Genotype × environment interaction and grain yield stability of quality protein maize hybrids under stress and non-stress environments

Abstract

Evaluation of maize varieties under multiple environments, including drought and low nitrogen (N) stressed sites is an important breeding approach, to identify well adapted and stable maize varieties. This study was undertaken to identify new quality protein maize (QPM) hybrids that have good agronomic performance and assess the presence of genotype by environment (G × E) interaction and grain yield stability of QPM hybrids under different environment conditions. Forty-five hybrids, including two QPM, two non-QPM and one local check were evaluated across 34 environments under stress and non-stress conditions in Ethiopia, Zimbabwe, Zambia, Mozambique, and Malawi during 2018 to 2020. Additive Main Effects and Multiplicative Interaction (AMMI) and Genotype main effects plus G × E interaction (GGE) bi-plots were used for stability analysis. Environment, genotype and G × E interaction effects were significant for grain yield and other traits in all management conditions. The top yielding hybrids were 44 (QS7646) 12 (CZH15099Q) under optimum; 14 (CZH15142Q), 44 (QS7646) and 23 (CZH17192Q) under random stress; 9 (CZH142237Q) and 10 (CZH142238Q) under managed drought; and 14 (CZH15142Q) and 34 (CZH17203Q) under low N conditions. Among these, 10 (CZH142238Q) and 14 (CZH15142Q) were the most stable hybrids and can be recommended for release in sub-Saharan Africa to improve food and nutritional security of smallholder farmers who depend on maize. Kwekwe (KWE), Bindura (BIN), Chokwe (CHO) and Bako (BK2) were identified as the most discriminating and representative for optimum, random stress, managed drought and low N environments, respectively and help to identify superior hybrids.

Keywords:

• AMMI
• GGE biplot
• G × E interaction
• grain yield
• quality protein maize

REVIEWING EDITOR:

• Manuel Tejada Universidad de Sevilla; Spain

SUBJECTS:

• Agriculture & Environmental Sciences
• Plant & Animal Ecology
• Soil Sciences
• Food Additives & Ingredients

1. Introduction

Malnutrition is a persistent problem in Africa, especially in rural areas where maize is used as the main source of carbohydrates and protein (Tandzi et al., 2017; Vivek et al., 2008). Biofortified crops like quality protein maize (QPM) have significant nutritional benefits in combating malnutrition if it is produced and consumed in sufficient quantities (Krivanek et al., 2007; Tandzi et al., 2017). It can provide a more balanced protein for human and other monogastric animals. The quality and biological value of its protein are much higher than that of conventional maize. For example, the biological value of common maize protein is equal to about 40% of the biological value of milk protein, whereas the biological value of QPM protein is about 90% of that of milk protein (Nuss & Tanumihardjo, 2011). Thus, the nutritional benefits of QPM approach those of milk protein, a common standard of nutritional excellence.

Nowadays, some QPM varieties have comparable productivity potential compared with normal maize (Abakemal et al., 2016; Mebratu et al., 2019; Setimela et al., 2017; Wegary et al., 2014). It is also difficult to distinguish them from normal maize using morphological characteristics such as ear height, silk and flower characteristics and kernel texture. However, normal maize generally yields more, though in some situations the difference with QPM genotypes can be marginal (Tandzi et al., 2017). Determination of lysine and tryptophan content of the grain is necessary to confirm whether the genotype is QPM or not (Teklewold et al., 2015).

Maize production and productivity in sub-Saharan Africa (SSA) are affected by biotic and abiotic factors such as drought, low soil fertility, high pressure of weeds, insect pests, diseases, low input availability (especially quality seed and fertilizer), high cost of agricultural inputs and limited use of inputs (Meseka et al., 2013; Shiferaw et al., 2011). Among these factors, drought and low N levels in the soil are major challenges in smallholder farmers^’ fields (Ertiro et al., 2017; Mebratu et al., 2019; Wegary et al., 2014). Combined effects of these stress factors can be devastating (Badu-Apraku et al., 2016). Drought affects maize at all stages of development, but the flowering and grain-filling periods are the most sensitive with yield losses ranging from 10–50% (Fisher et al., 2015). Nitrogen deficiency is also common in farmers’ fields in the tropics where fertilizer application is far below the optimal rate, or not used at all (Meseka et al., 2013). Under such conditions, it is important to develop maize varieties that are tolerant to abiotic stress conditions which is a more reliable and affordable option for poor farmers in marginal areas (Beyene et al., 2015; Ertiro et al., 2017).

Evaluation of genotypes in multi-environments can help to identify well adapted and stable genotypes before commercial release (Abakemal et al., 2016; Makumbi et al., 2015; Rezende et al., 2020). Hybrids that are superior in one environment during a season/year may not be superior in other environments in the same season or year, due to the existence of G × E interaction (Makumbi et al., 2015; Sserumaga et al., 2016, 2018). This makes it difficult to allocate a variety that is successfully adapted to a certain location or across locations (Mebratu et al., 2019; Wolde et al., 2018). As a result, measuring G × E interaction is important in optimizing breeding strategies for selecting cultivars that are well adapted to specific environments (Abakemal et al., 2016; Matongera et al., 2023). So far, several studies have been conducted to measure G × E interaction and stability of both conventional (non-QPM) and QPM germplasms under stress and non-stress conditions. However, compared to the conventional maize germplasm, limited information has been generated on G × E interaction and stability of QPM germplasms under managed drought and low N conditions. In addition, some studies also indicated that the impact of drought and low N on agronomic and nutritional traits of QPM germplasms is high (Ngaboyison et al., 2012; Wegary et al., 2011). Thus, information on G × E interaction and stability of QPM genotypes across different management conditions would provide a scientific basis to develop future breeding strategies that target the development of QPM germplasm for release and marketing across different countries in SSA (Mebratu et al., 2019). Several statistical methods have been developed and deployed to measure G × E interaction and stability. Additive Main Effects and Multiplicative Interaction (AMMI) (Zobel et al., 1988) and genotype plus G × E interaction (GGE) biplot (Yan & Kang, 2003; Yan & Tinker, 2006) analysis are the most commonly and widely used statistical models. AMMI model separates the additive effects for genotype and environment from the multiplicative effects for G × E interaction by applying principal component analysis (PCA) (Pacheco et al., 2015). GGE biplot is a linear-bilinear model that removes the environmental main effect and considers the genotype plus the G × E interaction. This model is powerful in depicting “which won where” pattern and easily identify stable and ideal hybrids as well as environments with representativeness and discriminating ability of the environments (Matongera et al., 2023).

CIMMYT in collaboration with the National Agricultural Research Systems (NARS), have been screening a number of QPM and non-QPM genotypes under drought and low N environmental conditions for a number of years (Ertiro et al., 2017; Mebratu et al., 2019; Sserumaga et al., 2016, 2018; Wegary et al., 2013, 2014). As a result, several stress tolerant maize varieties have been released and disseminated for commercial production in partnership with various public and private organizations in SSA (Abate et al., 2017; Fisher et al., 2015; Teklewold et al., 2015). Compared to the conventional maize varieties, however, a very limited number of drought and low N tolerant QPM varieties have been released and commercialized. Therefore, the objectives of this study were to (1) identify new QPM hybrids that have good grain yield and agronomic performance under optimum, random stress, managed drought and low N conditions and (2) explore G × E interaction and grain yield stability of experimental QPM hybrids across different management conditions to promote release for commercialize in eastern and southern Africa (ESA).

2. Materials and methods

2.1. Genetic materials and testing sites

Forty newly developed QPM hybrids were obtained from CIMMYT-Zimbabwe and partner seed companies for this study. These hybrids were selected based on grain yield, disease resistance and other agronomic trait from early-stage trials for further regional trial evaluation under different environmental conditions. Five commercial checks (two QPM, two normal maize and one locally developed QPM or normal check), were included in the trials (Table 1). The aim of including the best local check at each location within CIMMYT trials was to compare the performance of the CIMMYT hybrids with local check. In combined analysis, the local check was excluded from the analysis because different local checks were used at different locations. The experimental materials were evaluated across 34 environments in 14 optimum, 9 random stress, 3 managed drought and 8 low N environments in five countries: Ethiopia, Zimbabwe, Zambia, Mozambique, and Malawi during 2018 to 2020. The descriptions of these test sites are given in Table 2.

Table 1. Genotype name, seed source and protein quality traits of 45 hybrids used in this study.

