Asymmetric somatic hybridization between cauliflower “Korso” and black mustard “G1/1” resulted in the establishment of 117 individual hybrids in the field; of these, 13 fertile plants with preferred traits were selected for continued selfing and backcrossing (Wang G. X. et al., 2011). Hundreds of derivatives were obtained following year-by-year selection (since 2006) that combined phenotypic observation, cytological and molecular analysis, as well as pathogen-resistance assays (Figure 1). In this study, 28 putative ILs were chosen for phenotypic diversity and chromosomal constitution analysis, based on possession of cauliflower-like morphology and phenotypic stability over three generations. Next, 12 putative ILs (IL1-12) were used for genetic an epigenetic analysis. Chinese cabbage “Asko” (B. rapa, AA, 2n = 20) and tetraploid B. carinata (BBCC, 2n = 34) were chosen as reference materials for molecular and chromosomal analysis. The selfed seeds from ILs, parents, and B. carinata were sown in a greenhouse and are available upon request.

Multi-color FISH was conducted on meiosis metaphase I chromosomes of parent lines and putative ILs to confirm their chromosomal composition. B. nigra and CentBr (centromeric-specific tandem repeats of Brassica; accession numbers: CW978699 and CW978837, respectively; Wang G. et al., 2011) DNA were selected as probes and labeled with biotin (dig)-14-dUTP (Roche, Indianapolis, IN) using nick translation with an average length of 500 bp. Slides were prepared following a previously published protocol (Zhong et al., 1996) with minor modifications. To decompose the cell walls of pollen mother cells, anthers approximately 1–3 mm in length were digested at 37°C for 3 h in an enzyme mixture containing 6% cellulose R-10 and 6.5% pectinase (Sigma, solution in 40% glycerol). Previously described methods (Wang G. et al., 2011) were followed for FISH analysis. Hybridized probes were visualized using an FITC-conjugated avidin antibody or rhodamine-conjugated anti-digoxin antibody (Roche, Indianapolis, IN). Chromosomes were counterstained using 0.1 mg/mL DAPI (Vector Laboratories, Burlingame, CA). Images of the signals and chromosomes were captured using a CCD camera (QIMAGING, RETIGA-SRV, FAST1394) attached to a Nikon Eclipse 80i epifluorescence microscope (Tokyo, Japan). Image contrast and brightness were adjusted in Adobe Photoshop (8.0).

Simple sequences repeat (SSR), amplified fragment length polymorphism (AFLP), and methylation-sensitive amplification polymorphism (MSAP) fingerprinting were performed following previously described methods (Shaked et al., 2001; Liu et al., 2015). Sixty-five SSR, 77 AFLP, and 16 MSAP primer combinations were used; for SSR and AFLP, markers were collected from the A genome of B. rapa due to the limited number of cytological and genetic studies on the B genome of B. nigra (Tables S1–S3). For AFLP and MSAP, two independent technical replicates were performed for each of the three biological replicates. Only clear and reproducible bands larger than 100 bp were scored.

The 28 ILs were extremely diverse in multiple morphological and physiological traits (Table 1), including distinct variations in leaf or flower morphology, fertility, and resistance to pathogens (i.e., black rot or clubroot). Here we describe diagnostic differences in flower heads as an example. Four phenotypic groups were apparent based on flower curd characters (Figure 2). In the first group (type I), flower heads were white, compact, and hemispheric, with a wheel-like radial arrangement; flower buds were small, granular, and close. The second group (type II) possessed flat, loose flower heads with no obvious wheel-like arrangement; flower buds were light yellow, tiny, and soft, while the pedicel was green and slightly longer than Type I pedicels. The third group (type III) possessed relatively flat flower heads and pale yellow-green color; flower buds were small, loose, soft, and distributed in clusters, but with no obvious ball flower formation. The pedicel was green and long. The flower heads of the fourth group (type IV) were relatively flat, yellow green, with small flower buds that were fine and soft.

