Genetic architecture and candidate loci associated with growth and yield-related traits in sweet fig banana (Musa acuminata cv Sotoumon, AA)

Abstract

Background

Yield in Sweet fig banana is an economic trait for enhancing the production in the perspective of improving nutrition and income for smallholder farmers. Consequently, understanding its genetic basis is a major focus in breeding new varieties. To reveal loci linked to yield related traits in sweet fig banana, 238 accessions from Benin were characterized in field trials at two locations over 3.5 years. Genome-wide association studies (GWAS) were performed using 1,473 high-quality SNPs and five statistical models—GLM, MLM, BLINK, FarmCPU, and MLMM.

Results

Findings revealed significant variation in all phenotypic traits among accessions. LD decay of 5.392 Mb physical distance was noted in sweet fig banana chromosomes. GWAS identified 29 significant SNPs associated with growth traits and 16 with yield-related traits. Two SNPs were identified only in single-locus models, 36 exclusively in multi-locus models, and seven were shared by both. Moreover, 16 putative candidate genes associated with traits were identified. Plant height and yield-related traits exhibit a polygenic architecture, governed by the cumulative minor effects of multiple SNPs. Four multi-locus traits were detected. Given the extensive LD decay, marker-traits associations (MTA) for polygenic traits likely indicate haplotypic effects, and the multi-trait loci may be pleiotropic or tightly linked genes.

Conclusions

This study provides an initial understanding of the genetic control of growth and yield traits in sweet fig banana and offers a foundation for genomics-assisted breeding in dessert banana.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-026-13036-8.

Background

Current crop yield trends are not keeping pace with the sustenance needs of a growing global population projected to reach 10 billion by 2050, particularly in sub-Saharan Africa [1, 2]. The production increases must primarily come from yield in the face of ongoing water scarcity and fertile land [3]. Therefore, it is imperative to steadily develop diverse crop varieties with superior attributes and resilience to achieve sustainable food and nutrition security [4–6]. Bananas are a favourite dessert and fresh fruit worldwide, a staple food for hundreds of millions of people, and an important source of income for small-scale farmers in developing countries [7, 8]. They are a source of highly potent bioactive, nutritional, and pharmacological compounds with effective health and wellness properties [9, 10]. Furthermore, bananas are a high-value commodity that contributes significantly to international trade and holds great economic promise [7, 11]. In Benin, the average yield of banana dessert varieties is 4.8 t.ha^{− 1}, compared with 13.3 t.ha-1 in Africa and 22.7 t.ha-1 worldwide [12].

Local banana production is hampered by various pests, diseases, climate change challenges, and low soil fertility [13–16]. Among landraces, the sweet fig banana (Sotoumon in the Fongbé language) is the most commonly cultivated and holds great economic significance as a cash crop across Benin’s expanding regions [13]. However, the variety is better known worldwide for its exceptional taste qualities than for its agronomic performance [17], which is considered mediocre. Addressing the constraints of low yield and environmental pressures requires high-yielding and stress-resistant cultivars with quality attributes. The application of genetic engineering for yield, resilience to both biotic and abiotic stresses, and high-quality traits facilitates the development of new cultivars that demonstrate robust performance across various environmental conditions, thereby enhancing food security [18–20]. However, successful plant breeding for increased yield requires a thorough understanding of the genomic basis of the traits of interest. Unfortunately, the genomic regions associated with sweet fig banana growth and yield-related traits have not been investigated yet. The advent of genome-wide association studies (GWAS) offers a promising approach for elucidating relationships between phenotypes and genotypes [21–23]. GWAS enables breeders to efficiently identify and introduce desirable traits into plant species, accelerating the development of new crop varieties [24]. This approach has been successfully used to identify genomic regions associated with different agronomic traits in bananas [25–27], cassava [28], sugarcane [29], maize [30], rice [31, 32] and sorghum [33, 34]. Several GWAS statistical models have been developed, but the main challenge is to differentiate true positive marker-trait associations (MTA) from false positives [35]. Combining different GWAS models can improve the chances of detecting genuine MTA [33]. Previous studies on banana cultivars have examined loci linked to seedless traits [29], bunch weight, and component characteristics in East African Highland banana using the mixed linear model (MLM) [27], as well as morphology, fruit quality, and yield traits in 124 accessions from the Colombian Musaceae Collection employing MLM and Bayesian-information and linkage disequilibrium iteratively nested keyway (BLINK) models [28]. In this study, we examined 238 sweet fig banana accessions from the Republic of Benin to identify loci regulating key agronomic traits using five GWAS models. We conducted both single and multi-locus GWAS to achieve comprehensive results, pinpointing loci consistently found across models for more reliable breeding targets. The two single-locus models are the general linear model (GLM) and MLM. Multi-locus models include the multi-locus mixed model (MLMM), Fixed and random model circulating probability unification (FarmCPU), and BLINK. This study aimed to identify the causative loci underlying the agronomic performance of sweet fig banana to set the scene for marker-assisted breeding in the cultivar. We assumed that the variation in growth and yield-related traits among sweet fig banana accessions has a clear genetic basis.

