SGS; Ibadan: Forest; Zaria: NGS),

(iii)

B alone (Ife: Forest),

(iv)

Q alone (Yola: Sudan savanna),

(v)

GRMK (Port Harcourt: Mangrove; Nsukka: Forest; Asaba:

Mangrove; Abeokuta: Forest),

(vi)

all others (all in the savannas, except Ikenne: Forest).

For the two sets of trials, the biplots consistently showed locations H and P (Zaria and Samaru) as high yielding. In addition, the biplot of the late -maturing trial also picked locations A and E (Ibadan and

Ilorin) as high yielding. On the other hand, locations K, M, and three other locations (Port-Harcourt, Asaba and Nsukka) were

consistently low yielding.

Early varieties adapted to speci fic locations were not identified in

this study (Fig. 1a). Figure 1b, however, shows association of some late -maturing varieties with particular locations. Varieties 1, 2, and

3 (TZL Comp.4, TZL Comp.3 x 4, and DMR-LSR-W) are adapted to site B (Ife). Variety 4 is adapted to Ikenne, variety 6 is suited to Mokwa, while varieties 5, 7, and 8 are suited to Makurdi. Suitable late-

maturing varieties may have to be bred for low yielding sites, such as G, K, M, and R, and for site P with high interaction effect. The

144

three high yielding sites (A, E, and H) may also require special

varieties in order to realize their high yield potential (Table 1 and Fig. 1b). Otherwise, in the absence of location specific varieties, the two most stable late-maturing varieties (4 and 6) may be introduced

to these locations.

The ordination of locations and genotypes on the first two IPCAs for the two sets of maize varieties are presented in Figure 2. For the two sets of data, similarity of sites was fairly related to their

geographic proximity (Fig. 2, Table 1). In Figure 2a, 14 locations (A, B, C, D, E, G, I, J, K, L, M, N, Q, and R) formed a cluster, while in

Figure 2b, 11 locations (A, C, D, E, F, G, J, K, L, M, R) also clustered together. From Figures 2a and 2b it can be inferred that 10 locations (Ibadan, Orin-Ekiti, Ikenne, Ilorin, Port-Harcourt, Benin

City, Abeokuta, Mokwa, Asaba, and Nsukka) consistently fall within the same cluster for each set of data (Fig. 2). Apart from Ilorin and

Mokwa in the southern Guinea savanna, all the locations are within the mangrove and forest agroecologies of southern Nigeria. Therefore, these sites tended to induce similar adaptation patterns

on the maize varieties.

 

1.0-

 

1

   

F

I

 

O

P

 

5

C

 

3

QI

N

A

0.0-

 

P

D AGMC

 

7

2

 

2

J

L B

 

KR

E 6

H

 

-1.0-

 

4

-+---------+---------+---------+---------+---------+---------+

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

 

IPCA1

145

I E P P L A G Q C 8 CRMJ D 2 A 0.0-
I
E
P
P
L
A
G
Q
C
8
CRMJ
D
2
A
0.0-
7
6
+
4 K
2
F
B
O5
N
I
H
-1.0-
1

-+---------+---------+---------+---------+---------+---------+

-1.2

-0.8

-0.4

0.0

0.4

0.8

1.2

 

IPCA1

Figure 2. Ordination of locations and genotypes on the first two IPCA of the genotype-location interaction for (a) early maturing and (b) late maturing open pollinated maize varieties. Environment and genotype codes are as in Tables 1 and 2.

Because the other sites failed to cluster properly for both sets of

varieties, it is hard to determine their similarity from this study. However, earlier studies in Nigeria have estabished similarity of

some of the sites used in this study. Fakorede et al. (1989) subjected the four-year (1985—88) grain-yield data of the NCMRP to principal component factor analysis with varimax rotation for

orthogonality of similarity grouping of the test locations and adaptation of the varieties. Analysis of variance involving only

locations grouped into each factor eliminated, or at least considerably reduced the G x E interaction. On the basis of the multivariate statistical grouping of the locations, they identified

four distinct zones and suggested testing maize varieti es in only these zones:

i. Forest-transition-southern Guinea savanna;

ii. Northern Guinea savanna;

iii. Sudan savanna;

iv. Midaltitude.

