Availability of phenotypic simulation for fruit-related traits in F1 progenies of chili peppers (Capsicum annuum) using genomic prediction based solely on parental information

doi:10.21203/rs.3.rs-4945066/v1

Download PDF

Research Article

Availability of phenotypic simulation for fruit-related traits in F1 progenies of chili peppers (Capsicum annuum) using genomic prediction based solely on parental information

https://doi.org/10.21203/rs.3.rs-4945066/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Chili pepper (Capsicum spp.) fruits are used as vegetables, spices, and ornamental plants, necessitating various fruit characteristics. However, their genetic improvement is challenging through conventional cross-breeding due to the quantitative traits, which makes it difficult to predict phenotypes in the progeny. As a breakthrough, we focused on phenotypic simulation via genomic prediction (GP) and aimed to clarify its utility for fruit-related traits in chili peppers. The present study used 291 C. annuum accessions, including two populations: inbred lines and F₁ accessions derived from 20 inbred parents. We collected data of fruit length, width, shape index (length/width), weight, and pericarp thickness, and obtained single nucleotide polymorphism data via multiplexed inter-simple sequence repeat genotyping by sequencing. We simulated the fruit-related traits in the F₁ accessions by inputting their estimated genotypes (based on their parents) into the GP model using the GBLUP-GAUSS model, which was shown to be the most accurate regardless of population or trait differences in the present study. As a result, we observed strong positive correlations (r = 0.833 - 0.908) between the simulated and observed phenotypic values across all traits, suggesting that accurate ranking of F₁ progenies based on fruit-related traits can be achieved using parental information. This is the first report demonstrating the utility of phenotypic simulation via GP in chili pepper breeding, offering valuable insights for its application in this field.

genomic prediction

MIG-seq

chili pepper

phenotypic simulation

GBLUP-GAUSS

Phenotypic simulation of fruit-related traits in the F1 progeny of chili peppers was conducted using genomic prediction based solely on parental information, and the accuracies were assessed.

Chili pepper (Capsicum spp.) is one of the major autogamous crops in the Solanaceae family, with annual global production exceeding 38 million tons (FAOStat 2019). This underscores the economic importance of chili peppers, which have garnered significant attention from farmers and marketers worldwide. The various uses of chili peppers contribute to this importance. Unlike other major Solanaceae vegetables such as tomato (Solanum lycopersicum) and eggplant (S. melongena), whose fruits are primarily used as vegetables, chili peppers produce capsaicinoids, a type of pungent compound (Naves et al. 2019). As a result, chili peppers are not only used as vegetables, like sweet peppers, but also as a spice. Additionally, chili pepper plants with fruits are often used as ornamental plants. Thus, the diverse applications of chili peppers contribute to their economic significance. Given these various uses, a range of fruit characteristics is required. For instance, large-sized fruits with thick pericarps are desirable for vegetable use, with rounded or blocky shapes being preferable. In contrast, elongated small fruits with thin pericarps are favored for spice production because they dry easily and are suitable for making powders. Small fruits with round or elliptical shapes are necessary for breeding ornamental cultivars. Therefore, fruit-related traits (such as fruit size, shape, and pericarp thickness) are crucial in chili pepper breeding.

Due to their importance, genetic analyses of fruit-related traits have been extensively studied in previous research (Chaim et al. 2001; Han et al. 2016; Naegele et al. 2016; Lee et al. 2020; McLeod et al. 2023). These studies revealed that size-related traits (length, width, weight), shape, and pericarp thickness are quantitative traits, with phenotypic values determined by the genetic effects of multiple loci and environmental factors. Several genetic analyses have identified responsible genomic loci distributed across the genome through quantitative trait loci (QTL) analysis and genome-wide association studies (GWAS) (Chaim et al. 2001; Han et al. 2016; McLeod et al. 2023). However, fruit-related traits exhibit complex inheritance patterns involving numerous minor QTLs, making it impossible to explain the phenotypic values solely based on these identified QTLs. As a result, in crossbreeding, including F₁ hybrid and pedigree breeding of chili peppers, predicting fruit traits in progenies is challenging, and optimizing parental combinations to produce progenies with desirable fruit traits is difficult. Even in current breeding systems, this process largely relies on the intuition and experience of skilled breeders.

Genomic prediction (GP) is a method used to predict phenotypic values based on genotypic information using mathematical functions and machine learning approaches (Meuwissen et al. 2001). With this technique, phenotypic values can be estimated at the seedling stage without the need for cultivation and phenotypic evaluation, making GP an attractive new breeding strategy for various crops (Heslot et al. 2015; Spindel and McCouch 2016). Recently, the utility of GP has also been reported for fruit and yield-related traits and pungency levels in chili peppers (Hong et al. 2020; Kim et al. 2022; Tong et al. 2022; Lozada et al. 2023). Furthermore, it is theoretically possible to simulate progeny phenotypes based solely on parental lines. In this approach, the GP model is first developed using the phenotypic and genotypic data of the parental lines. The genotypes of the crossed progenies are then estimated from the genotypic data of their parental lines, and the progeny phenotypes are simulated by inputting these estimated genotypes into the constructed GP model. Empirical studies of this approach have been conducted in several horticultural crops, such as Japanese pear (Pyrus pyrifolia), tomato, and strawberry (Fragaria × ananassa), with promising results (Iwata et al. 2013; Yamamoto et al. 2017, 2021). If this simulation technique can be applied to fruit-related traits in chili peppers, it could simplify the optimization of parental combinations in crossbreeding to efficiently produce progenies with desired traits, even for breeders with limited experience.

For this approach to be feasible, several issues must be addressed. In chili peppers, GP for fruit-related traits in inbred populations (with high homozygosity) has already been conducted in previous studies (Hong et al. 2020), but it has not been applied to populations with relatively high heterozygosity, such as F₁ lines. Therefore, the accuracy of GP and the optimal model for different populations need to be identified. Additionally, no studies have performed phenotypic simulations via GP in chili peppers, so the accuracy and utility of this approach must be explored. Therefore, this study aimed to clarify these aspects by simulating fruit-related traits in F₁ progenies as a preliminary trial.

