SECTION II

Breeding & Seed Production

104

105

Variabilit gntique des ecotypes locaux de mas pour la tolrance la scheresse au Sngal
Abdou Ndiaye,
ISRACNRA BP, 53 Bambey, Sngal
Rsum La variabilit gntique est un lment essentiel pour tout programme de slection. Ainsi, 36 cotypes locaux de mas et 5 populations tolrantes la scheresse en provenance du CIMMYT ont t valus en station sous deux rgimes: i) sous irrigation complte et ii) sous stress hydrique induit la floraison pour lidentification de source de tolrance la scheresse. Plusieurs caractres agromorphologiques comme la hauteur moyenne du plant et de lpi, la date de floraison, le rendement et ses composantes ont t mesurs. Lutilisation de lanalyse de variance et de lanalyse en composantes principales a permis didentifier et de dcrire deux groupes rpartis en fonction de la nature du rgime hydrique. Quelques associations significatives entre caractres ont t trouves et pourraient tre dues la plotropie, au linkage chromosomique, au dsquilibre gamtique ou la slection. Etant donn le nombre important de paramtres prendre en considration pour la tolrance la scheresse, seuls ceux qui semblent plus importants seront utiliss dans le cadre du criblage au champ ou au laboratoire la fois sous rgime complet ou sous stress afin de combiner rendement lev et tolrance la scheresse. Abstract Genetic variability is essential for continued genetic improvement of any crop species f r stress tolerance. A total of 36 maize (Zea o 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 slection exige lexamen dun matriel aussi vaste que possible, dautant plus que le nombre de p roblmes dont on espre trouver la solution est lev. Dans quelque domaine que ce soit, les buts principaux atteindre sont : grande productivit, qualit et rgularit de la production : rsistance et/ou la tolrance aux diffrents alas climatiques dont la scheresse et les maladies. Ladaptation de la plante aux conditions du milieu physique tels les inconvnients climatiques (scheresse, fortes variations thermiques, vents desschants, etc.), le faible niveau de fertilit minrale des sols (teneur fai ble en azote) et la demande progressivement croissante sur le march lgitiment et commandent la recherche de techniques permettant une production accrue du mas. De nos jours, certaines limites dans les capacits damlioration des principales cultures ont t atteintes et de faon gnrale dans la recherche et le choix de nouvelles sources de variabilit pour la tolrance aux diffrents stress biotiques et abiotiques. Cest en particulier le cas du mas dont la base gntique des principales varits c ultives en France est trs troite, avec la prdominance de quelques lignes lites et de leurs drivs. Aux Etats-Unis, seuls quelques gnotypes issus des meilleures populations synthtiques ou traditionnelles (Iowa Stiff Stalk Synthtic, Lancaster Sure Crop, Reids Yellow dent et Krug) sont utilises. En Europe tous les hybrides cultivs sont, pratiquement issus dun parent dent amricain et dun parent corn europen, qui est en gnral apparent aux deux (2) lignes F2 et F7 issues de la population Lacaune. Au CIMMYT et en Afrique occidentale et centrale, le dveloppement de populations tolrantes la scheresse sest limit au criblage dun nombre plus ou moins restreint de varits performantes et dcotypes. Cest ainsi que, suivant diffrentes approches de slection pour la tolrance la scheresse, des populations sources ont t dveloppes telles Pool 16 sequia, Tuxpno sequia, DTP1, et DTP2 partir dun schma de slection rcurrente intrapopulation dune part, et dautre part par le cribl age et la combinaison de varits (lignes, populations, hybrides, cotypes etc) tolrantes et performantes en conditions de scheresse (Edmeades et al. 1997). En Afrique occidentale et

107

centrale, deux populations DRT-EW (blanc dent prcoce) et DRT-EY (jaune corn prcoce) ont t obtenues partir dcotypes locaux (Th et al. 1997). Cette approche est limite, entre autres, par les nombreuses interactions gnotype x environnement et le dterminisme plurignique de la tolrance au dficit hydrique. Cette limite dordre gntique, qui peut prsenter des risques agronomiques certains, est lie ou sinon due la non-utilisation de toute la variabilit actuellement disponible, soit par souci defficacit court terme, soit par manque de mthodologie pour pouvoir utiliser toute cette variabilit. Cest pour rpondre ces diffrents aspects et pour prparer le progrs gntique moyen et long terme que le programme sest fix comme objectifs de :
?? rassembler un grand nombre de dcotypes locaux adaptes aux conditions climatiques en particulier de scheresse ;

?? se donner les moyens dtude et de conservation de cette variabilit ; ?? prciser les mthodologies de gestion et dutilisation de cette variabilit en slection pour la tolrance la scheresse. En dfinitive, le but de ce travail consiste effectuer lvaluation de la variabilit des populations locales de mas pour la tolrance la scheresse afin dlargir la base gntique existante dune part et proposer dautre part des mthodes dintgration de ce germplasme dans le programme de recherche dune nouvelle variabilit.

Matriel et Mthodes Matriel vgtal

Le matriel vgtal tudi est compos de 36 cotypes locaux issus de la prospection de 1990 travers toutes les zones masicoles du Sngal et 5 populations tolrantes la scheresse en provenance du CIMMYT. La liste des cotypes locaux figure au niveau du tableau.
Tableau 1. Liste des cotypes locaux. No 1 2 3 4 5 6 7 8 Code TB1 TB2 TB4 TB5 TB57 TB58 TB60 VG6 No 9 10 11 12 13 14 15 16 Code VG9 VG10 VG11 VG12 KD13 KD14 KD16 KD18 No 17 18 19 20 21 22 23 24 Code KD19 KD20 KD22 KD23 KD26 KD27 KD40 KD41 No 25 26 27 28 29 30 31 32 Code KD43 KD46 SD28 SD29 SD30 SD31 SD32 SD34 No 33 34 35 36 37 38 39 40 41 Code NR50 NR54 NR55 NR56 DTPW DTPY TS6 PSC PSS

108

Mthodes
Le dispositif exprimental utilis est un essai en blocs complets randomiss avec 2 rptitions. Les lignes sont spares par des alles de 0,80 m, lcartement entre les poquets est de 0,50 m soit une densit optimale de 50 000 pieds/ha. Deux rgimes hydriques ont t appliqus : un rgime hydrique sans stress du semis la maturit (irrigation complte) et un rgime hydrique avec un stress simul pendant la phase de prfloraison floraison par arrt de lirrigation jusqu la maturit. Tous les essais ont t sems sur la mme bande durant la contre -saison de 1999 et 2000 et ont reu respectivement 300 kg/ha de N.P.K. (8.18.27) avant le semis, 150 kg d'ure la montaison et 150 kg/ha d'ure la floraison. Lessai a t mis en place la station de recherche de Nioro du Rip situ 14 8 latitude et 16 4 longitude au niveau de la zone de culture par excellence du mas. Elle se caractrise par un climat sahlo - soudanien, une pluviomtrie variant entre 400 et 600 mm, des tempratures minimale de 15 18 C, maximale de 35 40 C et moyenne de 25 28 C. Le choix des caractres mesurs a t effectu de faon ce quils apportent le maximum dinformations sur la plante. Il sagit des caractres agronomiques (rendement, prcocit), de caractres quantitatifs associs au rendement (mesures morphologiques) et des facteurs de rgularit du rendement telle la tolrance la scheresse et autres alas. Ltude de chacun des aspects de la variabilit gntique ncessite lemploi de techniques de description et danalyse diffrentes suivant que lon dsire tudier chaque caractre indpendamment ou leur ensemble, de faon synthtique. Lanalyse de variance a t utilise pour quantifier les variations inter et intrapopulations qui constituent la variation totale. La structure gnrale de la variabilit a t ensuite tudie par des analyses multivaries comme lanalyse en composantes principales. Cette analyse a t utilise par diffrents auteurs : Hugues de Cherisey (1983) sur le millet et Ndiaye (1985) sur le mas.

Il est noter que pour les analyses suivantes, les cotypes sont identifis par trois chiffres dont le premier indique le rgime hydrique (un nombre pa ir correspond au rgime sous irrigation complte ; un nombre impair au rgime stress hydrique) et les deux

109

derniers chiffres correspondent au numro du gnotype. Dans le cadre de lanalyse de variance combine, les essais sont identifis par le rgime complet(IR pour irrigation complte, ST pour le stress hydrique, ces lettres sont suivis par les deux derniers chiffres de lanne dexprimentation.

Resultats et Discussion Analyse de variance

Les rsultats de lanalyse de variance individuelle des essais montre que tous les caractres prsentent un effet gnotypique hautement significatif. Les rsultats de lanalyse de variance combine suite au regroupement des essais deux deux (Tableaux 2a et 2b) montrent un effet rgime dirrigation hautement significatif except les essais sous irrigation complte (IR9900) pour les caractres comme la dure semis-floraison femelle 50%, la hauteur du plant (HMP). Pour le cas de lessai Irrigu + Stress (IRST99), les effets rgime dirrigation et gnotype sont significatifs pour la hauteur moyenne du plant (HMP) et pour le rendement (RDT) au seuil de 5%. Cet effet rgime dnote de lefficacit du stress hydrique appliqu et qui se traduit par un indice de svrit de la scheresse variant entre 52 et 85% en fonction du gnotype.
Tableau 2a. Analyse de variance selon le modle interactif complet sur IRST99, sur IR9900 et sur ST9900 (Somme des Carrs des Ecarts). Rgimes Source ddl HMP 32012.2* 29317.2* 3301.5 9514.0ns 36885.9 IRST99 RDT (105) 3508 4438 2257 2117 6776 HMP IR9900 RDT (105) HMP 17329.8* 9225.0ns 4222.5 8875.0ns 29552.4 ST9900 RDT (105) 290* 932ns 150 198ns 332

Gnotype 40 Rgime 1 BlRgime 2 GxR Rsid. 40 80

17817.9ns 5686** 7994.0ns 498ns 3301.5 174 15499ns 32985 332ns 1000

Tableau 2b. Analyse de variance selon le modle interactif complet sur IRST00 et IRST9900 (Somme des Carrs des Ecarts). Rgimes Source Gnotype Rgime ddl 40 1 IRST00 HMP 12598.7* 27127.4* RDT(105) 391* 246* 25023.6* 73641.6* IRST9900 HMP RDT(105) 445* 6744*

110

Bl(Rgime) GxR Rsid.

2 40 80

4222.5 14397.5ns 25652.4

99 336ns 655

7524.1 26498.9ns 62538.4

325 844ns 1332

Pour le regroupement des essais de lan 2000 (IRST00), en plus de leffet gnotype, leffet rgime hydrique est significatif pour les variables comme la hauteur du plant et le rendement. Cette variabilit inter- rgime hydrique montre nouveau lefficience du stress appliqu comme cela a t le cas en 1999. Pour le regroupement des essais stresss (ST9900), les effets gnotype et rgime hydrique sont significatif pour les caractres tels le rendement et la hauteur moyenne du plant. Comme dans le groupe prcdent, les gnotypes semblent avoir ragi de la mme manire face au stress appliqu pendant les deux annes dexprime ntation. Pour lanalyse globale, leffet rgime est significatif pour le rendement et la hauteur moyenne du plant, ce qui confirme les rsultats des analyses prcdentes des essais regroups par anne et par rgime. Il faut cependant noter labsence d une interaction gnotype x rgime hydrique sur les groupes IRST99, IRST00 et sur lensemble des quatre essais regroups IRST9900 (cf. Tableau 2b).

Association des caractres

Seuls les caractres quantitatifs et dintrt agronomique ont d prsenter des corrlations hautement significatives. Ainsi observet-on : ?? de fortes corrlations positives existent entre la hauteur totale (HMP), la hauteur de lpi (HEPI), la date de floraison (FLOR) et la verse. Il semble donc quune floraison tardive saccompagne dun fort dveloppement vgtatif et dune sensibilit la verse ; ?? la vigueur au dpart et la floraison (r = -0,62). On en dduit que les populations prcoces paraissent donc mieux adaptes aux conditions en dbut de vgtation ?? la hauteur moyenne de l pi, la date de floraison et la pourriture des tiges. Un bon dveloppement vgtatif est li une diminution de la pourriture des tiges mais cependant, une floraison prcoce est statistiquement associe une

111

augmentation de la pourriture des tiges car les populations prcoces ont des tiges plus vieilles la rcolte. On peut ainsi suggrer les causes suivantes pour expliquer ces associations de caractres. Il sagit de la pliotropie gntique assimilable une corrlation physiologique, du linkage chromosomique, du dsquilibre gamtique, de la slection, de lassociation fortuite par drive gntique ou plus simplement par effet dchantillonnage et enfin du jeu des quatre premiers facteurs entre chacun des deux caractres et un troisime caractre.

Analyse en composantes principales

Lanalyse en composantes principales (A.C.P) a t effectue la fois sur les caractres quantitatifs et qualitatifs dans le but de mieux cerner la variabilit existante. Pour ce faire, les essais ont t regroups par anne et par rgime. Il sagit de IRST9900 (essai irrigu + essai stress en 1999 et en 2000) soit 4 essais ayant fait lobjet dune analyse globale, de ST9900 (regroupement des essais stresss de 1999 et de 2000), et de IR9900 (regroupement des essais sous irrigation complte en 1999 et en 2000). Pour lessai IRST9900, les quatre premiers axes reprsentent 74,5% de la variabilit totale, le cinquime axe avec 8,2% dinertie contribue moins quune variable mesure la description de la variabilit totale (Fig. 1). Pour lessai ST9900, les quatre premiers axes expliquent environ 79,6% de la variabilit totale (Fig. 2) et enfin pour le groupe IR9900, les quatre axes expliquent 74,2% de la variabilit totale (Fig. 3). Un rsum de lanalyse en composantes est donn par le Tableau 3. Ainsi : ?? La premire composante principale est un axe de rendement et darchitecture. Elle oppose principalement les deux rgimes hydriques cest--dire le rgime sous irrigation complte et celui rgime stress induit partir de la floraison. On trouve les gnotypes haut rendement, de grande taille du ct du rgime sous irrigation complte et loppos des gnotypes ayant des rendements faibles, une verse et une casse importantes et une incidence la scheresse trs leve.

112

n Valeur Pourcent Cumul 1 3.0499 2 1.4328 3 1.2431 4 0.9763 5 0.7370 6 0.5856 7 0.4498 8 0.4016 33.89 15.92 13.81 10.85 8.19 6.51 5.00 4.46 33.89 49.81 63.62 74.47 82.66 89.16 94.16 98.62

3.0499

Variance totale = 9.0

Figure 1. Valeurs propres du groupe dessais IRST99.

n Valeur Pourcent Cumul 1 2.7347 2 1.5816 3 1.2385 4 0.8128 5 0.5968 6 0.4767 7 0.4037 34.18 19.77 15.48 10.16 7.46 5.96 5.05 34.18 53.95 69.44 79.60 87.06 93.01 98.06

2.7347

Variance totale = 8.0

Figure 2. Valeurs propres du groupe dessais ST9900.

n Valeur Pourcent Cumul 1 2.2159 2 1.5719 3 1.1837 4 0.9632 5 0.7654 6 0.6851 7 0.4356 8 0.1791 27.70 19.65 14.80 12.04 9.57 8.56 27.70 47.35 62.14 74.18 83.75 92.32 0 2.2159

5.45 97.76 2.24 100.00

Variance totale = 8.0

Figure 3. Valeurs propres du groupe dessais IR9900

?? La deuxime composante est un axe de caractristiques de la prcocit. Elle oppose les populations de grande taille plus prcoces sous irrigation complte celles de taille souvent rduite et de dure semis-floraison femelle plus longue en raison du stress appliqu. ?? La troisime composante est un axe caractri stique de la sensibilit la verse et la scheresse. Elle oppose les cotypes

113

trop sensibles au stress hydrique et qui ont une verse importante ceux plus ou plus tolrants aux conditions de scheresse et ayant une incidence la scheresse relativement faible. ?? La quatrime composante principale est un axe caractristique de larchitecture avec des gnotypes ayant une bonne densit la leve et qui manifestent une casse relativement importante vers la rcolte.
Tableau 3. Rsum de lanalyse en composantes principales sur les moyennes. Axe Axe 1 Axe 2 Axe 3 Axe 4 Pourcentage dinertie explique 34.1 19.8 15.5 10.9 Variables contribuant la dfinition de laxe RDT (+0.161) HMP (+0.188) HEPI (+0.200) FLOR(+0.164) HUM (+0.126) VERS(+0.148) INCI (0.195) NPL (+0.600) CAS (+0.100)

Lobservation du Tableau 3 montre que les deux premiers axes sont caractriss par des caractres quantitatifs contrairement aux troisime et quatrime axes qui le sont par les caractres essentiellement qualitatifs. La Fi gure 4 donne la projection des caractres ou les cercles de corrlation sur les plans engendrs par les axes principaux 1 et 2.
1,0000 2

HMP RDT HE -1,6400

VER NPL CAS

1,6400 1

INCI HUM

FF50

-1,0000

Figure 4. Projection des caractres sur le plan engendr par les axes 1 et 2.

