Development of drought- and Striga-resistant population

The present study used S₁ lines derived from the drought-tolerant and Striga-resistant early maturing white population TZE-W Pop DT C₃ STR C₅, after five cycles of selection for Striga resistance followed by three cycles of selection for drought tolerance. The breeding methodology employed for the development of the Striga-resistant and drought-tolerant early-maturing (90–95 days to maturity) white population TZE-W Pop DT C₃ STR C₅ has been described in detail by Badu-Apraku et al. (Reference Badu-Apraku, Fakorede and Fontem Lum2007, Reference Badu-Apraku, Fakorede and Lum2008, Reference Badu-Apraku, Fakorede, Lum and Akinwale2009). Briefly, the population was developed from outstanding local and improved maize germplasm with white endosperm, identified through several years of extensive testing in WCA as resistant to Striga and tolerant to drought. Following one cycle of half-sib recombination of the drought-tolerant white materials, Pool 16 DT, Pool 16 Sequia C2 and DR-W Pool BC₁F₁, the resultant early white population was designated as TZE-W Pop. The Striga-resistant/tolerant IITA inbred line TZi 3 (1368 STR) was incorporated into TZE-W Pop to upgrade the level of Striga resistance and/or tolerance of the population. This was followed by two backcrosses, the generation of S₁ progenies, the selection of Striga-resistant S₁ lines from each population and two cycles of random mating under artificial Striga infestation and induced moisture stress to form TZE-W Pop DT STR C₀. Selection for drought tolerance was made under controlled conditions at Ferkéssedougou (Ferké) (9°30′N, 5°10′W; 325 m asl) and Sinématialli (Siné) (9°37′N, 3°04′W; 305 m asl) in Côte d'Ivoire and Kamboinse (12°28′N, 1°32′W; 303 m asl) in Burkina Faso. At Ferké and Siné, the crop was grown under irrigation during the dry season. At Kamboinse, a Sudan savanna site, varying levels of drought tolerance were induced using tied and untied ridges. The tied ridge is a water-harvesting technique capable of capturing water and holding it in place to minimize runoff and improve water infiltration (Boa Reference Boa1966; Badu-Apraku et al. Reference Badu-Apraku, Fakorede and Lum2008). At Siné in the Southern Guinea savanna, a relatively high plant density of about 80 000 plants/ha (compared with the standard density of 66 000 plants/ha) was used to induce stress conditions for selection purposes. At the initial stage, selection for drought tolerance was made under controlled conditions at Ferké and Siné in Côte d'Ivoire and Kamboinse in Burkina Faso. At the later stages, the screening sites were shifted to Ikenne and Bagauda in Nigeria. At Ferké, Siné and Ikenne the crop was grown under irrigation during the dry season using an overhead sprinkler irrigation system, which applied 17 mm of water/week. Induced drought stress was imposed by withdrawing irrigation water from about 2 weeks before anthesis to the end of the season. Another strategy that was often adopted at Siné in the Guinea savanna zone was to use high plant density to induce stressed conditions for selection purposes. At Bagauda, the materials were exposed to natural terminal drought which normally starts from flowering and continues till harvesting.

The number of progenies screened in each cycle ranged from 196 to 256. Based on the data across Striga-infested environments, 25–30% of the top ranking S₁ families of the population were selected using a base index that involved grain yield, emerged Striga plants, Striga damage syndrome rating and number of ears per plant (EPP) measured under infested and/or non-infested conditions (MIP 1996; Menkir & Kling Reference Menkir and Kling2007). Similarly, the best genotypes under induced drought stress were identified, based on the index values estimated using the equation:

$$I = [\left( {2 \times {\rm Yield}} \right) + {\rm EPP} - {\rm ASI} - {\rm PASP} - {\rm EASP} - {\rm SGC}]$$

where ASI was the anthesis-silking interval, EPP was ears per plant, PASP was the plant aspect, EASP was the ear aspect and SGC was the stay-green characteristic (leaf death scores). A minimum of three seasons were required per cycle of selection. Five cycles of selection for high grain yield and Striga resistance and three cycles of improvement for grain yield and drought tolerance had been completed in the population by the time the present experiment was initiated in 2011.