Materials and methods

Plant materials

The plant material used in this study, a collection of 238 accessions of sweet fig banana landraces, was sourced from diverse farmers’ fields and home gardens across 19 municipalities in the Guinean and Sudano-Guinean zones of Benin. This unique plant genetic resource is maintained as a field collection by CalaviGen, the University of Abomey-Calavi (Abomey-Calavi, Republic of Benin) genebank, and constitutes the core of our study. The accessions were propagated separately using the seedling-from-stem fragment (PIF) method [36], and the resulting plants were used to conduct phenotypic evaluation trials in the localities of Sékou and Dan. This vegetative propagation method involves a single parent and produces genetically identical offspring that retain the mother plant’s genetic traits, ensuring genetic consistency and stability.

Phenotypic data assessment

Experimental sites description

The field trials were carried out in two experimental sites of the Genetics, Biotechnology, and Seed Science Unit (GBioS) of the Faculty of Agricultural Sciences (FSA), specifically at Sékou (N 6°61’66’’and E 2°21’66’’E) and Dan (7°40’16’’N and 2°12’66’’E). Sékou is located on the ferralitic plateau of Allada District in the Atlantic Department of Southern Benin (between 6°30 and 7° North). The area falls within the Guinean zone, characterized by a sub-equatorial climate with two rainy seasons and annual rainfall around 1300 mm [37]. The temperature at the Sékou site typically ranges from 23 to 32 °C, with relative humidity between 73 and 79%. During the two-year experiment, the site recorded a cumulative precipitation of 1896.42 mm with a mean relative humidity of 79.4%. The measured temperature ranged from 21.5 to 30.02 °C, with an average of 28.25 °C [38].

Dan is located in the District of Djidja (Zou Department) in the Sudano-Guinean zone (7° to 9°30 North), marking the transition between humid sub-equatorial and Sudanian climates. The Sudano-Guinean zone lies in the Center of the Country. The climate is characterized by annual rainfall of 1200–1300 mm, with two rainy seasons. The average annual temperature varies from 22.5 to 35.2 °C, and the relative humidity is between 30 and 70%. The predominant soil type is leached tropical ferruginous, and the landscape is characterized by a plateau with granite rock formations. The total precipitation during the two-and-a-half-year experimental period was 2182.20 mm, and the mean relative humidity was 75.3%. The temperature ranged from 23.44 to 33.39 °C, with an average of 28.4 °C [38].

Experimental design and management

The experiments were conducted under rainfed conditions between August 2019 and July 2021 in Sékou and from October 2020 to March 2023 in Dan. In both locations, we implemented an alpha lattice design with two replicates, each containing 20 incomplete blocks, with a block size of 12. Three individuals of the same accession were planted per experimental unit. The planting plan was developed using the “Agricolae” package in the R software [39] with an efficiency factor of 0.89. The plant spacing for the field trials was 2 m between rows and 2 m within rows, with a density of 2500 plants ha^{− 1}. Each planting hole had a volume of 40 cm³. The trials were carried out following the technical itinerary described by GBioS [40]. Farmyard manure was applied in each hole at the rate of 5 kg/plant. Two fertilizer applications of 500 kg N₁₀K₂₀P₂₀ were made one and six months after planting, respectively. Additionally, potassium chloride (KCl) was applied six months after planting. Regular weeding was carried out every two to three months.

Data collection

Fifteen growth, phenology, and yield-related traits (Table 1) were collected during a single cycle on both fields. All data, except flowering time, were recorded at bunch harvest from January 2020 to July 2021 at the Sékou site and from September 2021 to March 2023 at the Dan site. Generally, the flowering and harvest periods are not synchronized in bananas.

Table 1.

Growth, phenology, and yield traits description and measurement techniques for phenotypic characterization

