Development of Brassica Napus L. Ogu-INRA CMS Restorers Using Recurrent Full-Sib Selection


 The Ogu-INRA CMS system in canola and rapeseed (Brassica napus L.) uses a cytological variant of the radish- (Raphanus sativus L.) derived Ogu CMS pollination control system introduced through interspecific introgression. The restorers (R-lines) contain an introgression that is associated with poor agronomic performance due to a large undesired segment of the radish chromosome that was introgressed along with the Rfo gene. The introgression contains pentatricopeptide (PPR) motif repeats that confer fertility restoration abilities to the R-lines. The objective of this research was to test the hypothesis that multiple cycles of intermating will result in R-lines with improved agronomic performance. A base population was developed by designing five R-line by R-line crosses. Twelve plants from each initial cross were grown and chain-crossed at random, without selection, other than the presence of the Rfo gene. Twelve flowers from each plant were crossed and the remainder of the plant was selfed. Three intermating crossing cycles (C0, C1 and C2) were completed and each was selfed three times for evaluation. Total pod number, seeds per pod, a visual pod rating, thousand seed weight and yield were evaluated. The visual pod rating showed a positive correlation with seeds per pod. Improvements for all traits were found at C0 and C1 when compared to the best parent. Individual families from two of the crosses showed a yield increase of over 78 % from the best parent. This suggests that improvements in yield components can be obtained from intermating R-lines.


Introduction
Brassica species exhibit both cross-pollination and high levels of self-pollination (Rakow and Woods 1987). Due to this variation in pollination strategies, several breeding and selection methods can be used to improve Brassica populations (Kumar et al. 2015). Some of the most widely used breeding methods for Brassica species include pedigree breeding, backcross breeding, development of synthetic and composite varieties, development of hybrids, recurrent selection, mass selection, doubled haploid (DH) development, in vitro mutagenesis and genetic transformation (Prakash et al. 2009). Most of these methods were rst evaluated in different crops such as maize and wheat. Around the 1940s, corn breeders developed the recurrent selection method to improve populations (Hallauer and Carena 2012). The rst time the "recurrent selection" term was used, it described a strategy for speci c combining ability aimed at high-yielding corn where one cycle of breeding was completed after selecting and intermating the best selfed hybrids that originated from a cross between a crossbred lot and a 'tester line' (Hull 1945). From this rst recurrent selection method, variations were developed including reciprocal recurrent selection for half-sib (Comstock et al. 1949) and full-sib progenies (Hallauer and Eberhart 1970). Reciprocal full-sib selection was proposed as an e cient breeding method in which non-additive genetic effects, as well as additive effects were important in the expression of hybrid superiority (Peiris and Hallauer 2005). This method was used for producing hybrids (full-sib progenies) and selfed seed on the same plant, as well as early testing of single cross combinations (Hallauer and Eberhart 1970).
Systems that repeat the same process for several cycles are referred to as recurrent selection systems (Norman 1992). The process involves, production of a set of genotypes (individuals or families), evaluation of those genotypes, identi cation of the superior ones to be used as parents for the next generation and intermating the selected genotypes to produce the next generation, obtaining an improved generation (Norman 1992). This method has continued to evolve as breeders modify it for different crops and traits. In 2010, a recurrent selection strategy was evaluated in a greenhouse study of dry bean as a means to pyramid white mold resistance genes (Terán and Singh 2010). Phyto-recurrent selection was another variation on the method developed to choose superior-performing Populus and Salix genotypes for phytoremediation purposes (Zalesny et al. 2010). In this case, recurrent selection was conducted both in greenhouses and growth chambers before moving to the eld to increase the speed of the evaluation and selection process. A marker-based recurrent selection strategy was successfully utilized in barley to increase heterozygosity and evaluate other yield components (Nandety and Geiger 2014).
Depending on the objective of each breeding program, there are other variables that can be modi ed within the recurrent selection method. Some of the most common recombination methods in cross-fertilizing species include diallel crossing, chain crossing, bulk entry and bulk plot (Darbeshwar 1992). The best methods for a small number of genotypes are diallel crossing and chain crossing (Darbeshwar 1992). In diallel crossing, the selected lines/families are intermated in all possible combinations by hand pollination (Darbeshwar 1992). Chain crossing involves the intermating between entries as described in Figure 1. This method is also referred to as a partial diallel where each entry is crossed to the same number of entries providing a balanced composite of the same number of crossed seed from each selected entry (family or genotype) to initiate the next cycle (Norman 1992). The combination of chain crossing and full sib recurrent selection produces a method to evaluate several combinations in a relatively small population and helps maintain a manageable number of entries in every cycle.  (Table 1). Twelve plants from each initial cross (full-sibs) were grown and chain-crossed ( Figure 1) at random, without any pre-selection other than the presence of an Rfo SCAR marker (Hu et al. 2008). Twelve owers from each plant were crossed (C 1 S 0 rst full-sib intermating) and the remainder of the plant was selfed (C 0 S 1 ). All pods were harvested from each of the twelve plants keeping the selfed and the intermated pods separate. One seed from the intermated pods of each plant was chosen at random and planted. Twelve owers from that plant were crossed again to create the next intermating cycle (C 2 S 0 second full-sib intermating). The twelve plants at each cycle formed 'families' since they originated from the same cross and were only intermated among genotypes with the same background. The number of plants per family remained consistent throughout the experiments. Two full-sib intermating crossing cycles (C 1 and C 2 ) were completed and each was selfed twice in order to compare all populations after their second sel ng (C 0 S 2 , C 1 S 2 and C 2 S 2 ) for experiment one. This generation was selfed to compare the intermating cycles at their third sel ng generation (C 0 S 3 , C 1 S 3 and C 2 S 3 ) (Fig. 2)