Entry	Genotype	Source	Hybrid type	Protein	Tryptophan	Entry	Genotype	Source	Hybrid type	Protein	Tryptophan
	name			%			name			%
1	ZS261	DR and SS	QPM check	8.3	0.085	24	CZH17193Q	CIMMYT	New QPM	10	0.093
2	Mama MQ623	DR and SS	QPM check	8.2	0.088	25	CZH17194Q	CIMMYT	New QPM	9.5	0.066
3	PAN53	Pannar	NM check	9.4	0.059	26	CZH17195Q	CIMMYT	New QPM	9.3	0.064
4	SC533	Seed Co	NM check	9.1	0.061	27	CZH17196Q	CIMMYT	New QPM	8	0.075
5	CZH04032	CIMMYT	New QPM	8.3	0.091	28	CZH17197Q	CIMMYT	New QPM	8.7	0.079
6	CZH132044Q	CIMMYT	New QPM	8.2	0.09	29	CZH17198Q	CIMMYT	New QPM	9	0.096
7	CZH132018Q	CIMMYT	New QPM	8.5	0.096	30	CZH17199Q	CIMMYT	New QPM	9.3	0.071
8	CZH142236Q	CIMMYT	New QPM	7.5	0.097	31	CZH17200Q	CIMMYT	New QPM	9	0.076
9	CZH142237Q	CIMMYT	New QPM	7.4	0.096	32	CZH17201Q	CIMMYT	New QPM	8.5	0.079
10	CZH142238Q	CIMMYT	New QPM	7.8	0.102	33	CZH17202Q	CIMMYT	New QPM	8.7	0.075
11	CZH15098Q	CIMMYT	New QPM	9.2	0.081	34	CZH17203Q	CIMMYT	New QPM	8.2	0.081
12	CZH15099Q	CIMMYT	New QPM	7.4	0.09	35	CZH17204Q	CIMMYT	New QPM	8.1	0.082
13	CZH15117Q	CIMMYT	New QPM	8.0	0.088	36	CZH17205Q	CIMMYT	New QPM	8.1	0.082
14	CZH15142Q	CIMMYT	New QPM	8.1	0.096	37	CZH17206Q	CIMMYT	New QPM	9	0.084
15	CZH16006Q	CIMMYT	New QPM	8.8	0.071	38	CZH17207Q	CIMMYT	New QPM	8	0.079
16	CZH16022Q	CIMMYT	New QPM	8.3	0.104	39	CZH17208Q	CIMMYT	New QPM	8.4	0.078
17	CZH16021Q	CIMMYT	New QPM	9.4	0.077	40	CZH17209Q	CIMMYT	New QPM	8.4	0.08
18	CZH17187Q	CIMMYT	New QPM	9.1	0.072	41	CZH17210Q	CIMMYT	New QPM	8.4	0.08
19	CZH17188Q	CIMMYT	New QPM	8.4	0.069	42	MH1633	Mukushi	New QPM	8.7	0.093
20	CZH17189Q	CIMMYT	New QPM	8.6	0.084	43	MH1634	Mukushi	New QPM	7.4	0.09
21	CZH17190Q	CIMMYT	New QPM	9.0	0.07	44	QS7646	Mukushi	New QPM	7.6	0.081
22	CZH17191Q	CIMMYT	New QPM	8.0	0.068	45	Local check	–	–	–	–
23	CZH17192Q	CIMMYT	New QPM	8.4	0.079

NM = normal maize; QPM = Quality protein maize.

Table 2. Description of the study sites, mean grain yield and CV of the trials evaluated during 2018–2020.

No	Country	Location	Location code	Management	Latitude	Longitude	Altitude (asl m)	Mean grain yield (t ha^−1)	CV
1	Ethiopia	Ambo season_1	AM1	Optimum	8°57’N	38°7’E	2225	8.9	7.64
2		Ambo season_2	AM2	Optimum	8°57’N	38°7’E	2225	7.6	8.84
3		Ambo season_1	AM1	Low N	8°57’N	38°7’E	2225	3.5	21.78
4		Ambo season_2	AM2	Low N	8°57’N	38°7’E	2225	2.6	11.09
5		Bako season_1	BK1	Optimum	9006’N	37°009’E	1650	8.0	9.35
6		Bako season_2	BK2	Optimum	9006’N	37°009’E	1650	8.9	9.3
7		Bako season_1	BK1	Low N	9006’N	37°009’E	1650	1.4	22.91
8		Bako season_2	BK2	Low N	9006’N	37°009’E	1650	2.7	17.76
9	Zimbabwe	Harare	HAR	Optimum	17°49’S	31°1’E	1489	7.9	8.35
10		Harare	HAR	Low N	17°49’S	31°1’E	1489	4.1	11.44
11		Gwebi	GWB	Low N	17°13’S	31°E	1406	2.1	15.61
12		Gwebi	GWB	Random stress	17°13’S	31°E	1406	5.2	10.71
13		Rattray-Arnold	RAT	Optimum	17°67’S	31°17’E	1452	8.1	6.91
14		Rattray-Arnold	RAT	Low N	17°67’S	31°17’E	1452	1.3	20.45
15		Bindura	BIN	Random stress	17°18’S	31°19’E	NA	5.2	19.63
16		Glandale	GLA	Optimum	17°08’S	31°3’E	1250	7.3	13.79
17		Chinhoyi	CHI	Optimum	17°21’S	30°11’E	1168	7.4	11.74
18		Lionsden	LIO	Optimum	19°19’S	30°58’E	1438	9.0	7.43
19		Kwekwe	KWE	Optimum	19°00’S	29°45’E	1187	10.1	8.3
20		AREX Harare	AREX	Low N	17°49’S	31°1’E	1489	1.5	21.12
21		Gweru	GWR	Random stress	19°28’S	29°48’E	1457	3.5	13.47
22		Makoholi	MAK	Managed drought	19°83’S	30°78’E	1111	2.2	25.35
23		Chisumbanje	CHS	Managed drought	20°48’S	32°14’E	455	3.0	22.49
24	Zambia	Mpongwe	MPO	Optimum	13°32’S	28°03’E	1300	8.3	7.69
25		Lusaka	LUS	Optimum	15°19’S	28°0’E	1153	8.1	7.83
26		Magobo	MAG	Random stress	14°3’S	32°7’E	1075	2.6	19.81
27		ZamSEED	ZAM	Random stress	15°19’S	28°18’E	NA	4.4	13.72
28		ETG-Farm	ETG	Random stress	15°42’S	28°26’E	1277	3.6	15.02
29		Chokwe	CHO	Managed drought	24°53’S	33° E	NA	2.9	13.88
30		Msekera	MSE	Optimum	13°38’S	32°39’E	1267	7.8	10.11
31	Mozambique	Lichinga	LIC	Random stress	13°30’S	35°23’E	1305	4.9	14.39
32		Umbuluzi	UMB	Random stress	26°58’S	32°38’E	23	5.4	13.1
33	Malawi	Meru	MER	Random stress	09°44’S	33°23’E	1231	4.2	21.35
34		Chitala	CHT	Optimum	13°13’S	34°07’E	733	8.5	11.74

NA = the data is not available; CV = Coefficient of variation.

2.2. Experimental design and trial management conditions

The trials were conducted using a 5 × 9 alpha lattice design with two replications. At all sites, each genotype was planted in two rows per plot of 5 m row length with spacing of 0.75 m between rows and 0.25 m between plants except at Bako in 2018 where both optimum and low N sites were planted at 4 m row length and 0.3 m spacing between plants and one row per plot. Due to shortage of land, Ambo (optimum site) in 2020 was planted at 5 m row length and 0.25 m spacing between plants and one row per plot. Two seeds were planted per hole and then thinned to one plant at three weeks after emergence, to obtain a final plant population density 53,333 plants per hectare. Fertilizer was applied based on recommendations for each location. For low N trials, all recommended rates of phosphorous, potassium and sulfur fertilizer were applied at planting, without further top dressing of urea. For optimum, random stress and managed drought trials, phosphorus, potassium and sulfur fertilizer sources were applied at planting, followed by a top dressing of urea fertilizer three weeks after planting. In Ethiopia, fertilizer was applied at a rate of 92 kg N and 69 kg diammonium phosphate (P₂O₅) ha⁻¹. Nitrogen was applied twice, at knee height and anthesis stage. All other agronomic practices such as weeding and pest management were done as needed. Optimum, random stress and low N trials were grown under rainfed conditions. However, random stress trials experienced random drought conditions where experimental sites normally received less than 600 mm rainfall per year (Setimela et al., 2017). Low N trials were conducted on N depleted sites. Nitrogen depletion was done by growing maize on it for more than 5 year and removing all crop stover from the field after harvest (Bänziger et al., 2006). Managed drought stress trials were also conducted during the dry/winter season with water applied entirely by irrigation. The trials were exposed to moisture deficiency by withholding irrigation supply 2 weeks before flowering up to 3 weeks post flowering so that drought stress would coincide with flowering and grain filling, which is the most vulnerable stages (Magorokosho et al., 2004). The aims of the managed drought and low N experiments were to had an average yield reduction of 25–40% compared to yields under optimal agronomic management conditions (Setimela et al., 2017). These environmental conditions simulate stresses that occurs in smallholder farmer’s fields where the grain yields are between 1 and 2 t ha⁻¹ (Magorokosho et al., 2010; Setimela et al., 2017). At harvesting, the first and the last plant of each row of low N and managed drought stress trials were discarded to minimize border effects.

