This study was carried out from August 2023 to November 2023 at the University of Yaoundé I (3.8528˚N and 11.5165˚E) located in the agroecological zone V (bimodal rain forest zone), with rainfall ranging from 1500 to 2000 mm/year and an average temperature of 23˚C at an altitude of around 797 m ( Figure 1 ). The climate is equatorial Guinean with four seasons including two dry seasons (December-February and July-August) and two rainy seasons (March-June and September-November).

The plant material consisted of 33 landraces of Zea mays L. from Cameroon and three from Kinshasa (Democratic Republic of Congo). This material was collected from farmers or local markets. Landraces were collected through direct observations and interviews with farmers, who provided information on the nature, performance, and origin of each landrace. Collection occurred either in the fields or from storage granaries, depending on material availability. Landraces were then chosen using two qualitative traits: grain colour and grain texture, following maize descriptors of [14] , allowing the identification of distinct morphotypes across Cameroon ( Table 1 ).

This material was collected from farmers or in the local markets. The experiment was conducted using an Alpha lattice design (6 × 6 = 36) with two replicates. Plants were spaced of 0.80 m between rows and 0.50 m between hills. The layout consisted of two blocks (replicates), each containing 648 plants, with 18 plants per ridge. The Alpha lattice design was chosen to improve the precision of the experiment by effectively controlling for spatial variation in the field. This design is particularly useful when dealing with many treatments, such as diverse landraces, where environmental heterogeneity may affect the performance of genotypes. By grouping treatments into smaller blocks, the design minimizes experimental error and enhances the accuracy and reliability of comparisons among genotypes.

CMR: Cameroon; DRC: Democratic Republic of Congo.

Phenotypic data were collected for 20 traits across the 36 maize landraces, covering key stages of plant development from germination to yield. These traits were selected based on their agronomic importance and potential to reveal variation in growth, flowering dynamics, and yield-related characteristics. For clarity, the traits were grouped into three categories: germination and early growth traits, morphological and growth traits, and reproductive and yield-related traits.

Measurements were taken at appropriate growth stages following standard protocols. Germination traits were recorded daily starting from the day of sowing until no further germination was observed. Morphological traits were measured at the milk stage, during the early grain-filling phase of maize development. Reproductive traits were recorded at flowering (anthesis and silking stages) and during harvest at physiological maturity. For each trait, data were collected from 5 randomly selected and clearly tagged plants per plot to ensure consistency. Measurement tools included graduated rulers for height (±0.5 cm precision), calipers for diameters (±0.1 mm precision), and electronic scales for kernel weight (±0.01 g precision). Leaf area was calculated using the formula of Raunkiaer as [15] : 2/3 (leaf length × leaf width). Tools were calibrated before use and operated according to manufacturer guidelines. Data was collected from representative plants within each plot, ensuring consistency across replications ( Table 2 ).

The collected data was subjected to comprehensive statistical analyses to assess the phenotypic diversity among the 36 maize landraces. Descriptive statistics, including mean, standard deviation, range, and coefficient of variation were calculated for each trait to summarize the overall variation.

To identify traits contributing most to the observed variability, principal component analysis (PCA) was performed to reduce dimensionality while retaining the traits with the greatest discriminatory power. The number of components retained was based on a scree plot inspection that shows variations in the 10 first components. Additionally, PCA facilitated visualization of trait associations and landrace distribution in reduced dimensions. Hierarchical clustering on principal components (HCPC) was then applied to group landraces based on their phenotypic similarities. This method combines PCA, hierarchical clustering (using Ward’s method and Euclidean distance), and partitioning around medoids (PAM) to optimize cluster assignment. HCPC improves clustering accuracy by eliminating redundancy and noise before clustering. The choice of Euclidean distance and Ward’s criterion is justified by their effectiveness in minimizing intra-group variance and maximizing inter-group separation.

Genetic parameters such as genetic variance (V_G), phenotypic variance (V_P), environmental variance (V_E), broad-sense heritability (H²), phenotypic coefficient of variation (PCV), genotypic coefficient of variation (GCV), genetic advance (GA), and genetic advance as a percentage of the mean (GAM) were calculated for each trait to assess the potential for genetic improvement. Genetic variance (V_G) was estimated by subtracting the environmental variance (V_E) from the total phenotypic variance (V_P). Broad-sense heritability (H²) was calculated as the ratio of genetic variance to phenotypic variance and was classified into three: low (0 - 0.3), medium (0.31 - 0.6) and high (>0.6) [16] . The PCV and GCV were computed using the formula of [17] . Genetic advance (GA) was estimated based on the expected response to selection, using the formula incorporating the heritability, selection differential, and the standard deviation (SD) of the population. Genetic advance as a percentage of the mean (GAM) was calculated to express the potential genetic improvement relative to the trait’s mean value. These parameters were calculated using standard formulas as outlined by Burton and De Vane [17] .

