Nine allohexaploid lines obtained from crosses between eight accessions of B. rapa from China and nine accessions of B. carinata from Ethiopia, as shown in Supplementary Table 1 (Tian et al., 2010), were used in this study. A total of 45 accessions of traditional B. juncea from various areas of the world were used to perform crosses with the nine hexaploid lines to obtain a new-type of B. juncea and to obtain intersubgenomic hybrids. Furthermore, analyses of genetic diversity were performed, in which an additional nine accessions of B. juncea were also used (Supplementary Table 2).

All of the DNA samples (included 82 plants from nine F_2:3 families, five hexaploid lines and eight accessions of traditional B. juncea used as parents of the new-type B. juncea lines, six accessions of traditional B. napus and two accessions of B. carinata of other related species) were extracted from young leaves by a modified version of the DNA extraction method (Doyle, 1990). A total of 208 pairs of fluorescence-labeled primers of simple sequence repeats (SSR, 104 from the A genome, 73 from the C genome and 31 from B genome of Brassica) with clear, reproducible and polymorphic amplification products (Supplementary Table 3) were used to evaluate the genetic diversity and population structure of the novel B. juncea lines and their parental lines (traditional B. juncea) used in this study. The synthesis of the SSR primers, conditions for PCR, and processing of the data generated by capillary sequencing were all described previously (Chen et al., 2010) using an AB3730xl capillary sequencer (Applied Biosystems, USA). An additional set of SSR markers anchored on 31 loci from the B genome in B. carinata (Zou et al., 2014) was used to evaluate the introgression of B genomic components (Supplementary Table 3). The markers were scored according to the polymorphism appeared in the parents and population. Those alleles newly appeared or missing in the population compared with the parents were noted as novel allelic variation.

The chromosome number in the plants was counted in cells from the wall of the ovary, as described by Tian et al. (2010). For meiotic analysis, young flower buds were fixed in fresh Carnoy's solution for 24 h and then stored in 70% ethanol. Slides of chromosomes from the pollen of mother cells were prepared for GISH (genomic in situ hybridization) and BAC-FISH (bacterial artificial chromosome fluorescence in situ hybridization) based on the procedures of Ge and Li (2007). Genomic DNA from B. nigra cv. Giebra (labeled with digoxigenin-11-dUTP; Roche, Basel, Switzerland) and plasmid DNA from BAC BoB014O06 (labeled with biotin-11-dUTP; provided by Susan J. Armstrong, University of Birmingham, Birmingham, UK, the probe distribute on all C chromosomes) were used to identify B- and C-genome chromosomes using GISH and BAC-FISH analysis. Probe preparation and in situ hybridization were performed using the methods described by Cui et al. (2012).

Plants were mainly grown in Wuhan (located in central China), but some plants in the F₄ generation were also grown in Hezheng (northwest China). The proportion of seed sets for each of the new-type B. juncea line and cultivars of traditional B. juncea was accounted for according to the average seed number of 10-30 siliques evaluated with 4-6 plants each line. A total of 445 F₄ plants from 72 F_3:4 families of new-type B. juncea, and a total of 81 F₆ plants from 20 F_5:6 families of new-type B. juncea were measured. The kilo-seed weight of each line was evaluated with the average weight of 1000 seeds from each of the investigated plants.

Inter-subgenomic hybrids between the lines of new-type B. juncea in early generations (F₃, F₄, and F₅) and traditional B. juncea were generated by hand crossing and planted along with their parents. The hybrids derived from the F₃ lines were grown in Hezheng as a spring-type crop in 2012 and in Wuhan as a semi-winter-type crop from 2012 to 2013 to evaluate heterosis potential for seed yield. All plants were grown in two rows, with each row 1.8 m in length and 0.25 m between rows. The hybrids derived from the F₄ lines and F₅ lines were grown in Wuhan from 2013 to 2014, and 2014 to 2015, respectively, with two replicates (24 plants with 0.3 m space among plants for each replication) and three replicates (21 plants with 0.3 m among plants for each replication), respectively. The middle parent heterosis and superior parent heterosis of the hybrids were calculated. Individual significant differences and overall significant differences among hybrids and their parents across multiple combinations in each year was evaluated by the F test at the significance level of P < 0.05, using the multiple comparison model under a single factor with repeats and the multiple comparison model under double factors with repeats in SAS 8.1 (SAS Institute, 1999), respectively.