Experimental protocol

Sixty S₁ families were extracted by selfing non-inbred (S₀) plants from cycles C₀, C₁, C₂ and C₃ of TZE-W Pop DT C₃ STR C₅ population. A total of 240 S₁ families were evaluated in two replicates using a resolvable incomplete block design with 16 incomplete blocks of 15 genotypes per replicate. The trials were evaluated under induced moisture stress at Ikenne (6°53′N, 30°42′E, 60 m asl, 1200 mm annual rainfall) during the 2010/11 and 2011/12 dry seasons and under well-watered conditions at Ikenne during the 2011 and 2012 rainy seasons in Nigeria. In addition, the S₁ families were evaluated under terminal drought at Kadawa (drought-prone environment; 11°45′N, 8°45′E, 468·5 m asl, 884 mm annual rainfall) during the 2011 growing season. The experimental units were single-row plots, each 3 m long with an inter-row spacing of 0·75 m and intra-row spacing of 0·4 m.

Uniform competition was ensured among plants between plots by planting at least two border rows at all the edges of each range and guard rows at the front and back of the experimental plots. Three seeds were planted and the seedlings were later thinned to two per hill about 2 weeks after emergence to give a final population density of about 66 666 plants/ha.

The induced drought stress was obtained by withdrawing irrigation water from 28 days after planting (DAP) until maturity, so that the maize plants relied on stored water in the soil for growth and development. During the first 4 weeks of growth, the plants were irrigated using a sprinkler irrigation system, which provided 17 mm of water each week. Thus the flowering and grain filling periods occurred under the managed drought stress with symptoms of stress appearing after about 25–30 days of induced drought stress. The soil at the experiment station at Ikenne is Alfisol (Soil Survey Staff 1999) and the experimental fields at the station are flat and fairly uniform, with high water-holding capacity. The well-watered (rainfed) experiment at Ikenne was conducted during the rainy season. At Kadawa, a drought-prone environment, the plants were subjected to natural terminal drought stress. Except for the amount of water applied in well-watered environments, all management practices were the same for both well-watered and drought stress studies. Fertilizer [nitrogen (N), phosphorus (P) and potassium (K)] was applied to the well-watered and drought-stressed plots at Ikenne and Kadawa, at the rate of 60 kg N/ha, 60 kg P/ha and 60 kg K/ha at planting. An additional 60 kg N/ha was top-dressed at 4 weeks after planting. The trials were kept weed-free with the application of atrazine and gramoxone as pre- and post-emergence herbicides at 5 litres/ha each of Primextra and Paraquat and subsequently by manual weeding.

Data collection

Data were recorded on both drought stress and well-watered plots for days to 50% anthesis (DA) and silking (DS) as the number of days from planting to when 50% of the plants had shed pollen and emerged silks. The anthesis-silking interval (ASI) was calculated as the difference between DS and DA. Plant and ear heights were measured as the distance from the base of the plant to the height of the first tassel branch and the node bearing the upper ear, respectively. Root lodging (the percentage of plants leaning more than 30° from the vertical), and stalk lodging (the percentage of plants broken at or below the highest ear node) were recorded. Stay-green characteristic (SGC) was recorded for the drought-stress experiments at 70 DAP on a scale of 1–9, where 1 = almost all leaves green and 9 = virtually all leaves dead. Plant aspect was recorded on a scale of 1–5 based on overall plant type, where 1 = excellent plant type and 5 = poor plant type. Husk cover was rated on a scale of 1–5, where 1 = husks tightly arranged and extended beyond the ear tip and 5 = ear tips exposed. Ear aspect was based on a scale of 1–5, where 1 = clean, uniform, large and well-filled ears and 5 = ears with undesirable features. Number of EPP was computed by dividing the total number of ears harvested per plot by the number of plants in a plot at harvest. For trials conducted under drought, harvested ears from each plot were shelled to determine the percentage grain moisture. Grain yield in kg/ha was computed from the shelled grain weight, adjusted to 150 g/kg moisture. For the well-watered experiments, a shelling percentage of 80 was assumed for all lines and grain yield (obtained from ear weight and converted to kg/ha) was adjusted to 150 g/kg moisture.

Data analysis

The analysis of variance (ANOVA) was conducted separately for data collected from drought-stressed and well-watered environments for grain yield, days to 50% silking, days to anthesis, ASI, EPP, plant height, ear height, SGC, ear and plant aspects, and husk cover using a RANDOM statement with the TEST option in the Statistical Analysis System (SAS, version 9.3) (SAS Institute 2011). The location–year combinations were considered as environments in the ANOVA for each trait. The effects of environment, replication, incomplete blocks, environment by cycle interaction and environment by genotype within cycle interactions were considered random effects; cycle and genotype within cycle effects were considered fixed when estimating means and s.e. per cycle, and were considered random when estimating variance components. The genotypes were nested within cycles in the present study in order to assess the performance of each genotype within each cycle. Means of cycles were separated using the least significant difference (LSD) test. The cycles of selection across location–year combination for each environment were combined to obtain the entry means.