Plant materials

A total of 291 C. annuum accessions were used in this study, consisting of two types: inbred accessions (n = 132) and F₁ accessions (n = 159). The inbred accessions included commercial cultivars and plant genetic resources from Asia and Central America, maintained either at the Agricultural Department of Shinshu University, Japan, or the Gene Bank of the National Agriculture and Food Research Organization, Japan. The F₁ accessions were obtained through test crosses among 20 of the 132 inbred accessions. Detailed information about the inbred accessions and parental combinations of the F₁ accessions can be found in Supplemental Dataset 1 and Fig. S1. These accessions were cultivated at the Faculty of Agriculture, Shinshu University, Minamiminowa, Nagano, Japan, over three years (2021–2023). Each year, one plant was cultivated, and five mature green fruits were used to collect phenotypic data on fruit-related traits.

Experimental design

This study was carried out in three steps: 1) Data collection (Fig. 1a), 2) Evaluation of GP model accuracy (Fig. 1b), and 3) Simulation of F₁ phenotypic values (Fig. 1c). First, we collected phenotypic data on five fruit-related traits: fruit length (FL), fruit width (FD), fruit shape index (FI) (FL/FD ratio), fruit weight (FWT), and thickness of pericarp (THP) from all accessions over three years. Concurrently, we also obtained single nucleotide polymorphism (SNP) data using multiplexed inter-simple sequence repeat genotyping by sequencing (MIG-seq). Using these data, we performed genomic predictions (GP) for the fruit-related traits across three populations (all accessions, inbred, and F₁) using 11 genomic models. The accuracy of each model was evaluated within each population. Finally, using the most accurate model, we conducted a phenotypic simulation of the F₁ accessions based solely on the information from the inbred lines, including their parental accessions, and assessed the accuracy of the simulation. Detailed methods for each procedure are described below.

Phenotypic and genotypic data collection

For phenotypic data, FL and FD were measured with a ruler, and FI was calculated as the ratio of FL to FD. FWT and THP were measured using an electronic balance (FA-2000, A&D Co., Ltd., Tokyo, Japan) and a caliper, respectively. The mean value for each trait was calculated from five fruits, with data collected over three years (2021–2023). To minimize environmental effects across the three years, we calculated the best linear unbiased prediction (BLUP) values using a mixed linear model in the R package ‘lme4’; genotypes (accessions) and year were set as random and fixed effects, respectively, as described by Minamikawa et al. (2017) For genotypic data, genomic DNA was extracted from young leaves using the DNeasy Plant Pro Kit (Qiagen, Hilden, Germany) and subjected to MIG-seq analysis by GENODAS Co., Ltd. (Miyagi, Japan). Briefly, the first polymerase chain reaction (PCR) was performed with the primer set described by Suyama and Matsuki (2015) to amplify the inter-simple sequence repeat (ISSR) region. The second PCR added barcode and adaptor sequences. DNA fragments longer than 300 bp were selected, and pair-end sequencing was performed on a NextSeq500 (Illumina, San Diego, CA, USA). After sequencing, we trimmed adapters, conducted quality control, mapped reads to the reference genome, and called variants following the procedures described in our previous study (Kondo et al. 2023), using the C. annuum reference genome ‘Takanotsume’ (Shirasawa et al. 2023). We obtained variant call format (VCF) files of SNPs, extracted bi-allelic SNP loci with a genotyping ratio above 70%, imputed missing SNP loci using Beagle (beagle.22Jul22.46e.jar; Browning and Browning 2016), and removed loci with a minor allele frequency below 0.05. The genotypic values at each SNP locus were then converted to numeric values (-1, 0, 1).

Genetic statistical analysis

We calculated the number and density of SNP loci detected via MIG-seq for each chili pepper chromosome. Using these genotypic data, two population genetic analyses were conducted. The first was phylogenetic and admixture analysis of the inbred accessions (n = 132). We calculated genetic distances among the inbred accessions using Euclidean distance and constructed a phylogenetic tree with the neighbor-joining method using the R packages ‘ape’ and ‘ggtree’. We also estimated the proportion of ancestral groups in the inbred accessions using ADMIXTURE v1.3 (Alexander et al. 2009), with the number of ancestral populations (K) ranging from 1 to 10, and calculated cross-validation (CV) errors for each K. The second analysis was principal component (PC) analysis for all accessions, calculating PC1, PC2, and PC3 scores. We also determined the heterozygosity ratio across all SNP loci.

Phenotypic statistical analysis

We calculated the mean, median, minimum, and maximum values for each trait. Broad-sense heritability (h_b²) for all accessions was calculated as follows: \(\:{h}_{b}^{\:2}=\:\frac{{\sigma\:}_{g}^{\:2}}{({\sigma\:}_{g}^{\:2}+{\sigma\:}_{\epsilon\:}^{\:2}/n)}\times\:100\) described by Schmidt et al. (2019), where σ_g² and σ_ε² represent genetic and environmental variances, respectively, obtained from BLUP and residue values, with n denoting the number of cultivation replicates (3 years). Pearson’s correlation coefficients were calculated to evaluate the relationships among the five fruit-related traits using BLUP values from all accessions. In the F₁ accessions, additive and dominance effects were calculated as follows: \(\:\text{A}dditive\:effect=\frac{\left|{P}_{P1}-{P}_{P2}\right|}{2}\), \(\:Dominance\:effect={P}_{F1}-\frac{{P}_{P1}+{P}_{P2}}{2}\), as described by Ukai (2002), where P_P1, P_P2, and P_F1 denote the phenotypic values of the mother parent, father parent, and F₁ progeny, respectively.

