We detected 95 large, intronic insertion/deletion events (indels) distributed across 86 of the 5,738 (1.5%) orthologous gene pairs: 80 genes have a single indel, three genes have two indels, and three genes have three indels (Additional data file 2). Since the 5,738 genes span 42.2 Mb of sequence in D. melanogaster, this suggests the rate of large insertion/deletion events is around 2.3 per Mb. The observed number of indel-harboring genes on each of the five major Drosophila chromosome arms is not significantly different from expectation (Table 1). The size of the inserted sequence ranges from 1,372 bp to 46,889 bp (mean 7,869 bp; standard deviation (SD) 7,347 bp), and 79/95 (83.2%) of the indels have the insertion in the D. melanogaster genome.

Large insertion/deletion events distinguishing orthologous genomic regions can indicate the presence/absence of transposable elements (TEs) 22. To examine whether the indels we detect represent insertions, we used TE annotations for the D. melanogaster genome sequence 2324, and also compared insert sequences against Drosophila TE sequences using BLAST (see Materials and methods for details). Of the 79 indel events showing the insertion in the D. melanogaster genome, 70 (88.6%) map to an annotated TE, and 69 of these also BLAST against known Drosophila TE sequences. For those indels where the insertion is in the D. pseudoobscura genome, 6/16 (37.5%) insertion sequences BLAST to TEs. Since D. pseudoobscura TEs are less well curated than those of D. melanogaster, it is possible that some/most of the remaining ten indels with insertions in D. pseudoobscura are also TEs. Thus, the majority of the indels we identify likely represent TE insertions.

In our analysis we detect TEs indirectly, and in an unbiased fashion, via the identification of large indels. Hence, our observation that the majority of indels have the insertion in the D. melanogaster genome suggests that D. melanogaster introns harbor more TEs than D. pseudoobscura introns. This corroborates the finding of Caspi and Pachter 22 that most of the identifiable TEs in a four Drosophila species genome alignment are present solely in the D. melanogaster lineage, and represent recent insertions in this species. Given these results, we might suspect that the size of orthologous introns would be greater in D. melanogaster than in D. pseudoobscura. Indeed, while the lengths of orthologous introns are highly correlated between these species 25, there is a very slight skew towards larger introns in D. melanogaster (see supplemental Figure S1 of Richards et al. 15). However, Yandell et al. 25 note that while some orthologous introns with highly divergent lengths in the two species may be due to TE insertions (validated by results presented here), most of the differences in the size are subtle and not easily explained by transposons.

We detected 121 small inversions within 93/5,738 (1.6%) orthologous gene pairs: 75 genes harbor a single inversion, 10 genes have two inversions, six genes have three inversions, and two genes have four inversions (Additional data file 3). On average, there are 2.9 microinversion events per Mb of transcribed sequence, suggesting that the rate of microinversion may be similar to the rate of large insertion/deletion - primarily TE insertion - events (2.3 events/Mb, see above). One of the intragenic inversions (CG31481_inv1) corresponds to the single D. melanogaster-D. pseudoobscura microinversion detected by Negre et al. 19 in the proboscipedia gene. The top panel of Figure 1 shows an example of a typical sliding-window BLAST profile, highlighting an inversion event. One possibility is that the events we identify as microinversions are in fact the result of genome assembly artifacts. To rule this out, three of the inversions (CG3578_inv1, CG3936_inv4, and CG32139_inv1) were confirmed by PCR/resequencing of the inversion breakpoints in both D. melanogaster and D. pseudoobscura. Also, for each of the 54 intragenic microinversion events less than 500 bp in size in both species we BLASTed the putatively inverted sequence, including 100 bp flanking each breakpoint, against databases of shotgun sequencing reads. When the orientation of the inversion observed in the assembled genome (relative to flanking sequences) is preserved in one or more reads, we can be confident that the microinversion events we detect are not due to errors in genome assembly. Over the 54 inversions, 51 (94.4%) correctly BLAST to at least one read for both species, and on average, inversions correctly BLAST to 10.0 (6.6) sequence reads in D. melanogaster (D. pseudoobscura). There were no BLAST hits to reads with sequences inconsistent with the inversion orientation in the genome assembly. We conclude that the microinversions we detect are likely real, and not caused by genome assembly artifacts.