114

La projection des cotypes sur le plan engendr par les axes 1 et 2 (Fig. 5 et Fig. 6) montre deux groupes principaux : ?? le groupe qui sidentifie au rgime sous irrigation complte matrialis par les numros dont le premier chiffre est impair. Il est compos des cotypes locaux potentiel de rendement lev, une hauteur moyenne du plant et de lpi importantes et,
??

le groupe qui caractrise essentiellement le rgime stress matrialis par les numros dont le premier chiffre est pair. Il est essentiellement compos dindividus rendement trs rduit en raison de leur sensibilit la scheresse. Ces derniers sont de taille relativement rduite en raison du rgime auquel ils ont t soumis.
3,4000 2

433 324 333 302 305

423

235 436 331 209 220 222 HMPT RD 211 216 229 5,3000 1 214

315

325 313

-5,3000 219

326 338

327

308 407 135

221 426 230

212

317

113 239 241 -3,4000

Figure 5. Dispersion des cotypes dans le plan engendr par les axes 1 et 2.
3,4000 3

133

108

114

214

134 -5,3000 315 326 327

HMP 220 219 216 229 212 5,3000 1

211 313 308 335 236 222

303

235 341 306 234 204 224

-3,4000

115

Figure 6. Dispersion des cotypes dans le plan engendr par les axes 1 et 3.

En reprenant lanalyse 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 tre les plus reprsentatifs compte tenu de leur contribution la dfinition des axes. Ainsi, avec la variable rendement (Fig. 7), laxe 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 sagit du groupe comprenant les gnotypes TB1, KD22, KD26, et VG11 et le groupe concernant les cotypes tels que KD26, TB57, VG10, KD16, DTPY, KD13 et La Posta squia.
4,5990 2 110 RDTcc N_Rep +2 365,3 - +3 833,4 +3 833,4 - +5 301,5 108 305 333 310

133 421 437 423 209 410436 235 414 220 413 211 222 216 234 215 219 415 221 230 224 426 212 239 430 214 138109

-5,6650

141 104 401 121 315 HMP HE 134 303 341411 327 338 326 335 132 135 203 408 123 317 440 308417 419 241

407

6,0460 1

113

-4,3000

Figure 7. Dipersion des cotype dan le plan engendr par les axes 1 et 2 sur la base du rendement corrig.

3,5000 2 HP M N_Rep +113,0 - +166,5 +166,5 - +220,0

305

333 310

HP M 121 -5,0000 101 33 1

35 1 5,0000 1

119

338

126 111 340

116

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

Il faut cependant noter la quasi-simulitude des rsultats entre lanalyse de variance et lanalyse en composantes principales quant la sparation des cotypes en fonction du rgime. Cela sest traduit par labsence dune interaction gnotype x environnement, le rgime hydrique pouvant tre considr comme un environnement. La sparation des individus sur le premier axe en deux groupes pose en fait le problme de lefficience du criblage sous rgimes hydriques diffrencis et/ou de lutilisation de ces cotypes pour la tolrance la scheresse. Cela revient dire que la slection doit seffectuer au niveau intragroupe dans un premier temps dans le souci didentifier les sources de tolrance dune part et de potentialit de rendement dautre part avant de passer la phase de slection intergroupe. Au vu des rsultats, la ncessit dune comprhension profonde des sources de variabilit devient imprative en vue de leur introduction dans le programme de slection. En dautres termes, la connaissance des relations entre les cotypes dune part, la meilleure prise en compte des ractions spcifiques aux environnements gntiques et cologiques (avec ou sans scheresse) dautre part, et enfin une plus large comprhension de leur potentiel de combinaison spcifique peuvent, non seulement rvler le taux de variabilit disponible mais aussi les mthodes de slection adquates. Parmi les mthodes prconises, on peut citer celles qui semblent pouvoir permettre de surmonter les problmes dadaptation aux conditions pdoclimatiques (scheresse, faible niveau de fertilit des sols etc.) et ceux lis lamlioration du niveau moyen de performance. Toutefois, des mthodes sont courantes et ont dj t utilises (Hallauer and Sears 1972). Il sagit de la slection massale, de la slection rcurrente avec descendances et du back-cross.

117

En pratique, toute slection est ncessairement multicaractre et les amliorateurs de plantes doivent choisir la stratgie la plus efficace compte tenu de leurs cont raintes et de leurs moyens. Gallais (1990) a montr lintrt et la supriorit dune slection multicaractre par index lorsque le nombre de caractres slectionner est grand, et quils sont dgale importance, ce qui est une condition defficacit de l slection long terme. Ceci est a dautant vrai que les caractres sont lis, et ce dautant que la corrlation est faible ou ngative. Mais quelle que soit la liaison entre les deux caractres, un progrs important sur un caractre limitera le progrs s ur lautre (Gallais 1990). Donc, plus la structure des corrlations est forte, plus le progrs simultan sur plusieurs caractres sera difficile. La connaissance de l intensit de la structure des corrlations dans une population, en plus de ses performances moyennes et de sa variabilit pour les diffrents caractres, pourrait donner des indications sur le progrs potentiel que lon peut esprer obtenir en slectionnant cette population.

References
Edmeades, G.O., M. Bnziger, and D. Beck. 1997. Development and per se performance of CIMMYT maize populations as drought tolerant sources. Pages 254262 in G.O. Edmeades, M. Bnziger, H.R. Mickelson, and C.B. Pena-Valdivia (eds.) Developing drought and low-N tolerant maize. Proceedings of a Symposium, March 2529, 1996, CIMMYT, Mexico. Gallais, A. 1990. Thorie de la slection en amlioration 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: 203206 Hugues de Chrisey. 1983. Contribution lvaluation des ressources gntiques du millet Setaria italica L. P.B. Variabilit des caractres quantitatifs. Thse de Docteur Ingnieur. Univ. Paris XI Orsay. Lavergne, V. 1988. Variabilit entre populations de mas (Zea mays L.). Approche biomtrique et approche enzymatique. Thse de Docteur de lINA-PG. Ndiaye, A. 1985. Etude de lorganisation de la variabilit dans les populations de mas (Zea mays L.) Mmoire de DEA, Univ. Paris Sud Orsay.

118

Th, C., C. Zonkeng, and S.K. Kim. 1997. Identification and development of drought tolerant maize cultivars in Cameroon. In G.O. Edmeades, M. Bnziger, H.R. Mickelson, and C.B. PenaValdivia (eds.) Developing drought and low-N tolerant maize. Proceedings of a Symposium, 2529 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. Rsum La dtection et la localisation des gnes qui contribuent l'expression de caractres quantitatifs est un problme pour les slectionneurs de plantes. Diffrents modles de slection et paquets statistiques ont t dvelopps afin didentifier les meilleures combinaisons qui rsoudront le problme. Les objectifs de l'tude taient dutiliser des simulations pour comparer trois

120

mthodes statistiques pour dtecter les loci du caractre quantitatif (QTLS) dans trois diffrentes tailles de population pour quatre caractres avec pour hritabilits 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 rgression employe par PLABQTL taient les mthodes utilises. La puissance tait mesure comme tant le nombre moyen de QTLS dtect pour chaque situation. Le pouvoir de dtecter un QTL tait reli lhritabilit de QTLS, le nombre de gnotypes analyss et la mthode de dtection. Quand lhritabilit du caractre tait trs bas (< 25%), plus que 300 gnotypes taient ncessaires pour obtenir le mme niveau de pouvoir dtect par 200 gnotypes lorsque lhritabilit du caractre tait lev (100%). Le MAPMAKER tait capable de dtecter le nombre maximum de QTLS dans tous les cas, tandis que les QTLS dtects par PLABQTL et ANOVA de SAS taient comparables seulement aux tailles de population leves. La mthode ANOVA de SAS tait capable de dtecter les QTLS mais le nombre de pseudo - QTLS a augment quand le nombre de gnotypes analyss sest accru audel de 200. Le nombre optimum de gnotypes employ dans ce type d'tude devrait tre entre 200 et 300 pour toutes les mthodes. La mthode ANOVA de SAS devrait tre la premire option avant que les deux autres mthodes ne soient utilises.

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 F2 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 de pend 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 an y 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 P1 and P 2 were crossed to produce the F1 progeny. Plants from the F 1 progeny were self pollinated to produce F2 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 F1 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 F2 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 marke r 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 MM Mm mm Mean expression (1-2r)a + 2r(1-r)d [(1-r)2 + r2]d (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: Yij = + gi + e ij where Yij is an observation of the jth line of the i th QTL genotype, is the population mean, gi is the QTL genotype, e ij is the ijth residual effect, i = 1,2,.,q; j = 1,2,.., n, q is the number of QTL genotypes. 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 F1 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 de rived 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/RSSfull). 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 om itted 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 estima ted as: LOD = - log10(Lparameters/Lno 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 pseudoQTLs (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 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 even at lower than 100% heritability. For any of the number of b
10 10

number of QTL

genotypes analyzed Mapmaker was always the best in terms of the number of QTLs detected. At lower heritabilities, Mapmaker was still better than both PLABQTL and single analysis of variance using
8 8

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

0 100 8 150 200 250 300 350

0 100 150 200 250 300 350

c
6

d
sas plabqtl mapmaker

5 number of QTL

2
2

0 100 150 200 250 300 350 population size

0 100 150 200 250 300 350

Population Figure 1. Power of test of difference between number of QTL size 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 number of QTL 4 3 2 1 0 0 0.5 1 1.5

12

10 8 6 4 2 0 0

0.5 heritability

1.5

12 number of QTL 10 c 8 6 4 2 0 0 1 heritability 2 sas plabqtl mapmaker Poly. (sas)

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

200 300

0.8 0.0

1.0 0.6

0.8 0.4

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 2n:q). Knapp and Bridges (1990) have shown that when s2n:q is equal to zero, then increasing the number of replications will result in the increase in the power to detect QTLs. However, when s2n: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 fi nal 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 67:3543. 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 86:411416. 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 48:305315. 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. Genetics 134:943951. Darvasi, A. and M. Soller. 1994. Optimum spacing o genetic f markers for determining linkage between marker loci and quantitative trait loci. Theoretical and Applied Genetics 89:351357. Doerge, R.W., Z.B. Zeng, and B.S. Weir. 1994. Statistical issues in the search for genes affecting quantitative traits in populations. Pages 1526 In Proc. Analysis of Molecular marker data. Dudley, J. W. 1993. Molecular markers in plant improvement: Manipulation of genes affecting quantitative traits. Crop Science 33:660668. Edwards, M.D., C.W. Stuber, and J.G. Wendel. 1 987. Molecular marker facilitated investigations of quantitative trait loci in maize: I Numbers, genetic distribution and types of gene action. Genetics 116:113125. Haley, C.S. and S.A. Knot. 1992. A simple regression method for mapping quantitative trait loci in line crosses using flanking markers. Heredity 69:315324. 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 132:11991210. Jansen, R.C. 1994. High resolution of quantitative traits into multiple loci via interval mapping. Genetics 136:14471455.

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 8:651655. Jian, C. J. and Z.B. Zeng. 1995. Multiple trait analysis of genetic mapping for quantitative trait loci. Genetics 140:11111127. 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 126:769777. Lander, E. S., and D. Botstein. 1989. Mapping Mendelian factors underlying quantitative traits using RFLP linkage maps. Genetics 121:185199. 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 90:6572. 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 138:13011308. 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 86:907913. Moreno-Gonzalez, J. 1992a. Estimates of marker-associated QTL effects in Monte Carlo backcross generations using multi ple regression. Theoretical and Applied Genetics 90:6572. 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. Markerassisted selection in backcross breeding. Pages 4143 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 34:4755. Staub, J. E., and Sequen. 1996. Genetic markers, map construction, and their application in plant breeding. Horticultural Science 31:729741. 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 92:92203. 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 91:200205. 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 73:25252537. Weller, J.I. 1986. Maximum likelihood techniques for the mapping and analysis of quantitative trait loci with the aid of genetic markers. Biometrics 42:627640.

134

135

AMMI analysis of genotype x environment interaction of open pollinated maize varieties evaluated in the major agro-ecologies of Nigeria
S.R. Ajibade1 , B.A Ogunbodede1 , and B.A. Oyejola2
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 latematuring varieties. Location effects accounted for 76 and 79% of the total variation in the early- and late-maturing varieties, respectively. Significant variety and variety x location (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, r espectively. 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 agroenvironments 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. Rsum Le model principal effet additif et l`interaction multiplicative (AMMI) a t utilis pour analyser les donnes de deux groupes de pollinisation ouverte (op) du mais (Zea mays L.) (varits prcoces et tardives). Chaque groupe a t valu sparment pour le rendement en grains dans 18 localits reprsentatives des principales zones agro/ cologiques du Nigria. Le rendement moyen des grains a t de 2,3 t/ha pour la varit prcoce et 2,7 t/ha pour les varits tardives. Les effets des localits sont de 76 et 79% de la variation totale des varits prcoces et tardives
1

136

respectivement. Une interaction significative des varits et des varits x location (GL) a t not dans chaque groupe d`essais. Linteraction GL compte pour 7,4 et 5,6% de la variation totale pour les deux groupes dessais respectivement. Les trois premires composantes principales dinteraction (IPCAs) sont significatives dans les deux groupes et comptent pour 82% et 79% de la somme des carrs (SS) des linteraction GL pour les varits prcoces et tardives respectivement. Dix sites dans la mangrove et les environnements agro forestiers ont form un cluster dans chaque groupe de donnes montrant ainsi que ces sites offrent les conditions dadaptations de ces varits de mas. Pour minimiser les cots, quatre sites de test dans les environnements de la mangrove forts sont suffisants pour conduire des essais de rendement des varits de mas semblables celles utilises dans cette tude. Lidentification des sites pour les essais mas dans dautres cologies sera faite par des tudes 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 littl e 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 earl ier 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 y ield of the 18 locations used for evaluating the performance of the early and late maturing maize varieties in Nigeria, 1996. Code A B C D E F G H I J K L M Site Ecology Forest Forest Forest Forest SGS Mid-alt. MG NGS Mid-alt. Forest Forest SGS MG Rainfall (mm) 1365.00 1610.60 1653.00 945.30 1382.00 2339.70 1043.20 2549.20 1471.60 Altitude (m) 224.00 -* 60.00 344.40 1290.00 29.00 655.00 78.00 106.00 302.00 97.60 Grain yield, t/ha Early OPs 3.36 b 3.06 bc 1.22 fg 1.19 fg 3.29 b 1.50 f 1.41 fg 4.47 a 2.16 e 2.27 de 1.25 fg 3.06 bc 0.32 h Late OPs Ibadan Ife Orin-Ekiti Ikenne Ilorin Heipang P. Hacourt Zaria Saminaka Benin City Abeokuta Mokwa Asaba 5.15 b 2.23 h 2.63 g 2.94 fg 4.63 c 1.50 I 0.61 j 5.72 a 2.64 g 2.64 g 1.33 I 3.27 ef 0.91 j

139

N O P Q

Gembu Makurdi Samaru Yola

Mid-alt. NGS NGS SS Forest**

1323.90 1043.20 935.60 2213.60

106.00 675.00 190.00 390.00

2.96 c 2.32 de 4.23 a 2.57 d 1.12 g 2.32 0.11

2.14 h 2.63 g 3.98 d 3.41 e 0.66 j 2.72 0.13

R Nsukka Mean SE

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:

Yger = + ? g + e + Sn ?n ?gn den + ?ge + ?ger

where:

Yger

= the observed yield of the g th genotype in e th location and rth replicate; = the grand mean; = the deviation of the mean of the gth genotype from ; = the deviation of mean of the e th environment from ; = singular value for the n th interaction principal component axis (IPCA); = genotype eigenvector values for the n th PCA axis; = environment eigenvector values for the n th PCA axis; = the residual effect; = the error term (Gauch and Zobel 1996).

?g
e ?n ?gn de n ?ge ?ger

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 Name Early-maturing varieties Source population Grain type Yield (t/ha)

140

1. 2. 3. 4. 5. 6. 7. Mean

TZE Comp. 4 DMR BC 1 TZE Comp. 3 C1 DMR-ESR-Y SUWAN-2-SSR EV8730-SR Acr 92 TZE Comp. 5-W Acr 90 Pool 16-DT

E8430-SR, 1K8149-SR TZE SR-W, DMR-ESRW DMR x TZSR SUWAN 2 POP 30 TZE Comp.5 Pool 16

D* D F F F/D D D

2.52 2.29 2.29 2.41 2.30 2.39 2.04 2.32 0.07

SE Late-maturing varieties 1. 2. 3. 4. 5. 6. 7. 8. Mean SE TZL Comp. - DMR TZL Comp. 3 x 4 C0 Akure 93 DMR-LSR-W Akure 9322- DMR-SR TZB-SR TZPB-SR-C2 Ikenne 9129-SR TZ 9043 DMR-SR TZL Comp. 4 TZL Comp. 3 x 4 DMR-LSR-W Pop. 22-SR TZB TZPB Pop. 29-SR Pop. 43-SR D F/D D D F/D D D D

2.93 2.85 2.56 2.67 2.53 2.80 2.76 2.68 2.72 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 7 9% 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 8090% 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 latematuring open pollinated maize varieties evaluated in 18 locations in Nigeria, 1996. S.V DF SS % total SS MS

141

Early-maturing maize Total 503 Location 17 Genotype GxL IPCA1 IPCA2 IPCA3 Residual Error 6 102 22 20 18 42 378

822.73 628.72 9.53 60.81 23.25 15.74 10.73 11.10 123.66

76.42 1.16 7.39 38.23 25.88 17.65 18.25 15.03

36.98 1.59 0.60 1.06 0.79 0.60 0.26 0.33

** ** ** ** ** **

Late-maturing maize Total 575 Location 17 Genotype GxL IPCA1 IPCA2 IPCA3 Residual Error 7 119 23 21 19 56 432

1502.62 1182.63 9.85 84.31 30.27 20.24 15.89 17.92 225.83

78.71 0.66 5.61 35.90 24.10 18.87 21.22 15.03

69.57 1.41 0.71 1.32 0.96 0.84 0.35 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.