2.3. Data collection

Grain yield and other agronomic data were collected from all sites for all management conditions. Grain yield was recorded in kilograms at harvesting by weighing the total ears per plot. This was adjusted to 12.5% moisture content to estimate grain yield expressed in ton per ha. The number of days to anthesis was recorded from planting date to when 50% of the plants in a plot started pollen shedding. The number of days to silking were recorded from planting date to when 50% of the plants in the plot produced 2–3 cm long silks. Anthesis-silking interval (ASI) was calculated as the difference between the number of days to silking and days to anthesis. The mean plant and ear height was measured in centimetre from five random plants in a plot, plant height was measured from the ground level to the first tassel branch, and ear height was measured from the ground to the top useful ear insertion base of the node. Ear position was calculated as a ratio of the ear height to plant height. At harvesting, the total number of harvested plants and ears were counted in all plots at each site. Number of ears per plant was calculated as the ratio of total ears in a plot to total number of plants harvested in a plot.

2.4. Data analysis

Individual and combined analysis of variance (ANOVA) was done for the collected data using PROC MIXED procedure of SAS® software (SAS, 2002). The hybrids were considered as a fixed factor while environments, replications within environments and the incomplete blocks within replications were considered random factors. Levene’s test was used to test the homogeneity of variances before analyzing the combined data (Gastwirth et al., 2009). In the combined environment analysis of ANOVA, genotype effects were tested for significance using the corresponding interaction with the environment as the error term, whereas the genotype × environment interaction was tested against the pooled error. Mean comparison among hybrid means was done by the least significant difference (LSD) test at 5% level of significance (Gomez & Gomez, 1984). After confirming the presence of significant G × E interaction in combined analysis for grain yield, further partitioning of the G × E interaction and stability analyses was conducted using AMMI and GGE biplot models. Data were combined across optimum, random stress, managed drought, low N stress and across all environments.

The combined ANOVA linear mixed model was followed: $$Y_{\text{ijk}}=\mu +{\text{G}}_{\text{i}}+{\text{E}}_{\text{j}}+{\text{R}}_{\text{k}}\left({\text{E}}_{\text{j}}\right)+{\text{B}}_{\text{o}}\left[{\text{R}}_{\text{k}}\left({\text{E}}_{\text{j}}\right)\right]+{\text{GE}}_{\text{ij}}+{\epsilon }_{\text{ijko}}$$

Where, Y_ijk is the observed mean of the i^th genotype (G_i) in the environment (E_j), in the k^th block (B_k), µ is the overall mean, G_i is the effect of i^th genotype, E_j is the effect of j^th environment, R_k is replication effect of the k^th replication in the j^th environment, B_o [R_k(E_j)] is the random effect of incomplete block B_o within replicate k^th and environment j^th, GE_ij is the interaction effect of the i^th genotype as well as j^th environment and Ɛ_ijko is the error term.

Broad sense heritability (H²) for a given trait across environments was estimated using the variance components (Alvarado et al., 2020). H² = $\frac{\sigma 2g}{\sigma 2g+\left(\frac{\sigma 2ge}{E}\right)+\left(\frac{\sigma 2e}{\text{Er}}\right)}$ where σ²g is genotype variance, σ²ge is genotype by environment interaction variance, and σ²e is error variance for r replicates and E environments of the genotypes at each testing site under optimum, random stress, manged drought and low N conditions.

The AMMI analysis was performed by using the raw data of a lattice design with GEA-R (Genotype by Environment Analysis with R) version 4.1 software (Pacheco et al., 2015). AMMI partitions the sum of squares into interaction principal component (IPC) axes.

The basic model is: $$\mathit{\text{yij}}=\text{μ}+g_i+e_{\mathit{\text{in}}}+{{\displaystyle \sum }}_{n=1}^N\left({\tau }_n{\gamma }_{\mathit{\text{in}}}{\delta }_{\mathit{\text{in}}}\right)+e_{\mathit{\text{ij}}}$$

Where yij is the yield of the i^th genotype (i = 1,…,I) in the j^th environment (j = 1,…,J); µ is the grand mean; g_i and e_j are the genotype and environment deviations from the grand mean, respectively; τ_n is the eigenvalue of the PC analysis axis n; γ_in and δ_jn are the genotype and environment principal components scores for axis n; N is the number of principal components retained in the model and "ij is the error term.

GGE biplot analysis was used for visualising patterns and interactions without environmental effects (Matongera et al., 2023). The GGE biplot graph was generated using singular value decomposition (SVD) of the first two principal component axis derived from environment-centered data (axis1 and axis2) using the Site Regression (SREG) model (Mekonnen et al., 2022; Yan & Kang, 2003). The first principal component (PC1) represented the variation due to genotype while the second principal component (PC1) represents the variation in the yield due to the G × E interaction (Yan, 2011).

According to Purchase et al., (2000), AMMI stability values (ASV) for grain yield stability analysis were calculated.

ASV = $\sqrt{{\left[\frac{\mathit{\text{IPCA}}1\mathit{\text{SS}}}{\mathit{\text{IPCA}}2\mathit{\text{SS}}}(\mathit{\text{IPCA}}1\mathit{\text{score}})\right]}^2+{(\mathit{\text{IPCA}}2\mathit{\text{score}})}^2}$ ASV = AMMI stability value; SS = sum of squares; IPCA1 and IPCA2= the first and the second interaction principal component axes, respectively

3. Results

3.1. Analysis of variance and hybrid performance

Combined analysis for each management condition showed significant difference among genotypes (G), environments (E), and G × E interaction for grain yield and almost all recorded agronomic traits (Table 3). High grain yield heritability values were observed for the combined environments (0.81), followed by random stress (0.69), optimum management (0.64), and low N stress conditions (0.61) while the lowest heritability was recorded under managed drought stress (0.08) (Table 3). All other traits showed high heritability values in combined environments analysis. Mean grain yield ranged from 7.3 t ha⁻¹ to 8.9 t ha⁻¹ with a mean of 8.3 t ha⁻¹ across optimum environments, 3.2 to 5.3 t ha⁻¹, with an overall mean of 4.3 t ha⁻¹ under random drought stress, 1.8 t ha⁻¹ to 3.6 t ha⁻¹ with a mean of 2.7 t ha⁻¹ under managed drought conditions, and 1.9 to 3.2 t ha⁻¹ with an average of 2.4 t ha⁻¹ under low N stress conditions (Table 4). Some QPM hybrids recorded better grain yield and agronomic performance than other QPM and normal maize checks under optimum, drought and low N stress conditions. The mean of top ten QPM hybrids under optimum, random stress, managed drought and low N stress conditions showed 15.8 and 6.4%, 32.0 and 13.3%, 20.1 and 27.6%, and 11.4 and 13.2% yield advantage compared to the mean of other QPM and normal maize checks, respectively. The first highest yielding QPM hybrid under each management condition also showed 13.7%, 30.3%, 29.3% and 15.4% grain yield advantage over the best QPM check under optimum, random drought stress, managed drought and low N stress conditions, respectively. Entries, 44 and 12 under optimum, 14, 44 and 23 under random drought, 9 and 10 under managed drought, and 14 and 34 were the first two top yielding hybrids.

Table 3. Mean squares for grain yield and agronomic traits for 40 QPM hybrids and four commercial checks evaluated under different environmental conditions in 2018–2020.