The genetic diversity and genetic distance between different plants was evaluated with molecular markers analyzed using Primer 6 software (Clarke and Gorley, 2006). The Shannon index was calculated using POPGENE 1.31 (Yeh et al., 1999). The phylogenetic tree was analyzed based on Nei's genetic distance (Nei et al., 1983) using the clustering method, employing a neighbor-joining tree with 1000 permutations, using POPTREE 2 (Takezaki et al., 2010). After eliminating improper markers with allelic frequencies less than 10 or >10% missing data, the population structure was analyzed by Bayesian clustering and the “admixture model” with STRUCTURE 2.2 (Pritchard et al., 2000; Falush et al., 2003). The length of burn-in time and replication number were both set to 100,000 in each run. To identify and determine the most probable population number (K), we calculated the ΔK values of K from 1 to 10 replicate runs for each K, and selected the K value corresponding to the peak of the ΔK graph after plotting (Evanno et al., 2005).

A set of Brassica allohexaploids in which the A^r and B^c subgenomes of B. carinata (B^cB^cC^cC^c) and B. rapa (A^rA^r) (Tian et al., 2010) were combined was used as donors for the introgression of exotic subgenomes into the traditional B. juncea genome. Subsequently, morphological/molecular/cytological identification and intensive selection for generations were performed (Figure 2).

Initially, nine hexaploid lines were successfully crossed with 45 accessions of traditional B. juncea, which generated thousands of hybrid seeds (approximately 15,000 seeds) derived from 108 combinations. The cross success rate for different pentaploid combinations varied from 0.1 to 1, with an average cross success rate value of 0.8 and 5.7 seeds per silique. The seeds per silique in the hexaploid seeds and traditional B. juncea parents varied from 0.9 to 3.7 and 9.6 to 16.7, respectively. The authenticity of the hybrids was verified using DNA markers after initial identification by morphological observation (Supplementary Figure 1), and the chromosome number of each hybrid was counted. After obtaining pentaploid progeny, self-pollination was performed across generations.

The seeds from 93 identified true hybrids gave rise to the F₂ generation. The F₂ generation was screened for plants with the correct number (or close to the correct number) of chromosomes (36) as well as good fertility (defined as an abundance of pollen in the flowers and more than eight seeds per silique). The 36 chromosomes were expected to consist of the total 2 A and 2 B genomes, without any of the C genome chromosome. When the genome of A, B and C was integrated in hexaploids, the C genome was more easily lost because the genome stability was B>A>C (Zhou et al., 2016). The pedigrees of the introgressed lines were established according to the original interspecific crosses for the pentaploids. Then, selfing of each pentaploid resulted in different F₂ families. Subsequently, the other round of phenotypic selection was performed at the F₃ generation with a focus on families descended from plants with 36 chromosomes. A genomic analysis was performed to evaluate whether the F₃ plants diverged from the traditional B. juncea lines due to the massive exotic introgression. In the F₄ generation, GISH and FISH analyses were used to confirm the correct genomic constitution of the lines of new-type B. juncea (AABB). Chromosomes were further assessed using the lines at the F₆ generation.

Because meiosis in the artificially synthesized hexaploids cannot be completely normal (Tian et al., 2010), the F₁ hybrids derived from the 108 interspecific crosses between the hexaploid and tetraploid B. juncea might have a segregated chromosome composition deviate from the expected 45 chromosomes. After removing the abnormal seedlings and maternal plants, more than 700 F₁ plants bearing self-pollinated seeds were obtained and subjected to chromosome counting. Only 6.73% of the plants had the expected number of chromosomes (45) for pentaploids (Figure 3A). To avoid incorrectly discarding plants with the appropriate number of chromosomes, the plants scored with 44 chromosomes were also regarded as potential pentaploid. Consequently, self-pollinated seeds harvested from 93 F₁ plants assumed to be pentaploid, were selected to produce the F₂ generation.