Genomic prediction for five fruit-related traits

Genomic predictions (GP) for the five fruit-related traits were performed in all accessions (n = 291), inbred accessions (n = 132), and F₁ accessions (n = 159), using 11 models: 1) Ridge, 2) Least Absolute Shrinkage and Selection Operator (LASSO), 3) Elastic Net (EN), 4) Genomic Best Linear Unbiased Prediction (GBLUP)-A, 5) GBLUP-AD, 6) GBLUP-GAUSS, 7) Random Forest (RF), 8) Bayesian Ridge Regression (BRR), 9) Bayes A, 10) Bayes B, and 11) Bayes C. Ridge, LASSO, and EN models were implemented using functions in the R package ‘glmnet’. GBLUP models were implemented as described by Ishimori et al. (2020), with additive relation distance matrix (A), dominance genomic distance relationship (D), and Gaussian kernel distance matrix (K) calculated using the R package ‘RAINBOWR’, and models developed using ‘EMMREML’. For GBLUP-A and -GAUSS, A and K were used as single kernels, respectively, while A and D were used for multiple kernels in GBLUP-AD. BRR, Bayes A, Bayes B, and Bayes C models were implemented using the R package ‘BWGS’. GP accuracy was assessed by calculating Pearson’s correlation coefficient between observed (BLUP) and predicted values using 10-fold CV. The mean values of the ten folds were treated as the representative values for each run of the 10-fold CV, and this process was repeated ten times. The mean values of these repetitions were compared across the 11 models.

Phenotypic simulation in the F₁ progenies

Using the GBLUP-GAUSS model, we developed a GP model based solely on the inbred accessions (n = 132), as previously described. We estimated the genotypic values of F₁ accessions (n = 159) based on their 20 inbred parents using the following formula: \(\:{g}_{F1}=\frac{{g}_{P1}+{g}_{P2}}{2}\), where gF₁, g_P1, and g_P2 represent the vectors of numeric genotypic values (-1, 0, 1) for the F₁ progeny, mother parent, and father parent, respectively. In this study, we did not account for the segregation of heterozygous loci in the F₁ generation. Therefore, when fractions (-0.5 and 0.5) were detected in the calculated gF₁ vector due to heterozygous genotypes in either parent, they were converted to -1 and 1, respectively. Using the developed GP model and the estimated genotypic values, we simulated the phenotypic values for the F₁ progenies without actual phenotypic and genotypic information of the F₁ accessions. To assess accuracy, we examined Pearson’s correlation coefficient between the observed and simulated values. To explore the factors contributing to simulation error, we first calculated the root square error (RSE) between the observed and simulated values for each F₁ accession. We then investigated Pearson’s correlation coefficient between the RSE and two statistical values: phenotypic values of the F₁ progeny and the absolute values of dominance effects. For comparison, the same values were calculated for the GP models using all (inbred + F₁) and only F₁ accessions. In these cases, RSE values for each F₁ accession were derived from observed and predicted values obtained through ten repetitions oF₁0-fold CV in the GPs. The mean RSE values of these ten replicates were used as the representative values for each F₁ accession.

Additionally, we simulated the fruit length (FL) and fruit diameter (FD) in the F₁ progenies derived from all possible combinations (₁₃₂C₂ = 8,646 combinations) of inbred accessions using the same method as described above, and presented these as scatter plots. In 2023, we cultivated eight new F₁ lines (F₁A - F₁H) along with six of their parental lines (P1 - P6) at the Faculty of Agriculture, Shinshu University. These parental lines were different from the 20 parents of the F₁ accessions. We collected the mean FL and FD values from five fruits of multiple plants (parental lines: n = 3, F₁ lines: n = 5). Detailed information about the accessions is provided in Table S1 and Supplemental Dataset 1. We then compared the mean values with the simulated values.

Population and genetic structure of the plant materials

A total of 3,194 SNPs were detected through MIG-seq in the 291 accessions. Figure 2a illustrates the SNP locus density (number of SNPs per 5 Mbp) across each chromosome, showing their genome-wide distribution. On average, 266 SNPs were detected per chromosome, with a density oF₁.0 SNPs per 1 Mbp (Table S2). Using these SNP data, we analyzed the population structure of the plant materials. Figure 2b displays the phylogenetic tree and the proportions of ancestral populations at K = 6 from the admixture analysis, where the mean values of CV errors were smallest at K = 6 (Fig. S2). The population structure of the inbred lines (n = 132) was characterized by proportions of six distinct populations, which largely corresponded to the clustering observed in the phylogenetic tree. The 20 parental accessions (indicated by black arrows in Fig. 2b) of the F₁ accessions also exhibited diverse population structures. In the F₁ accessions, these were placed to be surrounded by their parental accessions within the PC1-PC2 and PC2-PC3 scatter plots (Fig. 2c). Additionally, the mean heterozygosity ratio in F₁ accessions was 2.3 times higher than that in inbred lines (Fig. 2d).

Phenotypic characteristics of the plant materials

For this study, five fruit-related traits (FL, FD, FI, FWT, THP) were collected from all accessions over three years. Broad-sense heritabilities (h_b²) exceeded 90% for all traits (Table 1). Among these, FD and THP exhibited the highest and lowest heritabilities, respectively, with values of 98.1% and 90.6%. Correlations among the traits varied, with both weak and strong relationships observed (Fig. 3). Notably, FWT exhibited a strong positive correlation with FD (r = 0.925, P < 0.001) and THP (r = 0.736, P < 0.001). Conversely, the correlation between FL and FD was the weakest (r = 0.172, P < 0.01). Regarding inheritance-related statistical values in F₁ accessions, dominance effects were predominantly positive for FL and FI, but negative for FD, FWT, and THP (Fig. 4). This suggests that the phenotypic values of F₁ progenies tended to be closer to those of higher parents for FL and FI, but to lower parents for FD, FWT, and THP. Additionally, significant negative correlations were found between additive and dominance effects across all traits (Fig. 4), indicating that negative dominance effects increased with differences in phenotypic values among the two F₁ parents. Strong correlations were particularly observed in FD (r = -0.897, P < 0.001) and FWT (r = -0.941, P < 0.001).

