Our method compared to RECON

The bulk of the automated parts of our process, Steps 2 through 4, were very similar to RECON [4]. RECON does not appear to have analogues for Steps 1 (RepeatMasker), 5, 6, or 7. We utilized Step 1 to steer us away from previously characterized repetitive. We utilized the manually intensive Steps 5–7 to achieve a more parsimonious (smaller number of repeat families) than appeared to be possible with RECON alone. In this sense, we envision that our method is a complement to RECON, not a replacement.

Steps 1 – 4

Thirty-six percent of the sequence was masked by RepeatMasker. Comparing all unmasked sequence fragments (≥ 50 bp) produced 1,334,953 pair-wise alignments. One thousand five hundred seventy-nine highly redundant sequences (totaling 1.07 Mb) were identified that had a minimum of 10 hits for at least 50 contiguous bases. Sixty putative repeat element families resulted from clustering the 1579 highly redundant sequences. The repeat element families were labeled MPRE1 – MPRE60 (for M eat Animal Research Center P orcine R epetitive E lement). Their lengths ranged from 55 to 1059 bp and their copy numbers (across the 220 discovery BACs) ranged from 12 to 1102.

Steps 5 – 6

The 60 original MPREs were consolidated into 31 because of overlap or co-localization at multiple sites. Twenty-nine MPREs were absorbed into 31; the 31 original MPRE identifiers of the longer sequences were kept to maintain provenance. In addition, there were three combinations (MPRE20 and 57; MPRE15, 17, 19 and 26; MPRE44, 50 and 52) of repeats that frequently appeared together in the same order with some variation in their relative spacing. The most consistent group contained two elements – MPRE20 in reverse complement followed by a small gap, then MPRE57. All thirteen times that MPRE20 occurred, it occurred in this grouping. MPRE57 occurred 13 out of 14 times in this grouping. Naturally, we concluded that MPRE20 (600 bp) and MPRE57 (204 bp) were two parts of a longer RE that had a variable middle (100–250 bp range for all but one example). After examining the alignment in ClustalX, we could see that the middle was conserved except for an 84 bp deletion in one instance and a 67 bp insertion in another. Further review of the BACs showed that the 13 groups containing MPRE20 and MPRE57 sometimes occurred in overlapping regions between pairs of clones in the BAC collection, meaning that we only had 7 unique loci plus one very unique locus that had a PRE1a (Porcine Repeat Element 1a, as identified by RepeatMasker) inserted into the gap. There was no pattern to the gap in the other instances. We include this longer repeat element in our list of novel porcine repeat elements as MPRE61, which is more fully described in a later section.

The final alteration to the list of MPREs was the removal of MPRE48 due to its low copy number in the set of 275,595 porcine BAC-ends supplied by the Wellcome Trust Sanger Institute (hereafter shortened to "Sanger") [19]. Surprisingly, MPRE48 was found to appear less frequently, only six times, in BAC-ends (335.9 Mb) than in the much smaller portion of the genome spanned by the set of fully sequenced BACs (36.4 Mb) from which the MPREs were derived. That brings the final number of novel repeat elements reported here to 27, although we decided against removing MPRE48 from the fasta file of MPREs, see Additional file 1.

Step 7

The prevalence of these newly identified REs was compared to that of known REs. Three of the newly discovered porcine REs, MPRE11, 16 and 38, were more common than the LINE element L3 and one, MPRE42, was about as common as L3 (Table 1). The other 23 MPREs have lower count numbers. In the Sanger archive of 275,595 BAC-ends, the number of elements for all SINEs was 203,206, for all LINES was 116,107 and for all LTRs (Long Terminal Repeats) was 25,066 based on RepeatMasker. Looking specifically at the LINEs, the most common by far was L1 with 94,325, followed by L2 with 18,720 and L3 is third with 2,358.

These newly discovered repeat elements did not appear to be duplicated genes, LINE elements or expressed sequence that was transposed by a LINE element. To address these questions, the MPREs were translated and compared (BLAST) to the GenBank nr database and only one strong hit was found. MPRE1 hit Sus scrofa interferon alpha-1 precursor with a bit score of 352, so it was eliminated from further consideration as a novel RE. For comparison, the highest bit score of MPREs reported here was less than 50. The repeats were also compared (BLAST) to vectors, mitochondrial DNA, and tRNAs. The middle of MPRE58 did have high similarity to tRNA-GLU; otherwise, there were no substantial high-scoring pairs.

