Heterosis and inbreeding depression
Caraway breeders expected that inbreeding would cause inbreeding depression. They suggested sticking to population breeding (Toxopeus 1998). To the best of our knowledge, we showed now for the first time that strong inbreeding depression indeed occurs in caraway after inbreeding. We found some very low yielding inbred lines that can be described as highly inbreeding depressive lines. All inbred lines failed to reach the yield level of the standard population cultivar 'Sprinter'. A yield test of JKI breeding lines in the third inbred generation and of corresponding F1 populations did not yet reveal strong inbreeding depression and heterosis (Pank et al. 2007). Possibly, inbreeding depression was mainly generated in the subsequent step/steps of inbreeding. Finally, we can argue that the existence of inbreeding depression as reverse side of heterosis (Mackay et al. 2021) can be considered evident, if we can provide evidence for heterosis.
All and, importantly, even the best yielding inbred lines showed a yield increase after outcrossing in the polycross. Therefore, we can state that heterosis as better parent heterosis can be exploited in caraway. Whereas we can decidedly make this general statement, a statement on heterosis for each inbred line is more difficult. The prior evaluation of the polycross revealed that we could not assume completely random pollination in the polycross. Furthermore, inbred lines had different outcrossing rates (von Maydell et al. 2020). Without knowledge of the paternal side of the F1 populations, we can compute neither mid-parent heterosis, nor better-parent heterosis for each line. Nevertheless, such computations might be of scientific interest rather than important for the breeding progress. Not the heterosis, but the GCA is used to select components for a synthetic variety.
General combining ability & performance of lines per se
In theory, the GCA should comprise most information necessary to select components for a synthetic variety. The GCA should be constituted of the performance of a line per se and the heterosis generated by outcrossing. Based on this assumption, the performance of a line per se could be ignored in selection decisions. Wright (1973) and further authors discussed using the general varietal ability (GVA) instead of GCA, which gives additional emphasis on the performance of the inbred lines per se, in particular, if the number of components is low. Essentially, this should account for decreasing heterozygosity according to Hardy–Weinberg equilibrium in Syn2 or later generations. However, again we have to keep the mixed mating system of caraway in mind. Since outcrossing is likely to be incomplete in Syn1 generation, we rather expect that outcrossing proceeds and heterozygosity increases in Syn2 or later generations. In this case, the significance of the performance of the inbred lines per se for the later synthetic variety might actually be slightly overestimated within the GCA, determined in our study. The estimation of the specific combining ability (SCA) would provide information on non-additive effects, but the necessary procedure would be too laborious without an efficient system to produce F1 seeds in multiple crossings.
For breeding synthetics, it seems advisable to focus on inbred lines with a high yield per se. We found a strong correlation of the yield of inbred lines with the yield of the corresponding F1 populations and hence with GCA. In the end, selection based on GCA might make no large difference to selection based on performance of inbred lines per se in our dataset. Selection of high yielding lines should also be an advantage from a long-term perspective. Research in many species revealed that heterosis might be widely explained by additive dominance effects, whereas overdominance or epistasis merely play additional roles. (Kaeppler 2012; Mackay et al. 2021; Zhou et al. 2018). By recurrent selection within the gene pool of high yielding caraway breeding material, favorable dominant alleles could be accumulated. As alternative to single crossings, a synthetic population can provide the genetic basis for future breeding programs.
If we take into account that performance of lines per se strongly correlates with GCA and that the mixed mating system causes some statistical noise, one could question whether costly GCA testing is necessary, after all. It might be more cost efficient to test a few different combinations of high yielding inbred lines directly after a preselection of lines. This notion certainly applies to the present dataset of just about five valuable inbred lines. In the near future, the pool of valuable inbred lines might be considerably extended. In a larger pool of high yielding inbred lines, the effect of the performance of lines per se should decrease and other factors like genetic distance might increase in importance.
Besides, it did not surprise that the yield of inbred lines per se correlated with the yield of the F1 populations and that we found such correlations for most other traits. First, the maternal line genetically contributes 50% to a F1 population. Second, this proportion increases due to incomplete outcrossing. On average, about one third of individuals of an F1 population should genetically be identical with the maternal line. On the other side, we were surprised that the plots of F1 populations visually seemed to be rather uniform, i.e. they showed a homogeneous flowering, ripening and plant horizon. We suspect that inbreeding depressive plants were suppressed within the plot. Should this assumption be true, this could have an impact on our expectations for Syn1, Syn2 and later synthetic generations. First, more heterosis might already be exploited in Syn1 than we argued before, if outcrossed individuals occupy most of the space in a plot. Second, a synthetic population might virtually improve itself and favor outcrossing over generations. Synthetic populations should be investigated over several generations to gather evidence for such questions.
