Genomic selection and genome‐wide association studies in tetraploid chipping potatoes

Abstract

Potato is a major food crop in the United States and around the world. Most potatoes grown in the United States are destined for processing. Genomic selection can speed up breeding progress for important traits, including those with complex inheritance by guiding the identification of the best parents and guiding selection to advance clones in the breeding program. However, the application of genomic selection in polyploid species has been challenging. In this study, we obtained breeding values of 384 chipping clones evaluated in Texas between 2017 and 2020. The mean reliability of the genomic‐estimated breeding values obtained were 0.77, 0.41, 0.61, 0.71, and 0.24 for chip color, chip quality, specific gravity, vine maturity, and total yield, respectively. Potato clones with good chip quality, high yield, high specific gravity, and light‐color chips were identified using a multi‐trait selection index based on weighted standardized genomic‐estimated breeding values. Genome‐wide association studies identified quantitative trait loci on chromosome 5 for vine maturity and chromosomes 1, 3, and 7 for chip color. This research has laid the groundwork for implementing genomic selection in tetraploid potato breeding and understanding the genetic basis of chip processing traits in potatoes.

Core Ideas

• Developing improved potato chip cultivars would be extremely beneficial for the chip processing industry.

• Genomic‐estimated breeding values (GEBVs) should guide parental selection and advancement of clones.

• Potato chipping clones with high standardized weighted multitrait index values were identified.

• Standardized weighted multitrait indexes should be more effective than nonstandardized unweighted GEBVs.

• Genome‐wide association studies identified quantitative trait loci for vine maturity and chip color.

1. INTRODUCTION

Potato is the most important nongrain food on the planet and a highly recommended crop for ensuring food security for future generations (Devaux et al., 2014; FAO, 2009). Potatoes are used for fresh consumption, processing, and as seeds. Processing potatoes are the most important market class, representing 60% of the potatoes grown in the United States. Potato chips are the most consumed snack due to the convenience associated with consumers’ lifestyles and food consumption preferences (Liyanage et al., 2021). Potato breeding programs focus on developing varieties well suited to specific market classes. Important quality traits for processing potatoes (French fries and chippers) include tuber appearance, high specific gravity, post‐harvest quality, nutritional value, flavor properties, and sensory attributes like crispiness, and mouthfeel among others (Nacheva & Pevicharova, 2008). The industry prefers tubers with shallow eyes and a round‐oval shape for processing chips, whereas oblong‐long‐shaped tubers are favored for processing French fries. Likewise, in the potato industry, external and internal tuber defects are all undesirable.

Potato tubers contain, on average, 80% water and 20% dry matter, with starch accounting for most of the dry matter (Stark et al., 2020). However, the composition varies due to both genotype and environment (Zhou et al., 2017). Starch is the most important component of dry matter and determines tuber density, indirectly assessed as tuber‐specific gravity. For processing potatoes (French fries and chippers), high specific gravity is favored because they fry faster, give better texture, higher net weight to the final product, and use less oil while frying (Habyarimana et al., 2017). Although a smooth skin texture is preferred for making chips and a russet skin texture is chosen for making French fries (Stark et al., 2020), there are few exceptions. For example, Shepody has smooth skin and is used for making French fries, and Snowden, a chipping potato, has flakey skin rather than smooth.

Another important component that determines the quality of processing potatoes (French fries and chippers) is the reducing sugar content (Wiberley‐Bradford et al., 2014). During frying, reducing sugars (glucose and fructose) in potato tubers react with free amino acids in a nonenzymatic Maillard reaction, resulting in undesirably brown, bitter‐tasting products (Kumar et al., 2004; Wen et al., 2016). High‐temperature frying contributes to the aroma, flavor, and color of the fried potatoes, and the formation of acrylamide (when the amino acid asparagine is involved) during the Maillard reaction has raised health concerns due to potential toxicity and carcinogenicity (Hogervorst et al., 2010). For these reasons, reducing sugar content in processing potatoes should not exceed 0.2%–0.3% of fresh weight (Fiselier & Grob, 2005).

Internal tuber defects like internal heat necrosis (IHN), vascular browning, Verticillium wilt, stem end browning, hollow heart with discoloration, brown center, and internal black spot also contribute to browning and reducing chip quality. The potato processing industry is continuously searching for strategies to reduce browning. The accumulation of reducing sugar in tubers and internal defects varies with variety, storage temperature, physiological maturity, plant stresses, and tuber age (Leonel et al., 2017). At the industrial level, cold storage conditions facilitate long‐term storage and eliminate issues related to diseases, sprouting, and shrinking (Kleinkopf et al., 2003). However, cold temperatures promote the breakdown of starch and the accumulation of reducing sugars in potato tubers, a process known as cold‐induced sweetening (CIS). As a result, developing potato cultivars that can be stored at low temperatures without undergoing CIS would be extremely beneficial.

Core Ideas

• Developing improved potato chip cultivars would be extremely beneficial for the chip processing industry.

• Genomic‐estimated breeding values (GEBVs) should guide parental selection and advancement of clones.

• Potato chipping clones with high standardized weighted multitrait index values were identified.

• Standardized weighted multitrait indexes should be more effective than nonstandardized unweighted GEBVs.

• Genome‐wide association studies identified quantitative trait loci for vine maturity and chip color.