Counting repeat elements is challenging

Because of the degeneracy of repetitive elements it is difficult to arrive at an accurate count in the target genome. Another difficulty in the quantification of repeat elements is that REs are often composed of smaller repeat units that occur more frequently than the larger unit [21, 22].

To characterize the prevalence of MPREs, we went to an independent data set, the Sanger BAC-ends from the CHORI-242 library archived at Ensembl [19]. Table 1 lists three different measures of prevalence of MPRE within these BAC-ends. The first measure (BLAST hits to BAC-ends) gives the number of hits that were at least half the length of the repeat element. An issue here is the typical size of the traces – an average of 1219 bp. The longer REs will tend to be under-counted due to edge effects in the trace archive. The next two measures of count number were calculated by plotting the number of BLAST hits as a function of position on the RE. Some of the resulting plots were smooth and flat across most of the RE with an expected drop-off near each end. For these "regular" plots the count number was the average value of the middle 90% of the plot amplitude. Other plots varied quite a bit in amplitude across the RE. This was likely due to sub-repeats that hit in areas of the genome that the whole repeat did not. During this measure of count number there was no lower limit to the size of the hit other than that needed to get the expectation value below 0.1. These were considered irregular and the algorithm for determining their count number was to take the smallest value on the plot after ignoring the first and last 10% of the plot. A few plots were only mildly irregular, and for those both the regular and irregular algorithms were used with both numbers reported in Table 1.

Comparing the novel repeat element content across genomes

The sequences of novel porcine repetitive elements listed here were compared (BLAST [25]) to a recent build of the complete cow genome (AAFCO2 from [26]) as well as against the mouse and human genomes. In the case of mouse, there were no significant similarities found. The comparison to the human genome yielded only one significant hit – a 37 bp long section of MPRE17 (870 bp long) matched once in chromosome 9 thousands of bp away from any annotated features. The comparison to the cow genome yielded a variety of results. Five of the 27 MPREs did not hit at all (MPREs 6, 22, 28, 50 and 60), and three others (MPREs 44, 49 and 61) had ten or fewer hits (Table 1), despite the fact that the cow genome contains ten times more sequence than the collection of porcine BAC-ends tested. Fourteen of the 27 MPREs appeared frequently in cow as well as pig, as indicated by having at least 1000 BLAST hits to the cow genome.

Not surprisingly, the bovine hits tend to be shorter than the porcine hits because the MPREs were defined from pig sequence and as such would be expected to be more intact in porcine. What is interesting is that in both species the endpoints of the hits have a strong tendency to line up to particular spots in the MPRE, as shown in Figure 1 using MPRE12, 15, 17, 41, 51 and 58 as examples. Sometimes the common endpoints are the same in both species, sometimes not. This could be a result of the repeat elements being comprised of smaller repeat elements, not all of which have the same frequency of occurrence in either genome. The longer MPREs often had more than one sub-region with multiple extra hits. This, too, could be evidence of internal repeat structure.

Figure 2 shows that MPRE55 occurs in both swine and cattle in orthologous loci. The pig BAC lies along the x-axis, and the cow BAC lies along the y-axis. Also plotted are line segments of high similarity between the two BACs. The preponderance of these segments demonstrates little genomic rearrangement between species, which indicates that these are orthologous regions of likely common ancestry between the two species. This region is highly similar to the human contig NT_005403.16 and the locus of MPRE55 corresponds to the 3' UTR of the model gene LOC643405, which codes for a protein similar to TGF-beta induced apoptosis protein 2.

Because the collection of BACs spans only 1% of the whole pig or cow genome, we cannot rule out the possibility that all of the MPREs have at least one orthologous location in both species. The fact that 12 MPREs did not have blast hits in any of the cow BACs makes it seem likely that those 12 are relatively recent evolutionary occurrences. Of the 10 MPREs that appear most frequently in the cow, only two, MPRE55 and MPRE59, were observed to appear in orthologous locations among the tested set of fully-sequenced BACs.

