Environmental Variance and Genetic Differences in Nut Weight for Chestnut Breeding

Abstract

Large nut weight (NW) is an important target in chestnut breeding. The present study was conducted at the Nakatsugawa branch of the Gifu Prefectural Research Institute for Agricultural Technology, an important chestnut breeding site in Japan. We estimated the environmental variance components for NW by using nine cultivars/selections with three single-tree replicates for six years. The number of nuts evaluated for each tree was mostly over 100 nuts, with an average of 440 nuts. Subsequently, the dataset of average NW values for each tree was log-transformed and analyzed by ANOVA. The effects of genotype, year, and the genotype × year interactions were highly significant, whereas the tree effect was negligible. The resulting environmental variance components for the log-transformed values were as follows: variance among years = 5.2 × 10⁻⁴, variance among trees within genotype = 0, variance associated with the genotype × year interactions = 7.6 × 10⁻⁴, and residual variance = 11.2 × 10⁻⁴. The results suggest that tree replication is not necessary to evaluate the genotypic value of cultivars/selections or offspring in breeding and that year effect adjustment and yearly repeated measurements can effectively reduce environmental variance. The NW of 27 cultivars/selections with potential as cross-parents was also estimated with no tree replications and via measurements repeated for three years; the results ranged from 10.7 to 47.4 g, with high broad-sense heritability (0.93) based on a three-year evaluation for the log-transformed data. Based on the environmental variance estimates, the cultivars/selections used in chestnut breeding with NW above 22.3 and 26.7 g should be selected by evaluating one tree for three years at 95% probability and selecting those having nuts of 25 and 30 g or more, respectively.

Introduction

Japanese chestnut (Castanea crenata Sieb. et Zucc.) has been cultivated for food and timber for almost 6,000 years (since 3500 BCE), and it has been planted in hilly and mountainous areas in the past (Beccaro et al., 2019; Sawamura, 2006). Chestnuts are well adapted to the mild and humid climate in Japan. Even so, Japan has complex and diverse weather conditions because of the latitude distribution of its extended islands (from 20°N to 45°N), prevailing winds (westerlies and monsoons), and topography (≈75% of Japan consists of mountainous areas; MILT n.d.). Nowadays, flat land at low altitude is usually selected for plantations, although the Japanese chestnut production areas are often located in mountainous areas (Beccaro et al., 2019), where the larger diurnal temperature variation and the lower annual mean temperature may affect their growth. For example, the nut harvest time for the same cultivar could differ by several weeks between two area types (Saito et al., 2009, 2015).

Chestnut breeding studies in Japan have been conducted mainly at the Institute of Fruit tree and Tea Science, National Agriculture and Food Research Organization (NIFTS), which is located on flat land at a low altitude (23 m), and the Nakatsugawa branch of the Gifu Prefectural Research Institute for Agricultural Technology (GRIAT), located in a mountainous area at a moderately high altitude (390 m). They have developed and released several chestnut cultivars with important traits, including gall wasp resistance (Shimura, 1972) and an easy-peeling pellicle (Kamio et al., 2020; Saito et al., 2009).

Nut weight (NW) is an essential trait for the economic success of chestnut growers (Kang et al., 2019); therefore, the NW of the cultivars recently released in Japan has been very large, i.e., over 25 g (Saito et al., 2009, 2015). Environmental variance estimation is essential for efficient selection, especially in tree fruit breeding, because of the long selection time and the large space required for the trees (Hansche, 1983; Nyquist and Baker, 1991). Nishio et al. (2014) estimated the environmental variance of essential traits including NW, and found a low broad-sense heritability in the Japanese chestnuts developed at NIFTS. However, they analyzed only eight cultivars/selections. Therefore, the genetic differences and environmental variances among genetic resources, including many cultivars, are still unknown. Moreover, environmental variances should be estimated for each breeding site since environmental conditions (temperature, rainfall, etc.) differ among sites.

Therefore, environmental variances and genetic differences at GRIAT should be elucidated to allow chestnut breeding to efficiently achieve large NW. The objectives of the present study were as follows: to estimate the environmental variance of chestnut NW; to clarify the genetic differences for NW among 27 cultivars/selections with potential as cross-parents in chestnut breeding, and to estimate the critical values for NW in phenotypic selection for newly introduced cultivars and selections yielded from crosses in chestnut breeding.