Given our success identifying intronic microinversion events, we sought to examine those regions flanking the 5,738 conserved orthologs for microinversions that potentially disrupt upstream or downstream regulatory domains. We extended the sequence of each ortholog by 2 kb upstream and downstream in both D. melanogaster and D. pseudoobscura, and repeated our sliding-window BLAST procedure. In comparison with our scan of intragenic regions, an analysis of short regions flanking genes has lower power to detect microinversion events for three reasons. First, intergenic sequence is generally less conserved than transcribed intronic sequence, although this difference may be slight 26. Second, we only scan 2 kb regions, and thus can detect only microinversions below this size. Finally, outside of transcribed regions synteny between the two genomes can break down. Richards et al. 15 report that the average number of genes within a D. melanogaster-D pseudoobscura syntenic block is 10.7, or around 83 kb of sequence. Thus, the intergenic regions we compare may not always be orthologous.

We discovered 22 microinversions in the 19.7 Mb of unique intergenic sequence tested, or 1.1 events/Mb (Additional data file 4). This is proportionally far fewer inversions than we found in intragenic regions (121 microinversions were detected in 42.2 Mb of transcribed sequence, or 2.9 events/Mb), for the reasons stated above. Three of the 22 microinversions were upstream or downstream of genes also harboring an intragenic microinversion event. In total, over both of our sliding-window BLAST tests, we identify 143 unique microinversions distinguishing the genomes of D. melanogaster and D. pseudoobscura. These 143 events are in/near 112 different genes.

In D. melanogaster the frequency of nested genes, genes residing within introns of other genes, is around 7%, and the frequency of overlapping genes is around 15% 27. None of the microinversions overlap a host gene exon, but 7/143 microinversions overlap an annotated exon from a nested/overlapping gene in D. melanogaster (Additional data files 3 and 4). These seven microinversions were not identified by direct scanning of the nested/overlapping genes, presumably due to low sequence conservation of these genes between D. melanogaster and D. pseudoobscura. It is unclear what, if any, effect these seven microinversions may have on the ability of the orthologous D. pseudoobscura nested/overlapping genes to function correctly. To verify that the inverted sequences are single-copy in each of the tested genomes, we BLASTed the sequence of all 143 microinversions against the appropriate genome assembly. The sequences of 142/143 are single copy, while the remaining intronic inversion, CG1794_inv1, BLASTs six times to the genomes of both D. melanogaster and D. pseudoobscura. The inverted region in this case encompasses the cytosolic tRNA gene tRNA:met3:46A (CR30003) that resides in an intron of the Matrix metalloproteinase 2 (Mmp2) gene. We detect multiple BLAST hits for this sequence because tRNA genes are present in multiple copies throughout the fly genome.

The size of the 143 microinversions ranges from 46 bp to 4,006 bp (mean 628 bp; SD 635 bp) in D. melanogaster, and from 40 bp to 4,408 bp (mean 706 bp; SD 731 bp) in D. pseudoobscura. The difference in length between the species is due to insertion/deletion of nucleotides. There does not appear to be any strong directional change in microinversion length between the species, as the D. pseudoobscura arrangement is longer in just 86/143 (60.1%) of cases. Overall, the majority of microinversions are below 1 kb in both species (111/143, 77.6%). Using Clustalx version 1.83.1 2829 we aligned each D. melanogaster inversion event sequence with the corresponding, reverse complemented D. pseudoobscura sequence. Over the 143 events, ignoring alignment gaps, the average percent nucleotide identity is 74.9% (SD 12.8%). We expect a high level of conservation for the identified microinversions, as our ability to detect them was contingent on sequence conservation. Within the D. melanogaster and D. pseudoobscura genome alignment, only 46% of the D. melanogaster bases are identical 15, and this may generally obscure the signature of historical inversion events. Thus, the 143 detectable, conserved microinversions likely represent only a fraction of the events that have occurred since the divergence of D. melanogaster and D. pseudoobscura. Comparing the genomes of more closely related species of Drosophila may reveal much greater numbers of microinversions.