Breeding programs have made efforts to develop CIS resistance cultivars that produce lighter colored potato chips. Most improvements to date are associated with traits coming from the Lenape variety, which has been a common parent in chip crosses since its introduction (Love et al., 1998). Lenape is in the pedigree of popular chipping varieties including Atlantic and Snowden. Atlantic (Webb et al., 1978) has a high yield and high specific gravity, but it is sensitive to environmental stress and prone to internal defects, especially IHN (also known as internal brown spots). Its major processing weakness is the accumulation of reducing sugars in long‐term storage. Atlantic tuber dormancy is about 4 months (medium) in cold storage, and storage for longer periods will require the application of a sprout inhibitor. Snowden is like Atlantic except that it chips out of cold (10°C) storage without reconditioning. However, it produces undersized tubers, which results in lower yields. Lamoka is a CIS‐resistant variety with good chip color and resistance to the golden cyst nematode (De Jong et al., 2017). Mackinaw is a chipping potato with resistance to potato virus Y, late blight, and tolerance to common scabs. It is necessary to continue breeding for new potato cultivars with high chip quality, low acrylamide precursor contents, CIS resistance, high specific gravity, and free of diseases and tuber defects.

Reasons that have hindered potato breeding progress include the complex nature of the potato genome, the quantitative nature of the most important traits, rapid inbreeding depression, self and cross incompatibilities, and low intensity of selection in early generations (Ghislain & Douches, 2020). The current selective breeding process takes a long time (10–15 years) to produce a new potato cultivar following the initial cross (Halterman et al., 2016). Potato breeders consider several traits while developing new cultivars, hence the pace of breeding is slow (Slater et al., 2016). Speeding up the use of early generation clones as parents based on high breeding value for the traits of interest would shorten the breeding cycle and thus make recurrent selection breeding more effective (Slater et al., 2014). Studies have shown that the breeding cycle can be shortened by 2–3 years when marker‐assisted selection (MAS) and genomic‐estimated breeding values (GEBVs) are used together in potato (Slater et al., 2014).

The recent availability of dense single‐nucleotide polymorphism (SNP) arrays and software packages that can perform statistical genomic analysis in autopolyploid species presents opportunities to apply optimized analytical approaches to cultivated potato. An important step in developing new chip varieties that produce high‐quality chips is to identify the underlying genetic loci. Quantitative trait loci (QTL) mapping for chip color and sugar content has identified several regions in the potato genome, which contain genes influencing the traits (Li et al., 2008). Several of these QTLs were co‐localized with functional genes for carbohydrate metabolism and transport, including vacuolar invertase (Pain‐1/VInv) on chromosome 3, apoplastic invertases (Invap‐b and Invap‐a) on chromosomes 9 and 10, and sucrose synthases (Sus3 and Sus4) on chromosomes 7 and 12 (Gebhardt et al., 2014). Recently, Byrne et al. (2020) identified a major QTL on chromosome 10 for fry color and predicted fry color with moderate accuracy using genome‐wide markers. Frederick and Bethke (2019) identified nine QTLs for stem‐end chip defects, seven of which overlapped with QTLs for chip color traits. Other efforts to practically apply the SNP panel have focused on genome‐wide association studies (GWAS) to identify marker‐trait associations. The SolCAP SNP array was used for GWAS to identify marker‐trait associations for tuber yield and starch content (Schönhals et al., 2016), tuber quality, and fry color traits (D'hoop et al., 2014; Rak et al., 2017).

Furthermore, important progress has been made in recent years for rapid genetic gain in crops using pedigree or genomic information‐based methods such as genomic selection (GS) (Crossa et al., 2017). GS uses genome‐wide molecular markers and historical phenotype data from a breeding program to predict the performance of related individuals. GEBVs allow the selection of superior parents or individuals for next‐generation advancement faster than using phenotypes alone (Crossa et al., 2017; Meuwissen et al., 2001). Thus, GS enhances genetic gains by shortening the breeding cycle and/or enhancing testing efficiency. Habyarimana et al. (2017) demonstrated three GS models using SilicoDArT markers and indicated that GS can predict early clonal generations and can select traits with low heritability. Endelman et al. (2018), combining genotype and pedigree information with phenotype data for fry color, specific gravity, and yield, concluded that genome‐wide prediction is feasible in autotetraploid potato. Likewise, genotyping‐by‐sequencing‐developed genomic prediction models were used for starch content and chipping quality (Sverrisdóttir et al., 2017), and for dry matter content and chipping quality (Sverrisdóttir et al., 2018).

Additionally, the usage of selection indexes can aid in the selection of multiple traits at the same time. Using a multiple‐trait selection index, it is possible to identify superior and inferior genotypes by integrating several attributes (Bernardo, 2020). The best index to use depends on the crop and the traits that are important for selection. The sum of standardized variables (∑Z) or Z‐index (Mendes et al., 2009) is an alternative to other indices and has been used in common bean (Lima et al., 2015), upland rice (Ribeiro et al., 2016), and quince cultivars (Coutinho et al., 2019) to select superior lines.

The aim of this study was to (i) evaluate how well GS could predict chipping quality in tetraploid potatoes, (ii) generate multi‐trait selection indexes to identify superior individuals for important chipping traits, and (iii) perform GWAS for chip quality traits.