Materials and Methods

Expt. 1: Environmental variance estimation using nine cultivars/selections, with three single-tree replicates for six years

Three grafting clones of nine cultivars/selections (‘Arima’, ‘Ena 1’, ‘Kinka’, ‘Kunimi’, ‘Moriwase’, ‘Riheiguri’, ‘Tanzawa’, Tono 6, and ‘Tsukuba’; Table 1) were used over six years (2005–2010). They were adult trees (25–35 years old in 2013), planted in different rows at a spacing of 5 m × 5 m in a single field at GRIAT (Nakatsugawa, Gifu, Japan; 35°29'N, 137°28'E). For stable production, they were pruned every winter using a commercial method in Gifu prefecture.

Table 1

Cultivars/selections used in this study.

We harvested and selected intact nuts (i.e., without damage by split, insect infestation, etc.) every day. The NW was calculated by dividing the gross weight of the intact nuts by the number for each tree, as proposed by Nishio et al. (2014). The log-transformed NW data were subjected to analysis of variance (ANOVA) with the model used by Nishio et al. (2014), which estimated the environmental variance of nut traits in Japanese chestnut breeding.

The model for the phenotypic value of NW (P1_ijk) was   

$${P\mathit{1}}_{ijk}=\mu \mathit{1}+{g\mathit{1}}_i+{t\mathit{1}}_{ij}+{y\mathit{1}}_k+{(gy\mathit{1}}_{ik})+{r\mathit{1}}_{ijk}$$

where μ1 is a constant (the overall mean), g1_i is the effect of the i^th genotype, t1_ij is the effect of the j^th tree replicate within the i^th genotype, y1_k is the effect in the k^th year, (gy1)_ik is the interaction between the i^th genotype and the k^th year, and r1_ijk is the residual effect of the j^th tree replicate of the i^th genotype in the k^th year.

The ANOVA provided the genetic variance (σ_g1²) and the environmental variance; the latter consisted of the year variance (σ_y1²), the tree variance within genotype (σ_t1²), the genotype × year interaction (σ_gy1²), and the residual variance (σ_r1²). The distribution of the residual estimates approached normal distribution, and was not significant at the 5% level in the Kolmogorov–Smirnov one-sample test, indicating that the ANOVA is applicable (Campbell, 1974).

The residual variance normally consists of the tree × year interaction variance and the variance among nuts within a tree divided by the number of nuts from the tree (Yamada, 2011). In this study, the number of nuts from a tree was not constant, but mostly over 100, with an average of 440, possibly resulting in bias for the variance estimation. However, bias was estimated as minimal when the number of nuts evaluated is > 100 (Nishio et al., 2014).

Expt. 2: Genetic differences and environmental variance estimation using 27 cultivars/selections, with one tree for three years

The NW was evaluated as described in Expt. 1 for 27 chestnut cultivars/selections (Table 1) over three years of repetition (2012–2014) by using one adult tree per cultivar/selection, which was planted and grown as in the previous experiment. The NW was calculated by dividing the gross weight of intact nuts by the number for each tree, which was mostly over 100 with an average of 320.

The log-transformed NW data were subjected to ANOVA using the following model:   

$${P\mathit{2}}_{ij}=\mu \mathit{2}+{g\mathit{2}}_i+{y\mathit{2}}_j+{r\mathit{2}}_{ij}$$

where μ2 is a constant (the overall mean), g2_i is the effect of the i^th genotype, y2_j is the effect in the j^th year, and r2_ij is the residual effect of the i^th genotype in the j^th year.

The ANOVA provided the genetic variance (σ_g2²) and the environmental variance, which consisted of the year variance (σ_y2²) and the residual variance (σ_r2²). The distribution of the residual estimates approached normal distribution, and was not significant at the 5% level in the Kolmogorov–Smirnov one-sample test.

Results and Discussion

Environmental variance estimation of chestnut NW

In Expt. 1, we evaluated NW using nine cultivars/selections, with three single-tree replicates for 6 years to estimate environmental variance. The ANOVA showed that the effects of genotype, year, and genotype × year interactions were significant (P < 0.01), while the tree effect was not significant and negligible (P > 0.05; Table 2). The estimated σ_g1² value was 47.0 × 10⁻⁴; this parameter included not only the difference in genetic values, but also the effect of the pollinizer tree, which could influence the NW of the seed parent (Takada et al., 2010) because each cultivar was planted in a row (i.e., the pollinizer was different for each cultivar). Further study is necessary to determine the exact genetic variance of NW. The estimated variances, based on the log-transformed values, were as follows: σ_g1² = 47.0 × 10⁻⁴, σ_y1² = 5.2 × 10⁻⁴, σ_t1² = 0, σ_gy1² = 7.6 × 10⁻⁴, and σ_r1² = 11.2 × 10⁻⁴ (Table 3). The largest environmental variance component was σ_r1²; which accounted for 47% of the total environmental variance. Nishio et al. (2014) also reported that the residual variance was estimated to be the largest environmental variance component (17.8 × 10⁻⁴), accounting for 52% of the total environmental variance components. The year variance, the genotype × year interaction variance, and the tree variance within genotype in this study were slightly deviated from the estimates of 8.2 × 10⁻⁴, 4.3 × 10⁻⁴, and 4.0 × 10⁻⁴, respectively, in Nishio et al. (2014). In the present study, the genotype and age of trees used, years, and the location, including altitude, were different from those of Nishio et al. (2014). It is possible that slight deviations may be due to these differences in materials and environments.