In total, 112 genes harbor a microinversion within the transcribed region or just upstream or downstream. From Table 1 it is clear there is a significant excess of genes with microinversions on D. melanogaster chromosome 2L (Binomial test, P = 0.003), and a significant dearth on chromosome 2R (Binomial test, P < 0.001). What is not clear is why this might be the case, as within the major chromosome arms genes containing microinversions appear to be evenly distributed (Figure 2). If we consider the position of the intragenic microinversions within the host genes, they appear to preferentially reside within larger introns. Of the 121 intragenic microinversions, 82 (67.2%) are within the largest host gene intron, and 104 (85.2%) are within one of the largest two introns. Similar values are found when considering only those genes with greater than four introns (data not shown). However, within the host intron, the inversions show no positional preference: over the 121 intronic inversions, the distribution of the distance between the inversion breakpoints and the flanking exons (weighted by the size of the host intron) is approximately uniform (Additional data file 5). These observations are particularly interesting in light of the recent observation that longer introns diverge more slowly than shorter introns in Drosophila 30. If longer introns are under selective constraint, they may be expected to contain many functional motifs, which could be disrupted and/or shuffled around by an intronic microinversion event.

Comparative genomics seeks to identify functional elements by examining the pattern of sequence conservation across species. The rationale behind this approach is that over evolutionary time sequences will diverge, unless they are under some form of functional or selective constraint. Thus, the maintenance of sequence conservation despite inversion makes the microinversion events we describe particularly interesting, as they may be enriched for functional motifs. Since the microinversions are present both within introns and upstream of genes, this brings up the possibility that inversions might impact the regulation of splicing and gene expression. For example, shuffling transcription factor binding sites within regulatory domains could alter the ability of sets of factors to bind in a coordinated fashion, and thereby up- or down-regulate expression, or alter the timing or tissue-specificity of transcription.

We examined the position of the 143 microinversion events we identify relative to annotated regulatory regions in the D. melanogaster genome. We used two complementary resources: the DNase I footprint database is a systematically curated set of 1,362 Drosophila transcription factor binding sites 3132, and the REDfly database is a comprehensive collection of 628 known cis-regulatory modules (CRMs; sequences sufficient to regulate gene expression) in D. melanogaster 3334. None of the DNase I footprints overlap the sequence of any D. melanogaster microinversion. However, three microinversions are present within a CRM. Microinversion CG31481_inv1, initially detected by Negre et al. 19, resides in intron 2 of the gene proboscipedia (pb), and is present within a 10.4 kb sequence showing enhancer activity 35. Microinversion CG1030_inv1, situated just 3' of the gene Sex combs reduced (Scr), is present within a 6.7 kb region exhibiting enhancer activity 36. Finally, the inversion CG12287_inv1 resides in intron 3 of the gene POU domain protein 2 (pdm2), and overlaps a 1.3 kb enhancer region detected and validated by Berman et al. 37.

Of course, we do not know whether the microinversions we identify actually have an effect on transcriptional regulation in the two species. It is possible that in the three cases we describe the microinversions have no impact on the spacing/ordering of transcription factor binding sites. This may be particularly true for the two large enhancer regions, which at 10.4 kb and 6.7 kb likely do not represent the minimal enhancer. Work on the Sox21b gene, which shows a microinversion in intron 1 (Figure 1), has demonstrated that the pattern of Sox21b embryonic expression is conserved between D. melanogaster and D. pseudoobscura 38. Thus, for this gene at a particular stage in development, the transcribed microinversion appears to be neutral with respect to expression pattern. As the community begins to understand more about binding site biology and the gene regulatory 'code', we may also be able to determine if the inversions we identify generally have a significant impact on gene regulation.