Approximately 1000 F₂ plants were obtained, and approximately one-third of the plants exhibited abundant pollen (as observed by eye); they were selected as candidate tetraploid plants for chromosomal counting. In the F₂ population descended from the F₁ specimens with 45 chromosomes, only 2.69% of the plants exhibited 36 chromosomes; most of the plants had extra chromosomes (Figure 3B). To obtain more euploid progeny (A^r/jA^r/jB^c/jB^c/j, 2n = 36) in the next generation, F₂ plants with 37 or 38 chromosomes (which may be aneuploid with one or two extra chromosomes that could be lost in the next cycle of meiosis) were selected together with plants with 36 chromosomes to produce the F₃ generation via self-pollination.

More than 2000 F₃ plants were obtained from various 24 F₂ plants derived from different pentaploids with 36, 37, or 38 chromosomes, which were established as 24 derived F_2:3 families according to the origin of the pentaploid cross in the F₂ generation. The chromosome numbers were assessed for those fertile plants that generated abundant pollen at flowering and produced a large number of seeds. Of the fertile plants, 70% (119 of 170) were regarded as euploid (2n = 36), 85% of the inferred euploid specimens were derived from 13 F₂ plants with 36 chromosomes (Table 1, Figure 3C), and 5 of the 13 F₂ families constituted 69% of the inferred F₃ euploid plants. The plants derived from 37- or 38-chromosome F₂ plants accounted for 15% of the inferred euploid specimens in the F₃ generation (Supplementary Tables 4, 5).

Approximately 2500 F₄ plants derived from 87 F₃ plants with 36 chromosomes were obtained. Overall, 94% of the plants, counted from 116 F₄ individuals, were euploid (2n = 36) (Figure 3D, Supplementary Table 6). To verify that the “inferred euploid” specimens with 36 chromosomes exhibited AABB euploidy, the chromosome composition of five F₄ plants with 36 chromosomes was investigated by cytological observation. All of the plants exhibited normal behavior at meiosis in pollen mother cells with 20 chromosomes from the A genome and 16 chromosomes from the B genome (Figure 4). We further counted the chromosome numbers of 78 F₆ plants with normal fertility derived from 21 F₄ families and found that 96.2% of the F₆ plants had 36 chromosomes (Supplementary Tables 6, 7), and all of the plants descended from seven 36-chromosome F₃/F₄ families (25 of the 78 plants) consistently exhibited 36 chromosomes as well. It indicated that the lines of new-type B. juncea were genetically stable, at least, from the F₆ generation.

The seed set of plants in the F₁ generation varied widely among individually bagged plants, ranging from no seed set to close to a normal level. One-fifth of the F₁ plants were seedless, while approximately half of the plants yielded more than 100 self-pollinated seeds, and a small proportion of plants (approximately 4%) produced more than 500 seeds (Figure 5A).

Plants with 36, 37, and 38 chromosomes were used to investigate the fertility of the F₂ generation. The plants with 36 chromosomes exhibited the highest rate of seed set (6.04 seeds/silique), which was significantly higher (P < 0.05) than the seed set rate of plants with 37 or 38 chromosomes (Supplementary Table 5). Some of the 36-chromosome F₂ plants borne more than 10 seeds per silique (Figure 5B).

The majority of F₃ plants exhibited good fertility and flowered normally with abundant pollen (Table 1, Supplementary Figure 2). However, the F₃ progeny derived from the 36-chromosome F₂ plants had higher rates of seed set than the offspring derived from 37- or 38-chromosome F₂ plants, as judged by the number of seeds per silique and the seed yield per plant (Figure 5B, Supplementary Table 8). Compared with the F₂ generation, the number of seeds per silique in the plants with 36 chromosomes increased in the F₃ generation regardless of whether they were derived from 36-, 37-, or 38-chromosome F₂ plants.