2. MATERIALS AND METHODS

2.1. Plant materials and phenotype data

Three hundred and eighty‐four chipping potato clones were evaluated between 2017 and 2020 near Dalhart, TX (35°58′15′′N, 102°44′36″W) (Figure 1). These included entries from the National Chip Processing Trial (NCPT) and advanced chip selections from the western, southwestern regional, and Texas trials contributed by public potato breeding programs in the United States. The NCPT uses a two‐tier evaluation system, with one plot per location for tier one clones and two plots (replications) per location for tier two clones. Advanced chip selections from the western, southwestern regional, and Texas trials were planted in one to four replications. Plots in the NCPT trial contained 15 seed pieces, and advanced chip selections from the western, southwestern regional and Texas trials contained 24 seed pieces. Seeds were planted with 30 cm spacing between hills and 70 cm spacing between rows. Trials were planted in early May and harvested in early September, with vine desiccation 2–3 weeks before harvest. The soil type in Dalhart was Dallum fine sandy loam. Standard potato production practices were followed during the growing period in all years (https://potato.tamu.edu/reports/). The accumulated growing degree days (base temperature of 4.5°C) from 2017 to 2020 were 2368.7, 2452.3, 2428.9, and 2477.6, respectively. Plots were irrigated with center pivot as needed except during periods of sufficient rainfall throughout the growing season.

FIGURE 1.

Number of potato chipping clones tested from different breeding programs between 2017 and 2020 in Dalhart, Texas

Phenotypic data for chip color, chip quality, specific gravity, vine maturity, and total yield were included in this study. To evaluate fry color, 10 tubers from each plot were cut into two halves (stem to bud), and a 1.3 mm thick chip slice was used for each tuber. Chip slices were then rinsed in water and blot‐dried on a paper towel to remove extra starch. Slices were fried in vegetable oil at 182°C for 80 s or until bubbling ceased. The prevalent color of the chips (not considering internal defects) was scored using a subjective scale (1–5; 1 = light, 5 = dark) recommended for the NCPT based on the Snack Food Association color standards reference chart for potato chips. In this study, chip quality focused on external chip appearance based on the overall level of browning including internal defects. Sensory parameters were not considered. In Texas, in part due to heat stress during the growing season, chip defects such as browning due to reducing sugars, stem end, vascular discoloration, internal heat necrosis, brown center, and hollow heart are very common and were scored separately. For this study, all chip defects were combined as the overall level of browning to reflect external chip quality. Specific gravity was determined by the water displacement method as the weight of the tuber in air divided by the difference between the weight in air and the weight in water using 2–3 kg (113.4–170.1 g category) of tubers per plot (Wang et al., 2017). Vine maturity was scored on a 1–9 scale using standardized NE‐1014 rating codes based on the percent of green vines (1 = very early [vine 100% dead]; 4 = vine starting to senesce [yellowing evident]; 6 = green vine; done flowering; 7 = green vine; some flowers yet in bloom; 9 = very late; full bloom with buds evident as well). The total yield was based on the weight of all harvested tubers reported in Mg ha⁻¹.

2.2. Genotyping

Most of the clones in this dataset were genotyped with the Infinium 22 K V3 Potato Array on the Illumina iScan (Illumina Inc., San Diego, CA, USA), but some were genotyped with V2. The first set consisted of 126 clones genotyped with 6966 markers. The second set consisted of 258 clones genotyped with 9975 markers. The genotypic classes were based on the dosage (0–4) of the B allele. The fitPoly R package (Voorrips et al., 2011; Zych et al., 2019) was used to make genotype calls. The two datasets were merged based on best linear unbiased predictors (BLUPs), using the merge_impute function of the R package polyBreedR (https://github.com/jendelman/polyBreedR). This is equivalent to one iteration of the EM‐imputation algorithm described by Poland et al. (2012). After merging and imputing data for chipping clone genotypes with the SNP arrays, 14,401 markers were retained.

2.3. Two‐stage analysis of multienvironment trials

Data were analyzed using StageWise (version 0.26) (Endelman, 2022) based on a two‐stage approach (Damesa et al., 2017) in R (R Core Team, 2021) using ASReml‐R v4 (Butler et al., 2018) for REML estimation of variance components. In Stage 1, the best linear unbiased estimate (BLUE) for genotype was computed separately in each year, with a fixed cofactor for a block. The broad‐sense heritability was estimated on a plot basis as V _g/(V _g + V _e) from the Stage 1 variance components for genetic (V _g) and residual (V _e) variance using the same model, but with genotype as random.

Partitioning of variance in Stage 2 utilized the BLUEs from Stage 1 as the response variable. The Stage 2 model was given as

$$\mathrm{BLUE}[g_{ij}]=y_{ij}={\mathrm{\mu }}_j+\sum _k{\mathrm{\beta }}_k{w}_{ik}+a_i+d_i+s_{ij}+{\mathrm{\epsilon }}_{ij}$$ (1)

where $g_{ij}$ is the genotypic value for clone i in year j, ${\mathrm{\mu }}_j$ is the fixed effect for year j, $w_{ik}$ is the (centered) allele dosage for clone i at marker k, and β_k is the fixed effect for marker k. The vector of additive values (a) follows a multivariate normal distribution with covariance proportional to a genomic relationship matrix (G) estimated from markers (Endelman et al., 2018; VanRaden, 2008). Independent and identically distributed (i.i.d.) residual genetic values $d_i$ were included to capture nonadditive effects. The s_ij random effect follows a multivariate normal distribution with no free variance parameters: the variance‐covariance matrix $\mathbf{\Omega }$ is the direct sum of the variance‐covariance matrices of the Stage 1 BLUEs (Damesa et al., 2017). The model residuals ε_ij are i.i.d. and represent the genotype × year interaction. The variance for each term in Equation 1 was computed based on Legarra (2016) to estimate the proportion of variance explained.