Table 2

Analysis of variance for nut weight using nine cultivars/selections with three trees per genotype for six years (2005–2010).

Table 3

Estimates of the variance components in the analysis of variance for nut weight.

The results showed very little tree effect, indicating that tree replications are not necessary to evaluate the genotypic value of cultivars/selections or offspring in breeding programs in which NW is evaluated in the same way as in the present study. Also, the significant year effect and the negligible tree effect within genotype suggest that year repetition is more accurate than tree repetition for chestnut NW selection, similar to persimmon (Yamada et al., 1993) and grape (Sato et al., 2000). The evaluation cost is very high for tree replications compared with yearly repetitions (Yamada, 2011), and this conclusion may be very helpful for chestnut breeders.

The year effect is what all cultivars/selections have in common. The significant year effect observed in the present study suggests that its adjustment may effectively reduce environmental bias when evaluating NW genotypic values over several years, which is normal in the breeding of woody fruit and nut crops (Yamada, 2011; Yamada et al., 1994a, b). Moreover, a significant year effect was also observed by Nishio et al. (2014). The year effect can be adjusted, and neglected, by the yearly mean value of several cultivars/selections that are evaluated every year as control genotypes (Yamada, 2011; Yamada et al., 1994a). The efficiency of this yearly adjustment was evaluated based on the ratio between the year variance and the error variance of the yearly mean value of the control genotypes (adjustability coefficient); when n cultivars/selections in each tree are evaluated every year as control genotypes, the yearly mean value has an error variance expressed as (σ_gy1² + σ_r1²)/n, and the adjustability coefficient is given by σ_y1²/{(σ_gy1² + σ_r1²)/n} = 0.28n. When the adjustability coefficient exceeds 1, the year effect adjustment is effective; hence, at least four control genotypes had to be evaluated every year in the present study. In practice, if 10 cultivars/selections were evaluated every year, adjustability would be 2.8.

Genetic differences in NW among 27 cultivars/selections with potential as cross-parents in chestnut breeding

The 27 cultivars/selections used in Expt. 2 mainly consisted of Japanese chestnuts (Table 1), but also included some Japanese–Chinese hybrids (‘Hayashi 1’, ‘Hayashi 2’, ‘Mikuri’, ‘Riheiguri’, ‘Shuhou’, and 92-01) and Chinese chestnuts (‘Beijing 8070’, ‘Houji 360’, and 92-02; Nishio et al., 2020). No tree replication was used because of the negligible tree effect within genotype estimated in Expt.1. The average NW values based on the evaluation for 3 years ranged from 10.7 g (‘Beijing 8070’) to 47.4 g (‘Kannabe’), as shown in Figure 1. The Chinese chestnut cultivars had small NW values, while the Japanese ones were generally large and varied widely. Additionally, the Japanese–Chinese hybrids showed intermediate results, as reported by Nishio et al. (2020).

Fig. 1

Variation in the average values of nut weight (grams) among 27 cultivars/selections (2012–2014). The bars represent the standard error with the original scale transformed from the standard error calculated based on variance estimates with log-transformed values. Japanese chestnuts, Chinese chestnuts, and Japanese-Chinese hybrids are indicated by J, C, and H, respectively, in parentheses.

Significant genotype and year effects (P < 0.01) were observed in ANOVA (Table 4). The estimated σ_g2², σ_y2², and σ_r2² values were 131.9 × 10⁻⁴, 5.6 × 10⁻⁴, and 25.8 × 10⁻⁴, respectively. The σ_g2² value greater than σ_g1² (47.0 × 10⁻⁴; Table 3) confirms the large genetic variation among all the cultivars/selections used for chestnut breeding at GRIAT, while σ_y2² were nearly equal to σ_y1². Although σ_g2² included the tree effect within genotype due to single-tree replication, we regarded it as genetic variance because the estimated σ_t1² was 0, while we should consider that the pollinizer effect on NW was not removed from σ_g2² in the same way as in Expt. 1. The broad-sense heritability, defined as σ_g2²/(σ_g2² + σ_y2² + σ_r2²), was 0.81, indicating that the variation of phenotypic values of NW among the 27 cultivars/selections analyzed (Fig. 1) was mainly due to the large genetic variance, which was larger than in the previous study (broad-sense heritability = 0.27; Nishio et al., 2014). The mean value for three years should be used as the value of each cultivar/selection to assess the relative genetic difference among 27 cultivars/selections. In this case, broad-sense heritability is estimated as σ_g2²/{σ_g2² + (σ_y2² + σ_r2²)/3}, and at 0.93.