In analyzing the breakpoints between the syntenic blocks of D. melanogaster and D. pseudoobscura, Richards et al. 15 provided evidence for a D. pseudoobscura-specific breakpoint motif, which could in principle effect large inversions via ectopic exchange. The motif is virtually absent from intron sequences, and is thus unlikely to be the cause of the microinversion events we describe here. In bacteria, short (12 to 23 bp) inverted repeat elements have been shown to permit inversion of the intervening DNA segment 39. However, the precise mechanism by which very small inversion events occur in eukaryotes in unknown.

As an initial investigation into this problem, we examined whether the DNA sequence about the microinversion events showed any detectable signature. Richards et al. 15 noted that breakpoint junctions between syntenic blocks of D. melanogaster and D. pseudoobscura were AT rich. The top panel of Figure 3 shows data from a sliding-window analysis of average GC content across the flanking regions and breakpoints for the 143 D. melanogaster-D. pseudoobscura microinversions. It is apparent that in both species, GC content in the flanking region increases slowly towards the inversion breakpoints, and drops dramatically in the first/last 20 bp of the inversion. The average GC content for introns (where we identify most microinversions) is 40.0% in D. melanogaster and 44.0% in D. pseudoobscura, and 200 bp from the microinversion, GC content returns to this genome-wide average. Also, the GC content of the inversions themselves is similar to the intronic average (the average GC content for D. melanogaster inverted sequence is 42.5%, and for D. pseudoobscura is 44.6%).

One possibility is that the GC content pattern we observe across microinversion breakpoints is due not to inversions per se, but instead to a change in GC content between conserved and non-conserved sequence: the microinversions we detect essentially represent conserved sequence, present in opposite orientation in the two genomes. We extracted sequence from all 774 conserved non-coding sequence blocks in the 93 genes harboring intronic microinversions (see Materials and methods for details), and subjected these to the same sliding window GC content analysis we performed for the microinversions. As shown in Figure 3, the pattern of GC content across microinversion breakpoints (top panel), and the pattern across junctions between conserved and non-conserved sequence (bottom panel), is identical. The GC content patterns across conserved Drosophila sequence are very similar to those recorded by Walter et al. 40 for 1,373 blocks of non-coding sequence conserved between human and Takifugu rubripes (Fugu). The fact that the pattern is maintained across vertebrate and invertebrate systems is deserving of further work.

In an attempt to distinguish microinversions from conserved blocks based on nucleotide sequence data, we investigated the frequency of all 5-mer sequence motifs across the boundaries of the events, and examined the nucleotide compositional bias at the edges of the events 41. Neither test clearly distinguished microinversions from conserved blocks (data not shown), suggesting that if there is a general mechanism underlying Drosophila microinversion, it is not easily discernible from primary sequence data alone.

It is of interest to ask when the microinversions we identify occurred in the Drosophila lineage, and which arrangement (standard or inverted) is the ancestral state. Using data from the 12 recently sequenced Drosophila genomes 42 we extracted the orthologous regions surrounding 15 of the intragenic microinversions. For each region we then performed the same sliding-window BLAST procedure we describe above, in each case testing the D. melanogaster and the D. pseudoobscura orthologs individually against each of the other 11 species' orthologs. Figure 4 details the results of these analyses.

For nine of the events (CG6464_inv1, CG9019_inv1, CG9623_inv1, CG11354_inv1, CG12154_inv1, CG12287_inv1, CG31762_inv1, CG32139_inv1, and CG33529_inv1) the data are consistent with the inversion occurring prior to the divergence of the melanogaster group of species. For two events (CG3578_inv1 and CG3936_inv4) the inversion likely occurred prior to the divergence of the melanogaster subgroup of five species. Three microinversion events (CG2872_inv3, CG4220_inv1 and CG15455_inv1) occurred along the obscura group lineage. Finally, one event (CG4838_inv1) shows the inverted arrangement in the three species D. willistoni, D. persimilis, and D. pseudoobscura, and the standard arrangement in the remaining nine species. Three explanations are compatible with the phylogenetic distribution of CG4838_inv1. First, the same inversion may have occurred independently in the lineage leading to D. willistoni and in the lineage leading to the obscura group species. Second, the inversion may have occurred prior to the divergence of D. willistoni and the obscura group species, but re-inverted again in the lineage leading to the melanogaster group of species. Alternatively, the state of the CG4838_inv1 microinversion in D. willistoni may not be correct, and the inverted form may actually be present only in the pair of obscura group species. The latter possibility is conceivable as the current draft assembly of the D. willistoni genome has not been subject to the same scrutiny as the genomes of D. melanogaster and D. pseudoobscura.