Additionally, the Shannon-Weaver diversity index was computed for each trait to quantify the diversity within the landraces. This index accounts for both the richness and evenness of phenotypic variation, offering a robust measure of intra-population diversity. Multivariate analysis of variance (MANOVA) was conducted to partition the total phenotypic variation within and among landraces.

All statistical analyses were carried out using R software [18] . The significance level for all tests was set at p < 0.05.

Descriptive statistics for traits across the 36 maize landraces showed an important variability. Traits related to germination showed moderate variation, with the mean germination latency time being 5.61 ± 1.45 days, and a coefficient of variation (CV) of 25.83%. The time to 50% germination was on average 6.89 ± 2.01 days (CV = 29.19%), while the maximum germination time was lower, i.e., a CV of 9.53%, indicating that overall plant tends to germinate in a more consistent way for all the 36 landraces. The percentage of germination was on average 80.57% with a CV of 25.7% ( Table 3 ).

Vegetative growth traits also showed high differences. Plant height varied also moderately, with a mean of 240.39 cm and a CV of 18.49%, while root collar diameter and leaf area showed CVs of 21.26% and 24.88%, respectively. Ear height on the other part showed also an important variation, with a CV of 24.48%, demonstrating differences in plant architecture across landraces ( Table 3 ).

Reproductive traits for their part were relatively stable overall. Both male and female flowering at 50% had low CVs i.e., 5.11% and 4.15%, respectively. That suggests uniform flowering times across landraces. Panicle length exhibited a higher CV of 28.32%, while the number of primary and total panicle branches showed the highest variation among traits, with CVs of 37.26% and 40.09%, respectively. The number of ears per plant also presented notable variability (CV = 41.85%), reflecting differences in yield potential.

Finally, traits related to yield depicted moderate variation. Ear length and circumference had CVs of 17.58% and 9.52%, respectively. The number of kernel rows per ear and the average number of kernels per row exhibited CVs of 16.81% and 19.88%, respectively. Weight of 100 kernels had a CV of 19.52%, indicating moderate variability in grain size and weight ( Table 3 ).

These results reveal substantial variation among landraces, especially in vegetative and yield-related traits, suggesting rich potential for selection and improvement. The uniformity in flowering times is advantageous for breeding synchronization, while the diversity in germination and reproductive traits supports the adaptability of these landraces to varying environments.

The analysis of genetic parameters showed an important variation across the landraces ( Table 4 ). Broad-sense heritability (H²) ranged from 0.01 for maximum germination time to 0.69 for germination latency time, indicating diverse genetic control in the traits. High heritability was observed for traits such as germination latency time (0.69) and 100-kernel weight (0.62), suggesting an important genetic influence, while traits such as female flowering at 50% (0.20) and number of leaves above the ear (0.08) showed lower heritability, indicating a greater environmental impact. The phenotypic coefficients of variance (PCV) were higher than the genotypic coefficients of variance (GCV) for all traits, indicating a large influence of environmental factors. Indeed, traits like the number of primary panicle branches and total number of panicle branches showed high PCV (0.38 and 0.40, respectively) and moderate GCV (0.17 and 0.19), suggesting potential for selection. Genetic advance as a percent of the mean (GAM) varied a lot, with the highest values observed for germination latency time (36.66%) and number of primary panicle branches (14.96%), indicating that they are suitable for genetic improvement ( Table 4 ).

Overall, the observed variability in genetic parameters underscores the genetic diversity present among the landraces. Traits with high heritability, moderate to high GCV, and substantial genetic advance, such as germination latency time and 100-kernel weight, are then important for selection and breeding programs.

SD: standard deviation, Min: minimal value, Max: maximum value, CV: coefficient of variation.