Source variation	OP		RS		MD		LN
	Df	GY	Df	GY	Df	GY	Df	GY
E	13	48.88**	8	80.94**	2	17.78**	7	86.76**
Rep (E)	14	5.07**	9	4.73**	3	6.59**	8	3.99**
Block (Rep × E)	224	1.89**	144	1.79**	48	1.32**	128	0.91**
Genotype	43	3.84**	43	3.06**	43	0.90**	43	1.18**
G × E	559	1.45**	344	0.99**	86	0.74**	301	0.53**
Residuals	378	0.74	243	0.68	81	0.3	216	0.22
Broad-sense heritability		0.64		0.69		0.08		0.61
CV (%)		10.37		19.02		20.58		19.29
R-Square		0.88		0.9		0.9		0.95
Source variation	Across all environments
	Df	GY	DA	ASI	PH	EPO	EPP
E	33	647.02**	13990.74**	189.66**	145555.55**	0.24**	2.36**
Rep (E)	34	4.86**	29.48**	5.36**	2616.11**	0.01**	0.06**
Block (Rep × E)	544	1.58**	9.07**	3.10**	601.00**	0.00**	0.04**
Genotype	43	5.42**	66.93**	9.91**	2554.27**	0.01**	0.27**
G × E	1419	1.08**	4.05**	2.37**	257.99*	0.00**	0.03**
Residuals	918	0.56	2.66	1.34	225.26	0	0.02
Broad-sense heritability		0.81	0.95	0.74	0.91	0.87	0.9
CV (%)		13.97	2.27	55.99	6.83	9.28	14.13
R-Square		0.98	0.99	0.91	0.96	0.88	0.89

*P ≤ 0.05, **P ≤ 0.01; E = environment; Rep = replication; G × E = genotype × environment interaction; CV = Coefficient of variation; Df = degree of freedom; GY = grain yield; DA = days of anthesis; ASI = anthesis silking interval; PH = plant height; EPO = ear position; EPP = number of ears per plant; OP = Optimum management; RS = random stress; MD = managed drought; LN = low nitrogen.

Table 4. Grain yield and some agronomic traits of top ten QPM hybrids and commercial checks evaluated under different environmental conditions in ESA during 2018 to 2020.

Optimum			Random stress			Managed drought			Low N			Across all environments
Entry	Genotype name	GY	Entry	Genotype name	GY	Entry	Genotype name	GY	Entry	Genotype name	GY	Entry	GY	DA	ASI	PH	EPO	EPP
44	QS7646	8.9	14	CZH15142Q	5.3	9	CZH142237Q	3.6	14	CZH15142Q	3.2	14	6.0	70.6	1.6	218.6	0.47	1.11
12	CZH15099Q	8.9	44	QS7646	4.9	10	CZH142238Q	3.5	34	CZH17203Q	3.0	9	5.8	72.0	1.3	213.6	0.49	1.13
9	CZH142237Q	8.8	23	CZH17192Q	4.9	23	CZH17192Q	3.3	11	CZH15098Q	3.0	44	5.8	70.4	2.9	218.8	0.49	1.14
17	CZH16021Q	8.8	10	CZH142238Q	4.8	40	CZH17209Q	3.3	7	CZH132018Q	2.9	10	5.8	70.9	1.6	211.7	0.47	1.13
6	CZH132044Q	8.7	9	CZH142237Q	4.8	14	CZH15142Q	3.2	27	CZH17196Q	2.8	12	5.7	70.4	1.6	212.5	0.51	1.13
38	CZH17207Q	8.7	34	CZH17203Q	4.7	12	CZH15099Q	3.2	43	MH1634	2.7	34	5.7	72.5	2.3	221.8	0.52	1.07
10	CZH142238Q	8.7	8	CZH142236Q	4.7	35	CZH17204Q	3.1	12	CZH15099Q	2.6	23	5.7	72.5	2.3	231.4	0.50	1.00
20	CZH17189Q	8.7	38	CZH17207Q	4.7	22	CZH17191Q	3.1	9	CZH142237Q	2.6	11	5.6	69.8	1.6	213.1	0.49	1.14
14	CZH15142Q	8.7	18	CZH17187Q	4.6	8	CZH142236Q	3.1	29	CZH17198Q	2.6	17	5.6	72.4	2.4	226.1	0.49	0.96
11	CZH15098Q	8.6	16	CZH16022Q	4.6	7	CZH132018Q	3.0	10	CZH142238Q	2.6	6	5.6	72.9	1.7	226.8	0.49	1.07
ZS261 (QPM check)		7.3			3.2			2.6			2.2	1	4.6	70.3	2.1	214.1	0.47	0.98
Mama MQ623 (QPM check)		7.8			4.1			2.8			2.8	2	5.2	70.2	2.1	208.3	0.50	1.02
PAN53 (normal check)		8.8			4.9			3.0			2.8	3	5.9	71.9	2.0	228.5	0.49	0.92
SC533 (Normal check)		7.6			3.5			2.1			2.1	4	4.7	70.2	1.7	212.9	0.47	0.93
LSD		0.5			0.5			0.6			0.3		0.3	0.5	0.4	5.1	0.02	0.05
Mean		8.3			4.3			2.7			2.4		5.4	71.8	2.1	219.7	0.49	1.05
Minimum		7.3			3.2			1.8			1.9		4.6	69.8	1.3	207.2	0.47	0.92
Maximum		8.9			5.3			3.6			3.2		6.0	74.5	3.0	232.1	0.53	1.14
Mean of top 10 hybrids		8.7			4.8			3.3			2.8		5.7	71.44	1.93	219.45	0.49	1.09
Mean of QPM checks		7.5			3.6			2.7			2.5		4.9	70.26	2.10	211.21	0.48	1.00
Mean of normal checks		8.2			4.2			2.5			2.5		5.3	71.05	1.89	220.71	0.48	0.93
Best QPM hybrid		8.9			5.3			3.6			3.2		6.0	69.79	1.26	211.71	0.47	1.14
Best QPM check		7.8			4.1			2.8			2.8		5.2	70.2	2.1	208.3	0.5	1.0
Best normal check		8.8			4.9			3.0			2.8		5.9	70.2	1.7	212.9	0.5	0.9
Top 10 hybrids over QPM checks (%)		15.8			32.0			20.1			11.4		17.0	1.7	−7.9	3.9	1.2	8.7
Top 10 hybrids over normal maize checks (%)		6.4			13.3			27.6			13.2		8.0	0.5	2.1	−0.6	2.4	17.3
Best hybrid over best QPM check (%)		13.7			30.3			29.3			15.4		15.7	−0.6	−38.6	1.6	−0.8	12.3
Best hybrid over best normal maize check (%)		0.4			7.8			19.2			13.0		2.2	−0.6	−27.1	−0.5	−1.2	22.9

GY = grain yield (t ha⁻¹); DA = days to anthesis; ASI = anthesis silking interval; PH = plant height (cm); EPO = ear position; EPP = number of ears per plant; LSD = least significant difference.

In combined environments, grain yield ranged from 4.6 t ha⁻¹ for commercial QPM check (ZS261) to 6.0 t ha⁻¹ for entry 14 (CZH15142Q) with an overall mean of 5.4 t ha¹ (Table 4). The mean grain yield of the top ten QPM hybrids was 5.7 t ha⁻¹ and showed 17.0% and 8.0% yield advantage over the mean of other QPM and normal maize checks, respectively. Among the top ten hybrids, the top yielding hybrid, entry 14, showed a 15.7% yield advantage over the best QPM check and comparable yield performance with the normal maize check. Moreover, this entry showed a 38.6 and 27.1% lower ASI and 12.3 and 22.9% higher number of ears per plant, compared to the best QPM and normal maize checks, respectively, and recorded comparable anthesis and ear position with the mean of the best QPM and normal maize commercial checks. All top yielding hybrids showed mid ear placement, which can prevent lodging.

Compared to optimum environments, mean of grain yield was reduced by 48% under random drought stress, 68% under managed drought and 71% under low N conditions (Table 5). Days to anthesis and silking were also decreased by 7 and 6% under random stress, and 11 and 8% under managed drought stress conditions, and increased by 13% and 16% under low N conditions, respectively. ASI was increased by 80%, 218% and 203% under random stress, managed drought and low N compared to optimum conditions, respectively. Plant and ear height were reduced by 20% and 25% under random stress, 30% and 29% under managed drought and 23% and 27% under low N environments, respectively. Ear position was also reduced by 7% under random stress and 2% under managed drought and 6% under low N conditions. Number of ears per plant was reduced by 11% under random drought stress, 24% under managed drought and 20% under low N environments.

Table 5. Percentage of grain yield and agronomic performance reduction under random stress, managed drought and low N stress conditions compared to the performance under optimum management condition.