A phylogenetic analysis was performed on the different integration sites of MPRE55 from both the cow and pig BAC libraries using ClustalX (see Additional file 2 for the sequences), and the output (Additional file 3) was then input into R [27] to create Figure 3. The sequences that occurred at orthologous locations in swine and cattle are highlighted. As expected, the pig branches and cow branches tend to be separate. It is notable that the most similar sequences that occur in both species do not come from orthologous locations, but seem to be found in loci that originated after the cow and pig ancestral lines diverged. The evolutionary distance between them is represented by the sum of horizontal distances that one must travel along the tree to connect the two sequences. The leftmost part of that path represents a common ancestor. It is not surprising that the two sequences in question have individually diverged a significant amount from the original sequence of the common ancestor at that locus. The more surprising result is that some of the pig and cow sequences are more similar to each other than the sequences at the oldest loci. Coincidental convergence is an unlikely possibility. A more likely explanation is that enough copies of the old sequence were created that some of them experienced much less mutation than the diverged sequences at the ancestral locus. The most recent common ancestors (MRCA) occurred in a narrow window of time (evolutionary) relative to the full extent of the tree (< 1/5 of the distance from the root to the most peripheral branch). The MRCA among the orthologous sites occurred within the same time frame as the other MRCA. The tree clearly shows considerable radiation following speciation as evidenced by large genetic distances from MRCA to peripheral tips.

A closer look at MPRE61

Allelic differences or SNP can be identified from cases where MPRE61 sites coincide with overlaps among CVSI BACs. MPRE61 sites coincide with 3 pairs of overlapping BACs, 1 (AC145413 and AC144901), 2 (AC139879, AC140099) and 3 (AC146932 and AC087424). In addition, an MPRE61 site coincided with a group of 3 overlapping BACs, including AC138784, AC138788 and AC138786. Overlapping BAC pair 2 had two single base differences, and pair 3 had 3 single base differences and one 43 bp insertion/deletion. No sequence differences were observed within MPRE61 for pair 1 or the group of 3 overlapping BACs.

To put the apparent allelic diversity rates into context, we examined the genetic sources of the DNA used to construct the BAC library (RPCI-44). The source of DNA for RPCI-44 was a pooled sample with equal contributions from 4 male crossbred pigs each comprised of 3/8 Landrace, 3/8 Yorkshire and ¼ Meishan [28]. The probability of identifying SNP increases with the diversity of genomes sampled. For the cases of 2 overlapping BACs, the probability of sampling different genomes is 87.5%, different breeds is 65.7%, and one BAC of western (Landrace or Yorkshire) origin and the other of Meishan origin is 37.5%. The probability of sampling diverse genomes is higher for the case of 3 overlapping BACs. The probability of sampling more than one genome is 98.4%, more than one breed is 87.9%, and at least one BAC of western origin combined with one BAC of Meishan origin is 56.25%. The fact that we didn't observe SNP in one of the three pairs of overlapping BACs is not that unusual given that the probability of sampling identical genomes with at least one of the 3 pairs of overlapping BACs is 33% (1-.875³). On the other hand, the fact that we did not observe SNP within the group of 3 overlapping BACs given the relatively high probabilities of diverse genomes being sampled is unexpected.

To bolster the relatively small number of distinct MPRE61 loci (7) identified in the CVSI BACs, we further investigated the prevalence and diversity of MPRE61 by cloning and sequencing PCR amplification products derived from 16 pigs sampled from 10 breeds (Berkshire, Chester White, Duroc, Hampshire, Landrace, Meishan, Pietrain, Poland China, Spot, and Yorkshire). We used primers designed to match the highly conserved parts of MPRE61 to amplify and clone (see Methods for details) multiple and variable loci for the RE that are differentiable by size as well as sequence. The different breeds showed indistinguishable smears on denaturing PAGE gels including many different sizes. Too many fragments and too many sizes were present to identify allelic differences in sizes among animals. The PCR products were sequenced to yield 91 reads that were not bacterial or vector contamination. The 91 sequences (listed as a fasta file in Additional file 4) were analyzed with Clustal X (creating a dendrogram file, Additional file 5) and displayed in Figure 4 as a phylogenetic tree. The topology of the tree (number of diverse nodes) is consistent with the estimated copy number of 300 sites in the whole genome given in Table 1. We speculate that the more similar sequences represented as tips close (with few sequence differences) to their common ancestor are probably allelic differences at the same locus. On the other hand, the more diverse tips and peripheral nodes probably represent different sites or loci. The amount of sequence diversity presented in Figure 4 supports the idea that individual integration sites (loci) and alleles of repetitive elements can be uniquely identified by high-throughput array based assays by hybridizing samples to short probes. This demonstrates that repetitive DNA with similar properties to MPRE61 (i.e., prevalence and diversity) can be harnessed for genetic and physical mapping [29]. This dispels the long standing myth that repetitive DNA should always be avoided because it is intractable. Our results indicate that some classes (low to intermediate copy number and highly diverse) of repetitive DNA would be tractable with high-throughput technologies.