Well‐established formulas were used to compute the BLUPs and standard errors for both additive values and marker effects. For a generic random vector u, BLUP $[\mathbf{u}]=\widehat{\mathbf{u}}=\mathbf{CPy}$, where $\mathbf{C}=\mathrm{cov}[\mathbf{u},\mathbf{y}]$, $\mathbf{P}={\mathbf{V}}^{-1}-{\mathbf{V}}^{-1}\mathbf{X}{({\mathbf{X}}^{\prime }{\mathbf{V}}^{-1}\mathbf{X})}^{-1}{\mathbf{X}}^{\prime }{\mathbf{V}}^{-1}$, V is the variance–covariance matrix of the response variable, and X is the incidence matrix for fixed effects (Searle et al., 1992). The reliability of ${\widehat{u}}_i$ is the squared correlation with the true value, which equals $r_{\mathrm{i}}^2=\frac{\mathrm{var}[{\widehat{\mathrm{u}}}_{\mathrm{i}}]}{\mathrm{var}[{\mathrm{u}}_{\mathrm{i}}]},$ and $\mathrm{var}[\widehat{\mathbf{u}}]={\mathbf{CPC}}^{\prime }$.

2.4. Cross‐validation

To perform cross‐validation, the phenotypes for Texas crossed/selected clones in the dataset (80 clones that have identifiers “TX” in the name) were masked.

2.5. GWAS

Marker effects and GWAS scores were calculated by BLUP using StageWise (Endelman, 2022). For the vector a of additive values, $\mathrm{cov}[\mathbf{a},\mathbf{y}]={\mathrm{\sigma }}_{\mathrm{a}}^2{\mathbf{GZ}}^{\prime }$. For the m‐dimensional vector $\mathit{\alpha }$ of marker effects, $\mathrm{cov}[\mathit{\alpha },\mathbf{y}]={\mathrm{\sigma }}_{\mathrm{\alpha }}^2{\mathbf{W}}^{\prime }{\mathbf{Z}}^{\prime }$, where W is the centered allele dosage matrix and the marker effect variance parameter ${\mathrm{\sigma }}_{\mathrm{\alpha }}^2={\mathrm{\sigma }}_{\mathrm{a}}^2/(\mathrm{\varphi }{\sum }_{k=1}^m{p}_k{q}_k)$ for ploidy ϕ (Endelman et al., 2018). GWAS scores were computed from the standardized marker effects, which are approximately standard normal (Bernal Rubio et al., 2016; Duarte et al., 2014). The discovery threshold was based on a significance level of 0.05 and corrected for multiple testing by the effective number of markers (Moskvina & Schmidt, 2008).

2.6. Standardized multi‐trait selection indexes

Standardized multi‐trait selection indexes were calculated based on the Z values of GEBVs for all traits and clones using the following expression:

$$Z_{i\mathrm{GEBV}}=\left({\overline{y}}_{i\mathrm{GEBV}}-{\overline{y}}_{\mathrm{GEBV}}\right)/{\mathrm{\sigma }}_{\mathrm{GEBV}},$$

where $Z_{i\mathrm{GEBV}}$ is the standard GEBV trait value for clone i, ${\overline{y}}_{i\mathrm{GEBV}}$is the average GEBV trait value of clone i, ${\overline{y}}_{\mathrm{GEBV}}$ is the overall GEBV trait average (based on all clones), and σ is the standard deviation of the GEBV.

The weighted multi‐trait selection index for each clone (i) was calculated considering five traits according to the following expression:

$$\begin{array}{ccc}& & \mathrm{WMIS}=\left(\mathrm{W}1\right)\times Z_{i\mathrm{GEBV}\mathrm{Chip}\mathrm{color}}+\left(\mathrm{W}2\right)\hfill \\ & & \times Z_{i\mathrm{GEBV}\mathrm{Chip}\mathrm{quality}}+\left(\mathrm{W}3\right)\times Z_{i\mathrm{GEBV}\mathrm{Specific}\mathrm{gravity}}\hfill \\ & & +\left(\mathrm{W}4\right)\times Z_{i\mathrm{GEBV}\mathrm{Vine}\mathrm{maturity}}+\left(\mathrm{W}5\right)\times Z_{i\mathrm{GEBV}\mathrm{Yield},}\hfill \end{array}$$

where WMIS is the weighted multi‐trait index used for selection. The weight (W1‐W5) given to each trait is subjective and at the discretion of each breeder. The weight can be positive or negative depending on the scale and the desired values. In our case, chip color, chip quality, and vine maturity were given a negative sign because lower values are preferred for our breeding program (light chip color, good quality, and early maturity, respectively). $Z_{i\mathrm{GEBV}\mathrm{Chip}\mathrm{color}}$ is the standardized average GEBV for chip color for clone i, $Z_{i\mathrm{GEBV}\mathrm{Chip}\mathrm{quality}}$ is the standardized average GEBV for chip quality, $Z_{i\mathrm{GEBV}\mathrm{Specific}\mathrm{gravity}}$ is the standardized average GEBV for specific gravity, and $Z_{i\mathrm{GEBV}\mathrm{Yield}}$ is the standardized average GEBV for yield.