Table 1

Statistics and heritability estimates for five fruit-related traits across all plant materials
Trait	Statistics of phenotypic (BLUP ^a) values						h_b² (%)^b
Trait	Mean	±	SD	Median	Maximum	Minimum	h_b² (%)^b
FL (cm)	9.0	±	3.3	8.5	18.2	1.6	96.0
FD (cm)	2.2	±	1.3	1.8	8.1	0.6	98.1
FI	4.9	±	2.5	4.8	15.1	0.7	97.7
FWT (g)	12.8	±	12.9	8.5	93.2	1.1	96.7
THP (mm)	1.59	±	0.59	1.49	5.19	0.68	90.6
^a Best linear unbiased prediction. ^b Broad-sense heritability with a formula described by Schmidt et al. (2019)

Genomic prediction accuracy

To assess genomic prediction (GP) accuracies and their variations based on population, trait, and model, we performed GPs for five fruit-related traits across three populations: all (inbred + F₁), inbred, and F₁ accessions. These populations differed in terms of individual numbers, pedigrees, and heterozygosity levels (Fig. 2a, -b, -c). Accuracy was evaluated using Pearson’s correlation coefficient between observed and predicted values. Table 2 shows the GP accuracies for each population, trait, and model. In general, GP accuracies were highest in the F₁ accessions and lowest in the inbred lines, regardless of the model or trait. For each trait, the mean Pearson’s correlation coefficients across all models were highest for the F₁ accessions (r = 0.787–0.893) and lowest for the inbred accessions (r = 0.587–0.798). Among the five traits in each population, FD and FWT were more accurate traits in the inbred accessions (FD-inbred: r = 0.798, FWT-inbred: r = 0.705), while FI and FL were more accurate traits in the F₁ accessions (FI-F₁: r = 0.893, FL-F₁: r = 0.888), based on the mean Pearson’s correlation coefficients. Conversely, GP accuracies for FI and FL were lower in the inbred accessions (FI-inbred: r = 0.587, FL-inbred: r = 0.643), whereas FD and FWT had lower accuracies compared to FI and FL in the F₁ accessions (FD-F₁: r = 0.852, FWT-F₁: r = 0.787). In the overall population, FD and THP exhibited the highest and lowest accuracies, respectively, and their GP accuracy trends were not entirely consistent with neither inbred nor F₁ accessions. Lastly, GP accuracies varied depending on the model. Notably, GBLUP-GAUSS provided the highest accuracies for most traits and populations, except for FD in the inbred accessions. GBLUP-A and GBLUP-AD also achieved high accuracies but did not surpass GBLUP-GAUSS. Thus, GBLUP-GAUSS was determined to be the most optimal model. In contrast, RF and LASSO were generally the least accurate models in the inbred and F₁ accessions, respectively.

Table 2

Accuracya of genomic predictions (GPs) for fruit-related traits across three distinct populations.
Population	Trait	GP model
Population	Trait	Ridge	LASSO	EN	GBLUP -A	GBLUP -AD	GBLUP -GAUSS	RF	BRR	Bayes A	Bayes B	Bayes C	Mean
All (n = 291)	FL	0.770^a	0.751	0.749	0.776	0.782	0.804	0.732	0.787	0.786	0.775	0.786	0.772
	FD	0.856	0.832	0.838	0.846	0.848	0.876	0.814	0.869	0.869	0.814	0.869	0.848
	FI	0.750	0.723	0.727	0.758	0.767	0.791	0.664	0.757	0.768	0.741	0.775	0.747
	FWT	0.802	0.758	0.741	0.777	0.789	0.831	0.729	0.782	0.765	0.658	0.775	0.764
	THP	0.750	0.681	0.697	0.741	0.755	0.774	0.703	0.749	0.734	0.709	0.735	0.730
Inbred (n = 132)	FL	0.658	0.598	0.615	0.660	0.661	0.697	0.601	0.647	0.642	0.641	0.650	0.643
	FD	0.799	0.708	0.742	0.802	0.813	0.815	0.793	0.847	0.845	0.767	0.846	0.798
	FI	0.620	0.517	0.543	0.613	0.627	0.648	0.523	0.596	0.606	0.558	0.610	0.587
	FWT	0.741	0.660	0.660	0.729	0.723	0.779	0.703	0.734	0.705	0.586	0.737	0.705
	THP	0.685	0.560	0.586	0.661	0.685	0.700	0.604	0.669	0.644	0.469	0.637	0.627
F₁ (n = 159)	FL	0.895	0.856	0.863	0.907	0.906	0.915	0.825	0.903	0.905	0.892	0.903	0.888
	FD	0.884	0.823	0.836	0.892	0.887	0.893	0.757	0.863	0.847	0.827	0.859	0.852
	FI	0.896	0.859	0.872	0.910	0.908	0.915	0.831	0.907	0.910	0.909	0.911	0.893
	FWT	0.839	0.738	0.759	0.841	0.841	0.851	0.673	0.793	0.787	0.737	0.792	0.787
	THP	0.850	0.776	0.800	0.832	0.838	0.840	0.791	0.834	0.836	0.838	0.838	0.825
^a Mean Pearson’s correlation coefficients calculated from ten repetitions of 10-fold cross-validation.