MPRE61 size differences are not randomly distributed throughout the phylogenetic tree. Different sizes cluster on different branches of the tree; however, the clustering is not complete. This indicates that insertions and deletions (evolutionary events that cause size differences) occurred throughout the evolution of MPRE61, and in some cases while the element was still replicating. The incomplete clustering of sizes indicates evolutionary plasticity and as a result recurrent insertions and deletions.

MPRE61 was further characterized by plotting BLAST hits of it to the 275,595 sequences in the trace archive of BAC-ends submitted by Sanger. These were plotted along with the repeat elements recognized by RepeatMasker. The most interesting observations included the fact that three times among the 140 hits a PRE1 was incorporated into MPRE61. PRE1 is a porcine specific SINE that is included in the RepeatMasker library. Several other examples existed of PRE1 next to a section of MPRE61, but the trace end occurred next to the PRE1, so that it may or may not have had the continuing section of MPRE61 on its other side. No other REs were found to be incorporated into MPRE61, suggesting that MPRE61 replicated relatively recently. Another interesting observation was that the density of REs on the 3' side of MPRE61 was much higher than on the 5' side. To take a closer look at this, we collected the trace sequence 3' of the 62 hits that ended near (within 60 bp of) the 3' end of MPRE61 (length of 1059 bp). This flanking sequence, ranging in length from 12 to 1368 nucleotides, was analyzed for repeat content and distance of that content from the end of MPRE61 (Figure 5). Running RepeatMasker on the entire collection (275,595 sequences) of Sanger BAC-ends shows that the number of SINE elements is 75% greater than the number of LINE elements (203,206 vs. 116,107). The LINEs tend to be longer than the SINEs, so the total percentage of sequence occupied by the LINEs is actually larger (13.29% vs. 10.29%). The most obvious feature of Figure 5 is that LINEs are significantly over represented on the 3 prime side of MPRE61, particularly in the region closest to the end of MPRE61. For the 22 LINEs that occur within 80 bp of the end of MPRE61, 15 are oriented on the opposite strand and 7 on the same strand. At this point, there is no way to know which strand of MPRE61 might be transcribed. We arbitrarily chose one of the strands and used it consistently. Because the LINEs have a particular internal structure, the 5' and 3' ends are well defined. So another way of looking at the result would be to say that the LINEs occur on the 5' end of MPRE61 (or rather, its reverse complement) with 15 on the same strand and 7 on the opposite strand. Either way, there is less strand conservation than would be expected if MPRE61 used LINEs as a vehicle for either replication or integration.

Bioinformatics

Two hundred-twenty fully sequenced porcine BACs generated by the Comparative Vertebrate Sequencing Initiative [30, 31] were downloaded from the RPCI-44 clone library, totaling 36.4 Mb. RepeatMasker [17] masked out 36% of this sequence. All unmasked fragments of sequence that were at least 50 bp long were compared (BLAST) to the original data set. The BLAST parameters used were those recommended by Korf et al. (2003) for finding repeat elements, namely -r 1 -q -1 -G 2 -E 2 -W 9 -F "m D" -e 1 for NCBI-BLAST [32]. The output, which contained 1,334,953 hits, was analyzed using two similar methods. One was to use the RECON software [10] downloaded from its website [33] and the other used separate, original PERL scripts that performed several of the same functions included in the RECON package.

PCR and sequencing

Primer pairs for amplification of genomic DNA were designed from consensus MPRE61 sequences using Primer3 [34]. Primer sequences were 5'-TTTTCCTGTGGTGATTTGTGA-3' and 5'-GGGCGCTGGACTGCTCAAA-3' (positions 278–298 and 953-935 (5' to 3' on opposite strand) of MPRE61, respectively). PCR was performed in a PTC-225 DNA engine (MJ Research Inc, Watertown, MA) using 0.25 U Hot Star^® Taq polymerase (Qiagen, Valencia, CA, USA), 1X of supplied buffer, 1.5 mM MgCl2, 200 μM dNTPs, 0.8 μM each primer, and 100 ng of genomic DNA in 25 μl reactions. The PCR mixture was held at 94°C for 15 min, and cycled 44 times at 94°C for 20 sec, held at 57°C annealing temperature for 30 sec and extension at 72°C for 1.5 min, followed by a final extension at 72°C for 5 min. Five μl of the PCR reaction was electrophoresed in 1.5% agarose gels to determine quality of amplification and a portion (2–4 μl) was used for cloning in pCR4-TOPO vector (Invitrogen, Carlsbad, CA). Plasmid DNA was prepared using standard alkaline lysis and PTFE filter plates (Millipore, Bedford, MA) and was sequenced with T7 primer.