The accuracy of subgenome partition is very important for the genome evolution analysis in this study. The relatively recent genome triplication event that occurred in the early stage of B. rapa species origin is not as old as that in C. vinera (>80 MY) or the most recent tetraploidy in the A. thaliana lineage; the AK blocks in the genome of B. rapa's hexaploid ancestor are readily identifiable. Almost all the blocks, although they have been translocated and fused to form the 10 chromosomes of B. rapa, are retained intact. All three copies of each block can be separated, facilitating the partitioning of the genome into subgenomes. Using chromosomes of A. thaliana as the representative of B. rapa's diploid ancestral karyotype, and employing the rules of 1) nonoverlap of syntenic boundaries and 2) least translocation in B. rapa's chromosomes, we aligned and separated each AK block. This led to a reconstruction of the three subgenomes in B. rapa. Most blocks are conserved and are arranged in groups (many blocks existed continuously in all three subgenomes). However, only 11 breakpoints were found to be shared by all three subgenomes, such as the breakpoint located between blocks C and D (). These identical breakpoints probably reflect the chromosome boundaries of the real diploid ancestor of B. rapa; the A. thaliana genome is only a representative of the three true diploid ancestral genomes.
We have proposed a “two-step theory" to explain the B. rapa
genome evolution 
. According to this theory, there has been a diploid ancestral genome that contained one copy of the AK blocks. Step 1: two diploid genomes became a tetraploid with two new subgenomes (precursors of MF1 and MF2). Along evolutionary time progressed, loss of genes from the duplicate genomes finally resulted in a fractionated diploid genome (consisting of the two subgenomes MF1+MF2). Step 2: another diploid genome (LF) was added to the fractionated diploid genome, which initiated another round of gene loss. As a result, LF experienced one round of gene loss and retained more genes than MF1 and MF2, which experienced two rounds of gene loss. The two-step theory for genome triplication in B. rapa
is well supported by the obvious differentiation of gene density between subgenomes LF and MFs. However, we do not exclude the differential methylation hypothesis, because the differences in gene densities could also be maintained by different levels of subgenome methylation 
Using mRNA-seq data from both different tissues and pooled tissues of two different accessions of B. rapa
(Chiifu subspecies pekinensis
and L58 subspecies parachinensis
), we found that genes in subgenome LF are dominantly expressed over genes in two MFs (), and we further found that more genes are dominantly expressed in MF1 compared to MF2 (). The expression activitys of the three subgenomes was ordered as LF>MF1>MF2. Many studies have noted that methylation represses gene expression 
, thus gene methylation levels might be different among the three subgenomes. Consequently, methylation is likely to have played an important role in B. rapa
We observed ongoing biased gene fractionation in B. rapa
similar to that observed in maize 
. Using resequencing data of two B. rapa
accessions (L144, a rapid cycling laboratory accession and Vegetable Turnip VT117, subspecies rapa
), we found that subgenome LF accumulates significantly fewer non-synonymous SNPs and frameshift InDels than the other two MFs, which correlates with the dominant gene expression in LF compared to MFs. However, this ongoing gene fractionation was a result of the differentiation process of the two subspecies after LF, MF1, and MF2 were combined into one genome. The explanation of biased gene fractionation by genome dominance leaves unanswered questions. In maize, the most likely explanation was proposed to be differential epigenetic marking of subgenomes within an allotetraploid, possibly because allotetraploidy produces epigenetically inherited differentiation of parental genomes 
. If this is true, the “two-step theory" is consistent with the presence of an allotetraploid during B. rapa
evolution before LF was added to MFs. We identified many regions in which the gene densities in MF1 are higher than those in MF2 (). This could also be the result of differential methylation between MF1 and MF2. MF1 and MF2 were merged at the same time; therefore, the different gene density could not be explained by evolution time difference.
Evidence of ongoing biased gene fractionation together with the differential gene expression among the subgenomes of B. rapa could be better explained by differential subgenome methylation, although extensive whole genome methylation status data is needed to test this hypothesis. However, this does not exclude the hypothesis of the “two-step theory". Considering the current data, we tend to believe that the “two-step" evolution process, together with methylation differentiation, both played important roles during the evolution of the B. rapa genome.
has an ancient triplicated genome, which is old enough to have fractionated (many genes have been lost after polyploidization), but young enough so that most genes are clearly identifiable in the outgroup, A. thalian
a. It represents a good model for studying mesopolyploid genome differentiation and offers an opportunity to study evolutionary events on an intermediate timescale. We previously proposed a ‘two-step polyploidization’ hypothesis to explain the gene density difference in subgenomes of B. rapa
. Here, using more genomic datasets and accurate subgenome partition, we observed dominant expression of genes in a subgenome with higher gene density and ongoing biased gene fractionation between subgenomes of B. rapa
. We hypothesize that both differential methylation and ‘two-step polyploidization’ played important roles in B. rapa