Traits	Codes	Measurement techniques
Growth traits
Circumference of the pseudo-stem at harvest	Cir10	The girth of the pseudo-stem was measured using a meter tape at 10 cm from the base of the plant
Pseudo stem height at harvest	PHH	The pseudo-stem height was measured from the ground to the top of each plant
Number of functional leaves (at least half of the surface is still green) on each plant at the harvest.	NFLH	The number of standing functional leaves per plant at harvest was counted manually for all genotypes
The number of suckers emitted by each mother plant at harvest	NSEH	The number of suckers was recorded by counting the number of suckers produced by each plant
The number of suckers with more than 30 cm in length at harvest	NSLH	The length of suckers produced by each plant was measured using a meter tape, and the number of those that were more than 30 cm tall was recorded
The height of the successor sucker (tallest sucker)	HSS	The height of the tallest suckers on each plant was measured with a tape measure
Phenological traits
The number of days to flowering	NDF	The number of days to flowering was obtained by counting the days from planting to the appearance of the inflorescence
The number of days to fruit filling	DFF	The number of days to fruit filling was determined as the interval between flowering and bunch maturity (number of days from shooting to harvest)
The length of the growth cycle	LGC	The length of the growth cycle (number of days to maturity) was obtained by counting the number of days from planting to harvest of the bunch
Yield-related traits
The bunch weight at maturity	BW	The bunch was taken from each plant at full physiological maturity (when at least one finger or fruit had ripened) and weighed in kilograms using a 25 kg scale
The bunch diameter	BD	The diameter of the bunch was measured using a meter tape between the first and last hands in the middle of the attachment
The bunch length	BL	The length of the bunch was measured using a meter tape from the attachment point of the first hand to the last hand
The number of hands per bunch	NHB	The number of hands per bunch was counted and recorded
The number of fruits per bunch	NFB	The number of fruits per bunch was counted and recorded
The weight of the middle fruit on the middle hand of each bunch	WMF	The middle fruit of the middle hand was separated from the bunch and weighed in grams using the electronic scale with a capacity of 250 g

Phenotypic data analysis

The R software Version 4.1.3 (R Core Team, 2022) was used to perform descriptive statistics, analysis of variance, broad-sense heritability, trait correlation, and to generate all graphs. First, the raw phenotypic data for each trait were screened for potential outliers using the R package “outliers” [21, 41]. Removing outliers prevented them from skewing the data away from a normal distribution. The linear mixed model was employed to estimate phenotypic values using the lme4 package in R [42]. We estimated the best linear unbiased predictors (BLUPs) of random genetic effects using a linear mixed model. These BLUPs were used as corrected phenotypic values in subsequent variance and association analyses. Analysis of variance (ANOVA) was conducted for each trait, with genotypes as fixed effects and blocks as random effects. The following equation was used for the ANOVA:

where Yijk denotes the phenotypic response of the ith genotype in the kth incomplete block within the jth replication, µ is the overall mean, σi is the fixed effect of genotype i, ßj is the random effect of the jth replication, Ƴkj represents the random effect of the jth incomplete block nested within the kth replication, and εijk is the residual error.

Moreover, the distribution of phenotypic traits was assessed by constructing bar graphs and boxplots.

Furthermore, the broad-sense heritability (H²) of the target traits was calculated to estimate the proportion of phenotypic variation explained by genetic variation among accessions for each trait [21, 43]. The formula used for this calculation was as follows:

In this formula: is the variance of genotype, is the variance of genotype interaction with the environment, is the variance of residual error, r is the number of replicates in an environment, and n is the number of environments.

Pearson’s method was performed to determine interactions between the measured traits. The package psych in R [44] was used to visualize the correlation matrix.

Genomic data analysis

SNPs profile and genetic diversity analysis

The SNPs dataset used in this study was derived from a previously published study on genetic diversity analysis of sweet fig banana accessions from Benin [45]. To ensure uniformity in the accession line, one individual per accession was cloned and genotyped. In this earlier work, genotyping-by-sequencing of a panel of 273 sweet fig banana accessions, of which 238 were included in the present study, identified 8,457 DArTseq markers. After rigorous filtering, imputation, and removal of SNPs with polymorphic information content (PIC) below 0.05, a total of 1,992 high-quality and polymorphic SNPs were identified. The markers’ exclusion criteria included: missing data higher than 20%, minor allele frequency (MAF) below 0.05, call rate below 75%, and reproducibility rate below 95%. However, some plants in the core collection died after planting, so we had to use 240 accessions for field experiments. During the experiment, two accessions were lost, reducing the total to 238 at both sites. The SNPs from the 35 accessions that were lost were removed from the 1,992 high-quality, polymorphic SNPs identified in the previous study. Hence, the evaluation of agronomic performance in the present study was conducted on a subset of 238 sweet fig banana accessions, yielding 1,473 high-quality and polymorphic SNPs. Subsequently, these 1,473 SNPs were screened to exclude those not located on any chromosome, yielding a final set of 345 high-quality SNPs selected for genetic diversity and marker-trait association analyses. After updating the SNPs database, the genetic diversity of the remaining 238 accessions was re-evaluated to confirm the results of the prior study.

To assess the genetic diversity of the 238 accessions studied, we calculated the genetic diversity index, including observed heterozygosity (Ho), expected heterozygosity (He), total gene diversity (Ht), and inbreeding coefficient (Fis).

Linkage disequilibrium analysis

Software TASSEL 5.2.58 was employed to evaluate linkage disequilibrium (LD) across the chromosomes with 345 SNPs. LD between marker pairs was quantified by computing the squared allele frequency correlation (r²). The decay of LD was illustrated by plotting r² against the physical distance (bp) between SNP pairs, using the ggplot2 [46] package in R.