Accuracy of phenotypic simulation for F₁ accessions

We simulated the phenotypic values of five fruit-related traits in the F₁ accessions using genomic predictions (GPs). For the simulation, we utilized phenotypic and genotypic data from only the inbred accessions, including the F₁ parents, and constructed GP models using the GBLUP-GAUSS model, which had demonstrated high accuracy across all populations and traits. Genotypic values for the F₁ accessions were estimated from their parents and input into the GP models to generate phenotypic predictions for each F₁ accession. The simulated values exhibited significant positive correlations (r ≧ 0.833) with the observed values across all fruit-related traits (Fig. 5a). Specifically, FI and FL showed strong correlations (FI: r = 0.908, FL: r = 0.901), while FWT had the lowest correlation (r = 0.833) among the five traits. These results suggest that the ranking of simulated phenotypic values among the F₁ accessions was generally consistent with the observed values. For reference, Figs. 5b and 5c present the mean values of predicted phenotypic values in the F₁ accessions, obtained from ten iterations oF₁0-fold CV of GPs with both all accessions and only F₁ accessions, respectively. Differences in characteristics were evident in FWT, FD, and THP. In both cases of GPs with all accessions and only F₁ accessions, several F₁ accessions had predicted values that were lower than the observed values (Fig. 5b, 5c). However, in simulations based on inbred accessions, such discrepancies were not observed, and simulated values generally exceeded the observed values (Fig. 5a). To explore the factors contributing to these discrepancies, we examined the relationship between the root square error (RSE) of simulated versus observed values and two statistical measures: phenotypic values of the parents and F₁ accessions, and the absolute values of dominance effects. Although RSE was correlated with both statistical measures, the absolute values of dominance effects showed the strongest positive correlations across all traits (Table 3). This suggests that simulation errors were more pronounced in F₁ accessions with phenotypic values significantly deviating from the mid-point of their parents' phenotypic values. Particularly, FD, FWT, and THP displayed strong correlations (r ≧ 0.886, P < 0.001). In contrast, in the GPs with all and only F₁ accessions, such strong correlations were not observed. Instead, the phenotypic values of the F₁ accessions had a stronger correlation with the RSE compared to the dominance effects (Table 3). This indicates that GP errors were more strongly influenced by the degree of phenotypic values of the F₁ accessions rather than the dominance effects.

Table 3

Correlation between root square errors (RSEs)a in simulation/genomic prediction (GP) and F1 progeny-related values in F1 accessions.
Trait	Phenotypic simulation/prediction method	Pearson's correlation coefficient
Trait	Phenotypic simulation/prediction method	Phenotypic values of F₁ accession		Dominance effect (absolute value)
FL	Simulation with inbred accessions	0.181	*	0.528	***
	GP with all accessions	0.435	***	0.448	***
	GP with F₁ accessions	0.481	***	0.387	***
FD	Simulation with inbred accessions	0.408	***	0.939	***
	GP with all accessions	0.668	***	0.411	***
	GP with F₁ accessions	0.752	***	0.255	**
FI	Simulation with inbred accessions	0.519	***	0.679	***
	GP with all accessions	0.308	***	0.338	***
	GP with F₁ accessions	0.280	***	0.281	***
FWT	Simulation with inbred accessions	0.338	***	0.972	***
	GP with all accessions	0.707	***	0.448	***
	GP with F₁ accessions	0.786	***	0.333	***
THP	Simulation with inbred accessions	0.175	*	0.886	***
	GP with all accessions	0.388	***	0.255	**
	GP with F₁ accessions	0.464	***	0.060	N.S.
^a RSEs between simulated/predicted and observed phenotypic values in the F₁ accessions. For genomic predictions (GPs) with all accessions or F₁ accessions, phenotypic values were predicted using ten repetitions of 10-fold cross-validation. Mean RSE values were used as representative values for in each F₁ accession. , , *: significance: P <* 0.05, P < 0.01, P < 0.001, respectively. N.S.: No significance

Exhibition of Phenotypic Simulation Examples

In this study, we simulated the fruit length (FL) and fruit diameter (FD) in F₁ progenies resulting from all possible crosses among the inbred accessions as an example. We then compared these simulations with phenotypic data from eight new F₁ lines (F₁A - F₁H) derived from six parental lines (P1 - P6) (Fig. 6a). Figure 6b displays the simulated FL and FD for a total of 8,646 F₁ progenies, including the new eight F₁ lines (indicated by red circles) and their six parental lines (indicated by blue circles). The simulated FD and FL for the F₁ progenies were generally within the range of their parental values. Figure 6c shows the observed FL and FD for both the F₁ and parental lines. While their positional relationships largely corresponded to the simulation results, some discrepancies were noted.

Phenotypic simulation of crossed progenies using genomic prediction (GP) based on parental information is a promising technique for reducing costs and time in plant breeding. Its utility has been demonstrated in various crops (Iwata et al. 2013; Yamamoto et al. 2017, 2021). Fruit vegetables, in particular, require substantial space per plant and extensive management, including supporting structures and thinning of axillary buds, branches, and fruits. Consequently, the number of plants that can be cultivated is limited, making the development of such simulation techniques highly valuable. While chili peppers stand to benefit from these advancements, there has been no empirical study examining the utility of phenotypic simulation with GP in this crop. As a preliminary investigation, we implemented GP using different types of chili pepper populations (inbred and F₁) to assess the accuracy and identify the optimal model. We then conducted phenotypic simulations of F₁ progenies based solely on their parental information.

Initially, our study employed the MIG-seq technique for genotyping, detecting approximately 3.2k SNPs. These SNPs were distributed genome-wide, with an average density oF₁ SNP per Mbp (Fig. 2a, Table S2). This dataset provided adequate accuracy for GPs concerning fruit-related traits (Table 2). Hong et al. (2020) demonstrated that GP accuracy for fruit-related traits in inbred chili pepper accessions did not significantly deteriorate when the number of SNPs decreased from 18.7k to 2.8k. Therefore, a dataset of 3.2k SNPs is considered sufficient for performing GPs related to fruit traits in chili peppers. Compared to other next-generation sequencing (NGS)-based genotyping methods, MIG-seq offers lower costs and can be conducted with suboptimal DNA quality (Eguchi et al. 2020; Nishimura et al. 2022), making it a promising tool for genetic analysis and breeding. Our results provide empirical evidence supporting the use of MIG-seq in GP studies and offer valuable insights for future research in this area.

