Productivity under disease pressure
The line BRQ16-5409, used in this study in crossing with the line BR13-9499, was developed by the breeding program of Embrapa Soybean in Brazil through selection for good productivity, even presenting TAN lesions (susceptibility lesion) in the presence of ASR. All plants in the F2, F2:3 and F2:5 from this cross showed symptoms of TAN lesion, and the progression of severity in the parents and segregating generations was different from that shown by the resistant cultivar BRS 531 in F2:5 (Fig. 1c), indicating the absence of major resistance genes in the study population.
In addition to resistance, tolerant cultivars avoid productivity losses caused by ASR. Tolerance can be defined as the ability of the plant to support the pathogen development without presenting a significant decrease in productivity (Schafer 1971). Usually, yield losses are employed to assess tolerance, and the lower the loss, the higher the tolerance (Zambolim et al. 2022). Losses are calculated by the difference in productivity between severely infected and uninfected plots (Zadocks 1985). Zambolin et al. (2022) reported that rust-tolerant lines can be selected from segregating populations, growing under conditions of severe ASR attack, simply by selection for productivity. However, plants with TAN lesions in the F2, F2:3 and F2:5 (Fig. 1), but with good productivity, may not be tolerant. As we evaluate segregating generations, we do not have information on the loss of productivity of each genotype to know its tolerance level. In addition, one genotype may be more productive than other under disease pressure and have less tolerance. For example, a genotype ‘A’ may produce 2,000 kg/ha in the absence of the disease and 1,900 kg/ha under disease pressure, resulting in a 100 kg/ha yield reduction. Contrarily, genotype ‘B’ may produce 1,800 kg/ha in the absence of the disease and 1,750 kg/ha under disease pressure, resulting in a 50 kg/ha reduction. Thus, genotype A is more productive under disease pressure even though it is less tolerant than genotype B. Here, in this study, more productive genotypes are selected under disease pressure, which may be more or less tolerant (or non-tolerant) relative to those that non-selected.
Plants may also have an escape mechanism, showing rapid phenological development before the disease becomes severe and causes physiological damage (Zambolim et al. 2022). Although all plants had ASRS greater than 70% at the fourth evaluation in all generations (Fig. 1), earlier plants spent less time suffering the effects of rust attack. The escape mechanism was most evident in F2:5, where the correlations between SY and 50SW with DM were -0.77 and -0.75, respectively. In F2, although the Pearson correlation between 50SW and DM was -0.54, the correlation between SY and DM was non-negative (0.34).
The behavior of disease severity progression in generations in 2019, 2020 and 2022 (Fig. 1) was similar to that found by Lima et al. (2012), with an increase of more than 40% in severity level between the first and last evaluation, indicating that there were sufficient climatic conditions for disease development in the field. ASRS of more than 70% in the fourth evaluation observed in the F2, F2:3, and F2:5 generations (Fig. 1), allowed the construction of models to predict the GEBVs of soybean plants for yield components under disease pressure (Fig. 2 to 5).
Early-generation selection
Phenotypic selection was efficient in selecting 50SW, but mainly for DM and PH, which had higher heritabilities (0.79 and 0.84, respectively). However, it was unsuccessful for SY, whose correlation between F2 and F2:5 was low (r = -0.23). Several studies have indicated genomic selection use for traits with lower heritability to increase selection gains in soybean (Duhnen et al. 2017; Matei et al. 2018; Bandillo et al. 2022; Miller et al. 2023). For this, models obtained from the training and validation populations can be used to predict the GEBVs of a target population (Meuwissen et al. 2001; Sousa et al. 2019; Costa et al. 2022; Miller et al. 2023). In this study, the phenotyping of yield components of the F2:3 and F2:5 generations and the genotyping of the F2 generation from the crossing of the lines BRQ16-5409 and BR13-9499 were performed in soybean plants under ASR pressure. Such information we utilized to obtain models, described in Fig. 2 to 5, that can be used for early-generation selection in breeding programs aimed at developing soybean cultivars with higher genetic gain in yield components under ASR pressure.