Genome-wide association analysis

This study primarily focused on the growth and yield of the sweet fig banana and did not consider phenological traits for the GWAS. The association between SNP genotypes and adjusted agronomic performance traits values was tested with five GWAS statistical models: (i) two single-locus models, including GLM and MLM and (ii) three multi-locus models, such as BLINK, FarmCPU and MLMM. The Genome Association and Prediction Integrated Tool (GAPIT, Version 3) R package [47] was used to perform GWAS using these five models. The first three principal components (PCs) and the matrix « Q » were included as covariates in GAPIT to control for population structure. We implemented the Holm & Bonferroni correction method to determine the threshold for detecting significant SNPs. The GWAS results were visualized using Manhattan and quantile-quantile (Q-Q) plots generated with the CMplot and GAPIT packages in R (R Core Team 2022). The Manhattan plot provides an overview of the results by plotting the negative logarithm (-log10) of each GWAS model’s p-value against the genomic region of each SNP across the 11 banana chromosomes. The QQ plots assessed the fitness of GWAS models. They showed the observed negative logarithm (-log10) distribution of the p-values, compared with the expected distribution under the null hypothesis. The model should follow a uniform distribution.

Potential candidate gene identification

The genome sequences of the associated significant SNPs were blasted against the Universal Protein Resource (UniProt) database (https://www.uniprot.org) and the National Center for Biotechnology Information (NCBI) database (http://www.ncbi.nlm.nih.gov) to identify potential candidate genes underlying the causal variants and determine associated biological functions and processes they are involved in.

Results

Phenotypic traits variation in sweet fig banana accessions

Variance and heritability in traits

Significant variation was observed among the 238 sweet fig banana accessions evaluated in experimental fields for 15 traits, using an alpha lattice design with two replicates at two locations. The length of the growing cycle (LGC) ranged from 420 to 714.8 days, with an average of 557.0 ± 127.4 days. The average number of days to flowering (NDF) was 466.3 ± 123.8 days, ranging from 322 to 605.6 days. The mean number of days from flower to fruiting (DFF) varied from 84.5 to 146.7 days, with a mean of 100.5 ± 14.05 days. The coefficient of variation was low for these three variables: LGC (22.9%), NDF (26.5%), and DFF (14%).

The growth and yield-related traits in sweet fig banana accessions showed notable phenotypic variation (Table 2). These traits exhibited continuous distributions, indicating they are quantitative traits influenced by multiple genes.

Table 2.

Phenotypic variation of twelve Growth and yield-related traits in 238 accessions of the sweet fig banana landrace of Benin

Traits	Cir10 (cm)	PHH (m)	NFLH	NSEH	NSLH	HSS (cm)	BW (kg)	BD (cm)	BL (cm)	NHB	NFB	WMF (g)
Min	50	1.1	3	3.5	2.7	61	3.3	48.1	27.6	4.3	46.7	33.4
Max	55	4.6	4.7	5.1	4	103.4	8.7	55.4	58.9	8.9	87	57.9
Mean	52.5	3.9	3.4	4	3.1	68.4	4.2	52.1	32.4	5.2	57.3	39.9
SD	0.72	0.67	0.57	0.6	0.17	3.3	0.6	0.92	2.57	0.43	4.2	2.5
CV	1.37	17.17	16.76	15	5.48	4.82	14.28	1.76	7.96	8.26	7.32	6.26

Cir10 Plant’s circumference at 10 cm, PHH Plant height at the harvest, NFLH Number of functional leaves at harvest, NSEH Number of suckers emitted by the plant at harvest, NSLH Number of suckers with more than 30 cm of length, HSS Height of the tallest or successor sucker, BD Bunch diameter, BL Bunch length, BW Bunch weight, NHB Number of hands per bunch, NFB Number of fingers per bunch, WMF Weight of middle fruit, SD Standard deviation, CV Coefficient of variation, H Heritability

The variation ranges were 50–55 cm for Cir10, 1.1–4.6 m for PHH, 3–4.7 m for NFLH, 3.5–5.1 m for NSEH, 2.7–4 m for NSLH, and 61–103.4 cm for HSS. Likely variation ranges for yield-related traits were: 48.1–55.4 cm for BD, 27.6–58.9 cm for BL, 3.3–8.7 kg for BD, 4.3–8.9 for NHB, 46.7–87 for NFB, and 33.4–57.9 g for WMF. The most productive accessions in terms of bunch weight per season were EC267 (BW = 8.7 kg; NHB = 8.3; NFB = 87.0; WMF = 54.8 g). PHH showed the highest coefficient of variation among all traits, at 17.17%. Moderate heritability was observed for the Cir10 cm trait (62.4–62.7%) and HSS (67.6–69.9%), whereas the remaining growth traits exhibited heritability values below 20% across different accessions. We observed moderate to high broad-sense heritability for BL (32.2 to 39.2%), WMF (38.2 to 47.6%), BD (52.1 to 52.8%), and NFB (56.2 to 80.6%) traits. It was lower than 10% for BW (4.20 to 7.3%) and NHB (5.2 to 7.9%).