We chose to present four scenarios to calculate the WMIS. The first scenario corresponds to the priorities of the Texas A&M Potato Breeding Program, which places more emphasis on chip quality (W2 = −3) followed by specific gravity (W3 = +2), vine maturity (W4 = −2), yield (W5 = +2), and chip color (W1 = −1). The highest weight was given to chip quality as it is the most relevant trait for the chipping industry. The specific gravity, vine maturity, and yield were assigned a weight of 2. The lowest weight was given for chip color (considering the actual color of fried chips—excluding chip defects). Chip color received the lowest weight because the actual color of chips is based on regional preferences. In the second scenario, we chose to present WMIS corresponding to the priorities of the growers, which places more emphasis on yield (W5 = +3) and specific gravity (W3 = +3) followed by chip quality (W2 = −2), vine maturity (W4 = −2), and chip color (W1 = −1). The highest weight was given to yield and specific gravity because growers are paid based on a combination of these traits. In fact, growers are paid a premium as potato‐specific gravity increases. In the third and fourth scenarios, we chose to present WMIS considering only four traits, excluding vine maturity. The weight (W1–W4) for scenarios three and four were the same as scenarios one and two, respectively.

To be able to interpret the WMIS values in the context of the Z score distribution, the calculated WMIS was also converted into a Z value (Z _WMIS). The top 5% clones of the Z distribution were identified as potential parents or candidates to be advanced because they had the best‐combined breeding values.

$$Z_{i\mathrm{WMIS}}=\left({\overline{y}}_{i\mathrm{WMIS}}-{\overline{y}}_{\mathrm{WMIS}}\right)/{\mathrm{\sigma }}_{\mathrm{WMIS}},$$

where $Z_{i\mathrm{WMIS}}$ is the standardized WMIS average for clone i, ${\overline{y}}_{i\mathrm{WMIS}}$is the WMIS average of clone i, ${\overline{y}}_{\mathrm{WMIS}}$ is the overall WMIS average (based on all clones), and σ_WMIS is the standard deviation.

3. RESULTS

3.1. Phenotypic and genotypic data

Typically, in breeding trials, many clones are tested for one year and then dropped. However, repetitions across years are required to estimate genotype × year interactions. We found that 81 clones tested in DAL17 were also tested in other years. Likewise, 114 clones in DAL18, 106 clones in DAL19, and 84 clones in DAL20 were common across environments. After merging and imputing data for chipping clone genotypes with the V2 and V3 SNP array, 14,401 markers were retained. A total of 384 clones that had both genotypic and phenotypic data were used for further analysis. Fractional values of dosage were obtained by the imputation process where missing data in the input matrices were imputed with the population mean for each marker (Figure S1). These values are compatible with an estimation of the G matrix for predicting additive values. The typical, unimodal, zero‐mean distribution of G matrix coefficients for a homogeneous population is shown in Figure S2.

The presence of outliers and normality of the residuals for the traits were checked (Figures S3–S7) and broad‐sense heritability for each trait was obtained by environment (location–year combination) (Table 1). For chip color, the highest broad‐sense heritability of 77% was obtained in DAL19, and the lowest broad‐sense heritability of 39% was obtained in DAL17. Likewise, broad‐sense heritability of 54% was achieved for chip quality in DAL19. The broad‐sense heritability for vine maturity ranged from 74% in DAL19 to 61% in DAL17 and DAL20. Similarly, for yield, the broad‐sense heritability ranged from 62% in DAL20 to 36% in DAL17.

TABLE 1.

Broad‐sense (H ²) and narrow‐sense (h ²) heritability for each trait obtained by environment (location‐year combination) based on potato chipping clones evaluated in Dalhart (DAL), Texas in 2017–2020

Traits	H 2					h 2
	DAL17	DAL18	DAL19	DAL20	Overall
Chip color	0.39	0.57	0.77	0.59	0.58	0.55
Chip quality	0.24	0.29	0.54	0.41	0.37	0.17
Specific gravity	0.49	0.74	0.36	0.51	0.53	0.39
Vine maturity	0.61	0.71	0.74	0.61	0.67	0.54
Yield	0.36	0.49	0.6	0.62	0.52	0.08

3.2. Two‐stage analysis of multienvironment trials

Genomic narrow‐sense heritability, defined as the proportion of variance for the additive effects, was 55%, 17%, 39%, 54%, and 8% for chip color, chip quality, specific gravity, vine maturity, and yield, respectively (Table 1). The benefit of including covariance of the BLUEs, which accounts for the micro‐ and macroenvironmental variation in each of the estimates in the analysis, was reflected in a lower Akaike information criterion (AIC) compared to a model without it. The AIC was lowered by 87, 47, 48, 33, and 26 for chip color, chip quality, specific gravity, vine maturity, and yield, respectively. The results showed the progression of variance partitioning as more effects are included in the model. Including the covariance of the BLUEs enabled the splitting of the multienvironment residual into G×E and within‐environment residual, and we also observed some reapportionment of the main effect for genotype versus G×E. When the marker data were included, the genotype effect was further separated into additive and residual genetic (g.resid) components.

Breeding values for fried chip color, chip quality, specific gravity, vine maturity, and total yield were predicted using the mixed model equations (Table S1). The top‐performing clone for fry chip color was W14NYQ9‐2 with a GEBV of 0.87 and a reliability score of 0.68. COTX12235‐2W was predicted to have the best chip quality with a GEBV of 1.98 and a reliability score of 0.62. ATTX10333‐1W/Y was predicted to be superior for specific gravity with a GEBV of 1.082 and a reliability score of 0.69. NDTX13280CB‐3W was predicted to have the earliest vine maturity with a GEBV of 3.89 and a reliability score of 0.82. NYQ29‐1 was predicted to be superior for total yield with a GEBV of 65.06 and a reliability score of 0.27.