Menkir et al. (2000) subjected GIS spatial climatic data of 114 maize testing sites in sub-Saharan Africa to cluster analysis. Nearly all of

146

the sites used in our study were included in the analysis. The

analysis clearly demarcated four regions almost exactly as reported by Fakorede et al. (1989).

Considering these earlier results along with ours, it can be concluded that maize responds to the Forest, transition, and

southern Guinea savanna zones of Nigeria similarly. These ecologies may now be taken together for general maize testing purposes. A reduction in the number of sites for testing open

pollinated maize varieties in these zones is, therefore, warranted. The reduction in testing sites may free up some resources that can

be used for testing materials in new maize growing areas, especially in the northern part of the country. Although the choice of specific sites to be used for testing in this zone depends on many

factors, we propose four sites on the basis of our results. Ibadan may represent a high-yielding forest agro-environment, while Abeokuta,

Port-Harcourt, and Nsukka represent low-yielding forest (acid soil) environment. The sites to be recommended for other ecological zones in Nigeria must await further experimentation.

Acknowledgement

The authors are grateful to the Director, Institute of Agricultural

Research and Training (IAR&T), Ibadan, for permission to publish this paper. The International Institute of Tropical Agriculture (IITA), Ibadan is acknowledged for providing the seed of the maize varieties

for evaluation. The contributions of collaborators of the Nationally Coordinated Maize Research Project are also gratefully

acknowledged. We also thank staff of the Department of Meteorology Services, Oshodi, Lagos, for providing the rainfall and elevation data.

References

Annicchiarico, P. 1992. Cultivar adaptation and recommendation from alfalfa trials in Northern Italy. Journal of Genetics and

Breeding 46: 269–277.

Badu-Apraku, B., J.M. Fajemisin, and A.O. Diallo. 1997. The performance of early and extra-early varieties across

environments in West and Central Africa. Pages 149–159 in B. Badu -Apraku, M.O. Akoroda, M. Ouedraogo, and F.M. Quin (eds.) Contributing to food self-sufficiency: Maize research and development in West and Central Africa. Proceedings of a

147

Regional Maize Workshop, 29 May to 2 June 1995, IITA-Cotonou,

Benin Republic. WECAMAN/IITA.

Crossa, J, P.N. Fox, W.H. Pfeiffer, S. Rajaram, and H.G. Gauch. 1991. AMMI adjustment for statistical analysis of an international

wheat yield trial. Theoretical and Applied Genetics 81: 27–37.

Fakorede, M.A.B., J.E. Iken, S.K. Kim, and J.H. Mareck. 1989. Empirical results from a study of maize yield potential in the

different agroecologies of Nigeria. Pages 79–97 in J.M. Fajemisin, N. Muleba, A.M. Emechebe, and C. Dabire (eds.) Towards production technologies in semi-arid West and

Central Africa

SAFGRAD & IITA. 277 Pp.

Gauch, H.G. 1988. Model selection and validation for yield trials with interaction. Biometrics 44: 705–715.

Gauch, H.G. 1990. Full and reduced models for yield trials.

Theoretical and Applied Genetics 80: 153–160.

Gauch, H.G. 1992. Statistical analysis of regional yield trials:

AMMI analysis of factorial designs. Elsevier, Amsterdam, The

Netherland. Pages 53–110.

Gauch, H.G. and R.W. Zobel, 1996. AMMI analysis of yield trials. Pp 85–122 in M.S. Kang and H.G Gauch (eds.) Genotype by

environment interaction. Boca Raton, New York, USA. CRC Press.

Kempton, R.A. 1984. The use of biplots in interpreting variety by

environment interactions. Journal of Agricultural Science, Cambridge 103: 123–135.

M’Benga, M. 1989. The use of genotype x environment interaction

to enhance maize ( Zea mays L.) cultivar and test site selection in

the eastern part of Gambia. M.S. Thesis, Cornell University,

Ithaca, New York.