In the genomic prediction (GP) analyses of fruit-related traits across three populations (all accessions including inbred and F₁, inbred only, and F₁ only), we found that GP accuracies were highest in F₁ accessions and lowest in inbred lines (Table 2). One key reason for this difference is the genetic background of the populations. Generally, GP accuracy is known to depend on the genetic architecture of the training population, with accuracies often decreasing when training populations include genetically distant accessions (Spindel and McCouch 2016). In this study, the inbred accessions came from various regions (Supplemental Dataset 1) and were organized into six ancestral subgroups (Fig. 2b), contributing to a complex genetic background that likely lowered GP accuracies. In contrast, the genetic architecture of the F₁ accessions was relatively straightforward due to their derivation from a specific set of 20 parents, making GPs for F₁ accessions easier to achieve. Focusing on the GP accuracies of the five fruit-related traits, we observed that in inbred accessions, GP accuracies were generally high for FD and FWT but lower for FL and FI. This pattern is consistent with previous studies, despite differences in population background and size. Consequently, GPs for FD and FWT were consistently more reliable among fruit-related traits in the inbred population of chili peppers. However, in F₁ accessions, FD and FWT showed slightly lower accuracies compared to FL and FI. One reason for this could be the bias in the phenotypic values of the F₁ progenies. As shown in Fig. 5c, some F₁ accessions had predicted values significantly lower than observed values for FD and FWT, which likely contributed to reduced GP accuracies. These accessions had relatively high phenotypic values within the population but were relatively few in number. This phenotypic bias might make GPs for FWT and FD more challenging compared to FI and FL. Regarding GP models, GBLUP-GAUSS exhibited the highest accuracy across all populations and most traits (Table 2). Previous research indicated that the Kernel Hilbert Spaces (RKHS) model, which is equivalent to GBLUP-GAUSS in this study, achieved stable high accuracies for fruit-related traits in inbred chili pepper populations (Hong et al. 2020). Our results support these findings, but it was newly revealed that GBLUP-GAUSS remained the most effective model across different population sizes and heterozygosity levels. GBLUP-GAUSS is known for incorporating non-additive genetic effects, such as dominance and epistasis (Gianola et al. 2006; Gianola and van Kaam 2008). GBLUP-AD is a similar model but explicitly includes dominance effects (Nishio and Satoh 2014). We tried both models but the GP accuracies of GBLUP-AD never surpass that of GBLUP-GAUSS (Table 2). Previous research suggests that GBLUP-AD is advantageous for large populations with high heterozygosity or traits with significant dominance effects (Denis and Bouvet 2013; Zhao et al. 2013; Minamikawa et al. 2018). In this study, the population size and dominance effect levels of fruit-related traits in chili peppers were too small for GBLUP-AD to be more beneficial than GBLUP-GAUSS. Nonetheless, it is clear that GBLUP-GAUSS is a robust model for GPs of fruit-related traits across various chili pepper populations.

For the phenotypic simulation of fruit-related traits in F₁ accessions, we observed strong positive correlations (r = 0.833–0.908) between the simulated and observed phenotypic values for all five traits (Fig. 5a). This suggests that accurate ranking of fruit-related traits among F₁ progenies based on their parental information is feasible. Although the population size of the inbred accessions was relatively small (n = 132), it appeared sufficient for conducting phenotypic simulations. The high heritability of fruit-related traits (h_b² > 90%, Table 1), which typically enhances GP accuracy (Desta and Ortiz 2014), likely mitigates the impact of population size in this study. Interestingly, while the GP accuracies with the GBLUP-GAUSS model in inbred accessions ranged from 0.648 to 0.815 across the five traits (Table 2), the simulation accuracies were higher (Fig. 5a). This indicates that simulation accuracies did not directly reflect the GP accuracies within the training population (inbred accessions). Notably, there were differences in simulation accuracy among the traits, with the order as follows: FI > FL > FD > THP > FWT (Fig. 5a). This ranking did not align with the mean GP accuracies in inbred accessions (FD > FWT > FL > THP > FI), suggesting a weak connection between simulation accuracies and GP accuracies within the training population. Conversely, simulation accuracies correlated with the mean GP accuracies in F₁ accessions (FI > FL > FD > THP > FWT) (Table 2). This trend was also observed in a previous GP study in strawberries (Yamamoto et al. 2021), where GP accuracies for agricultural traits in F₁ progenies correlated more closely with those within the F₁ progenies rather than within inbred lines. This suggests that difficulties in GP within the simulated population (F₁ accessions) were more strongly reflected in simulation accuracies. However, the simulated phenotypic values for F₁ accessions did not perfectly match the observed values for any trait. Notably, outlier F₁ accessions were particularly evident in FWT (Fig. 5a). Analysis of related factors revealed that the RSE values (simulation error levels) strongly correlated with the absolute values of dominance effects in FD, FWT, and THP, but correlations were weaker in FI and FL (Table 3). In other words, simulations for FD, FWT, and THP became more challenging with higher levels of dominance effects (including negative values) in F₁ accessions. This suggested that phenotypic simulations based solely on inbred accessions could not adequately address the various dominance effects observed in each F₁ accession, leading to increased simulation errors. Supporting this, the correlation coefficients between RSE values and dominance effects (absolute values) were smaller in the following order: GPs with F₁ accessions, GPs with all accessions, simulations with inbred accessions, corresponding to higher heterozygous locus ratios (Table 3, Fig. 2d). This suggests that GP errors due to dominance effects were less pronounced in GPs with populations exhibiting higher heterozygosity. Given these insights, dominance effects may not have been adequately considered in the GP model for the present simulation based on phenotypic and genotypic data of inbred accessions. Consequently, simulated values for FWT, FD, and THP were often higher than observed values (Fig. 5a), likely due to the neglect of negative dominance effects in the GP model. These tendencies were not observed in GPs with all or only F₁ accessions (Fig. 5b, -c). Further studies using different populations and focusing on other traits are needed to confirm these hypotheses. However, it is crucial to be aware that simulation results may diverge from observed values due to differences in dominance effects among traits.