Correlation between traits

All traits showed significant (p < 0.05) correlations within and between growth, phenology, and yield-related traits (Fig. 1). Among growth traits, the highest positive association was observed between Circ10 and PHH (R² = 0.58). Concerning the phenological data, the strongest positive correlation (R² = 0.98) was recorded between NDF and LGC. For yield-related traits, NFB and NHB presented the highest correlation (R² = 0.88). The highest correlation between growth and yield-related traits was observed for PHH and BW (R² = 0.49).

Fig. 1.

Correlation between the fifteen agronomic performance traits measured on 238 accessions of sweet fig banana

The size and colour of the circles show the strength of the association, with larger, darker circles indicating values closer to 1 or -1. Circle size reflects the absolute value of the correlation coefficient. Blue represents a positive correlation, while red indicates a negative correlation. LGC: length of the growth cycle; NDF: number of days to flowering, DFF: number of days to fruit filling; Cir10: Plant’s circumference at 10 cm; PHH: plant height at the harvest; NFLH: number of functional leaves at harvest (NFLH); NSEH: number of suckers emitted by the plant at harvest; NSLH: number of suckers with more than 30 cm of length; HSS: height of the tallest or successor sucker; BD: bunch diameter; BL: bunch length; BW: bunch weight; NHB: number of hands per bunch NFB: number of fingers per bunch; WMF: weight of middle fruit.

Genetic diversity and linkage disequilibrium (LD)

The overall expected heterozygosity (He) was 0.03; the observed heterozygosity value (He) was 0.008. The total gene diversity (Ht) was 0.03, and the inbreeding coefficient (Fis) was 0.7.

The genomic pairwise kinship matrix (Additional file 1: Fig. S1) shows the distribution of co-ancestry coefficients in sweet fig banana and splits the accessions into three groups, of which one was largely dominant.

The LD computed on the 345 SNP revealed a decay at a physical distance of around 5. 39 Mb at the critical threshold (r² = 0.1) in the sweet fig banana chromosomes (Fig. 2). The LD is high at short distances (r² = 0.45) and declines rapidly with increasing physical distance. This relatively long-range LD indicates that GWAS mapping resolution is probably moderate, with significant SNPs likely tagging fairly large genomic regions.

Fig. 2.

Linkage disequilibrium (LD) decay plot in sweet fig banana diversity panel

The blue line is the LD decay curve fitted by non-linear least squares. The horizontal red line represents an LD threshold (r² = 0.1). The vertical black line indicates the specific LD decay.

Genetic loci controlling agronomic performance in sweet fig banana

We conducted a GWAS to unravel the genetic architecture of 12 growth- and yield-related traits in sweet fig bananas, using two single-locus and three multiple-locus models. Nine of the twelve traits evaluated showed significant marker-trait associations. Overall, 50 MTAs at 45 unique loci (Table S1) were identified across all nine traits and were significantly associated with at least one trait at the P < 0.01 threshold and an H&B error rate cutoff of 7.10^− 4 across two environments (Dan and Sékou). The results show that two SNPs were identified only in single-locus models, 36 solely in multi-locus models, and seven in both. The five models commonly detected five marker-trait associations, including 18,271,887, 20,341,671, 100,092,739, 100,191,715, and 100,251,706. Moreover, four SNP markers (18237471, 18263591, 100174550 and 100251706) were associated with more than one trait. These SNPs are simultaneously associated with traits such as mother plant and sucker height, sucker number, bunch length, number of hands, and fruit weight.

In this study, we display figures only from models with significant SNPs. Since the two SL models produced identical results, only the figures from the MLM model are included. Similarly, because all SNPs identified by MLMM were also detected by Blink, only Blink’s results are shown.

Marker-trait association for the growth traits

The GWAS scans revealed 36 distinct MTAs significantly linked to growth traits at 34 distinct loci (Table S1, Figs. 3, 4, 5, 6, 7 and 8, Fig. S3-8). For the plant’s circumference (Cir10) trait, a single marker 100,157,562 (position 3248877, chromosome 8) was co-detected only by BLINK and FarmCPU models (Table S1, Fig. 3, Fig. S2). This marker explained 67.3% of the phenotypic variance of circumference. The additive effect value was − 1.1.

Fig. 3.

Manhattan plots of the SNP-based association mappings for the circumference of sweet fig banana plants (Cir10) with BLINK and FarmCPU models

Fig. 4.

Manhattan plots of the SNP-based association mappings for plant height at harvest (PHH) in 238 sweet fig banana accessions with MLM, BLINK and FarmCPU

Fig. 5.

Manhattan plots of the SNP-based association mappings for the number of functional leaves at the harvest trait (NFLH) with BLINK and FarmCPU models

Fig. 6.

Manhattan plots of the SNP-based association mappings for the number of suckers emitted (NSEH) trait with MLM and BLINK models

Fig. 7.