Due to ascertainment bias (the microinversion must distinguish D. melanogaster and D. pseudoobscura) we identify only a particular subset of Drosophila microinversions. It will be extremely interesting to extend our analyses to all pairs of Drosophila species, and place identified microinversions on the Drosophila phylogeny. We predict that many more microinversions will be identified between other Drosophila species pairs, and show different phylogenetic patterns.

Finally, we note that the presence of both the standard and inverted arrangements of the 15 tested microinversions in multiple species provides independent support that microinversions are real features of Drosophila genome architecture.

A widely used method to examine sequence differences between/among diverged species is to use VISTA plots of aligned sequence data 43. This highly informative method allows the local nucleotide conservation between species to be assessed, and VISTA plots can be generated for arbitrary regions of aligned genomes using a web-based utility 44. However, while the combination of genome alignment and VISTA plots has been widely employed, the approach may miss some architectural sequence features. For instance, in a VISTA plot comparing two genomes, one is marked as the reference sequence, and the plot is drawn relative to that sequence. Thus, insertions/deletions distinguishing the sequences are not easily seen. This is demonstrated in Figure 1 - in the VISTA plot, using D. melanogaster as the reference sequence, it is not possible to determine that the D. pseudoobscura Sox21b gene is expanded relative to the D. melanogaster homolog. However, our BLAST approach shows that this is the case. Also, while there are methods available to identify rearrangements during genome alignment 45, these are not readily presented using the VISTA plot format. Generally, examining VISTA plots of aligned sequence data may capture much of the important differences between orthologous regions of diverged species. However, some ultrastructural features of the sequences may be missed in some cases. Sliding-window BLAST-based procedures such as that presented here, or those implemented in the GATA software package 46, are likely to prove a worthwhile addition to the armory of those examining the causes and effects of DNA sequence differences between diverged species.

The genome sequences of D. melanogaster (release 4.2.1) and D. pseudoobscura (release 1.04), and the annotation features for D. melanogaster (in GFF v.3 format) were downloaded from FlyBase 47. Details of all the D. melanogaster genes were extracted from the GFF annotation files using a custom perlscript. Orthologous regions of D. pseudoobscura were identified via BLAST, using the standalone BLAST executable function blastall 48.

Release 4.2.1 of the D. melanogaster genome harbors 13,667 annotated protein-coding genes, each represented by a unique CG identifier. We identified 11,011 D. melanogaster protein-coding genes having orthologs in D. pseudoobscura. For each D. melanogaster gene sequence we scanned through the sequence in 15 bp steps, at each step BLASTing a 31 bp query sequence against the putative D. pseudoobscura ortholog. This was accomplished using a custom perlscript calling the standalone BLAST executable function bl2seq 48. For each 31 bp D. melanogaster query sequence, we recorded the position, score, orientation and sequence of the best BLAST hit within the D. pseudoobscura ortholog. Only BLAST hits with scores above 45 were considered in further analyses. There were 5,738 orthologous pairs with at least two above threshold BLAST hits, and greater than 5% of the D. melanogaster gene sequence showing above threshold hits. Only these genes were retained for further analysis (Additional data file 1).