In conclusion, this study is the first to report the use of phenotypic simulation in crossed progenies through genomic prediction (GP) based on parental phenotypic and genotypic data in chili peppers. Most notably, the study demonstrated that this method can effectively rank the phenotypic values of fruit-related traits in F₁ progenies. This simulation technique allows for the estimation of fruit characteristics in a large number of F₁ progenies with a reasonable degree of accuracy, as illustrated in Figs. 6a, 6b, and 6c. This approach could be utilized to optimize parental combinations to obtain promising F₁ progenies without the need for extensive crossing and cultivation, thus saving both time and costs in chili pepper breeding. Furthermore, this technique has potential applications for phenotypic simulation in segregating generations, such as F₂ progenies, and further investigations in this area are anticipated.

Author Contributions

All authors contributed to the conception and design of the study. FK, YK, NA, and SH handled material preparation and the collection of phenotypic and genotypic data at Shinshu University. Genetic analysis, including GP, was conducted by FK, VP, and MD at the University of Molise. The first draft of the manuscript was written by FK, with all authors providing feedback on previous versions. All authors read and approved the final manuscript.

Acknowledgment

The authors thank Mr. Shinsaku Murayama (Agricultural Co. Peppers.jp, Gunma, Japan) for providing plant materials, and Dr. Motoyuki Ishimori (Graduate School of Agricultural and Life Sciences, The University of Tokyo, Tokyo, Japan) for his theoretical and technical advice on GP. Computations were partially carried out on the NIG supercomputer at ROIS National Institute of Genetics.

Funding

This research was financially supported by Yawataya Isogoro Inc and the Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number 22J13069.

Data Availability

The datasets generated in this study are available from the corresponding author upon reasonable request.

Competing Interests

The authors declare that they have no relevant financial or non-financial interests to disclose.