Figure 3 shows the results of multi-color FISH on chromosomes of “G1/1,” “Korso,” and 28 putative ILs. The two parent lines exhibited different hybridization patterns with both probes. As expected, when using “G1/1” genomic DNA, all 16 “G1/1” chromosome pairs exhibited red signals (Figure 3A). In “Korso,” red signals (2 strong and 2 weak; Figure 3B) were detected on only two pairs of satellite chromosomes that likely correspond to the hybridization signals of 45SrDNA (Fukui et al., 1998).

No signals were detected on “G1/1” chromosomes using the CentBr probe. In contrast, seven pairs of strong signals and one pair of weak signals were detected on “Korso” chromosomes (Figure 3B); the latter was hybridized to chromosomes that also exhibited strong red signals. Only 1 “Korso” chromosome pair exhibited no signals (Figure 3B).

Most cells in the putative ILs contained 18 chromosomes and exhibited hybridization patterns identical to “Korso.” The results of genomic in situ hybridization (GISH) found no “G1/1” chromosomal segments present in these putative ILs, likely because GISH cannot detect introgressions from Brassica chromosomes due to their small size and compact structure.

We analyzed the sequence introgression and variation in 12 ILs (IL1–12) using 65 SSR and 77 AFLP markers that were polymorphic between the parents. The results of SSR and AFLP profiling revealed 1799 loci from both parents, including 682 (37.9%) those were polymorphic. The IL profiles were similar to the “Korso” profile, but loss of 5–13 “Korso”-specific bands were common. Additionally, all putative ILs were confirmed to possess anywhere from 3 to 33 “G1/1”-derived fragments. However, we noted that in 3 ILs, SSR markers failed to detect “G1/1”-derived bands but AFLP profiling revealed 20–25 “G1/1”-specific loci, indicating that the latter method is more effective when the introgression size is small. Next, all 12 ILs contained B. carinata-specific loci (35–49), and we were able to amplify 1–7 new bands from them.

These results clearly showed that all 12 lines were ILs and that genomic-sequence variations were common among them. Notably, IL genomes were not simply similar to the “Korso” genome but with some “G1/1”-derived fragments added. For example, we noted that some ILs lacked bands present in “Korso,” or possessed fragments lacking in “Korso.” Moreover, several B. carinata-specific bands were present in all ILs, hinting at the occurrence of common changes in the early stages of both somatic hybridization and naturally occurring hybridization.

DNA profiling of the ILs revealed 1493 “Korso,” 1423 “G1/1,” and 1559 B. carinata fragments; 1117 of these were shared by both parents (Figure 4). The number of fragments present in the profiles of the 12 representative ILs ranged from 3 to 33, with an overall frequency of 7.5%, whereas the number of fragments lost ranged from 7 to 32, with an overall frequency of 5.1% (Table 2). Note that the percentages of “G1/1”-fragment presence and “Korso”-fragment loss were respectively based on the 306 “G1/1” fragments that did not co-migrate with “Korso” and the 376 “Korso” fragments that did not co-migrate with “G1/1.” In addition, the 12 ILs possessed 1–7 new fragments (1.4%) that were not present in the “Korso” profile, and 4–49 (15.5%) B. carinata-specific fragments from the 226 that did not co-migrate with “Korso” and “G1/1.”

The MSAP analysis used isoschizomers MspI and HpaII, which differ in their sensitivity to cytosine methylation (Shaked et al., 2001; Liu et al., 2015). Our comparison of the profiles generated from EcoRI-MspI and EcoRI-HpaII digestion revealed three kinds of band classifications (Figure 5). First, A type was characterized by fragments from both the EcoRI-MspI and EcoRI-HpaII digestions, representing non-methylated sites (or methylation within a single strand). Second, B type was characterized by half-methylation sites, present in EcoRI-HpaII digestion but not in EcoRI-MspI digestion. Third, C-type bands stemmed from complete methylation sites present in EcoRI-MspI digestion but not in EcoRI-HpaII digestion. Using 16 MSAP primer combinations amplifying clear and stable bands, we detected 401 fragments in parents, with 290 present in “Korso” but not in “G1/1.”