Table 4

Analysis of variance for nut weight using 27 cultivars/selections with one tree per genotype for three years (2012–2014).

Estimation of the critical values for NW for phenotypic selection of cultivars/selections

The inferred genotypic value provides the critical value in selection for given measurement repetitions; breeders may evaluate the genotypic value of a new cultivar/selection yielded from crossing in a similar way to the present study. When analyzing average NW of over 100 intact nuts per tree from a single tree per genotype for several year repetitions with a log-transformed value and year effect adjustment using 10 control genotypes, the lower one-tailed limit with 95% confidence (X_pl) of the genotypic value of NW (X_G) with σ_r2² based on normal distribution is calculated as   

$${\mathit{log}}_{10}\left(X_{pl}\right)={\mathit{log}}_{10}\left(X_G\right)-1.645\times \sqrt{\frac{{{\sigma }_{r2}}^2+{({\sigma }_{r2}}^2/10)}{y}}$$

where y is the number of year repetitions. We used σ_r2² instead of σ_gy1²+σ_r1² as it is more appropriate for the selection of unknown genotypes because of the larger number of cultivars/selections used in Expt. 2. The σ_r2²/10 is the error variance due to adjustment of year effects using the yearly mean value of the 10 control genotypes (Yamada, 2011; Yamada et al., 1994a, b). The standard error calculated from these values was multiplied by 1.645 (the approximate value of the one-tailed 95 percentile in the standard normal distribution) to calculate the confidence limit. For an original NW scale (g) based on the results, the cultivars/selections in breeding having NW above 22.3 and 26.7 g should be selected when evaluating one tree for three years at 95% probability to select those of 25 and 30 g or more, respectively (Fig. 2). To select genotypes with a 25 g NW, > 21.7 and > 22.8 g are critical values for selection in 2- and 5-year evaluations, respectively. Likewise, to select genotypes with a genotypic value of > 30 g, genotypes exhibiting > 26.0 and > 27.4 g should be selected, respectively.

Fig. 2

Lower one-tailed confidence limit at P = 0.95 (X_pl) of the nut weight of a genotype using a single tree (vertical axis) and the number of yearly replications (y: horizontal axis). The nut weights of genotype values (X_G) were assumed to be 25 (open circle) and 30 (solid circle). X_pl was calculated as follows: ${\mathit{log}}_{10}\left(X_{pl}\right)={\mathit{log}}_{10}\left(X_G\right)-1.645\times \sqrt{({{\sigma }_{r2}}^2 +\frac{{{\sigma }_{r2}}^2}{10})/\mathrm{y}}$, where the year effect is adjusted by a yearly mean of 10 control genotypes evaluated every year.

In conclusion, our results provided a selection strategy for chestnut NW in which yearly repetitions and adjusting year effects rather than tree replication effectively reduces environmental variance. Genetic differences in NW among 27 cultivars/selections with potential as crossing-parents in chestnut breeding were elucidated with high broad-sense heritability by evaluating NW for 3-year repetitions. Furthermore, we clarified a critical value in phenotypic selection to select genotypes with high genotypic values for NW in an evaluation over several years. These findings may be practically useful to succeed in chestnut breeding aimed at large NW.