1.57 I P C A 1 0.0M RCK GF D

O 6 IJ+ 1Q 34 5 2 NB E L A H

-1.5-

-+---------+---------+---------+---------+---------+---------+ 0.0 0.8 1.6 2.4 3.2 4.0 4.8 Mean grain yield (t/ha)

1.2B I P C A 1 K 0.0GR M F 5O 7 8 -1.2P -+---------+---------+---------+---------+---------+---------+ 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)

2 3 1

4 D N ?+ 6 L

A H

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. ii. iii. iv. v. M alone (Asaba: Mangrove), P alone (Samaru: NGS), H alone (Zaria: NGS), RCKGD together (Nsukka, Orin Ekiti, Abeokuta, Ikenne all in the Forest Zone and Port Harcourt- Mangrove), 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) (ii) (iii) (iv) (v) (vi) P alone (Samaru: NGS), EAH together (Ilorin: SGS; Ibadan: Forest; Zaria: NGS), B alone (Ife: Forest), Q alone (Yola: Sudan savanna), GRMK (Port Harcourt: Mangrove; Nsukka: Forest; Asaba: Mangrove; Abeokuta: Forest), 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). Varie ty 4 is adapted to Ikenne, variety 6 is suited to Mokwa, while varieties 5, 7, and 8 are suited to Makurdi. Suitable latematuring 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.

1.0F I P C A 2 0.0P 2

O 5 3 QI N D AGMC J LB KR E6 H 4 -1.0-+---------+---------+---------+---------+---------+---------+ -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 IPCA1 7

1.0-

145

I P C A 2 0.0-

P 8 7 O5

E L A G CRMJ D 6 + 4K F N I H

Q 2 B

-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 (198588) 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 invol ving 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. ii. iii. iv. Forest-transition -southern Guinea savanna; Northern Guinea savanna; Sudan savanna; 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: 269277. 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 149159 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: 2737. 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: 705715. Gauch, H.G. 1990. Full and reduced models for yield trials. Theoretical and Applied Genetics 80: 153160. Gauch, H.G. 1992. Statistical analysis of regional yield trials: AMMI analysis of factorial designs. Elsevier, Amsterdam, The Netherland. Pages 53110. Gauch, H.G. and R.W. Zobel, 1996. AMMI analysis of yield trials. Pp 85122 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: 123135. MBenga, 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 45:143150. 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 114127 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, 47 May 1999, IITA-Cotonou, Benin Republic. WECAMAN/IITA. Romagosa, I. and P.N. Fox. 1993. Ge notype x environment interaction and adaptation. Pages 373390 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: 985994. 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: 2740. Westcott, B. 1987. A method of assessing the yield stability of crop genotypes. Journal of Agricultural Science, Cambridge 108: 267274. Zobel, R.W., M.J. Wright, and H.G. Gauch. 1988. Statistical analysis of a yield trial. Agronomy Journal 80: 388393.

149

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

A.E. Salami1 , A. Menkir2 , M.O Akoroda1 , and V.O Adetimirin1
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 (h2 = 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 f rther progress from u 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 . Rsum Deux composites prcoces de mas (Zea mays L.), TZE COMP 3 et TZE COMP 4, sont en amlioration par la slection recurrente rciproque (SR8) linstitut international dagronomie tropicale (IITA), Ibadan Nigria. 50 lignes alatoires S1 de chaque composite original (C0) et deux cycles amliors (C1 et C2) ont t valus IKENNE en 1998 et 1999 et Zaria en 1999. L`objectif tait de dterminer le progrs de deux cycles de slection. La rponse la slection du rendement en grain a t de 10,4% par cycle dans TZE COMP3 et 5,7% dans TZE COMP4. La slection a prolong les jours 50% de floraison, la hauteur des pis et la hauteur des plants ont t de 1,18 3,76% par cycle de mas avec une rduction des aspects des plants et des pis. La variance gntique entre les lignes S1 pour le rendement en grain et dautres caractres agronomiques a t significative (P<0,05).

150

L`hritabilit estime dans le sens large a t haute (h2 = 0,41 0,88) avec une faible variation entre les cycles pour beaucoup de caractres. Ces rsultats indiquent que le RRS a t efficace dans lamlioration du rendement en grain et dautres caractres sans rduire la variance gntique. Ainsi, il est possible de faire plus de progrs partir de la slection pour le rendement en grain et dautres caractres dans ces composites. Les coefficients de correlation gntique du rendement en grain avec les aspects de la plante et de lpis sont ngatives. Cependant, la correlation gntique du rendement grain avec les jours 50% de floraison la hauteur des pis et des plants sont positives dans les deux composites, ceci suggre que les gains partir de la slection pour le rendement en grain est li aux changements de ces caractres.

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 S1 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 lines derived from each composite were 1 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 4050 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, C1 and C 2) of the two composites were sib -mated to increase seeds for evaluation. Evaluation of the S1 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 S1 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 fac tors 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 (P <0.01) linear increase in grain yield was detected after two cycles of selection (Table 2). The realised gain due to selection for grain yield was 10.4% for TZE Comp 3 and 5.7% for TZE Comp 4. Therefore, RRS has been effective in improving grain yield of the composites. The gains obtained in this study compare favorably with the average gains of 2 to 7% per cycle obtained for RRS programs conducted in other countries (Walter et al., 1991; Keeratinijakal and Lamkey, 1993; Holthaus and Lamkey 1995; Menz and Hallauer, 1997).
Table 1. Means for grain yield and some agronomic trait of S 1 lines derived from the original and two selection-derived populations of TZE Comp 3 and TZE Comp 4 evaluated in three environments. Cycle Traits Grain yield (Mg ha-1) TZE Comp 3 TZE Comp 4 2.7 3.0 3.1 3.1 3.3 3.4 0.05 0.06 Composite 0 1 2 s.e.

154

Days to 50% silk Ear height (cm) Plant height (cm) Plant aspect (1-5) Ear aspect(1-5)

TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4

51.7 52.8 61.1 59.1 149.7 146.9 2.7 2.6 3.0 2.8

53.1 53.0 67.2 62.4 156.6 152.8 2.4 2.3 2.9 2.8

53.5 54.1 65.5 63.3 154.9 157.9 2.3 2.3 2.8 2.7

0.14 0.13 0.63 0.66 0.84 0.91 0.03 0.03 0.03 0.03

Days to 50% silk increased significantly (P < 0.05) from cycle 0 to 2 by 1.8 and 1.3 days in TZE Comp 3 and TZE Comp 4, respectively. Gain per cycle for days to 50% silk in TZE Comp 3 was 1.65% and 1.18% for TZE Comp 4. In practical terms silking was delayed by about one and half days in the second cycle relative to the original cycle of selection in the two composites. This is presently negligible, but if the trend continues in later cycles of selection, the earliness of the two composites would gradually be lost.
Table 2. Linear response and percentage gain per cycle of reciprocal recurrent selection for grain yield from TZE Comp 3 and TZE Comp. 4 as measured by S1 lines evaluated in three environments. Response per cycle of selection TZE Comp 3 Traits Grain yield(Mg ha-1) Days to 50% silk Ear height (cm) Plant height (cm) Plant aspect (1-5) Ear aspect(1-5) Mean 0.290.05** 0.860.89** 2.230.48** 2.640.64** -0.160.02** -0.120.02** % 10.38 1.65 3.58 1.75 -5.97 -4.06 TZE Comp 4 Mean 0.170.05** 0.620.19** 2.070.49* 5.530.66** -0.120.02** -0.070.02** % 5.73 1.18 3.48 3.76 -4.70 -2.34

*, ** Significant at p=0.01 and p=0.05 levels, respectively.

Increased grain yield in these composites was associated with increased ear and plant heights (Table 2). On the other hand, the scores for plant and ear aspects decreased significantly (P <0.01) with selection in the two composites. These results corroborate those obtained by Menkir and Kling (1999) for the testcrosses of lines from the two composites.

155

Genetic variances among S 1 lines for grain yield, days to 50% silk, ear height, plant height and plant aspect of the selection cycles of both composites were significantly different from zero (P < 0.05) for each selection cycle (Table 3).
Table 3. Estimates of genetic variance (s.e.) for grain yield and some agronomic traits across three environments for S 1 lines derived from the original and two improved cycles of TZE Comp 3 and TZE Comp 4.
Cycle of selection Traits Grain yield (Mg ha-1) Days to 50% silk Ear ht (cm) Plant ht (cm) Plant aspect Ear aspect Composite TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 0 0.280.09** 0.300.10** 2.680.77** 2.570.69** 53.4914.78** 63.7417.00** 84.3123.50** 90.1425.26** 0.090.02** 0.060.02** 0.070.03** 0.080.03** 1 0.250.08** 0.230.09** 4.361.13** 1.500.44** 42.4812.90** 39.8012.91** 90.4325.41** 147.7838.79** 0.070.03** 0.060.02** 0.030.01* 0.070.02** 2 0.270.09** 0.280.10** 2.670.68** 1.990.58** 39.1012.61** 38.5512.12** 120.0032.90** 91.4926.09** 0.060.05** 0.050.02** 0.050.02** 0.050.02*

*, ** Significant at P = 0.05 and P = 0.01, respectively. Genetic variance estimates of the two composites showed 3.5 to 78.3% reduction from cycle 0 to 2 for all the traits except plant height, which increased. The slight difference in the genetic variance as a result of selection for grain yield implied that there is adequate genetic variance for future improvement of the two composites. However, the drastic reduction in genetic variance of grain yield between C1 to C2 for TZE Comp 3 and C0 to C1 for TZE Comp 4 were unexpected, because a fairly large number of S 1 lines (4050) were recombined to form each cycle of selection. Menkir and Kling (1999) in an S 1 x S 1 testcross evaluation involving the two composites made similar observation. Similar reductions in genetic variance of S 1 lines after the initial two or three cycles of selection have been reported in other studies; for example, in BSSS after three cycles of RRS (Walter et al. 1991). Heritability estima tes for grain yield, days to 50% silk, ear height and plant height were significantly (P < 0.01) high (h2 > 0.6) in all

156

the selection cycles indicating that further selection would be effective (Table 4).
Table 4. Broad-sense heritability estimates ( s.e.) for grain yield and some agronomic traits across three environments for S1 lines derived from the original and two improved cycles of TZE Comp 3 and TZE Comp 4. Cycle of selection Traits Grain yield (Mg ha-1) Days to 50% silk Ear height (cm) Plant height (cm) Plant aspect (1-5) Ear aspect (1-5) Composite TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 0 0.740.20 0.610.20 0.790.20 0.840.20 0.810.20 0.840.19 0.800.19 0.800.19 0.690.85 0.550.20 0.580.21 0.680.20 1 0.690.29 0.600.21 0.870.20 0.770.19 0.750.20 0.700.19 0.800.20 0.850.19 0.540.21 0.600.20 0.410.21 0.700.14 2 0.690.20 0.640.20 0.880.20 0.780.19 0.710.20 0.720.20 0.820.20 0.810.20 0.560.21 0.560.21 0.550.21 0.500.21

Heritabilities for days to 50% silk and plant height of C2 were similar to or higher than those of C0 for TZE Comp 3 while other traits had smaller heritability estimates after two cycles of selection. Reductions in heritability estimates were also observed for days to 50% silk, ear height and ear aspect after two selection cycles in TZE Comp 4 (Table 4). The overall changes in heritability estimates with selection followed trends similar to changes in genetic variance estimates. Schniker and Lamkey (1993) and Menkir and Kling (1999) have also shown that changes in heritability parallel those of genetic variance. Grain yield had positive genetic correlation coefficients with days to 50% silk, ear height and plant height (Table 5). This implies that:
Table 5. Genetic correlation (s.e.) between grain yield and and some agronomic traits across three environments for S 1 lines derived from the original and two improved cycles of TZE Comp 3 and TZE Comp 4. Cycle Traits Days to 50% silk Composite TZE Comp 3 0 0.660.08 1 0.280.12 2 0.180.13

157

TZE Comp 4 Ear ht (cm) Plant ht (cm) Plant aspect Ear aspect TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4 TZE Comp 3 TZE Comp 4

0.190.13 0.560.09 0.550.09 0.660.07 0.650.08 -0.990.00 -0.920.02 -0.450.15 -0.920.12

0.230.12 0.640.08 0.400.11 0.770.05 0.550.08 -0.650.09 -0.730.07 -0.650.12 -0.940.01

0.490.10 0.690.08 0.700.07 0.610.09 0.720.06 -0.730.08 -0.560.11 -0.540.13 -0.750.08

increase in grain yield with selection would result in taller plants and increased ear height. Consequently, efforts must be made to select plants with lower ear and plant heights in future cycles of selection. The trend in the genetic correlation for days to 50% silk indicated reduction in magnitude with selection, although this trend was significant (P < 0.05) only at cycles 0 and 1 for TZE Comp 3. Such reduction in magnitude indicates that more of the gains in selection for grain yield should be expected to be independent of the changes in days to 50% silk. In contrast, genetic correlation coefficient increased with selection for ear height, but did not follow any definite pattern for plant height. Grai n yield had negative genetic correlation with plant and ear aspects (Table 5). Therefore, selection based on low rating for these traits would lead to increased grain yield.

Conclusions
From the results of this study, we conclude that: 1. There was increase in mean grain yield, plant height, ear height, and days to 50% silk after two cycles of reciprocal recurrent selection in TZE Comp 3 and TZE Comp 4 maize composites. Plant and ear aspects decreased with selection. 2. Genetic variances and heritability estimates for grain yield and five other traits of the S 1 lines were sufficiently large to allow good response to further selection in the two composites. 3. Genetic correlation coefficients of grain yield with days to 50% silk, ear height and plant hei ght were positive whereas those of grain yield with plant and ear aspects were negative.

Acknowledgment

158

This study was by funded by Maize Improvement Program, International Institute of Tropical agriculture (IITA), Ibadan, Nigeria.

References
Darrah, L. L., S.A. Eberhart, and L.H. Penny. 1978. Six years of maize selection in Kitale Synthetic 11, Ecuadour 573, and Kitale Composite A, using methods of the comprehensive breeding system. Euphytica 27: 191204. Eberhart, S.A., M.N. Harrison, and F. Ogada. 1967. A comprehensive breeding system. Der Zuchter 37: 169174. Eberhart, S.A., S.K. Kim, J. Mareck, L.L. Darrah, and M.M. Goodman. 1991. A comprehensive breeding system for maize improvement in Africa. Pages 175193 in N.Q. Ng, P. Perrino, and H. Zedan (eds.) Proceedings of International Conference on Crop Genetic Resources of Africa Vol. II, IITA, IBPGR, UNEP, Ibadan, Rome, Nairobi. Everett, L.A., J.T. Eta-Ndu, M. Ndioro, and I. Tabi. 1994. Application of a modified comprehensive breeding system for maize (Zea mays L.) in Cameroon. Maydica 39: 231237. Falconer, D.S., and T. Mackey. 1996. Introduction to quantitative genetics. 4ed. Longman.Scientific and Technical Co. Exssex, England 464p. Hallauer, A.R. and J.B. Miranda Fo. 1988. Quantitative genetics in maize breeding 2nd ed. Iowa State Univ. Press, Ames Iowa, USA. Holthaus, J.F. and K.R. Lamkey. 1995. Composite means and genetic variances in selected and unselected Iowa Stiff Stalk Synthetic maize composites. Crop Science 35:15811589. Keeratinijakal,V and K.R Lamkey, 1993. Responses to Reciprocal recurrent selection in BSSS and BSCB1 maize populations. Crop Science 33:7377. Manyong, V.M, J.G. Kling, K.O Makinde, S.O. Ajala, and A.Menkir. 2000. Impact of IITA-improved germplasm on maize production in West and Central Africa. IITA, Ibadan, Nigeria. 11p. Manyong, V.M., J. Smith, G.K. Weber, S.S. Jagtap, and B.Oyewole. 1996. Macro characterization of agriculture systems in West

159

Africa: An overview. Resource and Crop Management Research Division Monograph No.21. IITA, Ibadan. Menkir, A., and J.G. Kling. 1999. Effect of reciprocal recurrent selection on grain and other traits in two early-maturing maize composites. Maydica 44: 159165. Menz, M.A., and A.R. Hallauer. 1997. Reciprocal recurrent selection of two tropical corn composites adapted to Iowa. Maydica 42:239 246. MIP (Maize Improvement Program-IITA) Part 1. 1996. Maize Improvement Program Archival Report, 1988-1992. Part 1: Maize Composite Improvement. Crop Improvement Division, International Institute o f Tropical Agriculture, Ibadan, Nigeria. Mode C. J. and H.F. Robinson. 1959. Pleiotropism and genetic variance and covariance. Biometrics 15: 518537 SAS Institute, 1995. SAS/STAT user's guide. Version 6, 4th ed. Vol. 1 and 2. SAS. Inst. Inc. Cary, N.C., USA. Schnicker, B.J. and K.R. Lamkey. 1993. Intercomposite genetic variance after reciprocal recurrent selection in BSSS and BSCB1 maize composites. Crop Science 33: (1) 9095 Walter, S.P., W.A. Russell, and K.R. Lamkey. 1991. Performance and genetic vari ance among lines and test crosses of Iowa Stiff Stalk Synthetic maize. Crop Science 31(1): 7180.