The contribution of each trait to the first principal components of PCA analysis showed distinct variation across the 36 landraces ( Table 5 , Figure 2 ). The first three principal components (PC1, PC2, and PC3) together explain 51.17% of the total variance in the dataset ( Figure 2 ), providing a significant overview of the phenotypic variation among the landraces. PC1 accounts for 22.9% of the variance and is largely driven by traits such as plant height, root collar diameter, and leaf area, indicating that larger plants and leaves are associated with higher positive values on this component ( Figure 3 ). PC2, which explains 15.8% of the variance, is influenced by traits related to germination and flowering time, with later-flowering plants contributing to its positive side. PC3, contributing 12.52% of the variance, further differentiates the landraces based on additional traits, although its specific contributions are less pronounced compared to PC1 and PC2 ( Table 5 , Figure 2 ).

V_G: Genetic Variance, V_p: Phenotypic (total) variance, V_E: Environmental variance, H²: Broad-sense Heritability, PCV: Phenotypic Coefficient of Variance, GCV: Genetic Coefficient of Variance, GA: Genetic Advance, GAM: Genetic Advance as percent of Mean. GA and GAM are computed for k = 2.06 i.e., at 5% of selection intensity.

The factor map of individuals ( Figure 4 ) illustrates the distribution of the 36 maize landraces in the two principal components, highlighting phenotypic similarities. The HCPC analysis ( Figure 5 ) further refines this by clustering the landraces based on their morphological characteristics, providing insights into genetic diversity patterns. Indeed, Cluster 1 is made up of Ako 1, Baf, Bar 2, Bar 4, Bar 6, Bar 8, and Kii 1, Kii 3, Kii 5, Kii 8, Kii 10 ( Figure 4 , Figure 5 ). These landraces tend to have shorter plants, smaller fruits, and fewer fruits per plant, with earlier flowering and lower yield potential, indicating adaptation to shorter growing seasons or resource-limited environments. Cluster 2 consists only of Ess 3, having unique traits such as higher fruit production and larger fruit dimensions. This difference suggests potential for productivity or specific adaptation patterns not shared with other landraces. Cluster 3 includes the majority of landraces, i.e., Ako 2, Ako 3, Bar 1, Bar 3, Bar 5, Bar 7, Ess 1, Ess 2, Ess 4, Ess 5, Evo 1, Evo 2, Kii 2, Kii 4, Kii 6, Kii 7, Kii 9, Kii 11, Kii 12, Kii 13, Kin 1, Kin 2, Kin 3, and Mba ( Figure 4 , Figure 5 ). These landraces are characterized by taller plants, longer fruits, and higher fruit numbers, indicating higher yield potential and adaptation to longer growing seasons.

Multivariate analysis of variance (MANOVA) showed significant differences among landraces when considering all traits simultaneously (Wilks’ Lambda = 1.44e−10, p = 1.595e−06). This suggests that there is a significant multivariate difference in the traits between the landraces ( Table 6 ). Univariate analyses revealed that traits such as germination latency time (1.102e−06), 100-kernel weight (1.905e−05), ear circumference (0.00104), germination percentage (0.0012), time to 50% percent germination (0.0022), male flowering at 50% (0.007) contributed most to these differences ( Table 3 ).

Shannon-Weaver diversity indices across the 36 landraces ranged from 1.815 to 2.141, indicating varying levels of morphological diversity ( Table 7 , Figure 6 ). Most landraces, i.e., 27 out of 36 displayed moderate diversity, while six landraces showed high diversity, including Kii 3, Kii 8, and Bar 8. In contrast, three landraces, i.e., Bar 1, Ako 2, and Evo 1, showed low diversity. The presence of landraces with high diversity suggests potential genetic richness, likely to be valued for breeding programs to improve traits of interest.

Significance codes. ***: p < 0.001, **: p < 0.01, *: p < 0.05, no symbol: p ≥ 0.1; df: Degree of Freedom, Wilks: Wilks lambda statistics; Approx. F: approximate F-value; Num df: Numerator degrees of freedom; Den df: denominator degrees of freedom.