Statistical analyses
Analysis of variance of the individual intermating cycles, sel ng generations and their interaction were conducted for each cross and the ve traits using Agrobase ® (Agronomix software, Winnipeg, Canada) as a complete block design. Experiments were analyzed separately due to the different generations (S2 and S3) evaluated in each experiment. Analyzing and comparing these experiments separately was facilitated by the control genotypes grown at each experiment. Least signi cant differences (LSD) were calculated to identify signi cant differences following the ANOVA results. The entire raw dataset was plotted using bivariate scatter plots for each pair of variables (10 panels pairing any two of the ve yield components) using the pairplot function from Seaborn in Python (Seaborn -Michael Waskom, New York, USA https://seaborn.pydata.org/).

Results
Data from all traits were analyzed to detect the effects of intermating cycles and crosses for all of the yield components evaluated. The combined analysis of all crosses demonstrated a signi cant effect on most variables for all traits. A representative example of this analysis is shown in Table 2 for mean pod rating at S 2 and S 3 (ANOVA tables for all other traits can be found in Appendix 1).
A bivariate scatter plot shows the relationship between the ve traits for all crosses combined (Fig. 4). As expected, yield has a linear positive correlation with and pod number (r = 0.85) (Fig. 4 d) and seeds per pod (r = 0.80) ( Fig. 4 j) ( Table 3). A negative correlation was found for seeds per pod and thousand seed weight (r = -0.25). The calculated correlation values between pod rating and the other traits was negative because the pod rating scale goes from 1 -9 where one represents a plant with the best pods and 9 a represents a sterile plant (in contrast to all the other traits in which a higher number indicated a better Regarding cycles of sel ng, differences were found between the two greenhouse experiments (S 2 and S 3 ). In the S 3 generation, the means for pod number, yield and seeds per pod for all three cycles were higher compared to the S 2 generation ( Table 4).
Highlights of the signi cant intermating effects for both experiments and each cross are presented for each trait individually (Table 5). Cross three had signi cant effects for the intermating cycles in all traits and both experiments; thus, cross three will be used to illustrate results for most traits. Data for all other crosses can be found in appendix 2.

Pod number
The analysis of the complete data set showed the largest mean pod number among all intermating cycles at C 0 S 2 (Table 4). Mean pod number was the highest among all intermating cycles after the rst cycle of full-sib intermating (C 1 ) in three of the ve crosses evaluated at S 2 ( Table 5). In the second experiment (S 3 ), cross 3 showed a larger mean pod number at C 2 . An evaluation of the individual number of families with improved mean pod number when compared to the best parent showed 7 improved families at C 1 S 2 and no signi cant improvements between cycles at S 3 except for C 1 in cross 2 (Table 6).
Data for the individual families at each intermating cycle in both experiments is shown in Table 7.