160

Progress in breeding for low nitrogen stress in maize in the northern Guinea savanna
S.O. Alabi1 , J.G. Kling2 , S.G. Ado1 , E.O. Uyovbisere1 , S.O. Ajala2 , F.A. Showemimo1 , B.D. Tarfa1 , A.O. Ogungbile1 and L. Omogui1
Institute for Agricultural Research, Ahmadu Bello University, Zaria. 2 International Institute of Tropical Agriculture (IITA), Ibadan.
Abstract Low-N varietal trials involving promising va rieties (G) from NARs, IITA, and CIMMYT were evaluated at three N levels (0, 30, and 90 kgha -1) from 1998 to 2000 at the low-N screening site in Samaru, Zaria, Nigeria. In addition, 172, 212, and 300 full sib families from the low N-tolerant pool (LNTP-Y) from cycles 1, 2, and 3, respectively, were evaluated at 20 and 120 kg N ha-1. The three cycles were evaluated under 0 and 90 kg Nha-1 in one season. Results showed significant differences among varieties for grain yield and most agronomic traits measured, but G x N interactions were generally not significant for the traits. Although there was a corresponding increase in grain yield of the varieties when N application was increased from 0 to 90 kg N ha-1, the yields obtained at 30 kg N ha-1 were comparable to that obtained at 90 kg N ha-1 for some varieties. In the selection experiment, grain yield improvement per cycle was 0.15 t ha-1 or 5.4% at 30 kg N ha-1 while across N levels, it was 0.13 t ha-1. Correlated responses to selection occurred for stay-green rating by about 20.35% and number of ears per plant by 2.9% per cycle. Grain yield under lowN conditions was associated with reduced ASI, ear number/plant and stay-green rating. Cycle 3 was more efficient in nitrogen use efficiency and utilization but was low in N uptake efficiency. We conclude that selection for performance under low N improved the efficiency with which N was utilized to produce biomass and grain. Selection for improved productivity under low N stress condition could be further enhanced by simultaneously selecting for high yield performance based on nitrogen use efficiency and utilization efficiency and on secondary traits such as ears per plant, staygreen rating and reduced ASI. Rsum Des essais sur le faible taux dazote comprenant des varits prometteuses des SNRA, de lIITA et du CIMMYT ont t conduits en 1998 et 2000 au site de criblage pour le faible taux dazote de Samaru, Zaria, au Nigria. Les varits ont t values trois
1

161

niveaux dN (0, 30 et 90 kg N/ha). Par contre, un total de 180, 212 et 312 prognies surs des pools de faible taux dazote, respectivement C1, C2 et C3, ont t values 20 et 110kg N/ha. Il y avait des diffrences significatives entre les varits pour le rendement grain et pour la plupart des caractres agronomiques mesurs, mais les interactions gnotype x azote ntaient gnralement pas significatives. Il y avait un accroissement du rendement grain des varits lorsque le taux dN appliqu tait port de 0 30 et 90kg N/ha. Le pool tolrant au faible taux dazote C2 a produit un rendement grain lev 0, 30 et 90 kgN/ha et avait aussi un bon aspect des plants, densit et aspect des pis. Il tait vidence que de bon progrs de slection avaient t accomplis au cycle prcdent de ce pool (pool c1 tolrant au faible taux dazote). Dans les essais impliquant les populations, des variations significatives ont t observes parmi les entres pour la plupart des caractres associs avec lefficacit d'utilisation de lazote. Gnralement, les prognies donnaient un rendement grain lev qui aurait pu tre obtenu 20 kg N/ha. Un pool blanc bien adapt au faible taux dazote a t dvelopp et est actuellement entrain dtre utilis pour gnrer de nouvelles familles surs qui seraient valus dans le site de criblage pour le faible taux dazote. Les prognies prometteuses seront values en milieu paysan comme un composant du systme intgr de gestion des cultures qui amliore l'efficacit dutilisation d'azote.

Introduction
Maize is one of the five major crops accounting for 7 to 70% of the total cereal production in 11 countries of West and Central Africa (Manyong et al. 1996; CIMMYT 1994). Nigeria produces 43% of all maize grown in West Africa (Ajala et al. 1999). The crop is especially important in the northern Guinea savanna (NGS) where the grain yield potential is high and it is one of the two major crops in about 40% of the area under agricultural production (Smith et al. 1997). In the past two decades, maize has spread rapidly into the savannas, replacing traditional cereal crops like sorghum and millet; particularly in areas with good access to fertilizer inputs and markets. In the northern Guinea and Sudan savannas, alfisols, ultisols, and inceptisols are the dominant soil types. These are predominantly leached acid soils, characteristically low in plant nutrients, especially in nitrogen (Ahn 1970; Jones 1973; Sanchez 1976), phosphorus (Bache and Rogers 1970) and in soil organic matter (Jones 1973).

162

Low soil nitrogen is one of the most important abiotic factors limiting maize yields in the tropics (Lafitte and Bnziger 1996). Initially, resource -poor farmers relied on shifting cultivation or bush fallow for soil fertility maintenance. In recent years, it has become increasingly difficult to sustain this system because of increasing population pressure. As a result, nutrients and organic matter in the soil are depleted with a corresponding decrease in yield. Although increase in maize production can be achieved through increased levels of fertilizer application, the nonavailability of these fertilizers and high price sometimes constitute limiting factors to achieving increased production. Considering the low soil N levels of the savanna soils and the tremendous capacity of these zones in maize production, developing maize cultivars capable of fully exploiting available nutrients, or that can be productive under low soil nitrogen levels and N stress conditions would broaden the cultivation of maize and also offer economic and environment al benefits. Lafitte and Bnziger (1996) summarized approaches employed at CIMMYT to develop improved maize cultivars for low N environments, including selection for improved yield potential, selection for yield under N stress, and selection for specific m echanisms that are expected to confer tolerance. We propose two other possible approaches to reduce the impact of N deficiency on maize production. The first is direct selection of cultivars that are superior in the utilization of available N. Theoretically, the plant mechanism being altered by selection in this approach would be either enhanced uptake capacity or a more efficient use of absorbed N in grain production. The second approach is the development and improvement of low N tolerant populations with appropriate and efficient selection scheme that would bring about the desired level of improvement in a relatively short period of time. We have combined these proposed approaches into a selection scheme and conducted research on it for several years. In this paper, we report the progress made in screening and identifying maize varieties that are tolerant of low N stress and the population improvement program using a full-sib recurrent selection method.

Materials and Methods

163

Experiment 1:Varietal trials

Varietal trials, involving promising varieties from NARS, IITA and CIMMYT, were conducted in 1998 and 1999 at the low-N screening site in Zaria, Nigeria. The treatments were varieties (12 and 13 entries in 1998 and 1999, respectively) and three N leve ls (0, 30 and 90 kg N ha-1) in a randomized complete block design arranged as a split-plot with four replications. N rates were assigned to the main plots while varieties were assigned to the subplots. Each plot consisted of four rows, 5 m in length with 75 cm spacing between rows and 25 cm spacing between hills within the row. Two seeds were sown per hill and the seedlings were later thinned to one per hill two weeks after planting (WAP). All plots received a blanket application of 60 kg P2O5 and 60 kg K2O ha-1 at planting. N was split applied, half at planting and the remaining half at 45 WAP. Soil analysis was done before planting the trials each year. Data were collected on the following agronomic traits: i. ii. iii. iv. v. vi. days to 50% tasseling, anthesis (pollen-shedding), and silking; plant height and ear height (cm); stalk lodging, husk cover (high N only); plants at harvest, ear weight, ear rot (high N only), shelled grain weight (Low N) and grain N content (%); plant aspect was rated on a 1-9 scale where 1 = excellent and 9 = poor plant aspect; stay green (low N only) was rated on a 1 scale of leaf -9 yellowing below the ear recorded at about 3 weeks after silking, where 1 = leaves still green and 9 = leaves badly fired. In addition, leaf chlorophyll measurements were taken four weeks after silking. The following efficiencies were calculated from the primary data: i. N-uptake efficiency = Nt/Ns (Moll et al. 1982);

ii. N-utilization efficiency = Yg/Nt; iii. N-use efficiency = Yg/Ns; where Nt=total plant nitrogen; Ns= nitrogen supplied;

164

Yg =grain weight per plant.

Experiment 2: Population trials

A total of 172, 212, and 300 full-sib progenies from full-sib recurrent selection cycles 1, 2, and 3, respectively were screened in 1998, 1999, and 2000 under low N stress (20kg Nha-1) and non-stress (120 kg Nha-1) conditions. The low N-tolerant pool-yellow (LNTP-Y) was the source population for the recurrent selection. The full-sibs were divided into two sets (A and B) and evaluated in single, 5 m-row plots, using lattice d esigns with two replications. Nitrogen was applied, half at planting and the remaining half at about 4 5 weeks after planting. All plots also received P and K as indicated in the varietal trials above. Data were taken on the same parameters as in the varietal trials.

Experiment 3: Progress from full-sib recurrent selection

A field study was conducted during the 2000 growing season at the low N screening site in Zaria to determine the N response of cycles 1, 2, and 3 of LNTP-Y. The low N pool with yellow semi-dent grains was developed using a full-sib selection scheme from which many families have been generated. The experimental design was a randomized complete block in a split-plot arrangement with four replications. Nitrogen rates were assigned to the main plots and selection cycles were assigned to the subplots. Each plot was made up of four rows, 5 m in length, with a spacing of 75 cm between rows and 25 cm between hills, giving a population density of 53 333 plants ha-1. Two seeds were sown per hill and later thinned to one seedling per hill at two weeks after planting (WAP). N, in form of urea, was applied twice, one half at two WAP and the remaining half was top dressed 45 at two rates, 30 and 90 kg N ha-1. All plots also received a blanket applicati on of 60 kg P2O5 and 60 kg K 2O ha -1 at planting. Soil analysis was done before planting. Grain yield, days to anthesis, days to mid silk, anthesis silking interval (ASI), plant height, ear height, stalk lodging, ears per plant, biomass, kernel weight, N u se efficiency (NUE), N uptake efficiency and N-utilization efficiency were recorded. Analyses of variance were computed for all traits. Cycles of selection and N levels were considered as fixed effects.

165

Results and Discussion

Analysis of the soil at the experimental plots indicated that only about 8.8 and 10 kg N ha-1 were available in the soils in 1998 and 1999, respectively (Tables 1 and 2). Therefore, we expected that the results of the trials would reflect the true response of the genetic materials to externally supplied N fertilizer.
Table 1. Chemical properties of the experimental site in 1998. Sample point Block 1 1 2 Block 2 3 4 Block 3 5 6 Depth (cm) 015 1530 015 1530 015 1530 015 1530 015 1530 015 Total N % 0.044 0.061 0.061 0.053 0.035 0.044 0.088 0.019 0.07 0.131 0.070 Avail. P Bray (mg kg-1) 47.2 23.8 38.8 16.3 21.6 10.5 25.7 20.2 34.3 13.9 25.3 Exch. K cmol kg-1 0.159 0.092 0.142 0.091 0.124 0.084 0.087 0.087 0.148 0.078 0.097

Org. C % 0.439 0.219 0.439 0.299 0.459 0.359 0.239 0.399 0.539 0.359 0.519

The analysis of variance (data not presented) revealed significant differences among varieties for grain yield and most agronomic traits. In 1998, the effects due to varieties were significant for several agronomic traits, including days to mid-silk, days to tasseling, plant height, stalk lodging, husk cover (Table 3). Varietal differences also existed for some traits related to sink capacity of maize; that is, stay green, number of ears at harvest, chloroph yll content, and grain yield (Table 4). In 1999, significant genotypic differences were observed for days to mid-silk, plant height, stay green, husk cover and grain yield under the three N levels (Table 5).
Table 2. Physical and chemical properties of the experimental site (Block 2, Field U4, IAR Farm), 1999. Parameter pH H2O pH(CaCl2) Value 5.25 4.90 Parameter Micronutrients Cu g kg-1 Value 0.40

166

Org. Carbon (%) Total nitrogen (%) Avail. P Bray kg kg -1 Exch. Ca cmol kg-1 Mg K Na Exch. Acidity ECEC

0.37 0.005 20.37 1.88 0.25 0.11 0.02 0.45 2.71

Zn Mn Fe Particle size Sand % Silt % Clay % Textural class

0.52 9.40 22.0 48 40 12 Sandy clay loam

Table 3. Means for agronomic traits of open -pollinated varieties and a hybrid check evaluated under three N levels at Samaru, 1998.

Variety IBP1WD Pool 27C10 Sint mazorca larga Acr 8328BNC7 La posta sequia C4 LOW N pool C1 Obatampa Cidaba Dababa Mamaba Oba super 2 (hybrid) TZB-SR Mean SE C.V.% N x G
aRating

Silk 68.3 68.2 69.0 67.7 67.4 67.5 67.1 67.0 67.1 67.3 69.3 69.4 67.7 0.34 1.6 NS

Anthesi s 65.2 65.3 65.8 64.8 64.8 64.7 63.3 63.6 64.4 63.7 63.8 66.5 64.6 0.46 2.5 NS

PHT (cm) 244.2 227.0 232.3 217.9 224.7 234.2 209.7 212.8 217.9 206.9 218.5 228.2 222.9 5.86 9.9 NS

Stalk lodg, % 3.0 3.7 1.8 2.8 1.8 2.8 4.2 2.0 2.5 2.0 3.0 3.2 2.7 0.56 72.9 NS

Husk (1-5) a 1.4 1.4 1.6 1.3 1.4 1.0 1.3 1.8 1.3 1.1 1.3 1.3 1.3 0.12 31.5 NS

from 1 to 5, where 1 = very tight husks extending well beyond the ear tip, and 5 =

ear tips exposed

These results are in agreement with those obtained by Sallah et al. (1996) who detected significant genotypic effects for grain yield, mid-silk, plant height and grain moisture content at both low and high N. G x N interaction mean squares were generally not significant except for number of plants at harvest in 1998 and plant height in 1999. Therefore, our discussion will emphasize the main effects, especially N and the variety means across N rates.

167

Relative to the 0 kg N ha -1 treatment, plants that received 90 kg N ha-1 were more vigorous and taller but did not lodge. Plants at 30 kg N ha-1 were similar in height to those at 90 kg N ha -1.
Table 4. Means for traits related to sink capacity of open-pollinated varieties and a hybrid check evaluated under three N levels at Samaru, 1998.
Stay Variety IBP1WD Pool 27C10 Sint mazorca larga Acr 8328BNC7 La posta sequia C4 LOW N pool C1 Obatampa Cidaba Dababa Mamaba Oba super 2 (hybrid) TZB-SR Mean SE C.V.% NxG
aRating

Ear No 26.4 27.2 28.7 29.3 25.3 28.0 29.9 24.6 31.8 24.3 28.9 26.2 26.9 1.51 19.3 NS

green (1-9)a 5.9 6.3 5.9 5.4 5.5 5.8 6.0 4.4 5.4 4.8 5.5 5.5 5.5 0.35 29.5 NS

Chlorophyll content 35.3 39.1 38.4 40.4 41.1 38.6 35.1 41.8 36.0 37.6 42.6 37.9 38.7 1.45 12.9 NS

Grain yield (kg/ha) 2055 2150 2589 2571 2216 2555 1972 2599 2812 2624 2671 1895 2392 187 27.0 NS

(low N only) from 1 to 9 taken at about 2 weeks after silking; 1 = excellent stay

green and 9 - leaves badly fired.

Table 5. Means for grain yield and agronomic traits of openpollinated varieties and a hybrid check evaluated under three N levels at Samaru, 1999.
Stay green (19)b 6.1 6.7 7.1 6.7 6.6 6.3 5.6 5.6 6.3 6.4 5.9 6.3 6.2 6.3 Grain yield (kg ha-1) 2538.8 1755.5 2111.1 2766.6 2900.1 2650.1 3266.7 3572.4 2561.2 2933.2 3158.9 3671.8 3544.4 2877.7

Variety Silk LNTP-W IBP1WD Pool 27C10 Sint Mazorca Larga Acr 8328BNC7 La posta sequia C4 LOW N pool C1 LOW N pool C2 Obatampa TZPB Pool C2 Tempx trop. pool Oba super 2 TZB-SR Mean 70.3 70.9 69.6 69.6 69.9 68.7 67.8 68.3 69.6 67.3 68.5 67.3 67.8 68.9

PHT (cm) 192.1 186.7 190.1 188.7 202.4 202.9 207.9 2.7.8 203.3 225.6 225.4 223.5 229.9 206.3

Husk (15)a 1.3 1.2 1.3 1.6 1.1 1.2 1.5 1.6 1.2 1.3 1.8 1.6 1.2 1.4

Ear No. 29.2 27.5 28.4 29.8 30.4 26.5 29.0 32.4 23.1 25.6 25.3 29.1 27.4 1.63

168

SE C.V.% NxG
aRating

0.69 3.5 NS

17.56 7.9 *

0.14 35.9 NS

0.22 12.2 NS

18.4 27.9 NS

179.92 21.6 NS

from 1 to 5, where 1 = very tight husks extending well beyond the ear tip, and 5 = ear tips exposed, bRating (low N only) from 1 to 9 taken at about 2 weeks after silking; 1 = excellent stay green and 9 = leaves badly fired.