Manhattan plots of the SNP-based association mappings for the number of suckers with more than 30 cm of length (NSLH) trait with MLM, BLINK and FarmCPU models

Fig. 8.

Manhattan plots of the SNP-based association mappings for the height of the tallest/successor sucker trait (HSS) with MLM, BLINK and FarmCPU models

The five GWAS methods used in this study identified 21 distinct SNPs significantly associated with PHH (Table S1; Fig. S3). The Manhattan plots (Fig. 4) showed significant SNP associations with PHH across all five models. These SNPs were found on nearly all eleven chromosomes, except chromosomes 8 and 10. The trait variation explained by each SNP varied from 0 to 56.7%. The additive effect varied from − 1.30 to 11.90. The markers 18271887,20341671 and 100,092,739 were co-detected by all five models. The marker 18,271,887 (position 5248569, chromosome 1) explained the highest percentage (56.7%) of the total phenotypic variance in PHH. The additive effect value of the M1 was − 17.54.

Considering the number of functional leaves (NFLH) trait, the BLINK and FarmCPU models simultaneously identified the SNP 18,312,344 (position 40702841, Chromosome 6) as a single significant marker-trait association (Table S1; Fig.S4; Fig. 5). This SNP explained the phenotypic trait variation at 35.9%. The additive effect value was 0.15.

One significant marker, 100,251,706 (position 21443886, chromosome 6) (Table S1; Fig.S5; Fig. 6), was found associated with the number of suckers emitted (NSEH) by both single and multiple locus models. This marker explained up to 85.5% of the phenotypic variation of the NSEH trait. The additive effect ranged from − 0.38 to -0.42.

Two marker-trait associations were identified for the number of suckers with more than 30 cm of length (NSLH) (Table S1; Fig.S6; Fig. 7). The SNPs 100,218,597 (position 26588220, chromosome 11) and 100,251,706 (position 21443886, chromosome 6). The phenotypic variance explained by SNP M55 ranges from 2.7% to 16.3%, with an additive effect of -0.28 to -0.31. The second SNP, M58, was simultaneously detected by all five GWAS models. The marker 100,251,706 explained 4.10^− 9 to 86.1% of the phenotypic variation in NSLH. The additive effect varied from − 0.36 to -0.54.

Regarding the height of the tallest sucker (HSS) trait, seven SNPs on chromosomes 1, 3,6, 7, 8, and 11 (Table S1; Fig.S7; Fig. 8) were found to be linked with the trait through five different models. These SNPs explained between 0% and 86.16% of the phenotypic variance. The additive effect ranged from − 7.09 to 34.98. The models did not identify any common SNPs for this trait.

Marker-traits association for yield-related traits

For the yield-related traits, 18 significant major trait associations (MTAs) at 16 loci were identified. Eleven major SNPs associated with banana bunch length (BL) were in different genomic regions on chromosomes 1,4,5 and 8 (Table S1; Fig.S8; Fig. 9). They explained the phenotypic variance from 0.09 to 43.4%. The additive effect varied from − 1.67 to 12.98. The marker 100,191,715 (position 36067757 on chromosome 4) was co-detected by all five models and explained 18.14–43.4% of phenotypic variance. The additive effect ranged from 3.05 to 10.33.

Fig. 9.

Manhattan plots of the SNP-based association mappings for the bunch length (BL) in 238 sweet fig banana accessions, with MLM, BLINK and FarmCPU models

Two SNPs were associated with the number of hands per bunch (NHB). These SNPs (100251706 and 18237471) were distributed across chromosomes 1 and 4 (Table S1; Fig. S9; Fig. 10). The phenotypic variance explained ranged from 1.28 to 38.94%, and the additive effect ranged from − 0.33 to 0.26.

Fig. 10.

Manhattan plots of the SNP-based association mappings for the number of hands per bunch (NHB) in 238 sweet fig banana accessions, with MLM and FarmCPU models

Five significant SNPs on chromosomes 4, 6, 8, and 11 were associated with fruit weight (Fig. 11; Table S1; Fig. S10). The proportion of phenotypic variation explained by these significant marker-trait associations ranged from 0.1% to 92.3%. The marker 20,351,495 (position 17276495, chromosome 11) was detected by four models, with phenotypic variance reaching 92.3%. The additive effect ranged from − 4.69 to 6.16.

Fig. 11.

Manhattan plots of the SNP-based association mappings for fruit weight (WFM) in 238 accessions of sweet fig banana with MLM, BLINK and FarmCPU models

Putative candidate genes associated with MTAs

To explore the potential role of MTA in growth and yield in sweet fig banana, genes located in or near significant SNPs were identified based on their positions and sequences in DH Pahang (v4), a reference genome, using databases such as UniProt and NCBI. Table S2 lists candidate genes with functional annotations from these databases for each trait category. Of the 45 significant SNPs associated with traits of interest, 23 had unknown functions, 4 showed no significant similarity with previous sequences, and 2 remained uncharacterized. The other 16 SNPs match homologous genes mainly involved in metabolic processes, cellular transport, transcriptional regulation, plant organ development, growth regulation, stress responses, and gene expression. The putative candidate gene GSMUA_302750.1 for SNP “18237471” has the molecular function “Protein serine/threonine kinase activity” and is linked to plant height, bunch length, and the number of hands per bunch.