Genetic variance and covariance component estimates were obtained for each cycle of selection from the combined ANOVA by restricted maximum likelihood (REML) method (Holland Reference Holland2006).

The statistical model used for the combined ANOVA is as follows:

$$Y_{klmi} = \mu _i + E_{ki} + R\left( E \right)_{kli}\, { + }\,G_{mi} + GE_{kmi} + \varepsilon _{klmi}$$

where Y _klmi is the observed measurement of trait i with mean effect μ _i , E_ki is the effect of environment k on trait i, R(E) _kli is the effect of replication l within environment k on trait i, G _mi is the effect of genotype m on trait i, GE _kmi is the effect of the interaction between genotype m and environment k on trait i, and ε _klmi is the experimental error effect associated with genotype m and replication l within environment k on trait i.

Broad-sense heritability (H ²) was computed as the ratio of the genetic variance to the phenotypic variance on a progeny-mean basis as follows:

$$H^2 = \displaystyle{{\sigma _g^2} \over {\sigma _g^2 + (\sigma _{ge}^2 /e) + (\sigma ^2/er)}}$$

where $\sigma _g^2 $ represents the component of variance estimate due to genotypes; $\sigma _{ge}^2 $ is the component of variance due to genotype × environment interaction (GEI); σ ² represents the estimate of experimental error variance; e is the number of environments; r is the number of replications per environment. Genetic variance was estimated by equating the observed mean squares to their expectations and solving for the desired components according to the method of Hallauer & Miranda (Reference Hallauer, Miranda, Hallauer and Miranda1988). The incomplete block effect of the model was dropped, the residual variance estimates computed and the error variance was estimated. Standard errors for genetic variance and heritability estimates were calculated using the method of Hallauer & Miranda (Reference Hallauer, Miranda, Hallauer and Miranda1988). Changes in the variances and heritability estimates among S₁ families were tested for significance by pairwise comparison of estimates using their s.e.

The predicted gain from selection (G _s ), based on C₃ alone, was estimated according to the method of Hallauer & Miranda (Reference Hallauer, Miranda, Hallauer and Miranda1988) as follows:

$$G_s = k\sigma _g^2 /\sigma _p$$

where k represents the standardized selection differential for S₁ families (20% selection intensity was used in this case, which is k = 1·3998), $\sigma _g^2 $ is the genetic variance, and σ _p denotes the square root of the phenotypic variance. The predicted gains may be biased upwards depending on the contribution of non-additive genetic variances to σ _g ². Means for each cycle were tested for significant differences using twice the s.e. and the difference between the means of C₃ and C₀ to provide an estimate of observed gain from selection. Entry means of the traits of the cycles across environments were regressed as dependent variables on cycle numbers (independent variable) to obtain estimates of realized gain per cycle of selection. The coefficient of linear regression (b-value) provided an estimate of the gain per cycle, which was divided by the intercept and multiplied by 100 to obtain the per cent response per cycle. The significance of the slope b was tested using orthogonal polynomial contrasts in the mixed model considering cycle number as fixed effect. The test was based on the linear degree of polynomial coefficients for comparison among four treatments with unequal intervals according to the method of Gomez & Gomez (Reference Gomez and Gomez1984). Specifically, the comparisons made were C₀ v. others, C₁ v. C₂, C₁ v. C₃ and C₂ v. C₃. Genotypic correlation coefficients among agronomic traits and their s.e. were computed with the REML method (Holland Reference Holland2006) using the MIXED (i.e. the effects of environments and replications were considered as fixed effects, while genotype and GEI were considered as random effects) and IML procedures of SAS (SAS Institute 2011). The correlation was calculated using the following formula:

$$r_{gij} = \displaystyle{{\sigma _{gij}} \over {\sqrt {\sigma _{gi}^2 + \sigma _{gj}^2}}} $$

where r _gij is the correlation coefficient, σ _gij is the genotypic covariance between traits i and j and $\sigma _{gi}^2, $ and $\sigma _{gj}^2 $ are variances of the two traits. Standard errors were obtained using the delta method as described by Holland (Reference Holland2006). A genetic correlation larger than twice its s.e. was considered statistically significant. Also, two times the s.e. was used to test for significant differences among the four cycles.