Parental selection should be based on breeding values, whereas clone selection (for advancement) should be based on genotypic values. The breeding value is two times the mean of the clone's progeny and reflects the potential of the clone as a parent. The reliability of genotypic values is typically higher when marker data are used since it improves additive component estimation (Figure 2).

FIGURE 2.

The reliability (squared correlation between the true and predicted values) of genotypic values (GV) with and without using marker data for chip color (a), chip quality (b), specific gravity (c), vine maturity (d), and yield (e). The reliability was typically higher when using marker data.

3.3. Cross‐validation

To illustrate cross‐validation, we masked the phenotypes for Texas crossed/selected clones in the dataset (which have identifiers “TX” in the name) and compared them with the original prediction. All else being equal, predictions for individuals without phenotypes, which are called marker‐based selection (MBS), had lower reliability than predictions for individuals with phenotypes, which is called MAS (Figure 3).

FIGURE 3.

Comparison of the reliability of the phenotypic predictions for Texas crossed/selected clones in the dataset (clones with “TX” in the name) when phenotypic values were predicted based on both observed phenotype and marker data (marker‐assisted selection [MAS]) and when only on marker data (no observed phenotype) (marker‐based selection [MBS]) (dotted line = slope of 1) for chip color (a), chip quality (b), specific gravity (c), vine maturity (d), and yield (e). Predictions for individuals without phenotypes for Texas crossed/selected clones had lower reliability than predictions with phenotypes.

3.4. Weighted standardized multitrait selection indexes (Z _WMIS)

Selection indexes are important in the process of breeding and selecting superior chipping cultivars. Weighted standardized multitrait selection indexes (Table S2) allowed the identification of chipping clones with high chip quality, yield, specific gravity, early maturity, and light‐colored chips. In all scenarios (differential weights), most of the clones with Z _WMIS > 2 trace back to Cornell University (NY in the clone's name). Several clones persisted at the top of the list independently of the scenarios used. For instance, seven clones (NYR102‐3, NYR102‐7, NYN24‐2, NDTX1246‐3W, NY169, W13NYP19‐2, and NYP116‐6) had high rankings in all scenarios. Due to the early maturity of NDTX13280CB‐3W, WAF13066‐2, and NDTX1246‐5W/Y, Z _WMIS in scenarios one and two (considering vine maturity) was higher than in scenarios three and four (excluding vine maturity). ND124C‐1 and BNC541‐5 appeared on the top list in scenario second based on the grower's priorities because of higher yield and specific gravity.

3.5. Marker effects and GWAS

Marker effects and GWAS scores were calculated by BLUP using StageWise (Endelman, 2022) using 14,401 SNP markers. A QTL on chromosome 5 was near the gene CDF1 for vine maturity (Figure 4a), which is known to have a large effect on potato maturity (Kloosterman et al., 2013). The fixed effect estimates of −0.8 for PotVar0079081 implied that, on average, each additional copy of the alternate allele reduced vine maturity by 0.8 (on a 1–9 visual scale). According to the proportion of variance, the PotVar0079081 marker accounted for 21% of the breeding value. Likewise, the QTL peaks were identified on chromosomes 1, 3, and 7 for chip color (Figure 4b). The maximum GWAS score was found for PotVar0120554 on chromosome 3 near the Y locus, which codes for beta‐carotene hydroxylase (Brown et al., 2006), with a fixed effect estimate of −0.46 which indicated that, on average, each additional copy of the alternate allele reduced chip color score by 0.46 (on a 1–5 visual scale). According to the proportion of variance, the PotVar0120554 marker accounted for 28% of the breeding value. However, no QTLs were detected for chip quality, specific gravity, and yield.

FIGURE 4.

Manhattan plots displaying the marker‐trait associations for vine maturity (a) and chip color (b) in 384 chipping clones evaluated from 2017 to 2020 in Texas. The horizontal axis indicates the chromosome number and the position of each single‐nucleotide polymorphism (SNP). The vertical axis indicates the negative logarithm of the p‐value for each SNP. Each dot signifies an SNP. The broken line indicates the threshold computed based on an effective number of markers.

4. DISCUSSION

Genomic selection can accelerate breeding for a range of traits, including those with complex genetic inheritance. To increase GS accuracy, the breeding program's historical data can be effectively used (Atanda et al., 2021). Proper biometric procedures are needed to combine all historical information while accounting for micro‐ and macroenvironmental variation. In this study, using data from chipping tetraploid potato, the genotypic estimates for each clone and environment combination were determined in stage one by assuming independent effects, and genomic covariance matrices were used in stage two.

GS can be incorporated into potato breeding, particularly in early clonal generations, to predict and select traits that are difficult or expensive to measure, and those with low heritability could benefit the most from GS. GS can overcome the need for evaluating many plants per clone over several locations and years. The broad‐sense heritability for fry chip color, ranged from 0.39 to 0.77, indicating a large environmental influence. Trait heritability is affected by changes in allelic frequencies, the introduction of new alleles (Latta, 2010), and genetic changes caused by altered genetic backgrounds, environmental factors (Chandler et al., 2017), or errors. Average broad‐sense heritability of 0.53 was observed in our study for specific gravity. However, Slater et al. (2014) estimated a higher (0.74) heritability for specific gravity. Temperature fluctuations between years affect specific gravity, and the DAL location is prone to year‐to‐year environmental variation and high‐temperature stress during the growing season.