Literature Cited

• Beccaro, G., A. Alma, G. Bounous and J. Gomes-Laranjo. 2019. The chestnut Handbook: Crop & Forest Management. CRC Press, Boca Raton, FL.
• Campbell, R. C. 1974. Statistics for biologists., 2nd ed. Cambridge Univ. Press, Cambridge, UK.
• Hansche, P. E. 1983. Response to selection. p. 120–128. In: J. N. Moore and J. Janick (eds.). Methods in fruit breeding. Purdue Univ. Press, Indiana.
•  Kamio,  S.,  Y.  Miyamoto and  H.  Isomura. 2020. Breeding the new chestnuts cultivar ‘Ena Hourai’ and its characteristics. Bull. Gifu Prefect. Res. Inst. Agric. Technol. 15: 20–26 (In Japanese).
•  Kang,  M. J.,  A. Y.  Shin,  Y.  Shin,  S. A.  Lee,  H. R.  Lee,  T. D.  Kim,  M.  Choi,  N.  Koo,  Y. M.  Kim,  D.  Kyeong,  S.  Subramaniyam and  E. J.  Park. 2019. Identification of transcriptome-wide, nut weight-associated SNPs in Castanea crenata. Sci. Rep. 9: 13161. DOI: 10.1038/s41598-019-49618-8.
• MILT. n.d. Land and Climate of Japan. <https://www.mlit.go.jp/river/basic_info/english/land.html> (Accessed: August 30, 2021).
•  Nishio,  S.,  S.  Ruan,  Y.  Sawamura,  S.  Terakami,  N.  Takada,  Y.  Takeuchi,  T.  Saito and  E.  Inoue. 2020. Genetic evidence that Chinese chestnut cultivars in Japan are derived from two divergent genetic structures that originated in China. PLoS One 15: 7. DOI: 10.1371/journal.pone.0235354.
•  Nishio,  S.,  M.  Yamada,  N.  Takada,  H.  Kato,  N.  Onoue,  Y.  Sawamura and  T.  Saito. 2014. Environmental variance and broad-sense heritability of nut traits in Japanese chestnut breeding. HortScience 49: 696–700.
•  Nyquist,  W. E. and  R. J.  Baker. 1991. Estimation of heritability and prediction of selection response in plant populations. Crit. Rev. Plant Sci. 10: 235–322.
•  Saito,  T.,  K.  Kotobuki,  Y.  Sawamura,  K.  Abe,  O.  Terai,  M.  Shoda,  N.  Takada,  Y.  Sato,  T.  Hirabayashi,  A.  Sato,  T.  Nishibata,  Y.  Kashimura,  T.  Kozono,  H.  Fukuda,  T.  Kihara,  K.  Suzuki and  M.  Uchida. 2009. New Japanese Chestnut Cultivar ʻPorotanʼ. Bull. NARO Inst. Fruit Tree Sci. 9: 1–9 (In Japanese with English abstract).
•  Saito,  T.,  K.  Kotobuki,  Y.  Sawamura,  N.  Takada,  T.  Hirabayashi,  A.  Sato,  M.  Shoda,  O.  Terai,  T.  Nishibata,  Y.  Kashimura,  K.  Abe,  S.  Nishio,  T.  Kihara,  K.  Suzuki and  M.  Uchida. 2015. New Japanese Chestnut Cultivar ʻMikuriʼ (Castanea crenata Sieb. et Zucc). Bull. NARO Inst. Fruit Tree Sci. 19: 1–9 (In Japanese with English abstract).
•  Sato,  A.,  M.  Yamada,  H.  Iwanami and  N.  Hirakawa. 2000. Optimal spatial and temporal measurement repetition for reducing environmental variation of berry traits in grape breeding. Sci. Hortic. 85: 75–83.
• Sawamura, Y. 2006. Chestnut, p. 82–85. In: Japan. Soc. Hort. Sci. (ed.). Horticulture in Japan 2006. Shoukadoh, Kyoto, Japan.
•  Shimura,  I. 1972. Breeding of chestnut varieties resistant to chestnut gall wasp, Dryocosmus kuriphilus Yasumatsu. Japan Agric. Res. Q. 6: 224–230.
•  Takada,  N.,  A.  Sato,  Y.  Sawamura,  S.  Nishio and  T.  Saito. 2010. Influence of pollen on pellicle removability and nut weight of Japanese chestnut (Castanea crenata Sieb. et Zucc.) ʻPorotanʼ. Acta Hortic. 866: 239–242.
• Yamada, M. 2011. Crossbreeding in woody fruit crops (In Japanese). Yokendo, Tokyo.
•  Yamada,  M.,  H.  Yamane and  Y.  Ukai. 1994a. Efficiency of use of control genotypes for reducing yearly fluctuations of quantitative characters of quantitative fruit characters in Japanese persimmon breeding. Bull. Fruit Tree Res. Stn. 26: 29–37.
•  Yamada,  M.,  H.  Yamane and  Y.  Ukai. 1994b. Genetic analysis of Japanese persimmon fruit weight. J. Amer. Soc. Hort. Sci. 119: 1298–1302.
•  Yamada,  M.,  H.  Yamane,  K.  Yoshinaga and  Y.  Ukai. 1993. Optimal spatial and temporal measurement repetition for selection in Japanese persimmon breeding. HortScience 28: 838–841.