Discussion

This study explores the genomic factors affecting growth and yield-related traits in a panel of 238 sweet fig banana landrace accessions. The findings provide insights into the genetic architecture and loci associated with these key traits.

Phenotypic variation in growth and yield-related traits

Our findings confirmed the hypothesis of significant phenotypic variation among the sweet fig banana accessions used as GWAS panels for each targeted trait. This trend of substantial variation across all growth, phenology, and yield-related traits during the growing season has been reported in previous research on banana cultivar performance [48–50]. The notable differences in yield and cycle duration observed between and within banana germplasms across cropping cycles are influenced by cultivars, planting material, plant density, environmental factors, altitude, pests, and diseases [51, 52]. Hence, the variability in agronomic performance traits among sweet fig banana accessions can be attributed to differential gene expression regulating these traits in response to environmental factors throughout the experiment. Furthermore, moderate to high broad-sense heritability was observed for girth, successor suckers’ height, number of fingers per bunch, bunch diameter, and finger weight. These results indicate that a significant portion of the phenotypic variation is heritable and can be linked to genotypic differences. However, collecting data over two consecutive crop cycles would have provided more comprehensive information for further analysis of stability, heritability, and the origin of phenotypic variations. Additionally, growth traits showed some predictive ability for most yield-related traits, as indicated by moderate positive correlation coefficients. These findings differ somewhat from those of Dagnew et al. [49] and Sora & Jibat Guji [53], who reported weak associations between growth and yield parameters across various banana cultivars. The strong, significant positive correlations between the number of hands per bunch, the number of fingers per bunch, finger weight, and bunch weight demonstrate that improving these traits can enhance yield. These results align with previous studies by Batte et al. [48] and Stanislas et al. [54]. The first study examined how growth parameters relate to the yield potential of six plantain landraces and identified the number of fruits and hands per bunch as key traits directly impacting yield. The second, based on allometric studies of forecasted yield for two edible diploid Musa cultivars, shows that the number of fruits per bunch, the number of hands per bunch, fruit length, and pseudo-stem volume can effectively predict yield.

Regarding the relationship between phenology and yield-related traits, the length of the crop cycle showed a moderately significant negative correlation with bunch weight, the number of hands per bunch, the number of fingers per bunch, and bunch length. Conversely, finger weight and bunch diameter had a positive correlation with the development cycle length. These results suggest that accessions with longer cycle durations produce smaller bunches with fewer hands and fingers, but with larger fingers. Our results contradict those of Batte et al. [48] and Stanislas et al. [54], who argued that a longer development cycle positively affects bunch weight and the number of fingers and hands, but negatively affects finger size in banana plantains and EAHB.

The phenotypic variation and the significant broad-sense heritability observed in sweet fig bananas highlight the importance of these resources in selecting genotypes with desired traits, breeding, and identifying genes that control these traits.

Linkage disequilibrium

To accurately identify genes influencing traits, we analysed how alleles are non-randomly associated by estimating LD. The rapid decline in LD, followed by long-range LD, observed in sweet fig banana germplasm aligns with previous findings from banana GWAS studies by Sardos et al. [25], Nyine et al. [26], and Osorio-Guerin et al. [27]. This relatively long-range LD is consistent with the reproductive biology and demographic history typical of outcrossing species [55], including clonal propagation, limited effective recombination, population structure, and historical bottlenecks [56]. From a GWAS perspective, in cases of long-range LD, significant MTAs are better interpreted as tag markers for genomic regions under genetic control rather than definitive causal loci. Long-range LD also increases the likelihood of co-localized MTAs across traits, which may be due to pleiotropy or to tightly linked, functionally distinct genes within the same LD block [57].

Genetic loci and candidate genes controlling agronomic performances in sweet fig banana

GWAS analysis identified 50 MTAs and 45 distinct SNPs distributed across diverse chromosomes, all of which were significantly associated with growth- and yield-related traits. These outputs corroborated our hypothesis that growth and yield-related trait expression in sweet fig banana has a genetic basis. Four SNPs are associated with several traits, such as mother plant and sucker height, sucker number, bunch length, number of hands, and fruit weight. However, the moderate LD extent observed in this study suggests that regions linked to multiple trait loci could be either pleiotropic or composed of tightly linked genes [58]. Differentiating these possibilities requires comprehensive multi-trait and functional analyses.