Partial replication over years allowed for explicit modeling of a g.resid effect in addition to the additive, dominance, and epistatic random effects. The amount of additive genetic variance captured by markers was 54% of the total genetic variance for vine maturity and fry chip color, compared with 39% for specific gravity. Specific gravity is an important component of chipping quality. Endelman et al. (2018) reported that additive genetic variance was 20% for specific gravity. Based on diallel studies, Tai (1976) reported significant specific combining ability (SCA)/general combining ability (GCA) of 0.60 for specific gravity. Lynch et al. (1992) found high GCA effects and nonsignificant SCA effects for specific gravity and suggested that mainly additive genetic factors account for the genetic variation. Wang et al. (2017) also showed relatively little genotype × environment interaction for specific gravity. Specific gravity as a proxy for dry matter content can be used for selecting chipping potatoes.

Breeding values can now be predicted by regressing phenotypic values on all available markers, because of the advent of higher density SNPs spanning the entire genome of many plants (Crossa et al., 2017). Using high‐throughput molecular marker platforms had a big impact on GS applicability. This study is based on the SNP arrays that provide high‐quality SNPs, but the cost per sample is higher. This could be expensive for small breeding programs since genomic selection often necessitates genotyping large populations. Although genotyping by sequencing (GBS) offers many markers, but there can be a lot of missing data (Elbasyoni et al., 2018). In potato, Endelman et al. (2018) used the SNP array, whereas Sverrisdóttir et al. (2017, 2018) and Byrne et al. (2020) utilized GBS to identify SNPs for genomic prediction models. Within the breeding program germplasm, several traits segregate that are influenced by both major and minor effect loci but lack causal markers. The identification of major effect loci for consideration as fixed effects could lead to improvement in the predictive ability of genomic selection models (Sarinelli et al., 2019). Marker density, a major concern for using GS, has been studied in some crops like wheat (Crossa et al., 2014; Elbasyoni et al., 2018) and pea (Tayeh et al., 2015) but has yielded mixed results. Although larger marker density could be advantageous, prediction accuracy typically plateaus as marker number increases (Lorenz et al., 2012). Sverrisdóttir et al. (2018) reported that the predictions were not constrained by the number of markers used, as equal prediction accuracies were attained with just 7800 markers in the tetraploid potato panels.

Other factors, including the size of the training population, the similarity of the training population to the validation population, and the genetic makeup of the traits under investigation, influence prediction ability. Generally, expanding the training set size will result in higher prediction accuracy (Lorenz et al., 2011). According to Lorenzana and Bernardo (2009), increasing the size of the training population might sometimes result in accuracy gains of up to 20%. Additionally, training sets with lines that are more closely related to the validation set provide predictions with higher accuracy than training sets of the same size with lines that are more distantly related (Edwards et al., 2019). Thus, a good training population for the Texas A&M University Breeding Program (that contributes to higher prediction accuracy) should include mainly Texas selected clones. The genomic selection framework for predicting genotype by environment interactions can also be improved by incorporating environmental covariates and crop modeling. This will increase prediction accuracy and shed light on the genetic architecture governing genotype by environmental interactions (Heslot et al., 2014). Nonadditive factors may have a significant impact on phenotypic expression and clone selection in clonal propagated crops (Andrade et al., 2019; Wolfe et al., 2016). Vegetatively propagated crop breeding can make use of nonadditive genetic effects by selecting the best clones to become varieties (Ceballos et al., 2015). The packages like StageWise (Endelman, 2022) to handle the various experimental designs, heritabilities, and spatial models for nongenetic variation are evolving for polyploid crops.

The reliability of genomic predictions is the expected squared correlation between the true and predicted values and indicates the proportion of the genetic variance that is explained (de Roos et al., 2009). Reliable prediction of complex quality traits will be essential to attain the objective of successful chipping potato breeding. In this study, we aimed to predict chipping quality with relatively high accuracy. Predictions were made for chipping quality traits like fry color, chip quality, and specific gravity using the mixed model equations. The mean reliabilities of the GEBVs obtained were 0.77, 0.41, 0.61, and 0.24 for chip color, chip quality, specific gravity, and total yield, respectively. For chip color, the reliability ranged from 0.41 to 0.97. Sverrisdóttir et al. (2018) obtained cross‐validated prediction correlations of 0.39–0.79 for chipping quality. Similarly, the reliability for the chip quality, specific gravity, and total yield ranged from 0.20 to 0.74, 0.31 to 0.86, and 0.10 to 0.41, respectively. The wide range of reliability of genomic predictions depends in part on the genetic makeup of the trait, particularly the number of loci influencing the characteristic and the dispersion of their effects. The superiority of the GS strategy over other MAS methods is determined not only by the ability to predict an individual's genetic merit with greater precision but also by the ability to shorten the breeding period and maximize genetic gains (Heffner et al., 2011). Breeders hope to pass useful traits from parental lines to their progeny by choosing parental lines with the highest GEBVs. Higher prediction accuracies will be obtained if there is a closer relationship between individuals in these populations and other unrelated populations of chipping potato clones. The relatedness among the regional/national breeding programs due to the exchange of breeding materials might be beneficial to use the training populations and prediction models for other breeding programs. Because numerous breeding programs work together to create the training phenotypic dataset, phenotyping expenses are drastically reduced (Rutkoski et al., 2015). Taken together, these factors make a genomic selection in potatoes feasible and attractive.