Menkir, A., J.G. Kling, S.S. Jagtap, and B.A. Aliu. 2000. GIS based classification of maize testing locations in West and Central

Africa. Maydica 45:143–150.

Ndiaye, A., S.K. Vasal, S. Mclean, and J. Crossa. 2001. Comparison of regression and AMMI analyses for assessing yield stability of maize inbred lines. Pages 114–127 in B. Badu-Apraku, M.A.B.

Fakorede, M. Ouedraogo, and R.J. Carsky (eds.) Impact, challenges and prospects of maize research and development

in West and Central Africa. Proceedings of a Regional Maize

148

Workshop,

WECAMAN/IITA.

4–7

May

1999,

IITA-Cotonou,

Benin

Republic.

Romagosa, I. and P.N. Fox. 1993. Ge notype x environment interaction and adaptation. Pages 373–390 in M.D. Haayward,

N.O. Bosemark, and I. Rosmagosa (eds.) Plant breeding:

Principles and prospects. Pretice Hall, London.

Saindon G., and G.B. Schaalje. 1993. Evaluation of locations for

testing dry bean cultivars in western Canada using statistical procedures, biological interpretation and multiple traits. Canadian Journal of Plant Science 73: 985–994.

Wallace, D.H, J. Baudoin, D.P Beaver, D.E. Coyne, P.N. Halseth, H.M.

Masaya, J.R. Munger, M. Myers, K.S. Silbernagel, Yourstone and R.W. Zobel. 1993. Improving efficiency of breeding for higher crop

yield. Theoretical and Applied Genetics 86: 27–40.

Westcott, B. 1987. A method of assessing the yield stability of crop genotypes. Journal of Agricultural Science, Cambridge 108:

267–274.

Zobel, R.W., M.J. Wright, and H.G. Gauch. 1988. Statistical analysis of a yield trial. Agronomy Journal 80: 388–393.

149

Progress from reciprocal recurrent selection in two early-maturing maize composites

A.E. Salami 1 , A. Menkir 2 , M.O Akoroda 1 , and V.O Adetimirin 1

1 Department of Agronomy, University of Ibadan. Ibadan, Nigeria 2 International Institute of Tropical Agriculture. Ibadan, Nigeria

Abstract Two early-maturing maize (Zea mays L.) composites, TZE COMP 3 and TZE COMP 4, are undergoing improvement by reciprocal recurrent selection (RRS) at the International Institute of Tropical Agriculture (IITA), Ibadan, Nigeria. Fifty random S 1 lines from each of the original (C0) and the two improved cycles (C1 and C2) of selection were evaluated at Ikenne in 1998 and 1999 and at Zaria in 1999. The objective was to determine the progress from two cycles of selection. Response to selection for grain yield was 10.4% per cycle in TZE COMP3 and 5.7% in TZE COMP4. Selection also increased days to 50% silking, ear height, and plant height by 1.18 to 3.76% per cycle, but reduced plant and ear aspects. Genetic variances among S 1 lines for grain yield and other agronomic traits were significant (P <0.05). Broad sense heritability estimates were high (h 2 = 0.41 to 0.88) with little variation across cycles for most of the traits. These results indicate that RRS was effective in improving grain yield and other traits without reducing genetic variance. Thus, it is possible to make further progress from selection for grain yield and other traits in these composites. Coefficients of genetic correlation of grain yield with plant and ear aspects were negative. However, genetic correlation of grain yield with days to 50% silk, ear height, and plant height were positive in both composites, suggesting that gains from selection for grain yield was related to changes in these traits .

Résumé Deux composites précoces de maïs (Zea mays L.), TZE COMP 3 et TZE COMP 4, sont en amélioration par la sélection recurrente réciproque (SR8) à l’institut international d’agronomie tropicale (IITA), Ibadan Nigéria. 50 lignées aléatoires S1 de chaque composite original (C0) et deux cycles améliorés (C1 et C2) ont été évalués à IKENNE en 1998 et 1999 et à Zaria en 1999. L`objectif était de déterminer le progrès de deux cycles de sélection. La réponse à la sélection du rendement en grain a été de 10,4% par cycle dans TZE COMP3 et 5,7% dans TZE COMP4. La sélection a prolongé les jours à 50% de floraison, la hauteur des épis et la hauteur des plants ont été de 1,18 à 3,76% par cycle de maïs avec une réduction des aspects des plants et des épis. La variance génétique entre les lignées S1 pour le rendement en grain et d’autres caractères agronomiques a été significative (P<0,05).