Plant aspect rating showed that at 30 and 90 kg N ha-1 the plants had better appeararance than plants that received 0 kg N. By the 810 WAP, the plants from plots that did not receive N were severely chlorotic and stunted. Generally, for all the varieties evaluated, increasing the amount of N from 0 to 90 kg N ha -1 resulted in appreciable increase in grain yield (Fig. 1). Mean yield was 1.6, 2.5, and 3.6 t ha1 for the three N rates, respectively, in 1998. Corresponding values in 1999 were 1.3, 2.6, and 4.7 t ha1, respectively.
6.0 1998 5.0 4.0 Yield, t/ha 3.0 2.0 1.0 0.0 0 kg N 30 kg N N rate/ha
Figure 1. Mean grain yield (with s.e. bars) of maize open -pollinated varieties evaluated under three N rates in Samaru, Nigeria in 1998 and 1999.

1999

90 kg N

Averaged across N levels in 1998, grain yield ranged from 2 t ha -1 for Obatanpa to 2.8 t ha-1 for Dadaba (Table 3), two of the four quality protein maize (QPM) variety in the trial. The other two QPM varieties in the trial (Cidaba and Mamaba) and several other varieties, including Acr 8328 BNC7, LN Pool C1, Sint Mazorca Larga, and Oba Super 2, a popular commercial hybrid in Nigeria, produced grain yield in excess of 2 t ha -1. TZB-SR that has for many years

169

produced outstanding yield in the savanna, had the lowest yield in this study. This is encouraging because it indicates that the effort of maize breeders to develop better varieties in recent years is producing positive results. In 1999, ranking of the varieties for mean yield across N rates changed drastically from that of 1998 (Table 5) thus indicating a strong genotype x year interaction. TZB-SR, Oba Super 2, Low N Pool C1, and Low N Pool C2 produced grain yield in excess of 3 t ha-1. Grain yield of Obatanpa remained about the same as in 1998, but the ranking among other varieties changed in 1999. 1BPILWD and Pool 27 C10 produced the lowest grain yields in 1999. Low N Pool C2 had one of the highest grain yields and out yielded the Low N Pool C1 by a margin of 10% across N rates. It also had good plant type, standability and ear aspect. These data provide evidence of good progress from selection for improved N use efficiency in this Pool.
Table 6. Performance of the best families from 172 full-sib families from the low N tolerant pool C1 evaluated under low N (120kg Nha-1) and high N (120kg Nha-1) in Samaru, 1998.
Stay Days to silk Entry SET A FS-3 FS-6 FS-25 FS29 FS-52 FS-72 FS-54 Checks TZB-SR Oba Sup 2 Mean (86 ent.) Mean (selected) SED C.V. % SET B FS-98 FS-106 FS-113 FS-122 High N 68.1 63.2 64.5 65.4 69.4 68.2 68.0 65.5 65.0 67.6 66.7 2.8 5.5 66.5 65.0 65.5 67.0 green Low N (1-9)a 3.9 4.5 3.8 5.0 5.6 5.3 5.5 6.0 5.8 5.6 4.8 1.3 19.2 4.5 5.5 5.3 4.5 No of ears harv. Low N 17.1 15.5 17.5 18.0 18.5 18.0 17.5 13.5 16.0 15.8 17.4 2.78 17.6 15.5 17.5 18.0 15.0 Ear aspect Low N (1-9)b 4.0 5.5 4.5 4.2 5.1 5.4 5.0 6.2 5.3 5.9 4.81 2.83 13.9 5.0 5.0 4.9 5.1 Grain yield Low N kgha-1 2634 2590 2928 3108 2526 2234 2696 1557 2379 2128 2674 633 29.7 2639 3020 2762 2899 Grain yield High N kgha-1 4042 3941 2979 3584 3402 4067 2920 1741 3107 2887 3562 805 27.4 3380 5052 4602 3720

170

FS-141 FS-142 FS-143 FS-148 FS-155 FS-165 Checks TZB-SR Oba sup 2 Mean (86 ent.) Mean (sel ent.) SED C.V. %

67.5 67.5 65.3 68.5 67.5 68.2 69.3 68.7 67.6 66.9 1.26 17.9

5.0 6.0 5.8 6.2 6.0 6.0 6.2 5.6 5.5 5.5 1.1 16.7

16.0 20.5 17.5 17.0 18.0 18.5 12.3 17.3 15.6 17.4 2.9 14.8

5.5 4.8 5.4 5.2 6.0 5.5 5.7 5.5 5.7 5.2 0.64 10.5

2864 2642 2150 2579 2432 2779 1096 2029 2034 2677 611.1 30.0

3030 3964 4317 4115 4157 3810 3109 3498 3411 4015 88.9 26.7

In screening genotypes for tolerance to N stress, apart from evaluating performance based on grain yield, other secondary traits such as anthesis-silking interval (ASI) and stay green rating have been shown to be useful indicators of low N tolerance when included with grain yield in a varietal improvement program (Lafitte and Bnziger 1997). The C1 and C2 of the Low N Pool, TZPB Pool C2, and Mazorca Larga showed shorter ASI of about 2 days across N rates (Table 5). The performance of selected full-sib families derived from three cycles of selection, C1, C2, and C3 are presented in Tables 6, 7, and 8, respectively. Generally, the full-sib families produced appreciable grain yield at the low as well as the high N levels except in 2000 when the grain yield under high N was lower than expected. The portion of the field containing the high N treatment was water logged and poorly drained making it difficult to obtain the expected N response. In both sets A and B, the genotypes produced higher yield than would have been expected at 20 kg N ha-1.
Table 7. Performance of the best families from 212 full-sib families from the low N tolerant pool C2 evaluated under low N (20 kg N ha-1) and high N (120 kg N ha-1) in Samaru, 1999.

Entry SET A FS-23 FS-29 FS-88 FS-89

Days to silk High N 62.5 65.3 68.5 67.2

Stay green Low N (1-9)a 4.5 5.0 5.0 5.0

No of ears harv. Low N 15.5 19.0 18.5 18.0

Ear aspect Low N (1-9)b 5.1 4.5 4.9 4.3

Grain yield Low N kgha-1 2670 3600 3335 2935

Grain yield High N kgha-1 5195 4685 5535 5265

171

FS-94 FS-102 FS-99 FS-104 FS105 FS-106 Checks TZB-SR Oba sup 2 Mean (212 ent.) Mean (Sel ent.) SED C.V. % SET B FS-113 FS-114 FS-162 FS-165 FS-167 FS-172 FS-178 FS-183 FS-184 FS-194 FS-195 FS-210 Checks TZB-SR Oba sup 2 Mean (212 ent.) Mean (of sel ent.) SED C.V. %

68.2 67.0 59.1 68.2 67.2 65.6 69.5 67.5 66.7 65.8 1.41 2.9 66.1 67.3 64.2 67.2 57.4 65.2 67.0 66.3 64.5 67.1 66.4 59.5 68.5 67.0 65.3 64.2 0.40 7.6

4.4 4.5 5.6 4.7 4.8 4.9 6.4 5.8 5.4 4.8 0.32 5.8 5.1 4.5 5.3 4.6 5.8 6.1 6.2 4.8 4.7 5.0 5.6 5.2 6.8 5.6 5.8 5.4 1.4 18.8

18.5 19.5 20.5 15.0 17.0 20.5 16.5 19.0 16.6 18.2 3.32 26.4 18.5 18.5 19.0 18.0 19.5 19.5 19.8 20.5 19.6 19.0 19.5 18.5 15.0 18.5 17.1 18.8 3.1 15.8

5.4 5.4 4.6 5.8 5.6 5.2 5.8 5.4 5.3 5.1 0.58 9.2 5.2 4.3 4.5 4.2 5.1 5.0 5.4 5.2 5.0 4.8 4.6 5.0 5.6 5.2 5.2 4.9 0.82 12.4

3865 3965 4265 3065 3665 4515 2935 3868 3055 3588 382.3 26.9 4135 4130 4265 4130 4000 4665 4535 5280 4265 4200 4465 4470 3965 3133 3451 4250 650.2 26.6

5270 4600 4800 3600 4670 5680 3133 3998 4184 4930 603.1 20.4 4600 4265 4730 4865 5200 5735 4865 5735 4285 4535 5535 4865 3900 4270 4265 4813 645.3 21.4

The genotypes that produced appreciable yields at this low N level would be considered to be tolerant of low N stress or to be N-use efficient. Possibly these genotypes developed deeper root systems in response to N stress and were able to maintain N uptake and photosynthesis longer.
Table 8. Performance of the best families from 300 full-sib families from the low N tolerant Pool C3 evaluated under low N (20 kg N ha-1) and high N (120 kg N ha-1) in Samaru, 2000.

Stay Days to Entry SET A FS-35 silk High N 68.2 green Low N (19)a 5.5

No of ears harv. Low N 15.8

Ear aspect Low N (19)b 4.0

Grain yield Low N kgha-1 2873

Grain yield High N kgha-1 3399

172

FS-129 FS-98 FS-77 FS-126 FS-123 FS-149 FS-55 Checks TZUTSRSG-1 Acr 8328BNC7 TZB-SRSGY Oba super 2 Mean (150 ent) Mean (sel ent.) SED C.V. % SET B FS-296 FS-236 FS-228 FS-175 FS-301 FS-244 FS-281 FS-259 FS-309 Checks TZUTSRSG-1 Acr 8328BNC7 TZB-SRSGY O ba super 2 Mean (150 ent) Mean (sel ent.) SED C.V. %

64.5 66.4 64.0 64.2 63.5 62.5 65.4 67.4 66.5 68.4 65.6 66.2 64.8 1.78 6.31 64.5 65.4 67.2 66.4 68.3 64.5 66.7 66.5 63.5 66.0 66.5 68.2 65.4 66.5 65.8 3.3 5.5

5.3 5.6 5.0 5.2 5.4 4.5 4.8 5.3 5.8 5.4 5.6 5.6 5.2 2.6 16.3 5.5 5.7 5.2 5.8 5.4 5.1 5.0 4.9 4.6 5.0 5.3 5.2 4.8 5.7 5.2 2.4 18.5

18.5 22.0 24.0 16.4 17.2 20.2 15.5 7.6 16.2 12.4 15.5 15.1 18.7 1.64 21.7 19.2 21.0 20.0 14.5 19.2 18.3 17.3 20.0 21.0 13.2 17.4 14.5 15.5 16.5 18.9 1.81 27.3

4.2 4.4 3.8 4.5 3.6 4.3 4.5 6.5 5.5 3.5 4.5 4.9 4.2 1.09 23.7 4.2 3.7 3.9 3.5 4.5 4.3 5.2 4.5 5.0 5.6 5.2 5.9 4.9 5.3 4.3 1.02 34.6

2751 3723 4370 2833 2914 2266 2589 725 1052 800 2347 3322 3040 709.6 33.4 4320 4922 3600 4000 4000 3500 3325 3360 3637 1840 2372 1780 2767 3912 3852 624.4 34.6

1457 1012 1780 1254 2994 3561 2833 567 1780 2185 1861 1522 2286 602.3 29.5 1818 2941 3937 3479 2293 2312 2132 3808 4370 1456 2670 1635 2913 2254 3077 529.1 38.3

Lafitte and Edmeades (1994c) have observed that selection for N stress tolerance increased root growth. Varietal differences in N utilization efficiency are more closely associated with yield performance under low than high soil N levels (Lafitte and Edmeades 1994b; Moll et al. 1982). Generally, the full-sibs that had good stay-green rating and short anthesis silking intervals were among those that produced the highest grain yield. For most traits, there were non-significant increases in response to selection under low N (data not shown). The non-significant improvement per cycle for grain yield was 0.15 t ha-1 (5.4%) at 30 kg

173

Nha-1 while across N levels it was 0.13t ha-1 (4.1%) per cycle. In a similar study, S allah et al .(1996) reported that between C o and C 6 of selection in La Posta maize population, yield increased by 0.26, 0.50, and 0.63 t ha-1 at 0, 80, and 160 kg N ha-1, respectively. In a tropical maize population, Acr 8328 BN, gains in grain yield over five cycles of full-sib recurrent selection averaged 84 kg ha -1 (45% per year) at low N and 120kg ha-1 per year (2.3%) at high N (Lafiette and Bnziger 1997). We expect significant increases in grain yield with continued selection in the Low N Pool. A few agronomic traits also showed non-significant desirable changes in response to selection for grain yield. The stay green rating increased by 20% and ear number per plant by 2.9% per cycle. ASI was reduced under low N conditions. Short ASI and staygreen rating under drought have been related to increased partitioning of biomass to the growing ear (Edmeades et al. 1999; Lafitte and Edmeades 1995). Thus, short ASI along with a longer stay green would lead to increased grain yield. For all cycles of selection, N-use efficiency (Y g/N s), N-uptake efficiency (Nt/N s), and N -utilization efficiency (Y g/Nt) decreased with increasing N levels (Table 9). N-use efficiency and N-utilization efficiency increased with selection cycles but N -uptake efficiency showed a decreasing trend.
Table 9. Observed genetic gain on N -use, N-uptake, and N utilization efficiency after three cycles of selection in the low N tolerant pool (LNTP-Y) evaluated in Samaru, 2000. Mean gain per cycle 5.2 1.4 3.3 -0.04 -0.03 -0.04 4.2 3.3 3.8 Relative change (%) 5.8 3.6 5.2 -2.6 -4.0 -3.0 6.9 6.3 6.6

C1 NUE (g grain/g Ns) Low N 87.9 High N 38.9

C2 93.2 39.2

C3 98.3 41.7 70.0 1.5 0.7

Across N 63.4 66.2 N-uptake efficiency (g Nt/g Ns) Low N 1.6 1.6 High N 0.8 0.7

Across N 1.2 1.2 1.1 N-utilization efficiency (g grain/g Nt) Low N High N Across N 61.3 53.2 57.2 59.9 60.1 60.1 69.7 59.8 64.8

174

Conclusion
It was concluded from these studies that three cycles of full-sib recurrent selection has been effective in improving the Low N Pool mai ze population under low N stress. Selection for improved productivity under low N stress condition could be achieved by simultaneously selecting for high yield performance based on N -use efficiency, N-utilization efficiency and secondary traits, especially ASI, ear number at harvest, and the stay green character.

References
Ahn, P. 1970. West African Soils. Oxford University Press, Ely House, London. Ajala, S.O., A. Menkir, and J.G. Kling. 2000. Fertility requirements of open-pollinated and hybrid maize genotypes. Pages 8795 in J.A. Valencia, A.M. Falaki, S. Miko, and S.G. Ado (eds.) Sustainable maize production in Nigeria: The challenges in the coming millennium. Proceedings of a National Maize Workshop, Ahmadu Bello University, Zaria, 2224 July, 2000. SG2000-IAR-FMARD-ADPs. Bache, B.W. and N.E. Roger. 1970. Soil phosphate values in relation to phosphate supply to plants from some Nigerian soils. J. Agric. Sci. Camb. 74: 383390. Bnziger, M., H.R. Lafitte, G.O. Edmeades, F.J. Bertran, D.L. Beck, and A. Elings. 1999. Recent advances in breeding for tolerance to low nitrogen in tropical maize. Pages 21 33 in B. Badu -Apraku, M.A.B. Fakorede, M. Ouedraogo, and F.M. Quin (eds.) Strategy for sustainable maize production in West and Central Africa. Proceedings of a Regional Maize Workshop, 2125 April, 1997, IITA-Cotonou, Benin Republic. WECAMAN/IITA. CIMMYT. 1994. CIMMYT 1993/94 World maize facts and trends. Maize seed industries revisited: Emerging roles of the public and private sectors. Mexico D.F., CIMMYT, Mexico Edmeades, G.O., J. Bolanos, S.C. Chapman, H.R. Lafitte, and M. Bnziger. 1999. Selection improves drought tolerance in tropical maize populations 1. Gains in biomass, grain yield and harvest index. Crop Sci. 39: 13051315.

175

Jones, M.J. 1973. The organic matter content of the savanna soils of West Africa. J. Soil Sci. 24:4253. Lafitte, H.R. and M. Bnziger. 1997. Maize population improvement for low N: Selection gains and the identification of secondary traits. Pages 485489 in G.O. Edmeades, M. Bnziger, H.R. Mickelson, C.B. Pena-Valdivia (eds.) Developing drought and low-N tolerant maize. Proceedings of a Symposium, 2529 March, 1996. CIMMYT, El Batan, Mexico D.F., CIMMYT. Lafitte, H.R. and G.O. Edmeades. 1994a. Improvement for tolerance to low soil nitrogen in tropical maize 1. Selection criteria. Field Crops Res. 39:114. Lafitte, H.R. and G.O. Edmeades. 1994b. Improvement for tolerance to low soil nitrogen in tropical maize II. Grain yield, biomass production, and N accumulation. Field Crops Res. 39:1525. Lafitte, H.R. and G.O. Edmeades. 1994c. Improvement for tolerance to low soil nitrogen in tropical maize III. Variation in yield across environments. Field Crops Res. 39: 2738. Lafitte, H.R. and G.O. Edmeades. 1995. Stress tolerance in tropical maize is linked to constitutive changes in ear growth characteristics. Crop Sci. 35: 820826. Manyong, V.M., J. Smith, G.K. Weber, S.S. Jagtap, and B. Oyewole. 1996. Macrocharacterization of agricultural systems in West Africa: An overview. Resource and Crop Management Research Monograph No. 21. IITA, Ibadan. Moll, R.H., E.J. Kamprath, and W.A. Jackson. 1982. Analysis and interpretation of factors which contribute to efficiency of nitrogen utilization. Agron. J. 74: 562564. Sallah, P.Y.K., N.J. Ehike and J.L. Geadelmann. 1996. Selection for response to low nitrogen in the La Posta maize population Pages 502507 in G.O. Edmeades, M. Bnziger, H.R. Mickelson, C.B. Pena-Valdivia (eds.) Developing drought and low-N tolerant maize. Proceedings of a Symposium, 2529 March, 1996. CIMMYT, El Batan, Mexico D.F., CIMMYT. Sanchez, P.A. 1976. Properties and management of soils in the tropics. Wiley-Interscience Publicaton, John Wiley and Sons, New York, USA.