The use of weighted standardized multi‐trait selection indexes is a useful approach when several traits are considered in the selection process. Standardized indexes account for the differences in scales and units of measurement of traits. The number of traits and weight of each trait should reflect selection criteria that consider growers’ priorities, market needs, industry, and consumer preferences (Magnussen, 1990). This strategy has been applied to a selection of early soybean inbred lines by Gesteira et al. (2018). In potato, Terres et al. (2015) estimated genetic gains using different selection indexes in potato populations. However, no reports were found on the use of this tool in the selection of chipping cultivars for chip color, chipping quality, specific gravity, vine maturity, and yield. For the Texas A&M Potato Breeding Program, selecting clones with early maturity is important because we have a short growing season in part of the state. Selecting for earliness could also be considered to avoid undesirable weather conditions near planting or harvesting. However, clones with early vine maturity may not be suitable if they produce small tubers. On the other hand, medium vine maturing clones with early tuber bulking would be preferred. It would be recommended to emphasize early tuber bulking instead of vine maturity when breeding for a short growing cycle (Bonierbale et al., 2020). The index was efficient in selecting cultivars for multiple traits simultaneously. Several New York clones ranked high based on selection indexes calculated. The indexes can be used to select superior clones to be used as parents and ultimately to enhance the efficiency of selection in breeding programs. Although this study focused on a few important traits, it will be necessary to consider additional traits including tuber size, tuber number, internal defects, and exterior defects in advancing selections.

Statistical mixed models are currently used for data analysis in association studies, with marker effects computed using a linear transformation of genomic breeding values. The examination of marker effects using their variance aids in the identification of specific genomic regions involved in additive variance, as well as in the reduction of false positives (Duarte et al., 2014; Bernal Rubio et al., 2016). The GWAS peak on chromosome 5 was near the vine maturity gene CDF1 (Figure 4a), which is known to have a significant impact on potato maturity (Kloosterman et al., 2013). QTLs for chip color (Figure 4b) colocalized with functional genes for carbohydrate metabolism and transport such as vacuolar invertase (Pain‐1/VInv) on chromosome 3 and sucrose synthases (Sus3 and Sus4) on chromosome 7 (Gebhardt et al., 2014). Also, the major QTL on chromosome 3 colocalized with the Y locus, which codes for beta‐carotene hydroxylase (Brown et al., 2006). Thus, GWAS allow the uncovering of specific genomic regions that affect an economically important trait.

5. CONCLUSIONS

Potato growers, the chip processing industry, and consumers would benefit greatly from cultivars with good chip quality, high specific gravity, high dry matter, and high yield. This research has provided valuable insights into the application of genomic selection in the context of breeding chipping clones. Weighted Z‐score‐based multi‐trait indexes (Z _WMIS) were used to identify superior clones as parents or to advance clones through the breeding pipeline, considering multiple traits together and assigning different weights to each trait. The clones NYR102‐3, NYR102‐7, NYN24‐2, NDTX1246‐3W, NY169, W13NYP19‐2, and NYP116‐6 had consistently high‐weighted multitrait indexes. We also uncovered the genetic basis of vine maturity and chip color in chipping potato clones.

AUTHOR CONTRIBUTIONS

Jeewan Pandey: Conceptualization; data curation; formal analysis; methodology; software; writing–original draft; writing–review and editing. Douglas C. Scheuring: Data curation; methodology; writing–review and editing. Jeffrey W. Koym: Data curation; methodology; writing–review and editing. Jeffrey B. Endelman: Conceptualization; methodology; software; writing–review and editing. M. Isabel Vales: Conceptualization; data curation; funding acquisition; methodology; project administration; resources; supervision; writing–review and editing.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Supporting information

Supplemental Figure S1. Fractional values of dosage were obtained from 384 clones with the imputation of 14,401 markers compatible with an estimation of the G matrix.

Supplemental Figure S2. Distribution of G matrix coefficients obtained using 384 clones and 14,401 markers.

Supplemental Figure S3. Boxplot and Q‐Q plots of chip color to check for outliers and normality of the residuals.

Supplemental Figure S4. Boxplot and Q‐Q plots of chip quality to check for outliers and normality of the residuals.

Supplemental Figure S5. Boxplot and Q‐Q plots of specific gravity to check for outliers and normality of the residuals.

Supplemental Figure S6. Boxplot and Q‐Q plots of vine maturity to check for outliers and normality of the residuals.

Supplemental Figure S7. Boxplot and Q‐Q plots of yield to check for outliers and normality of the residuals.

Supplemental Table S1. Genomic‐estimated breeding values and reliability score (r² ) of 384 chipping clones evaluated for fried chip color, chip quality, specific gravity, vine maturity and total yield in Dalhart, TX between 2017‐20.

Supplemental Table S2. Standardized (Z‐score) selection index for individual traits and weighted multi‐trait selection index based on the priority of the breeding program: including vine maturity (Z¹ WMIS), excluding vine maturity (Z³WMIS); and based on the priority of growers: including vine maturity (Z² WMIS) and excluding vine maturity (Z⁴ WMIS) for 384 clones evaluated in Dalhart, TX between 2017‐2020.

Pandey, J. , Scheuring, D. C. , Koym, J. W. , Endelman, J. B. , & Vales, M. I. (2023). Genomic selection and genome‐wide association studies in tetraploid chipping potatoes. The Plant Genome, 16, e20297. 10.1002/tpg2.20297