Traits	OP	RS	% change	MD	% change	LN	% change
Grain yield (t ha^-1)	8.3	4.3	48	2.7	68	2.40	71
Days to anthesis (days)	71.5	66.5	7	63.5	11	81.08	−13
Days to silking (days)	72.6	68.5	6	67.0	8	84.41	−16
Anthesis silking interval (days)	1.1	2.0	−80	3.5	−218	3.33	−203
Plant height (cm)	253.3	202.3	20	177.3	30	196.16	23
Ear height (cm)	129.3	97.3	25	91.3	29	94.14	27
Ear position (cm)	0.51	0.47	7	0.50	2	0.48	6
Number of ears per plant (#)	1.16	1.03	11	0.88	24	0.92	20

OP = Optimum management; RS = random stress; MD = managed drought; LN = low nitrogen; negative value of traits indicates an increase under random stress, managed drought and low N stress conditions.

3.2. AMMI analysis

Mean squares from the AMMI analysis showed highly significant difference among genotypes (G), environments (E) and G × E interaction for grain yield under optimum, random stress, managed drought, and low N conditions (Table 6). The proportion of total variations explained by G, E and G × E were 10, 36 and 54% under optimum, 8, 70 and 22% under low N conditions, 11, 54 and 35% under random stress and 28, 20 and 52% under managed stress, respectively. The G × E interaction source of variation was further partitioned into interaction principal components (IPCA). Under optimum conditions, G × E interaction was partitioned into eight significant IPCAs, which explained 86.1% of the total G × E interaction variation. Under random stress, G × E interaction was partitioned into four significant IPCAs. These IPCAs explained a total of 76.9% of the G × E interaction variation. Under managed drought, G × E was partitioned into two highly significant IPCAs, which explained 100% of the variation. Under low N conditions, G × E interaction was partitioned into six significant IPCAs, which explained 97.1% of the variation. The first two IPCAs in each management type explained 38.5%, 60.3%, 51.9% and 100% of variation under optimum, low N, random stress and managed drought, respectively.

Table 6. AMMI analysis of variance for grain yield of 40 QPM hybrids and four commercial checks evaluated under optimum and stressed environments in ESA during 2018 to 2020.

Optimum management							Low nitrogen stress
Source of variation	Df	SS	MS	Total variation	% G × E interaction	% total G × E interaction	Source of variation	Df	SS	MS	Total variation	% G × E interaction	% total G × E interaction
Environment	13	635.39	48.88**	35.8			Environment	7	607.31	86.76**	69.8
Genotype	43	183.02	4.26**	10.3			Genotype	43	66.64	1.55**	7.7
E × G	559	956.35	1.71**	53.9			E × G	301	196.36	0.65**	22.5
PC1	55	214.26	3.90**		22.5	22.5	PC1	49	70.07	1.43**		34.6	34.6
PC2	53	152.77	2.88**		16	38.5	PC2	47	51.92	1.10**		25.7	60.3
PC3	51	123.81	2.43**		13	51.5	PC3	45	27.69	0.62**		13.7	74
PC4	49	92.57	1.89**		9.7	61.2	PC4	43	20.29	0.47**		10	84
PC5	47	87.8	1.87**		9.2	70.4	PC5	41	13.69	0.33*		6.8	90.8
PC6	45	53.64	1.19**		5.6	76.1	PC6	39	12.78	0.33*		6.3	97.1
PC7	43	50.8	1.18*		5.3	81.4	Residuals	352	141.93	0.4
PC8	41	44.88	1.09*		4.7	86.1
Residuals	616	611.85	0.99
Random stress							Managed drought stress
Source of variation	Df	SS	MS	Total variation	% G × E interaction	% total G × E interaction	Source of variation	Df	SS	MS	Total variation	% G × E interaction	% total G × E interaction
Environment	8	647.54	80.94**	54			Environment	2	35.56	17.78**	20.4
Genotype	43	132.73	3.09**	11.1			Genotype	43	48.1	1.12**	27.6
E × G	344	418.89	1.22**	34.9			E × G	86	90.73	1.06**	52
PC1	50	109.47	2.19**		27.5	27.5	PC1	44	51.3	1.17**		59.2	59.2
PC2	48	96.87	2.02**		24.4	51.9	PC2	42	35.36	0.84**		40.8	100
PC3	46	53.67	1.17**		13.5	65.4	Residuals	132	71.37	0.54
PC4	44	45.74	1.04*		11.5	76.9
Residuals	396	384.24	0.97

*P ≤ 0.05; ** P ≤ 0.01; Df = degree of freedom; MS = mean squares; SS = sum of squares; G × E = genotype by environment interaction; PC = principal component.

Based on IPCA1 scores, hybrids such as 6, 10, 14, 20 and 24 under optimum, 3, 8, 9, 23 and 44 under random stress, 7, 19, 35 and 40 under managed drought and 3, 11 and 14 under low N had relatively close to zero IPCA1 scores and high mean grain yield. This suggested that these hybrids were relatively stable and high yielding in the respective environmental conditions (Table 7). On the other hand, hybrids 3, 12,23, 44, 11, 15, 17, 37 and 41 under optimum, 10, 14, 18, 34 and 38 under random stress, 8, 10, 15, 23, 3 (check), 9, 14 and 22 under managed drought, 7, 10, 27 and 34 under low N had high positive and negative IPCA1 values. This indicated that these hybrids were therefore unstable and adapted to specific environments under the respective environment conditions. Hybrids such as 1 (check) and 31 under optimum and 23, 30, 32, 33 and 36 under low N had low mean grain yield and close to zero IPCA1 values, indicating consistent, but poor yield performance across optimum and low N environments, respectively.

Table 7. Mean grain yield and IPCA scores and ASV values of 40 QPM hybrids and four commercial checks evaluated under stress and non-stress environments.