Descriptive statistics for traits across the 36 maize landraces showed an important variability in germination, vegetative growth, reproductive and yield traits based on morphological study. According to Salami et al. [19] for a good knowledge the genetic diversity must be performed prior morphological study. This agromorphological diversity of local maize landraces revealed significant differences between the traits analyzed, indicating strong phenotypic heterogeneity between the 36 landraces. This high level of heterogeneity was also observed in Côte d’Ivoire within 171 landraces [7] . In Their study on the morphological characterization of maize in Ivory Coast, N’Da et al. [7] observed a wide variation of agronomic traits such as plant height and ear height (>10%) and a low coefficient of variation (<10%) for the tassel length, male and female flowering. Our result showed low coefficient of variation for maximum germination time, male flowering at 50%, female flowering at 50% and ear circumference, while the order traits exhibit high variation. The results of Salami et al. [19] also showed that the plant height, ear height, tassel length exhibit high variation (>10%) while variable female flowering, male flowering, days to ear leaf senescence and germination days have a low variation (<10%) with the variation coefficient varies between 8.63% and 20.43%. In this study, variation range between 4.15 to 41.85 for female flowering at 50% and number of ears per plant respectively. As in Salami et al. [19] This difference can be explained due to the exclusive use of local maize landraces, the difference in the origin of these landraces and also the type of soil. This phenotypic farmer selection based on agronomic traits could explain the contribution of these variables.

The 5 principal components (PC) accounted for 88.40% of the total variability produced by all maize genotypes. In the work of N’Da et al. [7] , five PCs were obtained to explain 82.35% of the variance present in the variables. Our results indicated that all the traits showed positive loading on PC1 which contributed to 22.9%, with the highest loadings coming from plant height this is in contrary of the founding of Saleh et al. [20] where days to 50% tasseling, anthesis silking interval and cob diameter showed highest value on PC1 which contributed to 38.9%. The variation in diversity was captured across the principal components, each reflecting a distinct proportion of the total phenotypic variance among the landraces.

Hierarchical clustering analysis on principal components (HCPC) grouped the 36 landraces into three distinct clusters. Raman et al. [10] grouped maize genotypes in 6 clusters based on morphological variability. The numerical classification revealed four groups of landraces morphological similarity perspective both at the Centre and the North of Benin [19] . Also, the genotypes belonging to the cluster V, II, IV, and III were found most divergent. Accordingly, these genotypes could be selected as parents in future hybridization program for improvement of maize [10] .

Phenotypic coefficient of variation (PCV) was greater than the genotypic coefficient of variation (GCV), showing that the characters were more influenced by their surrounding environments. This founding is in accordance with those of Magar et al. [9] who founded that the GCV values were lower than PCV value. By indicating the high environmental influence on the studied traits, Al-Amin et al. [21] also found the PCV was higher than the GCV for all the yield contributing traits of maize. As in Singh et al. [22] , the traits evaluated in this study had low (>10%), moderate (10 - 20%), and high (<20%) phenotypic and genotypic coefficients of variation. Traits exhibited a wide range of variability, with high PCV and GCV observed notably for germination-related traits (germination latency time, time to 50% germination) and several morphological characteristics (leaf area, ear height, panicle length, 100-kernel weight). In contrast, low to moderate variability was recorded for traits such as male and female flowering at 50%, ear circumference, and number of kernel rows per ear. This suggests that while some traits are strongly influenced by genetic factors, others may be more affected by the environment. Low values of both GCV and PCV suggest environmental factors had more influence on the expression of traits than the genetic factors, indicating a limited scope for improvement on these traits by direct selection on high performance genotypes.

Our result showed low (<30%), moderate (30 - 60%), and high (>60%) estimates of heritability for the various traits studied, as defined by Yadesa et al. [8] . The same result has been obtained by Magar et al. [9] . According to Syahruddin and Suwardi [13] , the characters that have wide genetic variability will provide a great opportunity for selection to obtain superior genotypes and more effectively. High heritability, coupled with genetic advance, was observed for traits like grain yield per plant, plant population, ear height, plant height, and 100-seed weight, indicating additive gene action. These traits can be effectively improved through phenotypic selection [10] . High estimates of heritability for Germination latency time and 100-Kernel Weight suggested that variations were passed down to progeny, implying that a high-yielding variety may be developed by selecting desirable genotypes according to Magar et al. [9] . Genetic advance as per cent of mean was low for all characters except for Germination latency time. This founding is not in accordance with those of Raman et al. [10] who founded in their study high GAM for grain yield per plant (48.50), plant population (42.66), ear height (27.94), 100 seed weight (23.99) and plant height (23.03) and moderate to low for all the remaining traits. When additive gene effects controlled a characteristic, it usually resulted in both higher heritability and genetic advance as for Germination latency time, whereas when non-additive gene actions controlled a trait, it might result in high heritability but poor genetic advance as for 100-Kernel Weight.