150

L`héritabilité estimée dans le sens large a été haute (h 2 = 0,41 à 0,88) avec une faible variation entre les cycles pour beaucoup de caractères. Ces résultats indiquent que le RRS a été efficace dans l’amélioration du rendement en grain et d’autres caractères sans réduire la variance génétique. Ainsi, il est possible de faire plus de progrès à partir de la sélection pour le rendement en grain et d’autres caractères dans ces composites. Les coefficients de correlation génétique du rendement en grain avec les aspects de la plante et de l’épis sont négatives. Cependant, la correlation génétique du rendement grain avec les jours à 50% de floraison la hauteur des épis et des plants sont positives dans les deux composites, ceci suggère que les gains à partir de la sélection pour le rendement en grain est lié aux changements de ces caractères.

Introduction

Maize ( Zea mays L.) is one of the five most important crops in West

and Central Africa (WCA) accounting for 7 to 70% of the total cereal production (Manyong et al. 1996). Production of maize has expanded

considerably in many West African countries. This has been made possible by the adoption of improved varieties identified from

international trials coordinated by the International Institute of Tropical Agriculture (IITA), the Semi-Arid Food Grain Research and Development (SAFGRAD) project, and the West and Central Africa

Collaborative Maize Research Network (WECAMAN) (Menkir and Kling, 1999). Development of early and extra-early maturing

varieties in recent years has also enhanced the expansion of maize production into the drier areas of the savanna where the characteristic short rainy season had prevented its cultivation

(Manyong et al. 2000). There is also high demand for early maize in the forest zone because it allows farmers to market the early crop

as green maize at prime price. Furthermore, early maize provides farmers in different ecological zones with flexibility in date of planting and it is the crop most suitable for filling the hunger gap

after a long dry period (Menkir and Kling 1999).

To satisfy the need for improved early-maturing varieties, IITA developed several populations from various sources of germplasm. The limitation in the number of breeders working on maize at IITA

prompted the adoption of the comprehensive breeding system (CBS) proposed by Eberhart et al. (1967) to consolidate the gains achieved

in several populations and to integrate population improvement with hybrid development activities. The CBS involves the formation

of breeding population from diverse sources of germplasm and the

151

application of reciprocal recurrent selection (RRS) scheme. The

expected products of CBS are superior open-pollinated varieties, varietal crosses, inbred lines and hybrids. In Africa, this system has been successfully used in Kenya (Eberhart et al. 1967; Darrah et al.

1978; Eberhart et al. 1991) and Cameroon (Everett et al. 1994).

Two early-maturing composites, TZE Comp 3 and TZE Comp 4 were used in the CBS. These composites have gone through two cycles of reciprocal recurrent selection (RRS) using S 1 testcross progeny as

the selection unit (MIP, 1996). The progress from selection based on inter-population testcrosses of S 1 lines shows that RRS has been

effective in improving the composites with mid-parent heterosis of 4% for C1 and 7% for C2 relative to the original populations (Menkir and Kling 1999). However, the progress from selection based on the

performance of S 1 per se had not been studied. This information is

useful for the evaluation of the potential of the two composites as

sources of productive inbred lines for the development of hybrids and synthetics.

This study was, therefore conducted to evaluate the S 1 lines per se

for (i) changes in means and genetic gain from selection for grain

yield and other agronomic traits, and (ii) changes in heritability and genetic correlation between grain yield and other agronomic traits.