176

Progrs dans la slection pour la tolrance du mas au stress hydrique au Burkina

J. Sanou1 et F. Dabire
1

INERA, Station de Farako B, 01 BP 910 Bobo-Dioulasso 01 Burkina Faso

Rsum Le rendement du mas en zone semi-aride est faible en raison de frquentes priodes de scheresse et des maladies intervenant au cours de la culture. Les objectifs de notre tude ont t dune part la slection inter cultivars et dautre part la slection intra varitale pour la rsistance la scheresse. Les travaux ont t conduits la station de Farako-B en saison humide pour la slection contre les maladies et la valle du Kou, en saison sche o le criblage contre la scheresse a t ralis. Le dispositif de criblage contre la scheresse inclut une irrigation lETM et une privation deau de 30 jours autour de la floraison du mas. Le criblage contre la scheresse a concern une slection inter cultivars et une slection intra varitale. La slection inter cultivars a concern 21 cultivars locaux repartis en matriel extra prcoces et prcoces. Trois cultivars extra-prcoces (silmiri 1, Bondoukuy 1 et Ki 1) et cinq cultivars prcoces (Kouentou 3, Kouentou 7, Tenasso 1, Dand 2 et Nbia 1) ont t slectionnes pour leur bon comportement sous stress hydrique. Les cultivars locaux sont une source de rsistance la scheresse. La slection intra varitale a concern 604 lignes S1 de quatre varits amliores (FBC6, Pool 16 DT, DTE W SR BC3 C0 et DTE Y SR BC3 C0) permettant de slectionner 10, 6, 8 et 6 lignes comme matriel rsistant au stress hydrique respectivement dans les varits FBC6, Pool 16 DT, DTE W SR BC3 C0 et DTE Y SR BC3 C0. Ces lignes ont t brasses pour former respectivement 4 varits rsistantes la scheresse (FBDR1, FBDR2, FBDR3 et FBDR4). Abstract Maize grain yield in the semi-arid zone is low due to recurrent drought and high disease incidence during the growing season. The objectives of this study were to select (i) inter cultivars and (ii) intra varieties for drought tolerance. The study was carried out during the rainy season at the Farako-B research station for disease resistance, and during the dry season in the Kou valley for drought tolerance. The design used for drought tolerance screening involved one watering at MET and zero watering for a thirty-day

177

period around flowering. One inter-cultivar and one intra-variety selection were carried out. As for inter-cultivar selection, 21 local cultivars divided into extra-early and early maturing materials were tested. Three extra-early cultivars (silmiri 1, Bondoukuy 1 and Ki 1) and five early maturing cultivars (Kouentou 3, Kouentou 7, Tenasso 1, Dand 2 and Nbia 1) were selected for their good performance under moisture stress conditions. Local cultivars are a source of drought tolerance. As for intra-variety selection, 604 S1 lines of four improved varieties (FBC6, Pool 16 DT, DTE W SR BC3 C0 and DTE Y SR BC3 C0) were used to select 10, 6, 8 and 6 lines as moisture stress tolerant material respectively among the following varieties : FBC6, Pool 16, DTE, W SR BC3 C0 and DTE Y SR BC3 C0. These lines were recombined to form 4 drought tolerant varieties (FBDR1, FBDR2, FBDR3 and FBDR4).

Introduction
Le dficit hydrique touche prs de 50% des surfaces cultives dans le monde. Dans les zones semi-arides d'Afrique une tude du CIMMYT (1990) cite par Sanou (1991), indique que 40% des surfaces de mas subit une scheresse occasionnelle responsable d'une baisse de la production de 10 25%. De mme, 25% des surfaces y est soumis des stress frquents causant des pertes de rcoltes de 25 50%. Les priodes de scheresse imprvisibles d'au moins trois semaines y sont frquentes (Tesha 1991).

La scheresse dfinit l'tat de pnurie hydrique dont souffre un vgtal (Sarr 1975). Elle englobe plus un aspect climatique que physiologique (Wery 1987). Selon Ludlow et Muchow (1990), on distingue deux types de scheresse: une scheresse intermittente pouvant intervenir tout moment du dveloppement de la culture et une scheresse terminale ou de fin de cycle typique aux saisons sches des zones semi-arides. Le dficit hydrique ou stress hydrique est la perte importante d'eau par transpiration chez la plante par rapport au prlvement effectu au niveau du sol. A l'intrieur d'une scheresse intervient plusieurs stress hydriques d'intensits diffrentes, mais peut-on caractriser la scheresse ? Les agroclimatologues (Some 1989; Sivarkumar et Gnoumou 1987) partent des donnes pluviomtriques pour dterminer les annes sches sur la base d'un indice de dficit pluviomtrique quantifi par lcart de la moyenne pluviomtrique. Une saison est dficitaire en pluies si son cumul est infrieur au seuil de dficit

178

pluviomtrique de la saison. Ainsi dans les zones arides o il pleut moins de 300 mm/an, toute campagne qui enregistre une pluviomtrie infrieure la moyenne annuelle est systmatiquement une anne de scheresse alors que dans les zones humides, c'est surtout la rpartition saisonnire des pluies qui est dterminante. Lapproche agronomique pour caractriser les dficits hydriques repose sur la quantification des effets finaux sur le vgtal : elle consiste mesurer des paramtres de croissance et de productivit dun groupe de gnotypes en prenant comme situation de rfrence une situation favorable non limitative. La rsistance au dficit hydrique peut tre dfini comme laptitude d'une plante pousser correctement dans des conditions de dficit hydrique (Wery 1987). Les mcanismes de rsistance des plantes la scheresse sont divers. Laptitude rsister la contrainte hydrique peut-tre associe un systme racinaire abondant, une fermeture rapide des stomates, une grande efficience dans lutilisation de leau ou au maintien dun potentiel de turgescence leve ou enfin une modification des composs biochimiques de la plante. Les physiologistes (Blum 1988; Monneveux 1992 ; El Jaafari et al. 1993) regroupent les mcanismes de rsistance des plantes la scheresse dans lvitement (dveloppement de la plante en dehors des priodes de dficit hydrique), dans lesquive (la plante conserve un potentiel hydrique faiblement ngatif en condition de stress, vitant ainsi la dshydratation des tissus : systme racinaire dvelopp, enroulement foliaire), dans la tolrance (maintien de turgescence de la plante alors que son potentiel hydrique est trs ngatif grce au processus de lajustement osmotique). Globalement chez le mas, ds quun bourgeon axillaire se dveloppe pour donner un pi, toute altration de lappareil vgtatif de la plante se traduit par une diminution du nombre de rang dovules et ensuite par un raccourcissement de la longueur des rangs. Ds que la fcondation est effectue, si la plante souffre, elle rgule son rendement en faisant avorter les ovules les plus fragiles, donc les plus jeunes qui sont la pointe de lpi. Trois semaines aprs la fcondation, le nombre de grains ports par la plante est dfinitivement acquis. Au-del de ce stade, un stress hydrique nagit sur le rendement qu'en diminuant le poids des grains (Robelin 1983).

179

Au niveau des feuilles, un svre d ficit hydrique induit une baisse de la conductance stomatique et de la transpiration ainsi qu'une rduction ou une inhibition de la photosynthse (Tesha 1991; Hema et al. 1998). La fermeture des stomates est plus une raction de sensibilit la scheresse qu'un processus de dshydratation. La fermeture des stomates lve la temprature de 5c 6 c et acclre par ce fait la transpiration cuticulaire induisant alors une perte de l'eau rsiduelle de la plante. En raction cet accroissement de la temprature, les gnotypes sensibles au stress hydrique dveloppent des ncroses. Par contre l'enroulement foliaire (fltrissement) associ la capacit de reprise, est un bon indicateur de rsistance la scheresse; ceci peut tre utilis comme critre de criblage de gnotypes rsistants au stress hydrique.

La tige de mas est la source principale dassimilats pour le remplissage des grains en cas de longue scheresse (Gentinetta et al. 1986; Ouattar et al. 1987a). Elle est une alternative pour assurer le dveloppement du grain en labsence de photosynthse ; elle constitue une zone tampon entre le grain et le sol par le maintien de conditions hydriques favorables pour remdier temporairement au dficit hydrique dans les feuilles grce son stock de sucres. La section de la tige ainsi que sa taille sont des donnes trs importantes prendre en compte pour la slection de gnotypes rsistants au dficit hydrique. Au niveau de l'appareil reproducteur et du grain, le stress hydrique est responsable d'une augmentation de l'cart entre la floraison mle et femelle (Herrero et Johnson 1981), ce qui influence le pourcentage de fcondation et le nombre de grains forms. Monneveux (1992) propose la slection de gnotypes courte priode de remplissage du grain. Selon Ouattar et al. (1987a), le nombre de grain est un meilleur critre d'apprciation de l'adaptation au dficit hydrique que le poids des grains qui peut tre par ailleurs favoris par les rserves de la tige. La tolrance au stress hydrique s'exprimera chez un gnotype par un nombre lev de grain par rapport aux gnotypes sensibles.

Au niveau des racines, la rsistance la scheresse s'explique par une plus forte extraction de l'eau du sol par les racines. Cela n'est possible quen prsence d'une grande vi tesse de pntration en

180

profondeur et d'expansion des racines. Le nombre de racines est galement une caractristique prendre en considration dans la slection de gnotypes rsistants au stress hydrique. Fisher et al. (1981) ont mis en vidence quune slection pour une masse racinaire importante aboutissait une augmentation du rendement grain en condition de stress hydrique, alors que la slection pour l'augmentation de la longueur des racines tait plus intressante dans des cas de stress svres. Le s travaux de Fisher et al. (1983) sur Tuxpeno 1 pour obtenir une temprature plus faible du couvert, une plus grande longation foliaire, un meilleur synchronisme des floraisons, un retard la snescence aprs floraison, ont abouti un matriel gntique dot dun systme racinaire important pouvant extraire plus d'eau. Comment et quand slectionner? Blum (1988) estime que la slection pour la tolrance la scheresse base sur le rendement en conditions de stress hydrique nest pas efficace cause de la complexit de lhritabilit du rendement. Johnson et Geadelman (1989), ont montr pour le rendement que la slection en conditions de stress hydrique naccroissait pas plus la tolrance la scheresse chez le mas quun criblage en conditions irrigues. Ils ont not que dans certains cas, des caractristiques indsirables (verse) sont associes aux gnotypes obtenus en conditions de stress. Celles-ci n'apparaissent quen conditions hydriques favorables. Ces observations confirment l'intrt pour lequel Monneveux (1992) suggre une slection pour la tolrance la scheresse dans des conditions hydriques favorables. Les meilleurs gnotypes pour la productivit en condition irrigue sont slectionns et tests en situation de stress hydrique. Un couplage de parcelles irrigues et non irrigues aboutirait au mme rsultat. La priode de sensibilit au stress hydrique chez le mas s'tend du stade 1012 feuilles au dbut du stade grain pteux (Robelin 1983). Morizet et al. (1984) ont montr que la variabilit gnotypique de la tolrance la scheresse ne se manifeste que si le stress s'est dvelopp pendant la priode de floraison. Un stress plus prcoce ne conduit pas une diffrenciation gnotypique. Des tudes physiologiques sur la priode de floraison ont permis de confirmer l'importance de cette priode pour le criblage (Westgate et Boyer 1986; Tollenaar 1989; Ouattar et al. 1987a). Les travaux de Ouattar et al. (1987b) permettent d'tendre la priode de criblage la phase de dveloppement du grain. G entinetta et al. (1986) ont mis en

181

vidence la possibilit de criblage durant la phase de maturation physiologique du grain. Ces diverses connaissances bibliographiques ont permis

dentreprendre le criblage varital du mas au Burkina en vue de mettre la disposition des agriculteurs des zones risque de scheresse, des varits prouves par rapport au dficit hydrique en cours de culture. Dans cette communication, nous prsentons les principaux rsultats acquis dune part (i) sur la slection inter cultivars locaux et dautre part sur (ii) lamlioration intra varitale de varits lites par rapport au stress hydrique.

Slection inter varitale pour la rsistance la scheresse Matriel vgtal

Le matriel soumis au criblage contre le stress hydr ique est constitu de 21 cultivars locaux slectionns en 1998 et 1999 pour leur adaptation aux conditions de culture de saison des pluies. Ils proviennent de divers villages du Burkina do ils ont t obtenus par prospection auprs des paysans. Les donnes de collecte et dvaluation en station ont mis en vidence leur grande variabilit (couleur, cycle, forme de lpi) et leur rusticit (adaptation aux conditions difficiles de culture : sol pauvre, scheresse). Ces cultivars sont repartis en cycle vari ts extra prcoce (7584j) et prcoce (8594 j). La liste nominative des cultivars locaux est cidessous prsente ; chaque nom comporte le nom du village de collecte suivi dun numro de collecte.La liste des cultivars tudis est la suivante : Cultivars extra prcoces : Finland 1, Yrfoula 1, Klesso 1, Silmyri 1, Hound 3, Sobarani 2, Carrefour 1, Bondoukuy 1, Ki 1 Cultivars prcoces : Noumoudara 1, Samandni 1, Kouentou 3, Kawara 3, Konandougou 1, Djonkl 2, Kouentou 7, Tenasso 1, Dand 2, Kimini 1, Yasso 2, Nbia 1,

Mthodes
Dispositif : Chaque groupe de prcocit est valu dans un essai diffrent utilisant un dispositif factoriel deux facteurs (facteur varital, facteur hydrique). Le facteur irrigation comprend deux

182

niveaux : le niveau To correspondant des parcelles irrigues lETM donc sans stress hydrique au cours de la culture ; le niveau Ts correspond des parcelles subissant 30 jours de privation deau (16 jours avant la floraison mle et 14 jours aprs cette dernire). Chaque nive au de ce facteur est regroup dans un seul bloc pour viter les interfrences entre To et Ts. Le facteur varital est randomis dans les rptitions. La parcelle dtude est reprsente par 48 plantes semes selon la densit de peuplement de 50 000 plantes/ha. Chaque essai comporte quatre rptitions ; les essais ont t implants en saison sche (13 janvier 2000) la valle du Kou. Lirrigation de type gravitaire est effectue tous les 7 jours. Les conditions de fertilisation pour chacun des essais a consist apporter aux plantes durant le cycle, 97 units fertilisantes d'azote, 46 units fertilisantes de phosphate, 28 units fertilisantes de potassium. Ces apports ont t raliss en utilisant l'engrais NPK (14-23-14) raison de 200 kg/ha et du K2SO4 50 kg/ha 15 jours aprs semis (jas). L'apport d'ure a t fractionn en deux (100 kg/ha 30 jas et 50 kg/ha 45 jas). Les engrais sont pandus en side -dressing dans des raies ctoyant les plantes. Variables: Les variables suivies sur 10 plantes concernent, lcart en jours entre les floraisons mle et femelle (ASI) au sein dun mme niveau dirrigation ; plus cet intervalle est important, moins le matriel considr fleuri de manire synchrone. La hauteur des plantes (HP) et la hauteur dinsertion de lpi principal (HIE) sont mesures en centimtres ; la notation de la snescence foliaire (SNF) est une estimation de la surface foliaire dessche des plantes sous leffet du stress hydrique. Elle est note sur une chelle de 1 10 o chaque uni t de lchelle reprsente 10% de surface foliaire dessche (chelle de notation prconise par l'IITA). Les observations concernant la hauteur des plantes et la snescence sont ralises au 14 m e , 21 m e et 28 m e jour aprs larrt de l'irrigation.). La reprise des plantes (RP) est une estimation de la mortalit des plantes due au stress hydrique ; elle est effectue au quatorzime jour aprs la reprise de l'irrigation suivant une chelle de 1 5 o 1 correspond une reprise totale de toutes les plantes et 5 une mortalit totale des plantes.

Analyses statistiques: Elles concernent une analyse de variances faites en prenant en compte les effets rptition, irrigation, varital

183

et linteraction de ces deux derniers. Cette analyse permet dune part de noter lexistence effective dun effet des diffrents niveaux dirrigation dans lexprimentation et dautre part, dvaluer la variabilit gntique inter varitale. Une analyse en composantes principales (ACP), est faite pour l'identification des principales variables discriminantes du matriel. Ces variables serviront de critres de base la slection des meilleurs gnotypes. Les meilleurs matriels par rapport la scheresse sont ceux ayant prsent le moins de variation de leur comportement entre la culture irrigue optimale et la culture stressante (Monneveux 1992). Les analyses statistiques sont ralises en utilisant des programmes conus sur le logiciel S.A.S. (1987).