Pod Rating
During analyses of the entire data set, no signi cant differences were found between the intermating cycles in either experiment for pod rating (Table 4). Differences in the overall mean pod rating were found in three of the ve crosses at S 2 and one cross in S3 (Table 5). At S 2 , cycles 0 and 1 had the best mean pod rating scores. In contrast at S 3 , cross 3 showed an improvement in pod rating when compared to the best parent after two full-sib intermating cycles (C 2 ).
Cross 5 included a parent with the worst eld rating score (Table 1). However, a signi cant improvement was found at S 2 after one cycle of full-sib intermating (C 1 ) and at S 3 after one intermating cycle (C 0 ) ( Table 5). Comparison of the mean pod rating for the individual families in cross 5 showed a similar trend with the improvement of one out of 12 families at S 2 after one cycle of full-sib intermating and 5 families out of 12 when compared to the best parent at C 0 S 3 (Table 8).
Thousand seed weight The analysis of the complete data set showed the largest mean TSW among all intermating cycles at C 0 S 2 ( Table 4). Cross 1 had a higher mean thousand seed weight at S 2 after one cycle of full-sib intermating (C 1 ) ( Table 5). No signi cant mean TSW differences were found in the other four crosses for any intermating cycles in either experiment (Table 5). For cross 5, C 2 had the highest number of improved families (5/12) at S 2 . The same number of improved families (2/12) was observed in C 0 and C 1 at S 3 (Table 9).

Seeds per pod
No signi cant differences for SPP were found between the intermating cycles on either experiment when analyzing the total data set ( Table 4). The cycle means for SPP showed one cycle of full-sib intermating (C 1 ) was better than the other intermating cycles for 2 of the 5 crosses at S 2 (Table 5). When comparing the means of individual families in every cycle at S 2 , only C 1 and C 2 had an improved number of seeds per pod when compared to the best parent (Table 10). No signi cant differences were found for the overall cycle means or the number of improved families when compared to the best parent at S 3 (Table 10).

Yield
The analysis of the complete data set showed the highest mean yield among all intermating cycles at C 0 S 2 (Table 4). Mean yield was highest following the initial intermating cycle (C 0 ) for crosses 1 and 2. Cross 2 showed an improved mean yield of 1.09 g. while the best parent had a mean yield of 0.77 g. in S 2 (Table 5). Crosses 3 and 5 exhibited higher mean yield at C 1 . The cycle with the highest number of improved families from all crosses was C 0 at S 2 ( Table 11).
The largest number of improved families for a single cross was found at C 1 S 2 for cross 3. Data for the individual families of cross 3 at each intermating cycle in both experiments is shown in Table 12.

Improvement of individual families
Two of the crosses with the greatest improvements were crosses three and ve. At least one of the twelve families from each of these crosses showed an increase of >36 % in pod number and yield when compared to the best parent after one cycle of full-sib intermating (C 1 S 2 ) (Table 13).