Materials and Methods

Two broad based early-maturing maize composites, TZE Comp 3 and TZE Comp 4, identified from a combining ability study (MIP 1996)

were used for the study. TZE Comp 3 was developed by crossing selected S 1 lines from TZESR-W C 3 to DMR-ESR-W while TZE Comp 4 was formed by crossing EV8430-SR to Ikenne (1) 8149-SR. The two

composites have unde rgone two cycles of RRS for improved grain yield, resistance to foliar diseases and other desirable agronomic

traits using the reciprocal S 1 selection scheme. In each selection cycle, 500 to 1000 S 1 lines derived from each composite were evaluated for agronomic traits. Selection of S 1 lines for

recombination was based on yield potential, resistance to foliar diseases, ear rot, stalk rot and root lodging of testcross progenies. A

base index was used to identify the best 40–50 S1 lines for recombination to form the next cycles. Five seasons (two seasons per year) were needed to complete one cycle of selection for each

composite. Details of the RRS procedure are reported in MIP Annual Report (1996).

152

In 1997, 50 random S 1 lines derived from each cycle of selection (C0, C 1 and C 2 ) of the two composites were sib-mated to increase seeds for evaluation. Evaluation of the S 1 lines was conducted in three

environments (Ikenne 1998, Ikenne 1999, and Zaria 1999). A total of 300 S 1 lines (50 lines x 2 composites x 3 cycle s) were divided into

10 equal sets; each set contained 5 lines from each cycle of each composite. A set consisted of five randomly selected S 1 lines from each cycle of each composite. Each set was planted in a randomized

complete block design with two replicates in each of the three environments. The S 1 lines were planted in single row plots, 5 m

long with 75 cm between rows and 25 cm within a row. Fertilizer and field management practices recommended for optimum maize production were used for each environment.

The number of days from planting till 50% of the plants had

incipient silk extrusion in each plot was recorded as days to silk. Plant and ear heights were measured in cm as the distance from the base of the plant to the first tassel branch and the node bearing

the upper ear, respectively. Plant aspect was rated on a scale of 1 to 5 based on overall plant type, where 1 = excellent plant type and 5 =

poor plant type. Ear aspect was also scored on a 1 to 5 scale, where 1 = clean, large, uniform and well-filled ears and 5 = ears possessing

undesirable features. All ears harvested from each plot were weighed and representative ear samples were shelled to determine percent moisture. Grain yield was computed from ear weight (EWT,

kg/m 2 ), adjusted to 15% moisture content (MOIST) and 80% shelling percentage (SHELL), using the following formula:

Grain yield (Mg ha -1 ) = EWT*[(100-MOIST)/85]*(10000*SHELL)

Data were subjected to the analysis of variance (ANOVA) using the General Linear Model procedure (GLM) for randomised complete

block design SAS (SAS Institute, 1995). ANOVA was computed for each selection cycle of the two composites. Environments, S 1 lines (hereinafter referred to as genotypes), and genotype x environment

interaction were considered as random factors in determining the expected mean squares for the analysis. Genotype x environment

interaction mean square was used as the denominator in the F-test for significant genotypic effects. Response to selection was calculated by regressing means of genotype s averaged over

environments on the cycles of selection. Percentage gain per cycle was obtained by dividing the linear regression coefficient by the

153

intercept multiplied by 100 (Hallauer and Miranda 1988). Genetic

variances for the traits in each cycle of selection were estimated from the combined ANOVA by equating the observed mean squares with the expected mean squares. Broad-sense heritability was

estimated as the ratio of the genetic variance to phenotypic variance on progeny mean bases. Standard errors for the estimates

of genetic variance and heritability were computed using the method described by Hallauer and Miranda (1988). Genetic correlation between grain yield and other traits were calculated

from variance and covariance estimates by using the formula of Falconer and Mackey (1996). Standard errors of genetic correlation

values were computed based on the method of Mode and Robinson

(1959).

Results and Discussion

Average grain yield of S 1 lines across the three environments was 3.03 Mg ha -1 for TZE Comp 3 and 3.17 Mg ha -1 for TZE Comp 4. Mean

grain yield in the test environments were considerably wide ranging from 2.37 Mg ha -1 at Ikenne in 1998 to 4.12 Mg ha -1 at Zaria for TZE Comp 3 and from 2.39 at Ikenne in 1998 to 4.26 Mg ha -1 in

Zaria for TZE Comp 4. Mean grain yield increased with selection in the two composites (Table 1).

Significant (