Table 3 summarizes variation in methylation patterns and their heredity in parents and the ILs. The average frequency of hypermethylated loci in different ILs was 12.4%, while the average frequency of hypomethylated loci was 4.8%.

The low genetic diversity in modern Brassica varieties is concerning because it reduces potential genetic gains in breeding programs. To combat this problem, ILs containing fragments from related species can be used to generate improved cultivars. This method has proven successful in many crops, including wheat (Liu et al., 2007), rice (Rangel et al., 2008), potato (Chavez et al., 1988), eggplant (Mennella et al., 2010), B. napus (Primard-Brisset et al., 2005; Leflon et al., 2007), barley (Johnston et al., 2009), tomato (Menda et al., 2014), and rye grass (Roderick et al., 2003).

In this study, we generated a set of ILs using somatic hybridization between cauliflower and black mustard. The original objective was to select disease- and drought-resistant ILs for use in a Brassica breeding program. In fact, these ILs showed considerable genetic diversity in many morphological and physiological traits beyond resistance. Some of these traits are useful for increasing cauliflower adaptation to stressors. For instance, IL2 and IL6 respectively exhibited strong resistance to black rot and clubroot, major Brassica diseases with worldwide distribution. Other ILs exhibited commercially desirable traits such as early maturation and high nutritional quality, especially high phosphorus and potassium content (IL3 and IL9), as well as less desirable traits such as loose flower heads, Early inflorescent, and long pedicel. This diversity indicates that all of the ILs could be a source of new alleles for developing Brassica cultivars. Thus, somatic hybridization provides an effective and efficient means of achieving introgression into a crop species from its wild relatives and should continue to be important in producing novel germplasm for breeding programs.

Although GISH has been successfully used to investigate genomic relationships in several agriculturally important genera, including Brassica (e.g., Lysak et al., 2005), and has detected introgressions in crops such as wheat (e.g., Liu et al., 2015), GISH was unable to detect black mustard chromosomal segments in our ILs. We believe this outcome was likely due to the small and compact Brassica chromosomes or else very small introgression sizes. We dealt with this issue by first using multi-color FISH to identify chromosome composition and then combining SSR and AFLP markers to detect “G1/1” sequence introgression. Primer sources and amplification results from the genomes of B. rapa, B. nigra, B. oleracea, and B. carinata are shown in Table S1. As those data indicate, the primers are usable in multiple Brassica species and can detect abundant polymorphism. Moreover, compared with the SSR primers, AFLP markers were more effective when working with small introgressions, such as those in our ILs (Table 2). In sum, our study demonstrated that Brassica ILs can be clearly identified with a combination of cytological and molecular methods.

The extensive phenotypic variation of the ILs was caused by genetic and epigenetic changes in the ILs relative to their parents. However, black mustard introgression was actually more limited (7.5%; Table 2) than other ILs, such as the somatic hybrids between wheat and tall wheatgrass (Liu et al., 2015). We believe the low introgression was probably due to suboptimal UV treatment. Thus, treatment factors such as irradiation time and intensity will require further testing and validation to achieve higher introgression rates. At the same time, the amount of introgression observed may be enough to achieve considerable genomic diversity. For example, in stable rice introgression lines derived from intergeneric hybridization (followed by successive selfing) between rice (Oryza sativa L.) and a wild relative (Zizania latifolia Griseb.), very low introgression (0.1%) by foreign DNA was enough to trigger up to 30% of the genomic changes (Wang et al., 2005). The mechanism underlying these changes is unclear, but some researchers have suggested that they happen very early, through a cryptic pathway that differs from conventional or unorthodox meiotic recombination of homeoalleles between rice and Zizania (Zhang et al., 2013).