Discussion
The goal of this study was to determine whether multiple cycles of intermating with recurrent selection would result in R-lines with improved agronomic performance. The cycle effects on all the crosses combined, would suggest that C 0 is the best for improving pod number, TSW and yield. However, when evaluating each cross individually and the number of improved families per cross and cycle, C 1 showed the most improvements when compared to the best parent. . This is to be expected from 'poor pods' (6 -8 in the pod rating scale (Fig. 3) which produce a maximum of 5 seeds that are larger and heavier than the average (Olsson 1959). A positive correlation was observed for yield (Fig. 4 j) and seeds per pod (Fig. 4 f) with the pod rating. This a good indicator for the accuracy of the pod rating scale (Fig. 3). Future studies can utilize this pod rating to predict high yielding restorer plants. The use of a pod rating scale facilitates selection of improved genotypes and potentially decreases the time needed to make selections on restorer genotypes.
The time of the year when the experiments were grown in the greenhouse had an effect on the performance of all the genotypes evaluated. The rst experiment was planted in June and grew during the summer months where temperatures were higher (on average 10 °C higher for 2-3 hours a day at S 2 due the cooling and shading limitations of the greenhouse), which had a negative effect on the plants. Parental genotypes (controls for both experiments) also exhibited lower scores for all the previously mentioned traits when compared to the S 3 experiment that was planted in December.
The initial step in any plant breeding strategy, is to identify plants with desirable traits among existing plant genotypes (or developing novel phenotypes if they are unavailable) (Ahmar et al. 2020). Due to the nature of the traits evaluated, selection could not take place before intermating the restorer genotypes. Random mating among the twelve genotypes per family was an important factor in this research. As observed in a soybean recurrent half-sib selection study, random mating can have a positive effect on the oil content of seeds but no signi cant results were found for yield or seed weight (Feng et al. 2004). Similar results were found in our study where mean seeds per pod and mean pod number were improved when compared to the best parent at cycles C 1 and C 2 ( Table 5 and Table 10). Mean yield improved when compared to the best parent at C 0 in cross 2 and at C 1 in cross 3 at S 2 (Table 5). These results suggest that up to one cycle of random full-sib intermating can improve yield components in fertility restorer genotypes.
As described earlier, full-sib recurrent selection is a plant breeding method that has been well established and can break of random full-sib intermating in select crosses. One family from cross 3 and one from cross 5 showed an improvement of over 70 % in yield when compared to the best parent after one cycle of full-sib intermating (C 1 ) and over 36 % in pod number (Table 13). This indicates that intermating restorer genotypes can produce improved restorer genotypes.
The ability to generate an improved restorer genotypes is of the utmost importance to produce improved canola and oilseed rape hybrids. Dozens of improved   Tables  Table 4. Means for all yield components and pod rating measured for the complete data set of Brassica napus L. fertility restorer crosses at all intermating cycles (C 0 , C 1 , C 2 ) in the two greenhouse experiments (S 2 and S 3 ). *Significantly best among all intermating cycles at p = 0.05. a Pod rating scale from 1 -9 where 1 represents a full straight pod with more than 25 seeds; 2 represents a straight pod with 15 -25 seeds; 3 represents a curved pod with 10-15 seeds; 4 represents a straight pod with 7 -9 seeds; 5 represents a curved pod with 6-8 seeds; 6 represents an uneven pod with 4-5 seeds; 7 represents a curved pod with 2-3 seeds; 8 represents a pod with only 1 seed; 9 represents an aborted pod with no seed. b LSD least significant difference. c SPP represents seeds per pod. d TSW represents thousand seed weight. Table 5. Means for all yield components measured for the two parental genotypes and all intermating cycles (C 0 , C 1 , C 2 ) with 12 famili cycle and all 5 Brassica napus L. fertility restorer crosses in two greenhouse experiments (S 2 and S 3 ). * Significantly best among intermating cycles at 0.05. a Pod rating scale goes from 1-9 where 1 represents a full straight pod with more than 25 seeds; 2 represents a straight pod with 15 -25 seeds; 3 represents a curved pod with 10-15 seeds; 4 represents a straight pod with 7 -9 seeds; 5 represents a curved pod with 6-8 seeds; 6 represents an uneven pod with 4-5 seeds; 7 represents a curved pod with 2-3 seeds; 8 represents a pod with only 1 seed; 9 represents an aborted pod with no seed. b LSD least significant difference. c SPP represents seeds per pod. d TSW represents thousand seed weight. Table 7. Individual family mean pod number values for each cycle (C 0 , C 1 , C 2 ) of cross 3 at the second (S 2 ) and third selfing (S 3 ). c Parents were grown within the same experiment as control samples without any intermating. Table 8. Number of Brassica napus L. restorer families a with mean pod rating significantly better than the best parent per cycle of intermating (C 0 , C 1 , C 2 ) at S 2 and S 3 .    Table 11. Number of Brassica napus L. restorer families a with mean yield significantly higher than the best parent per cycle of intermating (C 0 , C 1 , C 2 ) at S 2 and S 3 .  c Parents were grown within the same experiment as controls without intermating. a Data for each family is based on a mean of 3 replicates each with four individuals.
b Data for the best parent is based on a mean of 3 replicates each with four individuals. Figure 1 Chain crossing. Entry number 1 is crossed with entry number 2, which is in turn crossed with entry number 3 and so forth, until the nth entry, which is then crossed with entry number 1 to complete the chain.

Figure 2
Development of three Brassica napus L. subpopulations using recurrent selection. Each population began with a cross at C0S0. A total of 12 plants were grown per cross. From each plant, 12 owers were crossed to develop the next cycle (C1S0) while the remainder of the plant was selfed (C0S1). One seed from the 12 intermated pods of each plant was chosen at random and planted. Twelve owers from the new plant were crossed again to create the next intermating cycle (C2S0). Replicated experiment 1 analysed the 3 cycles after two sel ngs and replicated experiment 2 followed the same populations after three sel ngs.

Figure 3
Pod rating scale for Brassica napus L. pods. The score assigned represents a mean visual rating of the pod development on the entire plant. 1 represents a full straight pod with more than 25 seeds; 2 represents a straight pod with 15 -25 seeds; 3 represents a curved pod with 10-15 seeds; 4 represents a straight pod