Entry no	Hybrids	Optimum				Random stress				Managed drought				Low N stress
		GY	IPCA1	IPCA2	ASV	GY	IPCA1	IPCA2	ASV	GY	IPCA1	IPCA2	ASV	GY	IPCA1	IPCA2	ASV
1	ZS261	7.3	−0.01	−0.02	0.02	3.2	−0.24	−0.32	0.42	2.6	−0.16	0.41	0.47	2.2	−0.28	0.32	0.50
2	Mama MQ623	7.8	−0.31	−0.38	0.57	4.1	0.30	−0.25	0.42	2.8	0.42	0.16	0.63	2.8	−0.42	0.22	0.61
3	PAN53	8.8	0.50	0.36	0.79	4.9	−0.12	−0.28	0.31	3.0	−0.61	0.12	0.90	2.8	0.08	−1.00	1.01
4	SC533	7.6	0.25	−0.31	0.47	3.5	0.37	−0.29	0.50	2.1	−0.14	−0.39	0.44	2.1	0.34	−0.51	0.69
5	CZH04032	7.8	0.18	−1.00	1.03	3.7	0.22	0.03	0.25	2.6	0.36	0.22	0.57	2.4	0.17	0.78	0.81
6	CZH132044Q	8.7	−0.01	0.19	0.19	4.5	0.05	−0.39	0.40	2.3	−0.15	−0.06	0.23	2.6	0.39	−0.06	0.52
7	CZH132018Q	8.2	−0.20	0.12	0.30	4.1	−0.50	−0.26	0.62	3.0	0.09	−0.17	0.21	2.9	−0.84	0.41	1.21
8	CZH142236Q	8.4	0.19	0.19	0.33	4.7	0.15	0.09	0.19	3.1	0.81	0.48	1.27	2.2	0.54	−0.12	0.74
9	CZH142237Q	8.8	−0.22	−0.09	0.32	4.8	0.16	0.79	0.81	3.6	−0.81	−0.12	1.18	2.6	0.30	−0.14	0.42
10	CZH142238Q	8.7	−0.19	0.84	0.88	4.8	0.62	−0.01	0.70	3.5	0.69	0.13	1.01	2.6	0.63	−0.18	0.87
11	CZH15098Q	8.6	−0.74	0.15	1.05	4.4	−0.50	−0.66	0.86	2.6	0.08	0.31	0.33	3.0	−0.29	0.29	0.49
12	CZH15099Q	8.9	0.43	0.41	0.73	4.5	−0.01	0.18	0.18	3.2	−0.27	−0.17	0.43	2.6	0.32	0.02	0.43
13	CZH15117Q	8.5	0.46	−0.08	0.65	4.4	−0.60	−0.61	0.91	3.0	−0.30	0.41	0.60	2.6	−0.26	0.52	0.63
14	CZH15142Q	8.7	0.11	0.13	0.20	5.3	0.77	−0.20	0.89	3.2	−0.49	−0.14	0.72	3.2	−0.29	0.51	0.64
15	CZH16006Q	8.5	−0.68	0.38	1.02	4.2	−0.27	0.15	0.34	2.8	0.77	−1.00	1.49	2.3	−0.30	0.42	0.59
16	CZH16022Q	8.4	0.52	−0.47	0.87	4.6	0.23	−0.32	0.41	2.3	−0.01	−0.07	0.07	2.6	0.32	−0.17	0.47
17	CZH16021Q	8.8	−0.41	−0.56	0.80	4.5	−0.23	0.43	0.50	2.5	0.18	0.27	0.37	2.5	−0.38	0.06	0.52
18	CZH17187Q	8.1	0.13	0.22	0.28	4.6	0.65	0.20	0.76	2.3	−0.21	0.02	0.30	2.1	0.86	0.12	1.17
19	CZH17188Q	8.1	−0.13	−0.50	0.54	4.6	−0.17	0.90	0.92	3.0	−0.01	0.68	0.68	2.4	0.43	0.02	0.59
20	CZH17189Q	8.7	0.01	−0.09	0.09	4.0	−0.66	0.29	0.81	2.7	0.29	0.35	0.54	2.4	−0.22	0.24	0.38
21	CZH17190Q	7.9	0.25	−0.33	0.48	3.8	−0.48	0.60	0.81	2.5	−0.37	0.37	0.65	2.1	0.34	0.57	0.73
22	CZH17191Q	8.4	0.37	0.23	0.56	4.3	−0.10	0.42	0.43	3.1	−0.63	−0.04	0.92	2.3	0.28	0.08	0.39
23	CZH17192Q	8.6	0.40	−0.19	0.59	4.9	−0.06	−0.05	0.09	3.3	0.67	−0.54	1.11	2.2	−0.03	0.47	0.47
24	CZH17193Q	8.6	−0.07	0.09	0.13	4.4	0.68	0.10	0.77	2.2	0.18	−0.10	0.28	2.0	0.46	0.15	0.63
25	CZH17194Q	7.9	−0.20	0.50	0.57	3.9	−0.51	0.34	0.67	1.8	0.28	−0.42	0.58	2.5	−0.35	−0.38	0.60
26	CZH17195Q	7.7	0.29	0.03	0.41	4.1	0.04	−0.03	0.05	1.9	−0.05	−0.11	0.13	2.1	0.16	−0.59	0.63
27	CZH17196Q	8.2	−0.43	0.09	0.61	3.8	−0.38	−0.18	0.46	1.9	−0.31	−0.66	0.80	2.8	−0.58	−0.38	0.88
28	CZH17197Q	8.3	0.11	−0.03	0.16	4.2	0.05	−0.59	0.60	2.8	0.38	−0.35	0.65	1.9	0.31	0.15	0.44
29	CZH17198Q	8.4	−0.26	0.01	0.37	4.3	0.09	−0.48	0.49	2.7	−0.06	−0.36	0.37	2.6	−0.87	−0.72	1.37
30	CZH17199Q	8.1	0.26	−0.15	0.40	4.2	0.10	−0.25	0.28	2.5	0.18	0.33	0.42	2.0	−0.12	0.11	0.19
31	CZH17200Q	8.0	0.01	−0.74	0.74	4.1	−0.02	0.09	0.10	2.3	−0.54	−0.65	1.02	2.1	0.27	−0.01	0.37
32	CZH17201Q	7.8	0.36	0.32	0.60	4.2	−0.83	−0.19	0.95	2.3	0.21	−0.46	0.55	2.4	0.04	−0.25	0.26
33	CZH17202Q	8.3	−0.53	0.03	0.74	3.8	−0.49	0.29	0.62	2.5	0.06	0.12	0.15	2.2	0.02	−0.53	0.53
34	CZH17203Q	8.5	0.75	0.17	1.07	4.7	0.79	−0.17	0.91	2.3	−0.27	0.02	0.39	3.0	−0.36	−0.44	0.66
35	CZH17204Q	8.1	−0.10	−0.02	0.14	4.2	0.07	1.00	1.00	3.1	0.14	0.38	0.44	2.5	−0.24	0.15	0.35
36	CZH17205Q	8.1	−0.37	−0.25	0.57	4.5	0.43	−0.54	0.73	2.4	−0.46	−0.06	0.67	1.9	0.09	0.56	0.57
37	CZH17206Q	8.5	−0.75	−0.18	1.06	4.5	−0.08	0.17	0.19	2.4	−0.02	0.37	0.37	2.2	−0.21	0.13	0.32
38	CZH17207Q	8.7	−0.21	0.30	0.42	4.7	0.57	0.60	0.88	2.8	−0.47	0.02	0.68	2.2	0.59	−0.09	0.81
39	CZH17208Q	7.8	0.45	0.33	0.71	4.6	0.61	0.26	0.74	2.4	0.32	0.16	0.50	2.0	0.13	−0.18	0.25
40	CZH17209Q	7.8	−0.28	0.19	0.43	4.1	0.09	0.05	0.12	3.3	0.08	0.09	0.15	2.4	0.17	−0.40	0.46
41	CZH17210Q	8.4	−0.73	0.11	1.03	4.0	−0.24	0.00	0.28	2.7	−0.21	−0.19	0.37	2.0	0.35	0.04	0.47
42	MH1633	7.9	0.32	0.16	0.48	4.3	0.23	−0.53	0.59	2.7	0.35	0.22	0.56	2.4	−0.17	0.00	0.23
43	MH1634	8.3	0.13	−0.40	0.44	4.0	−0.60	−0.05	0.68	2.6	0.28	0.06	0.41	2.7	−0.52	0.26	0.75
44	QS7646	8.9	0.32	0.24	0.51	4.9	−0.19	−0.32	0.39	2.8	−0.26	0.34	0.50	2.5	−0.82	−0.44	1.19
	mean	8.3				4.3				2.7				2.4

GY (t ha⁻¹) = mean grain yield ton per hectare; ASV = AMMI stability value; IPCA = interaction of principal component analyses.

AMMI stability values showed variation in grain yield stability among the 44 hybrids under different management conditions (Table 7). Hybrids with ASV values closer to zero are considered more stable than hybrids with higher positive or negative ASV values (Purchase et al., 2000). Across optimum conditions, hybrids 1 (ZS261) check, 6, 20, 24, 35, 14 and 28 had the lowest ASV values and are considered as more stable. Among these hybrids, 6, 14, 20 and 24 showed above overall trial mean grain yield while hybrids 1 and 35 showed the lowest mean yield. Across random stress trials, hybrids 37, 12, 8 and 23 had relatively lower ASV values and high mean grain yield and these hybrids were considered more stable and widely adaptable across random stress environments. On the other hand, hybrids 3 (check), 9 and 44 were top yielding, but with the high ASV values. These hybrids were considered unstable but adapted to specific environments under random stress conditions.

Hybrids 16, 26, 40, 33, 7, 6 and 24 showed the lowest ASV values compared to the remaining hybrids, indicating these hybrids were relatively more stable under managed drought environments. Hybrid 40 was also both stable and a widely adaptable hybrid under managed drought environments. Hybrids 9 and 10 exhibited the highest ASV values, but were top yielding which indicates that these hybrids were unstable, but could adapt to specific environments under managed drought environments. Across low N environments, hybrids 30, 42, 39, 32 and 37 showed relatively low ASV values and mean yield below the average mean of the trial whereas hybrids 3, 14, 11 and 34 had relatively high ASV values and showed high mean grain yield. This indicates that these hybrids were adapted to specific testing environments under low N stress conditions.

3.3. GGE biplot analysis

3.3.1. Which won where pattern

GGE biplot analysis of grain yield for 40 QPM hybrids and four checks are presented in Figure 1a–d. The first two principal axis explained 39.74% (PC1 = 20.98% and PC2 = 18.76%) under optimum, 49.41% (PC1 = 30.85 and 18.56%) under random stress, 74.86% (PC1 = 41.95 and PC2 = 32.91%) under managed drought and 59.15% (PC1 = 39.11 and PC2 = 20.15%) under low N of the total G × E interaction variations. The which-won-where polygon view of GGE biplots in Figure 1a–d showed which genotypes performed well at which site. The line that starts from the origin of the biplot and passes perpendicular to the side of the polygon was divided into different sectors. Under optimum conditions, ten sectors; under random stress, seven sectors; under managed drought, six sectors and under low N conditions six sectors were formed. Hybrids that formed the vertex of the polygon were the highest yielding in the corresponding environments within each sector. On the other hand, hybrids located far away from all the test environments were considered as poor yielding at each testing site. The vertex hybrids 15, 41 and 12 were the highest yielding at AM2, BK1 and CHI sites, respectively, whereas hybrids 1, 4, 5 and 34 were low yielding under optimum conditions (Figure 1a).