Selection for higher essential oil content
Next to yield, the essential oil content has to be considered as most important quality trait. The heritability of the essential oil content was only medium, partially due to a high genotype-year interaction effect. We assume this corresponds with the contradictory findings on correlations between essential oil content and other traits comparing both years of growing. We observed that most early flowering genotypes had a lower essential oil content in 2020 than in 2019, whereas most late flowering genotypes had a higher essential oil content in 2020. We assume that the essential oil content to a considerable degree depends on weather conditions at a certain stage of development. Literature indicates that environmental conditions in the ripening period play a major role (Bouwmeester et al. 1995; Toxopeus and Bouwmeester 1992).
Although high genotype-year interaction effects are detrimental to the breeding progress, we should not neglect the equally high genotype effect. Most importantly, we found inbred lines that combine acceptable yields with high essential oil content. Hence, we can select for higher essential oil content within the given gene pool. Usually, quality traits do not show strong inbreeding depression and heterosis (Mackay et al. 2021). Our results are in accordance with this general observation because across both years we found no significant difference between inbred lines and corresponding F1 populations. Yet, F1 populations are earlier in development, which can affect essential oil content in both directions probably depending on weather conditions. This should explain the contradictory effects of outcrossing comparing both years of growing.
Significance of other traits for selection decisions
The beginning and the end of flowering showed a very strong correlation with yield, a high heritability and can easily be estimated. They could be valuable traits to select for higher yield particularly in breeding stages that do not allow reliable yield estimations. The strong correlations with yield were already detected within the polycross data (von Maydell et al. 2020). Probably, late developing genotypes will not play any role in future breeding processes because of the low seed yields of these genotypes. Besides, in a selected pool of early flowering genotypes, genotype-year interaction effects of the essential oil content, as discussed above, might be reduced. Maturity showed a lower correlation with yield and a lower heritability. Hence, this trait might be a bit less suitable for selections. Pank (2012) mentioned early ripening as breeding goal for annual caraway. Early flowering and ripening might be advantageous due to better water availability, better pollination conditions or avoidance of pathogens like powdery mildew. However, we do not claim a definite causal relationship, i.e. that early development per se leads to increasing yield. Some genotypes might just have a higher 'general vitality', e.g. due to a more efficient production of metabolites, that goes along with faster vegetative growth, earlier generative development and higher yields. This might also be the likeliest explanation for the earlier development of F1 populations compared to inbred lines.
Thousand-grain weight and height
It does not surprise that yield was positively correlated with thousand-grain weight. Acimovic et al. (2015) found a strong positive correlation between thousand-grain weight and harvest index for annual caraway. Thousand-grain weight might be less suitable for selection processes than the beginning or the end of flowering, because correlation with yield and heritability were lower and estimation is more laborious. From experience, we know that usually the early flowering umbels within plants dominate yield formation in caraway, whereas late flowering umbels produce only few additional seeds. This goes together with a reduction of female organs in later umbels (Pank 2012). Hence, selection towards higher thousand-grain weight might be a better option than selecting for higher numbers of umbels and seeds per plant.
Results for height showed a higher heritability than thousand-grain weight, but a lower correlation with yield. Although estimation of height is easy, it might be no preferable trait for selection processes. We found a large residual error in the analysis of variance components that was connected to a high variance within plots. For both, thousand-grain weight and height, we found a heterosis effect that goes along with the positive correlation with yield.
Shattering rate and stalk attachment rate
Selection of non-shattering genotypes is a general breeding goal for caraway (Pank 2012). Toxopeus and Lubberts (1994) detected severe yield losses after too late harvesting of shattering biennial varieties, whereas yield of non-shattering varieties was unaffected by late harvesting. In our study, only the cultivar Aprim and two inbred lines showed a clearly reduced shattering rate. Here, it must be considered that the shattering rate, as we measured it, was neither a measure of the real nor a measure of the potential loss of seeds before or during harvest, but an approximate measure of the firmness with which mature seeds are connected to the umbel. Unlike Toxopeus and Lubberts (1994) we found no negative correlation between shattering rate and yield. This might be due to the unbalanced dataset or the method of measurement, but probably we also harvested the plots in time so that severe losses by shattering were avoided. It remains to be seen in future practical trials with standard harvesting machines whether shattering can be a problem within the selected breeding material.
Some farmers stated that strong non-shattering genotypes might tend to higher stalk attachment rates. High stalk attachment rates are detrimental to marketing. Since we found a negative correlation between shattering rate and stalk attachment rate, that statement might be true. We measured a maximum stalk attachment rate of 7.38% that would yet be no issue for marketing. However, the stalk attachment rate can strongly depend on harvesting and post-harvesting techniques (Pank 2012) and could be a problem in practical growing.