F2:3 phenotyping and F2 genotyping may be beneficial in generating models if their predictive abilities are greater than those generated by models using phenotyping and genotyping in F2. The use of F2:3 in predicting SY, DM, and HP is not advantageous, because the predictive abilities of the best models using F2 and F2:3 were similar (Fig. 2 to 4). The predictive ability in F2:3 for SY was reduced by up to 54% compared to that obtained in F2, depending on the model used. On the other hand, the predictive ability of the best model using phenotyping in F2:3 for 50SW was 68.29% higher than that obtained using phenotyping in F2 (Fig. 5). The greater this superiority, the more advantageous the use of F2:3 phenotyping may be, increasing the possibility of higher selection gains. Although it is necessary to advance one generation and increase the number of plants phenotyped to build the models, this strategy does not increase the cost of genotyping, as this activity is performed in the F2 generation of the training, validation, and target populations. The results presented in Fig. 2 to 5 are related to the heritability differences estimated between generations. The difference in heritabilities for SY (0.59, 0.43, respectively), DM (0.79, 0.95), and PH (0.84, 0.93) in F2 and F2:3 were closer to those found for 50SW (0.53, 0.84). Heritabilities were obtained at the individual level for F2 and the family average level for F2:3 because data information from individuals in F2 and family averages in F3 were used to estimate the models. Costa et al. (2022) also found increased predictive ability with increasing heritability. The authors simulated data from F2 populations for twelve traits with heritabilities ranging from 0.5 to 0.8. The predictive capacity of the model generated for a trait with a heritability of 0.8 was 81% greater than that with a heritability of 0.5.
Replacing the phenotype of the F2 plants by the average of the lines of each F2:5 and F2:5(50%) family in the construction of the predictive models made it possible to obtain their predictive ability between 0.61 ± 0.01 to 0.84 ± 0.01 in F2:5 and 0.57 ± 0.03 to 0.79 ± 0.02 in F2:5(50%), depending on the trait evaluated (Fig. 2 to 5). The correlations between F2:5 and F2:5(50%) were higher than 0.90 regardless of the trait evaluated, demonstrating the similarity of their results. The high predictive ability of models using F2:5(50%) phenotyping means that the breeder can select the F2:5(50%) families of the target population based on the genotyping of individuals of this population in F2. Zhong and Jannink (2007) report that the value of a cross in a breeding program depends on the performance of its best progeny and can be considered as a linear combination of the mean of the cross’ progeny and its standard deviation. When two crosses result in populations with the same mean and different standard deviations, the greater the deviation, the greater the likelihood that the cross will produce superior lines. In the present study, models we created to select F2 genotypes that result in families F2:5 with the highest average of their 50% best lines, expressed as F2:5(50%) (Fig. 2 to 5). The use of F2:5(50%) considers the mean and the standard deviation, being unnecessary to estimate these parameters.
Models using different approaches showed different predictive abilities (Fig. 2 to 5), as confirmed in other genomic selection studies (Azevedo et al. 2013; Duhnen et al. 2017; Costa et al. 2022; Gebremedhin et al. 2024). However, regardless of these differences, the models were able to detect the same inter-generational variations, such as the increase in predictive ability in F2:3 for 50SW, the increase in predictive ability in F2:5 and F2:5(50%) for SY, and the similar inter-generational predictive abilities for DM and HP.
Breeding Program
Even though the F2, F2:3 and F2:5 evaluations in the training and validation populations were made under field conditions (Fig. 1 to 5) and the conduct of F2:4 in a greenhouse, the generation of the F2:5(50%) models in breeding programs can be done by conducting the first generations (F2, F2:3 and F2:4) in a greenhouse and only F2:5 in the field when the yield components, under ASR pressure, are evaluated. That can reduce the time to advance generations. The inoculation with the pathogen is done in F2:5 and the genotyping will be done in F2. In this study, the numbers of 230 plants genotyped in F2 and 1380 F2:5 lines (230 families x six lines per family) evaluated for ASRS have proven adequate for building the models. Alternatively, the evaluation of yield components under disease pressure can be done in F2:4 by sowing lines within families, which will reduce one generation cycle.
After obtaining the models from the training and validation populations, F2 plants from the target population are genotyped. The number of genotyped F2 plants may vary depending on the structural capacity of the breeding program. Considering a hypothetical case of 1000 genotyped F2 plants and a selection percentage of 10%, it is possible to obtain 100 F3 lines (1 F2 plant: 1 F3 line) and 10,000 F4 lines (1 F3 line: 100 F4 line). A new assessment of ASRS can be made in F2:4. Since genomic selection eliminated 90% of the plants in F2, this evaluation would correspond to one in 100,000 lines in a selection based on phenotype alone. In the latter case, the breeding program would require a larger logistical structure to conduct and harvest the lines, inoculate the pathogen, and maintain appropriate environmental conditions for allowing field disease development. The ASRS assessment in the target population may not be necessary since the models were generated from genotypes under disease pressure in the training population. However, that evaluation can be performed to enhance selection accuracy.
Genomic selection showed efficiency in predicting yield components of soybean genotypes under ASR pressure in subsequent generations (F2:3 and F2:5) from the cross between BRQ16-5409 and BR13-9499. For this purpose, phenotyping of advanced generations (F2:3 and F2:5) under disease pressure and genotyping of the F2 generation were used to obtain genomic prediction models. The use of these models for early-generation selection in a target F2 population will reduce the number of segregating lines and consequently optimize the mechanical and financial resources of genetic breeding programs to develop productive soybean genotypes under ASR pressure and increase selection gains.