Figure 1. “Which won where” polygon view of the GGE biplot for 40 QPM hybrids and four standard checks evaluated across (a) optimum management, (b) random stress, (c) managed drought and (d) low nitrogen stress. Environment codes are explained in Table 2.

Hybrids 14 and 19 were on the corners of the polygon and the highest yielding at BIN and MER, and UMB testing sites, respectively under random stress conditions (Figure 1b). Similarly, hybrids 9 and 10 were top yielding at CHI, and MAK and CHO respectively, while hybrids 31, 27, 25 and 15 were on the lower yielding vertex of the polygon under managed drought conditions (Figure 1c). Under low N conditions, hybrids 3, 7 and 34 were top yielding at the corresponding sites of HAR, AM1 and BK2, respectively, whereas hybrids 18 and 36 were on the low yielding vertex of the polygon (Figure 1d). Hybrids like 23, 7, 18, 43, 35 and 25 under optimum; 12, 31 and 24 under random stress; 33, 37, 29 and 11 under managed drought; 5, 40 and 35 under low N stress were relatively closer to the origin of the biplot and considered as more stable than hybrids far from the origin, in either positive or negative direction.

3.3.2. Mean vs. stability under different management conditions

Average environment coordination (AEC) view of the GGE biplot was drawn to see the stability and adaptability of hybrids evaluated under optimum, random stress, managed drought, and low N conditions (Figure 2a–d). The axis of the AEC ordinate is a line that passes through the origin of biplot and is perpendicular to the AEC abscissa to separate genotypes that had below average grain yield from those showing higher grain yields. The hybrids were ranked along the average environment axis with the arrow pointing to a higher value based on mean performance across all testing environments. The top yielding hybrids according to their projections onto the AEC, in descending order were 10, 12, 3, 15, 38 and 6 under optimum (Figure 2a); 14, 34, 38, 10, 39, 23, 9 and 3 under random stress (Figure 2b); 9, 10, 19, 40, 8 and 35 under managed drought (Figure 2c); and 34, 14, 3, 7, 27 and 1 under low N (Figure 2d). Among these top yielding hybrids, 3, 12 and 15 under optimum; 14 and 9 under random drought stress; 9, 8 and 10 under managed drought and 3 and 7 under low N conditions had large projections onto the AEC abscissa. Hence, these hybrids were unstable and might be adapted to specific environments. Hybrids like 10, 14, 44, 6, 9, 38 and 20 under optimum; 38, 23, 10, 3 (check) and 39 under random stress; 19, 40, 35, 13 and 12 under managed drought; 34, 14, 1 (check), 23 and 35 under low N environments had above average grain yield, located on the AEC abscissa with low projections. Hence, these hybrids were considered as stable and might have wide adaptability across corresponding environments. Among checks, hybrid 3 (PAN53) was the highest yielding normal maize hybrid in this study but it was not stable under all management conditions except under random stress. In contrast, hybrid 1 (ZS261) and 4 (SC533) were the lowest yielding and most stable QPM and normal maize checks, respectively under most conditions.

Figure 2. Mean vs. stability view of GGE biplot based on grain yield of 40 new QPM hybrids and four standard checks evaluated across (a) optimum management, (b) random stress, (c) managed drought and (d) low nitrogen stress. Environment codes are explained in Table 2.

3.3.3. Genotype ranking

The small circle on the AEC axis, with an arrow pointing to it and drawn based on the genotype-focused biplot, represents the ideal hybrid. In this study, hybrid 10 under optimum (Figure 3a); hybrids 14 and 10 under random stress (Figure 3b); hybrids 19 and 40 under managed drought (Figure 3c); and hybrids 14 and 34 under low N (Figure 3d) stress conditions were the most ideal. Contrary to this, hybrids 1 (ZS261) QPM check, 4 (normal maize check) and 5 were the least ideal genotypes under optimum and random drought conditions and hybrids 25 and 36 were the least ideal under managed drought and low N conditions, respectively.

Figure 3. GGE biplot view showing ranking of hybrids based on grain yield of 40 new QPM hybrids and four standard checks evaluated across (a) optimum management, (b) random stress, (c) managed drought and (d) low N environments. Environment codes are explained in Table 2.

3.3.4. Representativeness and discriminating ability, and association

The AEC axis indicated the most representative and discriminating testing environments among the testing sites (Figure 4a–d). A test environment with a small angle with the AEC axis is more representative than other environments and a longer vector length of the environment is considered as more discriminative (Yan & Tinker, 2006). In this study, BK2 was the most discriminating, but it was the least representative environment under optimum conditions, whereas KWE, BK1 and CHI were the most representative and discriminating environments. In contrast, MSE and CHT were non-representative and non-discriminating environments (Figure 4a). Under random stress, LIC was the most representative and non-discriminating environment. UMB and MER were the most discriminating and non-representative environments (Figure 4b). Under managed drought, CHO was the most representative, while CHI was the most discriminating testing site (Figure 4c). Under low N conditions, RAT was the most representative and the least discriminating environment, whereas HAR was the most discriminating, while AM1 was also a representative and non-discriminating environment (Figure 4d).

Figure 4. Discriminativeness vs. representativeness view of the GGE biplots based on grain yield of 40 new QPM hybrids and four standard checks evaluated across (a) optimum management, (b) random stress, (c) managed drought and (d) low nitrogen. Environment codes are explained in Table 2.

The angle between two environments shows the relationship among testing environments and this was visualized by vectors drawn by a dotted line that connects each testing site with the biplot origin (Figure 4a–d). An angle between two environments of <90° indicates a strong positive correlation between them, whereas an angle of equal to 90° indicates no correlation between them. A degree angle of >90° and <270°, indicates negative correlation between the environments (Yang et al., 2009). Under optimum conditions, testing sites such as KWE, RAT, LUS, MPO, BK1, AM1 had strong positive association because they had a less than 90° angle among them, whereas BK2 and HAR had a negative association (>90° angle).

Under random stress, MER and UMB showed negative association, while strong positive association was observed among LIC, BIN, MAG, GWB, GWE testing sites. Across managed drought testing sites, CHO and MAK showed positive association while CHI with MAK showed negative association. No correlation was observed between CHI and CHO sites. Under low N environments, BK2, GWB, AREX and HAR sites had strong positive association, whereas negative association was observed between AM1 with BK2 and AM1 with HAR. In this study Ethiopian sites Bako and Ambo were included for two seasons under both optimum and low N environmental conditions. Both Bako and Ambo sites showed positive association during season one and two under optimum and low N conditions (Figure 4a and d).

4. Discussion

In this study, significant variability was shown for yield and other important agronomic traits among experimental QPM hybrids across all environments. This indicates the existence of different genetic responses among hybrids for measured traits under non-stressed and stressed environments. Previous research also showed the presence of considerable variation among QPM hybrids under different management conditions for grain yield and other agronomic traits (Abakemal et al., 2016; Mebratu et al., 2019; Setimela et al., 2017; Wegary et al., 2014). The highest grain yield was recorded under optimum conditions, followed by random stress, and managed drought, while low N had the lowest grain yield. Even though all stress types had high negative effects on grain yield and other agronomic traits, the effects of low N and managed drought stress were more severe than random stress conditions. Grain yield reduction was also recorded up to 68% under managed drought and 71% under low N environments. This indicated that when N level decreases and drought conditions become more severe, grain yield is reduced (Bänziger et al., 2006). Drought and low N stress conditions that occur before or at anthesis, delaying silk emerging time, results in an increase in ASI, causing incomplete or zero fertilization and resulting in absence of kernel development (Beyene et al., 2013; Ertiro et al., 2017; Magorokosho et al., 2004). As a result, it is useful using secondary traits, like ASI and number of ears per plant as an efficient approach for indirect selection to improve maize grain yield under these stress conditions (Bänziger et al., 2006; Beyene et al., 2013; Ertiro et al., 2017; Magorokosho et al., 2004).