Rsultats
Nous prsentons successivement les rsultats danalyse de variances pour apprcier limpact du stress hydrique dans lexprimentation et les rsultats de lanalyse en composantes principales permettant didentifier les variables principales qui serviront de critres de choix inter matriel. Rsultats danalyse de variances : Le Tableau 1 regroupe les rsultats des analyses de variances effectues sur les cultivars extra prcoces et les cultivars prcoces. Les variables prsentes sont lcart entre la floraison mle et femelle (ASI), la hauteur des plantes (HP), la hauteur dinsertion de lpi (HIE), les notes de snescence foliaire (SNF) et la reprise des plantes aprs le stress hydrique (RP). Le Tableau 1 montre au niveau des cultivars extra prcoces quexception faite de lcart de floraison il ny a pas deffet rptiti ons. Par contre, des diffrences trs hautement significatives (p = 0.001) sont observes pour lcart la floraison (ASI), la snescence foliaire (SNF) et la reprise (RP) pour les cultivars prcoces. Ces effets sont toutefois contrls par le dispositif exprimental.
Tableau 1. Rsultats de l'analyse de variance des cultivars extraprcoces et prcoces. Sources de Rptitions (A) Irrigation (B) variation Varital (C) Cultivars extra prcoces Interaction (BxC)

184

ASI HP HIE SNF RP ASI HP HIE SNF RP

5.23** 1.38 2.39 1.51 1.58 7.90*** 1.79 1.42 6.67*** 6.04**

159.03*** 98.06*** 65.61*** 111.41***

1.55 6.54*** 15.90*** 1.59

1.09 0.86 1.40 1.75 6.51*** 0.34 1.54 1.85 0.57 1.03

109.23*** 5.65*** Cultivars prcoces 32.80*** 270.40*** 64.24*** 171.02*** 58.61*** 0.76 3.68*** 4.19*** 0.58 0.90

*, **, *** significatif 5%, 1% et 0,1% respectivement.

Pour le facteur irrigation, les diffrences statistiques sont trs hautement significatives pour tous les caractres, dmontrant limpact effectif du stress hydrique davec lirrigation optimale. Au niveau varital, des diffrences trs hautement significatives sont observes pour les caractres lis l'architecture de la plante (HP, HIE) et la reprise des plantes (RP) au sein des cultivars extra pr coces. Ces diffrences ne concernent que les caractres darchitecture dans les cultivars prcoces. Ceci montre bien lexistence dune variabilit gntique. L'interaction entre les effets varitaux et les niveaux dirrigation ne prsente pas de signification dans les cultivars extra prcoces tandis quelle nest significative que pour la reprise des plantes (RP) au sein des cultivars prcoces. Cette quasi -absence dinteraction indique que le stress hydrique a le mme impact sur les diffrents cultivars, exception faite des aspects de variabilit gntique dtermine. Ces rsultats indiquent dune part que les cultivars locaux sont gntiquement diffrents et dautre part, quils ont tous subit les effets du stress hydrique. Leur comportement vis vis de ce stress dpendra de limpact du stress hydrique sur les variables principales dtermines par lanalyse en composantes principales. Moins limpact du stress sera important sur ces variables, plus tolrant seront les varits. Rsultats de lanalyse en composantes principales. Les rsultats de lanalyse en composantes principales sont prsents dans le Tableau 2; il sagit de la contribution des diffrentes variables la

185

constitution des axes de lACP et de la contribution de chaque axe lexplication des informations gnres par lACP. Au niveau des cultivars extra prcoces, lanalyse en composantes principales indique que les quatre premiers axes expliquent 98% de linformation. Le plan 1/2 reprsent par les axes 1 et 2 explique 87% de la variabilit observe. La variable HIE est la variable principale discriminant les cultivars extra prcoces; les variables secondaires sont la snescence foliaire (SNF) et la reprise (RP).

Au niveau des cultivars prcoces, lanalyse en composantes principales montre que les quatre premiers axes justifient 96% de la variabilit observe. Le plan 1/2 explique 86% de linformation observe. Laxe 2 est caractris par une contribution importante de la variable, HP1 (80%), tandis que laxe 1 est dtermin par SNF2 (38%). Ces variables sont dterminantes dans la sparation des cultivars par rapport aux deux traitements. Lidentification des diffrentes variables principales par lACP permet une slection inter varitale base sur ces variables. Les meilleurs cultivars sont ceux prsentant le moins de variation entre le T0 (irrigation optimale) et le Ts (traitement sous stress) pour les variables considres. Slection inter cultivars extraprcoces. Les cultivars extra prcoces slectionns sont ceux ayant subi le moins d variation entre le traitement l'irrigation e optimale et le traitement stressant en se basant sur les caractres darchitecture (HP et HIE). Ainsi, les cultivars Silmiri 1, Bondoukuy 1, et Ki 1 ont t retenus pour leur bon comportement.
Tableau 2. Contribution des diffrentes variables dans lACP. Variables Axe1 Axe2 Axe3 Axe4 principales 0,37 0,05 0,10 0,63 -0,24 -0,55 0,51 0,35 -0,33 -0,36 0,31 -0,15 -0,32 0,38 0,41 -0,02 -0,24 0,63 0,38 0,02 0,37 0,37 0,05 0,06 0,29 0,27 0,07 0,06

Matriel Extra prcoces

Variables ASI HP1 HP2 HP3 HIE SNF1 SNF2 SNF3 RP Cumul

HIE

0,38 0,04 0,35 -0,15 0,73 0,87

0,21 -0,05 0,34 -0,66 0,96 0,98

186

Prcoces

ASI HP1 HP2 HP3 HIE SNF1 SNF2 RP Cumul

0,35 0,37 0,04 0,77 -0,29 0,80 -0,01 -0,33 -0,39 0,13 -0,06 -0,03 -0,38 -0,11 0,31 0,17 -0,33 0,07 0,71 0,25 0,36 0,02 0,50 -0,44 0,38 -0,05 0,37 -0,13 0,34 0,78 0,44 -0,01 -0,04 0,86 0,93 0,96 HP

Slection inter cultivars prcoces. Le tableau 3 regroupe les donnes moyennes obtenues par les cultivars prcoces. Les variables prsentes sont ASI, HP, HIE, SNF et RP. On remarquera que le dcalage entre les floraisons est rduit pour ces cultivars dans le traitement Ts. Ces cultivars possdent galement une bo nne capacit de reprise ds le retour des conditions hydriques favorables. Au niveau des varits prcoces, cinq dentre elles ont montr un comportement acceptable en condition de scheresse. Le tableau 3 indique le comportement des varits slectionnes. Elles prsentent une bonne synchronisation des floraisons malgr le stress hydrique de 30 jours. Ces varits prsentent galement une bonne capacit de reprise ds le retour des conditions hydriques favorables. Le comportement diffrentiel des cultivars par rapport la scheresse est lie aux caractristiques de base de ces cultivars (HP, HIE). La valeur de ces caractres dtermine le comportement des cultivars vis vis de la scheresse. Gentinetta et al. (1986) et Ouattar et al. (1987) ont indiqu en effet que la tige constitue une zone tampon en cas de stress hydrique. Lexistence dune variabilit gntique inter cultivars pour la hauteur de ce caractre est en mme temps responsable du comportement diffrentiel inter cultivar pour le stress hydrique.
Tableau 3. Varits extra-prcoces slectionnes. ASI Cultivars Silmiri 1 Bondoukuy 1 Ts 5,3 8,5 T0 4,0 5,0 D 1,3 3,5 Ts HP3 T0 D Ts 21,8 68,6 Extra prcoce 53,4 104,3 -50,9 107,8 162,1 -54,4 41,6 -19,9 84,3 -15,6 HIE T0 D

187

Ki 1 Kouentou 3 Kouentou 7 Tenasso 1 Dand 2 Nbia 1 Cultivars

9,0 6,3 5,8 5,3 5,3 6,8

8,0 4,5 2,0 1,0 2,3 2,0

1,0

102,9 150,3 -47,4

62,3

72,6 -10,4

Prcoce 1,8 121,4 170,9 -49,5 3,8 129,1 164,2 -35,1 4,3 3,0 4,8 135,0 170,1 -35,1 106,1 176,9 -70,8 108,3 179,8 -71,6 RP D Ts T0

78,8 101,2 -22,5 79,8 91,4 -11,6 90,0 106,6 -16,6 64,1 100,1 -36,0 64,3 105,1 -40,7

SNF3 Ts T0 Extra prcoce

Silmiri 1 Bondoukuy 1 Ki 1 Kouentou 3 Kouentou 7 Tenasso 1 Dand 2 Nbia 1

6,3 3,6 4,1 4,4 4,1 4,4 3,3 3,8

0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0

6,3 3,6 4,1 Prcoce 4,4 4,1 4,4 3,3 3,8

3,6 1,8 2,3 2,0 1,4 1,4 1,5 1,9

1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0

2,6 0,8 1,3 1,0 0,4 0,4 0,5 0,9

D= Ts-To

Discussion
Les travaux de Sarr (1975) et de Tesha (1991) ont montr galement lexistence de cette variabilit entre les cultivars pour la rsistance la scheresse. Cette variabilit se manifeste pour les caractres lis la hauteur des plantes et la hauteur dinsertion de lpi pour les deux groupes de cultivars. En effet les rsultats de l'analyse de variances rvlent que des diffrences significatives existent entre les gnotypes pour les paramtres.

Conclusion
La rvlation de la variabilit inter varitale nous a permis de slectionner des varits dans les deux groupes de cultivars. Ainsi dans les cultivars extra prcoces, les varits Silmiri 1, Bondoukuy 1, et Ki 1 ont montr un bon comportement sous stress hydrique, tandis que dans les cultivars prcoces, Kouentou 3, Kouentou 7, Tenasso 2, Dand 2, et Nbia 1 se sont distingus.

Slection intra varitale pour la rsistance la scheresse

188

Lamlioration varitale comporte une slection inter varitale pour ladaptation aux conditions normales de culture (maladies notamment) et une slection intra varitale pour la rsistance au stress hydrique. Le criblage de varits par rapport aux diffrentes maladies est une ncessit si l'on veut amliorer la productivit du mas en conditions de stress hydrique. En effet une plante malade rsistera moins aux effets de la scheresse comparativement une plante saine. Compte tenu que la slection pour la rsistance la scheresse est effectue en saison sche (faible pression des maladies), une valuation du matriel choisi est ncessaire en saison humide. Dans le criblage contre la scheresse, il sagit dvaluer le comportement par rapport au stress hydrique de matriels slectionns en saison humide pour leur bon comportement par rapport aux maladies. Cette exprimentation a t mene en saison sche la valle du Kou. La recombinaison des entres retenues pour la cration de nouvelles varits performantes achve une tape du processus damlioration intra varitale (Fig 1).

Matriel vgtal
Le matriel soumis au test de rsistance la scheresse est constitu de 123 lignes S1 slectionnes en saison humide 1999 parmi 604 lignes S1. Ces lignes ont t slectionnes par rapport aux maladies (helminthosporiose, curvilariose) et ont obtenu une note de 1 2.5 sur une chelle 1-5 (1 : rsistante ; 5 : sensible). Les familles de lignes S1 sont extraites par autofcondations des varits lites : FBC6, Pool 16 DT, DTE W SR BC3 C0 et DTE Y SR BC3 C0 rputes pour leur bon comportement vis vis de la scheresse. FBC6 (Farako-B Composite n6), est u varit ne slectionne par l'INERA. Elle est issue dun brassage de 8 composites. Cest une varit prcoce avec un cycle semismaturit de 91 jours. Adapte aux zones pluviomtrie suprieure ou gale 900 mm, son rendement potentiel est de 5.6 tonnes lhectare. Cette varit grain jaune corn dent prsente un complexe de rsistance aux diffrentes maladies du mas. Les trois autres varits sont des obtentions du WECAMAN.

Les diffrents groupes de lignes S1 ont t codifis par la lettre DR pour " Drought Resistance" suivi de deux numros. Le premier identifie la varit source (1 pour FBC6, 2 pour Pool 16 DT, 3 pour

189

DTE W SR BC3 C0 et 4 pour DTE Y SR BC3 C0) et le second la famille. Ce sont respectivement 39, 29, 35 et 20 lignes S1 de FBC6, Pool 16 DT, DTE W SR BC3 C0 et DTE Y SR BC3 C0 qui sont values par rapport au stress hydrique.

Mthodes
Les mthodes utilises dans cette partie de notre tude sont analogues celles prcdemment dcrites dans la slection inter varitale pour la rsistance la scheresse. Chaque groupe de lignes est valu dans un essai diffrent utilisant un dispositif factoriel deux facteurs (facteur hydrique, facteur varital) et trois rptitions.

Rsultats
Rsultats danalyse de variances. Le Tableau 4 contient le rcapitulatif des rsultats danalyse de variances effectues sur les diffrentes variables. Il peut y tre not lexistence dune diffrence statistique trs hautement significative pour le facteur hydrique au niveau de tous les caractres. Cest di re que les niveaux dirrigation ont eu un impact sur les caractres agronomiques suivis.

Slection inter varitale

Slection intra varitale

Matriels choisis

Cration de lignes rsistantes

Cration de varits rsistantes

190

Figure 1 : Schma du criblage contre le stress hydrique

Tableau 4. Rsultats des analyses de variances sur les familles DR1, DR2, DR3, DR4. Variables Rptitions (A) 2.41 5.49** 3.10* 0.15 0.53 4.32* 4.67* 15.14*** 9.81*** 11.73 0.85 0.75 4.92** 1.08 12.66 0.65 4.63* 9.26*** 8.38*** 1.40 Irrigation (B) 38.59*** 207.68*** 63.56*** 291.58*** 143.81*** DR2 459.85*** 263.08*** 35.36**** 180.20*** 290.20*** 116.14*** 428.82*** 18.68*** 400.15*** 347.11*** DR4 1.39 2.21** 2.69** 1.34 2.83*** 1.18 1.97* 2.38 1.34 2.83*** 282.96*** 380*** 165.65*** 320.06*** 214.06*** 1.39 1.69* 3.69**** 0.64 1.07 1.57* 2.46*** 4.46*** 1.30 3.03*** 28.00 0.96 1.18 0.64 1.06 0.81 1.29 1.24 1.38 3.06*** Gnotypes (C) Interaction (BxC) 1.22 2.13*** 2.09*** 1.22 1.43 1.23 0.91 0.77

Lignes extraites de FBC 6 DR 1 ASI HP HIE SNF RP ASI HP HIE SNF RP ASI HP HIE SNF RP ASI HP HIE SNF RP

1.22
1.14

Lignes extraites de Pool 16 DT

Lignes extraites de DTE W SR BC 3 C0 DR3

Lignes extraites de DTE W SR BC 3 C0

*, **, *** significatif 5%, 1% et 0,1% respectivement.

Au niveau gntique, des diffrences statistiques significatives sont observes entre les lignes pour quelques caractres. Les lignes DR1, DR2, DR3 et DR4 sont variables pour les caractres darchitecture (HP, HIE) ; les lignes DR3 et DR4 sont variables pour la reprise (RP) tandis que les lignes DR4 seules sont variables pour lcart de floraison (ASI). Ce rsultat est d dune part au fait que les lignes proviennent de varits lites dj homognises, et dautre part au fait quune slection inter ligne a t effectue pour la rsistance aux maladies dans chaque groupe. Linteraction gnotype*irrigation nest significative que pour quelques caractres (RP, HP) dans les familles de lignes DR3 et DR4. Labsence dune signification gnralise pour cette

192

interaction permet dutiliser comme critre de slection chacun des caractres tudis, pour peu quils soient identifis comme variables discriminantes dans lanalyse en composantes principales. Rsultats de lanalyse en composantes principales. Le Tableau 5 regroupe des rsultats de lanalyse en composantes principa les concernant la contribution des diffrentes variables lexplication de la variabilit densemble observe au sein des lignes.
Tableau 5. Contribution des diffrentes variables la constitution des axes de lACP. Variables Familles de lignes Variables AXE1 Lignes DR 1 ASI HP1 HP2 HP3 HIE SNF1 SNF2 SNF3 RP Cumul Lignes DR 2 ASI HP1 HP2 HP3 HIE SNF1 SNF2 SNF3 RP Lignes DR3 Cumul ASI HP1 HP2 HP3 HIE SNF1 SNF2 SNF3 RP Lignes DR 4 Cumul ASI 0.29 0.23 0.40 0.40 0.35 0.32 0.40 0.38 0.10 0.57 0.35 0.24 0.31 0.37 0.27 0.36 0.36 0.36 0.36 0.77 0.32 0.28 0.36 0.36 0.19 0.36 0.38 0.37 0.34 0.71 0.34 AXE2 0.45 0.59 0.17 0.23 0.37 0.35 0.22 0.23 -0.09 0.72 0.20 0.67 0.28 0.09 0.46 0.25 0.23 0.24 0.22 0.90 0.14 0.45 0.19 0.22 0.67 0.24 0.23 0.24 0.28 0.84 0.09 AXE3 0.20 0.28 0.03 0.19 0.10 0.07 0.12 0.13 0.90 0.84 0.02 0.25 0.61 0.13 0.73 0.06 0.06 0.04 0.03 0.95 0.48 0.53 0.34 0.06 0.55 0.14 0.06 0.01 0.22 0.91 0.11 AXE4principales 0.52 0.29 0.33 0.16 0.37 0.52 0.10 0.01 HP 0.32 0.89 0.78 0.15 0.25 0.22 0.16 HP 0.13 0.31 0.35 0.04 0.97 0.73 0.31 0.16 0.06 0.35HIE 0.32 0.20 0.27 -0.03 0.95 0.70HP

193

HP1 HP2 HP3 HIE SNF1 SNF2 SNF3 RP Cumul

0.28 0.33 0.35 0.33 0.34 0.35 0.35 0.33 0.87

0.73 0.41 0.01 0.09 0.22 0.21 0.22 0.37 0.94

0.19 0.20 0.35 0.80 0.22 0.27 0.18 0.03 0.97

0.20 0.08 0.02 0.07 0.15 0.00 0.07 0.66 0.99

Caractre gras=contribution maximale la constitution de laxe ; Cumul : contribution cumulative des axes

Les quatre premiers axes de lACP expliquent 89 99% de linformation gnre dans lanalyse. Les axes 1 et 2 du plan expliquent seuls 72 94% du comportement de chaque groupe de lignes. Ces deux axes sont les axes principaux sur lesquels il faudrait rechercher les principales variables discriminantes des lignes. Les axes 3 et 4 contribuent un degr plus faible dans la discrimination des lignes. La contribu tion des diffrentes variables la constitution des axes principaux permet didentifier les variables principales discriminantes des lignes. Notons que les mesures effectues plusieurs dates sur les variables hauteur (HP1, HP2, HP3) et la snescence foliaire (SNF1, SNF2, SNF3) sont redondantes, car contribuant la mme contribution explicative des axes de lACP. Ainsi, nous ne retiendrons que les dernires mesures effectues sur ces variables dans la suite de notre interprtation. La variable cart de floraison nintervient que sur laxe 4 de lACP, qui contribue faiblement lexplication de la rpartition des lignes. La contribution de ASI sur cet axe est cependant importante (52 78%). Cette variable sera tout de mme utilis comme critre de slection en prenant le soin dviter les trop grands carts de floraison entre les deux niveaux dirrigation. Dans les lignes DR1, la variable principale discriminante est la hauteur des plantes (HP) ; secondairement, la hauteur dinsertion de lpi (HIE) et la snescence foliaire (SNF). Dix (10) lignes ont t slectionnes parmi les 39 lignes S1 extraites de FBC 6 ; elles ont par rapport la scheresse prsent le moins de variation de leur comportement entre la culture irrigue optimale et la culture

194

stressante. Le Tableau 6 prsente le comportement des lignes slectionnes.