Alexander DH, Lange K (2011) Enhancements to the ADMIXTURE algorithm for individual ancestry estimation. BMC Bioinformatics 12:246. https://doi.org/10.1186/1471-2105-12-246
Browning BL, Browning SR (2016) Genotype Imputation with Millions of Reference Samples. Am J Hum Genet 98(1):116–126. https://doi.org/10.1016/j.ajhg.2015.11.020
Chaim A, Paran I, Grube R, Jahn M, Wijk RV, Peleman J (2001) QTL mapping of fruit-related traits in pepper (Capsicum annuum). Theor Appl Genet 102:1016–1028. https://doi.org/10.1007/s001220000461
Denis M, Bouvet JM (2013) Efficiency of genomic selection with models including dominance effect in the context of Eucalyptus breeding. Tree Genet Genomes 9:37–51. https://doi-org.kyoto-u.idm.oclc .org/10.1007/s11295-012-0528-1
Desta ZA, Ortiz R (2014) Genomic selection: genome-wide prediction in plant improvement. Trends Plant Sci 19(9):592–601. https://doi.org/10.1016/j.tplants.2014.05.006
Eguchi K, Oguri E, Sasaki T, Matsuo A, Nguyen DD, Jaitrong W, Yahya BE, Chen Z, Satria R, Wang WY, Suyama Y (2020) Revisiting museum collections in the genomic era: potential of MIG-seq for retrieving phylogenetic information from aged minute dry specimens of ants (Hymenoptera: Formicidae) and other small organisms. Myrmecological News 30:151–159. https://doi.org/10.25849/myrmecol.news_030:151
FAOStat FA (2019) Food and agriculture data. Crop Statistics. Available online: http://www.fao.org/fast (accessed on 14 August 2020)
Gianola D, Fernando RL, Stella A (2006) Genomic-assisted prediction of genetic value with semiparametric procedures. Genetics 173(3):1761–1776. https://doi.org/10.1534/genetics.105.049510
Gianola D, van Kaam JB (2008) Reproducing kernel Hilbert spaces regression methods for genomic assisted prediction of quantitative traits. Genetics 178(4):2289–2303. https://doi.org/10.1534/genetics.107.084285
Han K, Jeong HJ, Yang HB, Kang SM, Kwon JK, Kim S, Choi D, Kang BC (2016) An ultra-high-density bin map facilitates high-throughput QTL mapping of horticultural traits in pepper (Capsicum annuum). DNA Res 23(2):81–91. https://doi.org/10.1093/dnares/dsv038
Heslot N, Jannink JL, Sorrells ME (2015) Perspectives for genomic selection applications and research in plants. Crop Sci 55:1–12. https://doi.org/10.2135/cropsci2014.03.0249
Hong JP, Ro N, Lee HY, Kim GW, Kwon JK, Yamamoto E, Kang BC (2020) Genomic Selection for Prediction of Fruit-Related Traits in Pepper (Capsicum spp). Front Plant Sci 11:570871. https://doi.org/10.3389/fpls.2020.570871
Ishimori M, Hattori T, Yamazaki K, Takanashi H, Fujimoto M, Kajiya-Kanegae H, Yoneda J, Tokunaga T, Fujiwara T, Tsutsumi N, Iwata H (2020) Impacts of dominance effects on genomic prediction of sorghum hybrid performance. Breed Sci 70(5):605–616. https://doi.org/10.1270/jsbbs.20042
Iwata H, Hayashi T, Terakami S, Takada N, Saito T, Yamamoto T (2013) Genomic prediction of trait segregation in a progeny population: a case study of Japanese pear (Pyrus pyrifolia). BMC Genet 14:81. https://doi.org/10.1186/1471-2156-14-81
Kim GW, Hong JP, Lee HY, Kwon JK, Kim DA, Kang BC (2022) Genomic selection with fixed-effect markers improves the prediction accuracy for Capsaicinoid contents in Capsicum annuum. Hortic Res 9:uhac204. https://doi.org/10.1093/hr/uhac204
Kondo F, Umeda K, Sudasinghe SP, Yamaguchi M, Aratani S, Kumanomido Y, Nemoto K, Nagano AJ, Matsushima K (2023) Genetic analysis of pungency deficiency in Japanese chili pepper ‘Shishito’(Capsicum annuum) revealed its unique heredity and brought the discovery of two genetic loci involved with the reduction of pungency. MGG 298:201–212. https://doi.org/10.1007/s00438-022-01975-2
Lee HY, Ro NY, Patil A, Lee JH, Kwon JK, Kang BC (2020) Uncovering Candidate Genes Controlling Major Fruit-Related Traits in Pepper via Genotype-by-Sequencing Based QTL Mapping and Genome-Wide Association Study. Front Plant Sci 11:1100. https://doi.org/10.3389/fpls.2020.01100
Lozada DN, Sandhu KS, Bhatta M (2023) Ridge regression and deep learning models for genome-wide selection of complex traits in New Mexican Chile peppers. BMC Genom Data 24(1):80. https://doi.org/10.1186/s12863-023-01179-6
McLeod L, Barchi L, Tumino G, Tripodi P, Salinier J, Gros C, Boyaci HF, Ozalp R, Borovsky Y, Schafleitner R, Barchenger D, Finkers R, Brouwer M, Stein N, Rabanus-Wallace MT, Giuliano G, Voorrips R, Paran I, Lefebvre V (2023) Multi-environment association study highlights candidate genes for robust agronomic quantitative trait loci in a novel worldwide Capsicum core collection. Plant J 116(5):1508–1528. https://doi.org/10.1111/tpj.16425
Meuwissen TH, Hayes BJ, Goddard ME (2001) Prediction of total genetic value using genome-wide dense marker maps. Genetics 157(4):1819–1829. https://doi.org/10.1093/genetics/157.4.1819
Minamikawa MF, Nonaka K, Kaminuma E, Kajiya-Kanegae H, Onogi A, Goto S, Yoshioka T, Imai A, Hamada H, Hayashi T, Matsumoto S, Katayose Y, Toyoda A, Fujiyama A, Nakamura Y, Shimizu T, Iwata H (2017) Genome-wide association study and genomic prediction in citrus: Potential of genomics-assisted breeding for fruit quality traits. Sci Rep 7(1):4721. https://doi.org/10.1038/s41598-017-05100-x
Minamikawa MF, Takada N, Terakami S, Saito T, Onogi A, Kajiya-Kanegae H, Hayashi T, Yamamoto T, Iwata H (2018) Genome-wide association study and genomic prediction using parental and breeding populations of Japanese pear (Pyrus pyrifolia Nakai). Sci Rep 8(1):11994. https://doi.org/10.1038/s41598-018-30154-w
Naegele RP, Mitchell J, Hausbeck MK (2016) Genetic Diversity, Population Structure, and Heritability of Fruit Traits in Capsicum annuum. PLoS ONE 11(7):e0156969. https://doi.org/10.1371/journal.pone.0156969
Naves ER, de Ávila Silva L, Sulpice R, Araújo WL, Nunes-Nesi A, Peres LEP, Zsögön A (2019) Capsaicinoids: Pungency beyond Capsicum. Trends Plant Sci 24(2):109–120. https://doi.org/10.1016/j.tplants.2018.11.001
Nishimura K, Motoki K, Yamazaki A, Takisawa R, Yasui Y, Kawai T, Ushijima K, Nakano R, Nakazaki T (2022) MIG-seq is an effective method for high-throughput genotyping in wheat (Triticum spp). DNA Res 29(2):dsac011. https://doi.org/10.1093/dnares/dsac011
Nishio M, Satoh M (2014) Including dominance effects in the genomic BLUP method for genomic evaluation. PLoS ONE 9(1):e85792. https://doi.org/10.1371/journal.pone.0085792
Schmidt P, Hartung J, Bennewitz J, Piepho HP (2019) Heritability in Plant Breeding on a Genotype-Difference Basis. Genetics 212(4):991–1008. https://doi.org/10.3389/fpls.2020.01100
Shirasawa K, Hosokawa M, Yasui Y, Toyoda A, Isobe S (2023) Chromosome-scale genome assembly of a Japanese chili pepper landrace, Capsicum annuum 'Takanotsume'. DNA Res 30(1):dsac052. https://doi.org/10.1093/dnares/dsac052
Spindel JE, McCouch SR (2016) New Phytol 212(4):814–826. https://doi.org/10.1111/nph.14174. When more is better: how data sharing would accelerate genomic selection of crop plants
Suyama Y, Matsuki Y (2015) MIG-seq: an effective PCR-based method for genome-wide single-nucleotide polymorphism genotyping using the next-generation sequencing platform. Sci Rep 5:16963. https://doi.org/10.1038/srep16963
Tong H, Nankar AN, Liu J, Todorova V, Ganeva D, Grozeva S, Tringovska I, Pasev G, Radeva-Ivanova V, Gechev T, Kostova D, Nikoloski Z (2022) Genomic prediction of morphometric and colorimetric traits in Solanaceous fruits. Hortic Res 9:uhac072. https://doi.org/10.1093/hr/uhac072
Ukai Y (2002) Genetic analysis of quantitative traits. IGAKU SHUPPAN, Tokyo. (in Japanese)
Yamamoto E, Kataoka S, Shirasawa K, Noguchi Y, Isobe S (2021) Genomic Selection for F₁ Hybrid Breeding in Strawberry (Fragaria × ananassa). Front Plant Sci 12:645111. https://doi.org/10.3389/fpls.2021.645111
Yamamoto E, Matsunaga H, Onogi A, Ohyama A, Miyatake K, Yamaguchi H, Nunome T, Iwata H, Fukuoka H (2017) Efficiency of genomic selection for breeding population design and phenotype prediction in tomato. Heredity 118(2):202–209. https://doi.org/10.1038/hdy.2016.84
Zhao Y, Zeng J, Fernando R, Reif JC (2013) Genomic prediction of hybrid wheat performance. Crop Sci 53:802–810. https://doi.org/10.2135/cropsci2012.08.0463

Download PDF

Version 1

posted

You are reading this latest preprint version

Availability of phenotypic simulation for fruit-related traits in F1 progenies of chili peppers (Capsicum annuum) using genomic prediction based solely on parental information

Status:

Version 1

Abstract

Figures

Key Message

Introduction

Materials and Methods

Results

Discussion

Declarations

References

Supplementary Files

Status:

Version 1