The number of seeds per silique increased again in the F₄ generation, and the best F₃-derived lines with 36 chromosomes had a good seed set rate that was comparable to that of traditional B. juncea lines (Table 2). The results of cross-pollination with traditional B. juncea accessions showed that the lines of new-type B. juncea were not sexually isolated from but were instead compatible with traditional cultivars of B. Juncea (Supplementary Figure 2).

A set of 978 SSR markers was used to analyze the genetic diversity of the new-type B. juncea lines compared with their parental species. Of them, 44.2, 17.5, and 9.6% markers could be assigned to A, B, C genome according to the origin of the alleles amplified in different parental species, respectively. However, 28.7% of the markers could not be distinguished because of multiple amplified products in different genomes. For the plants from the same family, we estimated the percentage of the introgression according to the origin of the alleles from both parents (Supplementary Table 9). Those alleles were not originated from both of the parents, but newly appeared or disappeared in the new-type B. juncea lines were noted as novel allelic variation. Averagely 39.2, 25.5, and 35.4% of the genome of new-type B. juncea were constituted with the introgression of the traditional B. juncea parents, hexaploid parents and novel allelic variation (Supplementary Table 9). For the introgression from hexaploid parents, 55.5 and 15.1% of the alleles were originated from the A and B genome, respectively, but 29.4% of them could not be assigned. The new-type B. juncea lines exhibited substantial genetic distance from their hexaploid parents, as assayed via principal ordination analysis (Figure 6A). The F_2:3 family, C34J27-1, along with most of other F_2:3 families, exhibited clear genetic differences relative to their tetraploid parents and traditional B. juncea accessions. Most of the F_2:3 families of new-type B. juncea, especially C18J05-2, also exhibited marked polymorphisms within their families and demonstrated increased genetic variation as a whole compared with traditional B. juncea accessions. An estimate of the gene diversity index between the different populations indicated that the new-type B. juncea lines exhibited the richest genetic diversity and the highest Shannon Index (0.49), which was considerably higher than that of traditional B. juncea accessions (0.31). A phylogenetic tree revealed that the new-type B. juncea was obviously separated from traditional B. juncea and had a greater genetic distance from other Brassica species than traditional B. juncea (Figure 6B). Population genetic structure analyses indicated that three apparent genetic clusters existed among the investigated lines and their parents (Figure 6C, Supplementary Figure 3). The new-type B. juncea lines inherited major genetic components from both parents (the hexaploids and traditional B. juncea).

In total, 119 of the best F₃ plants with marked variation in morphology and agronomic traits were selected from 24 F_2:3 families, which represent 14 combinations involving five hexaploid lines (each synthesized from different accessions of B. carinata and B. rapa) and 11 accessions of traditional B. juncea (Supplementary Table 10). After testing the seed yield and yield components, seeds from 76 F₃ plants representing all of the founder parents and cross combinations were ultimately sown in the field to generate the F₄ generation, which resulted in approximately 2500 plants that exhibited distinct differences from their parental species and presented comparable agronomic traits to traditional B. juncea (Table 2, Supplementary Figures 4A–D). The agronomic and morphological traits of the new-type B. juncea lines were further improved at the F₆ generation through selection, resulting in an obviously increased seed number per pod.

We estimated the heterosis potential for the seed yield of the inter-subgenomic hybrids between the new-type B. juncea lines in different early generations (F₃, F₄, and F₅) and traditional B. juncea accessions. The seed yield of the hybrids was significantly increased compared with parental accessions when grown as a spring crop in the northwest of China and as winter crop rapeseeds in central China (Figure 7, Supplementary Table 11 and Supplementary Figure 4E). The average over superior parent heteoris and mid-parent heterosis of the tested hybrids across environments were 60 and 96%, respectively. However, the heterosis potential of different lines also varied extensively.

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.