Among the top ten QPM hybrids, some hybrids showed high or comparable grain yield potential to the best non-QPM commercial checks. And almost all of them also showed higher grain yield than the best QPM checks under different management conditions. Compared to the mean of all checks and best QPM check, the top ten QPM hybrids showed high grain yield advantage across all management conditions. However, these top ten QPM hybrids also showed comparable or slightly higher grain yield compared to the best normal maize check under stressed and non-stressed environments. For more than three decades, CIMMYT, in collaboration with other institutes, have been developing QPM germplasms that are superior to commercial QPM and normal maize checks under stressed and non-stressed environments (Mebratu et al., 2019; Setimela et al., 2017; Wegary et al., 2014), and these findings were confirmed in the yield progress reported in this study. Setimela et al., (2017) reported that QPM genotypes showed varying performance in different management conditions. However, in this research, QPM hybrids such as 9 (CZH142237Q), 10 (CZH142238Q) and 14 (CZH15142Q) performed the best under both drought and low N stress conditions. These findings confirmed the results reported by Beyene et al., (2013) who found that maize varieties developed for drought tolerance also showed good levels of tolerance to low N stress. This suggesting the presence of common adaptive mechanisms for both water and low N stresses.

The results of the AMMI ANOVA indicated genotypic difference in grain yield performance among new QPM hybrids across testing sites under stress and non-stress environment conditions. Significant environmental variation and G × E interactions reflected the diversity of the testing environments and the reaction of the hybrids to this, indicating the existence of ranking change of QPM hybrids across different environments. This could significantly limit the selection of superior QPM hybrids that have good grain yield performance in multi-environments. A suitable breeding strategy should follow to develop genotypes that have good performance in multi-environments or genotypes adapted to a specific environment (Mebratu et al., 2019). Similar findings were reported in the previous studies conducted in tropical and sub-tropical maize genotypes evaluated under stress and non-stress environmental conditions (Abakemal et al., 2016; Mebratu et al., 2019; Matongera et al., 2023; Setimela et al., 2017).

Some QPM hybrids showed high positive and negative IPCA1 scores and yielded above the mean, indicating the existence of variation among hybrids in response to the different environments. These kinds of hybrids were considered as unstable and adapted to specific environments. On the other hand, some QPM hybrids also produced good mean grain yield and showed better stability than the rest of the hybrids (had close to zero IPCA1 and ASV values) across different management conditions. Similar findings have been reported in the previous research conducted on QPM hybrids (Abakemal et al., 2016; Mebratu et al., 2019; Setimela et al., 2017), on normal maize hybrids (Ertiro et al., 2017; Sserumaga et al., 2016; Wolde et al., 2018) and on PVA genotypes (Mengesha et al., 2019).

The GGE biplot was provided a polygon view to visualize “which-won-where” pattern (Figure 1a–d) by dividing the biplot into different sectors. When different testing environments occurred in different sectors, it indicated that they have different winner hybrids for those sectors (Mebratu et al., 2019; Yan & Tinker, 2006), indicating the presence of crossover (G × E interaction). This suggesting the possibility of dividing the environments into different mega-environments (Yan et al., 2007) and help to deploy different hybrids in different mega-environments (Abakemal et al., 2016; Mebratu et al., 2019; Wolde et al., 2018). Under optimum conditions, the polygon was divided into ten sectors, but testing sites fell into three of them that were represented by BK2, AM2 and CHI (Figure 1a). Under random stress, the polygon was divided into seven sectors, but the testing sites fell only into two of them that were represented by MER and UMB sites (Figure 1b). Similarly, under managed drought, the polygon was divided into six sectors, but the testing sites fell into two of them that were represented by CHO and CHI (Figure 1c). Under low N conditions, the polygon was also divided into six sectors, but the testing sites fell into three of them that were represented by AM1, BK2, HAR (Figure 1d). On the other hand, genotypes that were placed far away from all the testing environments indicated that they were poor yielding at those testing environments (Makumbi et al., 2015; Wolde et al., 2018). In this study, hybrids 5 (CZH04032), 4 (SC533), and 34 (CZH17203Q) under optimum, 1 (ZS261), 25 (CZH17194Q) and 35 (CZH17204Q) under random stress, 25 (CZH17194Q), 27 (CZH17196Q) and 31(CZH17200Q) under managed drought and 36 (CZH17205Q) and 18 (CZH17187Q) under low N stress were poor performing hybrids. In general, the results of this study gave initial information on mega-environment delineation of ESA maize testing locations. To validate this information, multi-location and season data are needed.

Based on genotype ranking GGE biplot analysis, genotypes that are closest to the concentric circle are considered as ideal genotypes that have both high mean grain yield and are stable in multi-environments (Yan & Kang, 2003; Yan & Tinker, 2006; Wolde et al., 2018). As indicated in Figure 3a–d, the most desirable hybrids closest to the concentric circle under each management condition were identified. Entries 10 and 14 were identified as ideal genotypes across non-stressed and stressed environments, indicating that these hybrids had inherently high yielding potential and wide adaptability across different management conditions. Because of the genetic background of the genotypes or nature of the trait (grain yield), the ideal genotype does not always exist in the reality. However, such kind of hybrids can be used as a reference for further studies. On the other hand, entry 1 (ZS261), the QPM check, was far from the concentric circle under optimum and random drought environments compared with experimental QPM hybrids. This tells, this hybrid was inherently low yielding and poorly adapted to some or all the testing sites, which is similar with results reported by Mebratu et al. (2019).

Environments that have small angles with the AEC are more representative, whereas long vectors from the biplot origin is considered as more discriminative environments (Yan & Tinker, 2006). Environments that have both long vectors and small angles with the AEC abscissa are more discriminating and representative environments and are considered “ideal testing sites” and are important to identify desirable hybrids (Mebratu et al., 2019). In this study, KWE under optimum (Figure 4a), BIN under random drought (Figure 4b), CHO under managed drought (Figure 4c) and BK2 under low N stress (Figure 4d) were both discriminating and representative environments. These sites were recognized as ideal environments in the respective management conditions. An environment which could discriminate the genotypes and is representative of the target environment is considered as the ideal environment (Yan & Tinker, 2006). So these testing environments can be used to select the top yielding hybrids that will perform consistently across environments (Mebratu et al., 2019; Tolessa & Gela, 2014). Testing environments which had strong negative association indicates the presence of cross-over among genotypes (Yan & Tinker, 2006). Even though most of the environments had positive association, negative association was shown between BK2 and HAR environments under optimum conditions, MER and UMB under random drought, CHI and MAK under managed drought and AM1 with BK2 and AM1 with HAR under low N conditions. If two testing sites associated consistently across seasons/years, one could be dropped without significant loss of information about the genotypes (Lule et al., 2014; Tukamuhabwa et al., 2012). In this study, Bako and Ambo testing sites showed positive association during season one and two both under optimum and low N conditions, but negative correlation was observed between BK2 and AM1 under low N. This kind of inconsistency might have been due to variation in weather conditions like rainfall, temperature, soil conditions, and other unpredictable factors.

5. Conclusions

In this study, promising QPM hybrids under stressed and non-stressed environments were identified. Based on AMMI and GGE biplot analysis, the most desirable hybrids were 6, 10, 14 and 20 under optimum; 3, 14 and 23 under random stress; 19, 35 and 40 under managed drought and 14 under low N conditions. GGE biplot analysis identified siting sites for proper selection of QPM hybrids on the basis of their discriminating ability and representativeness. Accordingly, Kwekwe (KWE), Bindura (BIN), Chokwe (CHO) and Bako (BK2) testing sites were the most suitable environments in discriminating the experimental QPM hybrids and being representative test environments for optimum, random stress and low N stress conditions, respectively. Generally, further evaluation of the selected QPM hybrids across multi-location and season would help to promote into variety verification for commercialization and could contribute to improve food and nutritional security problems in sub-Saharan Africa.

Author contributions

This work was conducted as part of BTE’s PhD thesis in plant breeding at University of the Free State, South Africa in collaboration with CIMMYT. The trials were planned and designed by AT, DW and BTE. The first draft of the manuscript was written by BTE and revised the manuscript by MTL. All authors have read and approved the final manuscript for submission.

Acknowledgments

The authors acknowledge the International Maize and Wheat Improvement Centre (CIMMYT), National Agricultural Research System and private seed company scientists and technicians who participated in conducting the trials and data collection at all testing sites. The National Research Foundation is also acknowledged for supporting the thesis work.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are included in the manuscript and further inquiries can be forwarded to the corresponding author.

Correction Statement

This article was originally published with errors, which have now been corrected in the online version. Please see Correction (http://dx.doi.org/10.1080/23311932.2024.2355806).

Additional information

Funding

This research was supported by the National Research Foundation (NRF) through the SARChI chairs initiative (UID 84647).