Lcart maximal entre les floraisons est denviron 9 jours sous stress hydrique. Il est denviron 5 jours en irrigation optimale. Le pourcentage de surface foliaire dessche aprs lapplication du stress est infrieur 60% ; ce qui indique une bonne capacit de rsistance de ces lignes la scheresse de 30 jours ferme. Les caractres darchitecture prsentent galement une variation pouvant atteindre 66 cm sur la hauteur des plantes et 33 cm sur la hauteur dinsertion de lpi ; ceci sexplique par le fait que le stress hydrique a t appliqu lmergence de la panicule alors que la plante tait toujours en croissance.

Pour les lignes DR2, la variable principale utilise dans la slection est la hauteur des plantes (HP) ; la hauteur dinsertion de lpi (HIE), la snescence des feuilles, la reprise de plantes (SNF, RP), lcart de floraison (ASI) sont maintenus au niveau le plus faible possible. Notons que lcart la floraison de ces lignes sous le stress hydrique est plus important que celui observ prcdemment sur les lignes DR1. Les meilleures lignes sont celles prsentant le moins de diffrence entre le traitement To et le traitement Ts (Tableau 6). En ce qui concerne les lignes DR2 de Pool 16 DT, six lignes se sont montres intressantes sous stress hydrique. Ces lignes n'ont pas subi de diminution significative de taille (hauteur d'insertion de l'pi) malgr les conditions de stress. Dans les lignes DR3, la hauteur dinsertion (HIE ) est la variable principale ; les variables de snescence foliaire (SNF) et de reprise (RP) sont des variables secondaires.
Tableau 6. Comportement des lignes S1 slectionnes. S1 Code DR1 1 9 10 11 18 Ts 5,7 6,7 4,7 5,3 4,0 ASI To 1,7 2,7 2,7 2,3 2,0 D 4,0 4,0 2,0 3,0 2,0 Ts 121,3 92,3 113,7 112,7 85,0 HP T0 141,7 130,0 156,7 155,0 130,8 D 20,3 37,7 43,0 42,3 45,8

195

21 25 26 34 39 DR2 2 5 8 23 25 29 DR3 1 7 18 20 23 25 29 31 DR4 3 9 10 11 12 15

6,3 5,3 6,0 5,0 6,5 6,0 6,3 7,7 8,7 7,3 6,3 5,3 5,7 5,0 7,3 7,7 6,3 3,7 5,7 6,0 7,0 8,3 7,0 6,0 7,7

3,7 2,3 2,3 2,3 3,7 2,7 1,3 2,0 3,3 1,7 2,0 3,5 5,0 3,7 3,7 3,7 5,3 2,7 1,7 5,0 5,0 5,0 4,0 4,0 2,7

2,6 3,0 3,7 2,7 0,8 3,3 5,0 5,7 5,3 5,7 4,3 1,8 0,7 1,3 3,7 4,0 1,0 1,0 4,0 1,0 2,0 3,3 3,0 2,0 5,0

131,0 109,7 101,0 91,8 117,3 87,3 71,7 53,7 68,0 76,7 80,3 88,0 68,0 85,3 69,7 93,3 86,0 74,3 81,0 81,7 73,3 67,3 82,0 67,3 68,0

166,7 149,7 136,7 109,0 152,0 111,7 99,3 97,7 90,7 99,0 100,3 112,5 115,0 129,7 117,7 118,7 121,3 116,0 121,3 126,7 117,0 133,7 125,3 132,0 103,7

35,7 40,0 35,7 17,3 34,7 24,3 27,7 44,0 22,7 22,3 20,0 24,5 47,0 44,3 48,0 25,3 35,3 41,7 40,3 45,0 43,7 66,3 43,3 64,7 35,7

D= Ts - To

Tableau 6 continued. HIE Code DR1 S1 1 9 10 11 18 21 25 26 34 39 Ts T0 D 1,0 Ts 1,9 4,7 3,3 2,4 2,7 4,0 3,8 4,3 4,7 2,3 74,7 75,7 SNF T0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 D 1,9 4,7 3,3 2,4 2,7 4,0 3,8 4,3 4,7 2,3 Ts 1,3 3,0 1,5 1,0 1,9 1,2 1,8 1,8 2,2 1,0 RP T0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 D 0,3 2,0 0,5 0,0 0,9 0,2 0,8 0,8 1,2 0,0

65,7 72,3 6,7 66,0 84,0 18,0 70,3 87,7 17,3 53,7 68,0 14,3 87,0 98,0 11,0 77,3 84,7 62,7 64,0 7,3 1,3

54,4 57,0 2,6 72,0 85,5 13,5

196

DR2

2 5 8 23 25 29

55,3 56,3 40,3 43,3 31,0 36,0 37,7 42,0 43,3 45,3 44,3 44,7 48,0 50,0 36,7 44,3 57,0 58,7 40,0 41,3 49,7 51,0 38,0 42,0 46,7 48,0 51,0 53,7 54,7 64,0

1,0 3,0 5,0 4,3 2,0 0,3 2,0 7,7 1,7 1,3 1,3 4,0 1,3 2,7 9,3

3,1 3,9 4,7 4,8 2,3 3,7 1,7 1,2 2,7 1,9 2,1 2,2 2,0 3,0 3,7 3,5 4,2 5,7 5,3 4,3

0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0

3,1 3,9 4,7 4,8 2,3 3,7 1,7 1,2 2,7 1,9 2,1 2,2 2,0 3,0 3,7 3,5 4,2 5,7 5,3 4,3

2,7 2,2 2,8 2,8 3,3 2,7 1,5 1,5 2,7 1,7 2,7 2,8 2,3 3,0 2,5 1,7 2,5 2,3 2,0 1,7

1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0 1,0

1,7 1,2 1,8 1,8 2,3 1,7 0,5 0,5 1,7 0,7 1,7 1,8 1,3 2,0 1,5 0,7 1,5 1,3 1,0 0,7

DR3

1 7 18 20 23 25 29 31 3 9 10 11 12 15

DR4

42,7 55,7 13,0 43,0 76,0 33,0 53,3 64,0 10,7 46,0 70,3 24,3 38,3 51,7 13,3

D= Ts - To Huit lignes se sont bien comportes sous stress hydrique. Ces lignes n'ont pas subi de variations importantes entre le traitement l'irrigation optimale et le traitement stressant pour les paramtres considrs (Tableau 6). Pour les lignes DR4, la variable hauteur de plante (HP) caractrisant laxe 2 est la variable principale discriminante. La reprise (RP) et la snescence foliaire (SNF) sont des variables secondaires. Six lignes prsentent un bon comportement sous stress hydrique. Elles ont en particulier une capacit de reprise ds le retour des conditions hydriques favorables.

Discussion
Le degr de fanaison a t not au cours de lapplication du stress au travers de la variable snescence foliaire. Sobrado (1987) confirme lintrt de cette variable en tant que critre lors dun travail de slection pour la rsistance la scheresse ; en effet, le fonctionnement de lappareil foliaire conditionne une bonne activit

197

photosynthtique, elle -mme favorable lobtention dune bonne production.

Laptitude une reprise acclre chez les gnotypes contribuera un rendement important par suite d'un niveau lev de photosynthse tardive. Ceci est report par Morizet et al. (1984) par mesure directe pendant la priode sche et galement aprs, et en conditions de fort clairement ; ils ont montr que la varit rsistante Liza a une activit photosynthtique suprieure celle de la varit sensible LG11. La hauteur des plantes et la hauteur dinsertion de lpi sont les paramtres les plus variables entre les diffrents gnotypes. Les gnotypes sensibles ont subi une diminution significative de leur taille alors que les gnotypes rsistants ont continu crotre malgr le stress hydrique de 30 jours. Cette aptitude des gnotypes rsistants crotre sous stress hydrique serait lie selon Gentinetta et al. (1986), Ouattar et al. (1987a), l'existence d'un gne majeur dominant prsent chez les gnotypes rsistants et responsables de laccumulation de sucres dans la tige de mas. Cette accumulation constitue une rserve importante en cas de stress hydrique. En effet la tige de mas est la source principale d'assimilats pour le remplissage des grains en cas de longue scheresse. Elle se prsente donc comme une alternative pour assurer le dveloppement du grain en l'absence de la photosynthse. En outre, la tige constitue une zone tampon entre le grain et le sol (ou l'atmosphre) par le maintien des conditions hydriques favorables, remdiant temporairement au dficit de l'eau dans le sol. Ces diffrentes constatations indiquent que les meilleurs gnotypes sont les lignes qui ont subi le moins de diminution de leur taille. Le dcalage entre la floraison mle et la floraison femelle demeure un indice intressant de criblage contre la scheresse. Herrero et Johnson (1981) ont indiqu que le stress hydrique est responsable dune augmentation de lcart entre la floraison mle et femelle chez les gnotypes sensibles. Ce rsultat est observ dans notre tude. Robelin (1983) a montr quun stress hydrique intervenant avant le stade de floraison n'a aucune consquence significative sur la production du mas. Morizet et al. (1984) ont montr que le stade floraison est le seuil auquel la variabilit gnotypique de la rsistance la scheresse se manifeste ; lapplication du stress hydrique a t propos dans nos exprimentations, ce qui permet

198

dobserver les variations significatives du comportement des lignes sous le stress hydrique.

Conclusion
Les critres ci -dessus cits (la snescence, laptitude la reprise, la hauteur des plantes et la hauteur dinsertion de lpi, lcart entre la floraison mle et femelle) nous ont permis de slectionner un ensemble de lignes performantes en situation de stress hydrique dans les varits FBC6, Pool 16 DT, DTE W SR BC3 C0, et DTE W SR BC3 C0.

Conclusion gnrale
Nos rsultats indiquent que des caractres morphologiques simples peuvent servir de critres de slection de gnotypes rsistants la scheresse comme l'ont montr Edmeades et al. (1995) et Tollenaar (1989) sur la taille des plantes et des panicules et sur la surface foliaire. Les cultivars identifis pour le bon comportement vis vis du stress hydrique serviront de source de gnes pour lamlioration de la rsistance la scheresse. Les famille s de lignes obtenues ont t respectivement brasses pour crer les varits rsistantes au stress hydrique, nommes respectivement FBDR1, FBDR2, FBDR3, et FBDR4 au cours de la saison humide 2000. Le second brassage de ces nouvelles varits est effectu la Valle du Kou en saison sche 2001 (Fig. 1). La poursuite de la cration de lignes fixes rsistantes au stress hydrique est en cours (stade S3) ; lextraction de lignes S1 dans les cultivars locaux slectionns pour leur bon comportement la sche resse a t faite en saison humide 2000.

Rfrences
Blum, A. 1988. Methods of plants breeding for drought resistance. Pp 153190 in Agriculture: Drought resistance in plants. Physiological and genetic aspects.

199

Edmeades, G.O., M. Banziger, S.C. Chapm an, J.M. Ribaut, and J. Bolanos. 1995. Pp 22-41 in B. Badu-Apraku, M.O Akoroda, M. Ouedraogo, and F.M. Quin (eds.) Recent advances in breeding for drought tolerance in maize. Proceedings of a Regional Maize Workshop 29 May2 June, 1995, Cotonou, Benin Rep ublic. WECAMAN/IITA. El Jaafari S., R. Paul, P. Lepoivre, J. Semal, et E. Latitat. 1993. Rsistance la scheresse et rponse lacide abscissique : analyse dune approche synthtique. Cahier dAgriculture 2 :256263. Fischer, K.S., E.C. Johnson, and G.O. Edmeades. 1981. Breeding and selection for drought resistance in tropical maize. Principles and methods in crop improvement for drought resistance with emphasis on rice. IRRI, 48 May 1981. 19p. Fischer, K.S., E.C. Johnson, and G.O. Edmeades. 1983. Breeding and selection for drought resistance in tropical maize. CIMMYT, EL Batan, Mjico. Gentinetta, E., D. Cepi, G. Leporic, M. Motto, and F. Salamini. 1986. A major gene for delayed senescence in maize. Pattern of photosynthate accumulation and inheritance. Plant Breeding 97:193206. Hema, D., S.K. Kim, F. Mondeil, et B. Tio-Toure. 1998. Tolrance la scheresse du mas en Afrique: aptitude la recombinaison et hritabilit du phnomne de transpiration. Science et Technique, Sciences Naturelles 23(1):2937. Herrero, M.P. and R.R. Johnson. 1981. Drought stress and its effects on maize reproduction system. Crop Sci. 21:105110. Johnson, S.S. and J.L. Geadelman. 1989. Influence of water stress on grain yield response to recurrent selection in maize. Crop Sci. 29:558564. Ludlow, M.M. and R.C. Muchow. 1990. A critical evaluation of traits for improving crops yield in water limited environments. Advances in Agronomy 43:107153. Monneveux, P. 1992. Amlioration gntique de la tolrance au dficit hydrique des crales d'hivers. ENSA Montpellier. INRA. Morizet, J., A.M. Tribo, et M. Pollacsek. 1984. Rsistance la scheresse chez le mas : quelques mcanismes impliqus. Physiologie et Production du Mas. INRA (Pau) France. 167 174.

200

Ouattar, S., R.J. Jones, and R.K. Crookston. 1987a. Effect of water deficit during filling on the pattern of maize kernel growth and development. Crop Sci. 27:726730. Ouattar, S., R.J. Jones, R.K. Crookston, and M. Kajeiou. 1987b. Effect of drought on water relation of developing maize kernels. Crop Sci. 27:730735. Robelin, M. 1983. Fonctionnement hydrique et adaptation la scheresse. Colloque Physiologie du mas. Royan 1517 Mars 1983, Paris, INRA. S.A.S. 1987. Guide for personal computers. Version 6. SAS Instute Inc. SAS Instute, Cary, NC, USA. Sanou, J. 1991. Slection du mas (Zea mays L.) pour la rsistance la scheresse. Synthse bibliographique, ENSA Montpellier. 17p. Sarr, A. 1975. Modle d'tude d'une structure de population : Analyse de la variabilit gntique de populations "naturelles" de mas (Zea mays L.) du sngal. Thse de Docteur- Ingnieur. Universit de Paris-Sud. 155p. Sivakumar M.V.K. et Gnoumou. 1987. Agroclimatologie de l'Afrique de lOuest: Le Burkina Faso. ICRISAT Bulletin d'information n23. Sobrado, M.P. 1987. Leaf rolling : a visual indicator of water deficit in corn (Zea mays L.). Maydica 32:918. Some, L. 1989. Diagnostic agroclimatique du risque de scheresse au Burkina Faso. Etude de quelques techniques agronomiques amliorant la rsistance pour les cultures de sorgho, de mil, et de mas. Thse de doctorat, Universit de Montpellier II. Sciences et techniques du Languedoc. 312p. Tesha A.J. 1991. Testing for drought resistance in maize cultivars grown in Tanzania. Pages 219224 in Influence du climat s ur la production des cultures tropicales. Compte rendu du Sminaire rgional organis par la Fondation International pour la Science (IFS) et le Centre Technique de Coopration Agricole et Rurale (CTA), Ouagadougou, B.F. Tollenaar, M. 1989. Response of dry matter accumulation in maize to temperature. Crop Sci. 29:12751279.

201

Wery, J. 1987. Adaptation des plantes cultives au dficit hydrique. Cours UB 4 (matrise de leau et du bioclimat). ENSA Montpellier. Westgate, M.E. and J.S. Boyer. 1986. Silk and pollen water potentials in maize. Crop Sci. 26 : 951956.