According to the genetic architecture of the examined traits, growth parameters, including circumference, number of leaves, and suckers, exhibited a simple genetic architecture, with only one locus showing a minor negative or positive effect. In contrast, the heights of the plants and the suckers displayed a complex genetic architecture, characterized by multiple loci with small effects. Likely, the complex genetic architecture underlying the yield-related traits. For the yield traits, 18 marker-trait associations distributed on six distinct chromosomes were identified by the five models in our study. Our results were in agreement with those reported by Osorio-Guarin et al. [27] who used SL and ML-GWAS methods to detect candidate genes associated with morphology, fruit quality, and yield in 124 diploid, triploid, and tetraploid bananas (A and B genomes). This previous study identified 23 SNPs associated with yield in A-genome bananas and demonstrated that the targeted traits, especially yield-related traits, were controlled by multiple genes distributed across different chromosomal regions. Yield is a complex quantitative trait governed by a network of many small-effect minor genes [59]. On the other hand, Nyine et al. [26] identified 25 significant markers associated with bunch weight and component traits, mainly localized on chromosome 3, by performing the MLM (SL-GWAS) model in a panel of 307 triploid EAHB bananas (A group) and concluded a few large-effect QTLs in the studied population control that yield. This contrast in results would be explained by the five comparative GWAS models (SL and ML) applied in this study and by the three models used by Osorio-Guarin et al. [27]. ML-GWAS models demonstrated superior performance and accuracy compared with SL-GWAS, offering the potential to identify novel, reliable loci [32, 35]. Furthermore, in polygenic traits, long-range LD indicates that markers stay correlated over large genomic distances, forming extended haplotype blocks [60]. This can lead GWAS to detect fewer signals than loci, as multiple loci are statistically confounded within the same LD region. According to Qian et al. [61], extended LD may inflate SNP effects since significant markers often reflect the influence of several linked loci rather than a single causal variant. Therefore, MTAs for polygenic traits could be considered haplotypic effects, and gene identification should consider broad genomics intervals around associated SNPs.

Implications and perspectives for sweet fig banana breeding

Our findings revealed that multiple loci/SNPs with small effects-controlled mother-plant and suckers’ height growth and yield-related traits. The complex genetic architecture of the sweet fig banana genome, as revealed by genome-wide scanning calls, first calls for balancing phenotypic trade-offs by selecting loci with favourable or synergistic effects on yield. This selection of favourable, validated functional loci enabled improvements in traits beyond those initially targeted [62]. Multi-trait analyses could be conducted to improve the ability to identify SNPs and detect pleiotropy [63]. Further studies should include a larger population of sweet banana sizes to ensure that significant associations are reliable and relevant to a broader breeding population. Furthermore, epistasis interactions also influence polygenic traits, and their bias in trait prediction slows down the rate of crop improvement [64]. Hence, the need to further elucidate the form, function, and extent of epistasis interactions and pleiotropy effects in polygenic trait variation will help breeders predict and control trait expression to develop relevant selection strategies. Therefore, a better understanding of the complex and subtle interactions underlying the genetic architecture will help accurately identify gene combinations for selection in breeding programs to develop cultivars with superior traits. The favourable SNPs/genes linked to the targeted loci in this study could be utilized for introgression or pyramiding in newly developed cultivars. Likely, the marker-trait associations identified in our study could be used for genomic or marker-assisted selection to accelerate genetic improvement in sweet fig banana yield.

Conclusion

Sweet fig banana has significant economic and nutritional value for grassroots communities in Benin, yet its productivity remains low. Deciphering the architecture of genomic regions regulating growth and yield-related traits facilitates the development of new crops with optimised performance. This study performed two single-locus (SL-GWAS) and three multi-locus (ML-GWAS) GWAS models on a panel of 238 sweet fig banana accessions, phenotyped for 12 growth- and yield-related traits. Significant variability, moderate to high heritability and correlation were observed for most traits. GWAS based on BLUP identified 50 significant marker-trait associations (MTA) at 45 loci across eleven chromosomes. Several loci with relatively minor effects were identified as regulating plant height and yield-related traits. Our findings indicated that the productivity of sweet fig banana could potentially be improved through molecular breeding for plant height, number of functional leaves, bunch length, number of hands per bunch, and fruit weight as target traits in the development of new high-yield varieties. However, a prior functional analysis should be conducted to validate the influence of the detected SNPs on traits of interest in sweet fig banana and other banana cultivars. These could facilitate improvements in banana yield and other desirable characteristics. The present study offers first insights into the architecture of genomic regions regulating a range of traits of interest in sweet fig banana, thereby paving the way for molecular breeding in bananas.

Authors’ contributions

DBEC and EGAD conceptualized the study and designed the methodology. DBEC and JH conducted field experiments and collected data. DAT and DBEC performed data analysis. DBEC wrote the manuscript. EGAD, DAT, DEOS and VNFH review & editing of the manuscript. EGAD supervises the research. All authors reviewed and approved the final manuscript version.

Funding

The author(s) received no specific funding for this work.

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.