Similar to previous findings (Wang et al., 2005), we also observed sequence loss and new bands (Table 2) following somatic hybridization, backcrossing, and self-pollination (Figure 1). The frequency of fragments lost (5.1%) was higher than presence of novel bands (1.4%). It is not entirely clear why the percentage of novel bands was so much lower, but one reason may be that some of them were B. carinata-specific, and B. carinata-specific fragments occurred at a relatively high frequency (15.5%). Thus, future investigations may need to confirm the exact nature of the lost and novel bands, as well as exclude any B. carinata-specific loci.

Our results were consistent with previous reports in artificial hybrid B. carinata (BBCC, x = 17), produced via crossing and polyploidization between B. nigra (BB, x = 8) and B. oleracea (CC, x = 9). Around 47% of the hybrid B. carinata (BBCC) genome possessed isoenzyme sites specific to natural allotetraploid B. carinata (Jourdan and Salazar, 1993). Given this similarity, we propose that the early generations of artificial synthesis and natural polyploid evolution of both genomes may have experienced the same change events. Supporting this idea, previous analyses of recombination in other genera, such as the homoploid hybrid sunflower (Helianthus anomalus), showed that newly synthesized hybrids converged on the linkage pattern of wild hybrids within five generations (Rieseberg et al., 1996).

Based on our results and those of previous studies, we propose that genetic changes in Brassica ILs occur via the following mechanisms. First, alien genetic elements are incorporated into cauliflower early in hybridization, through the activation of transposable elements or DNA methylation, as suggested in rice introgression lines (Wang et al., 2005). Next, chromosomal rearrangements occur through intergenomic translocations or transpositions (Udall et al., 2005) and homoeologous pairing (Leflon et al., 2006; Nicolas et al., 2007). During Brassica evolution, ancestral group rearrangements contributed to the high homology found especially in the three linkage groups of the B genome (B5, B6, and B4) and the A genome (A5, A6, and A4) (Panjabi et al., 2008). Finally, consecutive selection results in rapid sequence elimination, as reported in synthetic hybrids and hybrids less related to the original parents (Osborn et al., 2003; Pires et al., 2004; Gaeta et al., 2007; Zhang et al., 2013).

Cytosine methylation plays an important role in epigenetic gene regulation at both the transcriptional and the post-transcriptional levels (Paszkowski and Whitham, 2001). Our ILs exhibited fairly high proportions of change to methylation patterns (17.2%; Table 3), compared with several other plant hybrid/allopolyploid systems analyzed to date using MSAP. For instance, in resynthesized allotetraploid Arabidopsis suecica, 8.3% of the fragments experienced methylation changes (Madlung et al., 2002), whereas in newly synthesized allohexaploid wheat, ~13% of the loci saw alterations to cytosine methylation (Shaked et al., 2001). In contrast, a 23.6% change to cytosine methylation was observed in asymmetric somatic wheat introgressions (Liu et al., 2015). Taken together, these data allow us to conclude that somatic hybridization induced a broad spectrum of cytosine-methylation changes that perturbed gene expression to a larger extent than allopolyploidization.

Consistent with previous findings (Zhang et al., 2013), changes to hypermethylation (9.7–13.5%; Table 3) occurred more frequently than changes to hypomethylation (2.4–7.6%). Zhang et al. (2013) proposed that this pattern may be caused by the activation and subsequent silencing of some transposable elements after hybridization, but this hypothesis requires more data for confirmation. Additionally, the methylation patterns in ILs may be related to changes in enzymatic machinery, as well as the expression of siRNAs and long noncoding RNAs (Goll and Bestor, 2005; Wierzbicki et al., 2008; Zhang et al., 2013).

In conclusion, somatic hybridization mimics many of the genetic alterations induced by polyploidization or sexual wide hybridization, but to a stronger extent and with considerably less time. Therefore, somatic hybridization is both effective and efficient in achieving introgression of a crop species and its wild relatives. Moreover, this approach provides a potential means to explore the genetic and epigenetic events induced by “somatic genomic shock” (Liu et al., 2015).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.