A custom script written in the freely available statistical programming language R 49 was applied to each of the resulting sliding-window ortholog BLAST files. Inversions were recognized as at least two consecutive, above-threshold BLAST hits, where the D. melanogaster query sequences BLAST D. pseudoobscura in reverse orientation, and the order of the hits in the two genomes is reversed (that is, the D. melanogaster query sequences A-B-C-D-E are reverse complemented in D. pseudoobscura, and in reverse order E-D-C-B-A). We placed no restriction on the distance between BLAST hits defining a microinversion to avoid identifying only small events with high levels of nucleotide conservation throughout their length. This means that the threshold of nucleotide conservation required to detect a microinversion is not a constant across the genome. Large insertion/deletion events distinguishing the two genomes were also identified. To be detected, the endpoints of the BLAST hits flanking the insertion had to be separated by greater than 1 kb, and be 10 times more distant than the endpoints flanking the deletion. Plots for all 5,738 genes were manually checked to ensure the accuracy of our automatic scripts (Additional data file 6 50). Also, since we analyzed each gene independently, and genes can overlap in the Drosophila genome 27, we ensured that the inversion and insertion events we describe are unique.

To test whether the large insertion/deletion events we observe are the result of TE insertion, we performed two tests. For those events where the insertion is in the D. melanogaster genome, we compared the position of each insertion with the positions of 6,013 TEs annotated in the D. melanogaster genome 2324. No corresponding database exists for D. pseudoobscura. Second, using BLAST we compared each insertion sequence to a set of TE sequences identified in Drosophila. These sequences are present in the file 'D_mel_transposon_sequence_set.fasta' (version 9.4.1) available from the BDGP natural transposable element project website 51.

To ensure that inferred inversion events are not generally the result of genome assembly errors we designed 1 kb PCR amplicons about three of the inversion events: CG3578_inv1, CG3936_inv4, and CG32139_inv1. Products were amplified in the fly strains used for genome sequencing, that is, D. melanogaster stock number 2057 (Bloomington stock Center) or D. pseudoobscura stock number 14011-0121.94 (Tucson Drosophila species stock center). Accuracy of the genome assemblies was confirmed via dideoxy sequencing. PCR/sequencing oligos are available in Additional data file 7.

The orientation of a putatively inverted sequence in a genome assembly is likely correct if, relative to the flanking sequences, the orientation is preserved within one or more single shotgun sequencing reads. Hence, for the 54 intragenic microinversion events less than 500 bp in size in both species, we extracted the sequence of the inversion and the 100 bp flanking each breakpoint, and BLASTed against the appropriate genome shotgun trace archive database using Mega BLAST 52.

For each breakpoint of the 143 D. melanogaster-D. pseudoobscura microinversions we extracted a contiguous segment of 220 bp (200 bp flanking the breakpoint, and 20 bp internal to the inversion) from each species. For each species, independently for each breakpoint, across all sequences we calculate GC content for all overlapping 5 bp windows.

Conserved blocks were defined on the basis of the D. melanogaster-D. pseudoobscura sliding-window BLAST procedure described above. In the microinversions we identify, the average number of above-threshold BLAST hits per 100 bp of D. melanogaster sequence is 1.6. We therefore defined a conserved block as a sequence having at least 1.6 BLAST hits per 100 bp of D. melanogaster sequence. All of these hits must be between sequences having the same orientation in the two genomes. Furthermore, the 200 bp flanking each edge of the block must be free of above-threshold BLAST hits. Finally, the conserved blocks must be at least 200 bp in length in both species, and no part of the conserved blocks or flanking sequence can be exonic. Using these rules we identified 774 blocks of non-coding sequence conserved between D. melanogaster and D. pseudoobscura in the 93 genes harboring intronic microinversions. To examine GC content change across the boundaries of conserved and non-conserved sequence, using the 774 blocks we performed an analysis identical to that described for the microinversion breakpoints above.

For 15 intragenic microinversion events (CG2872_inv3, CG3578_inv1, CG3936_inv4, CG4220_inv1, CG4838_inv1, CG6464_inv1, CG9019_inv1, CG9623_inv1, CG11354_inv1, CG12154_inv1, CG12287_inv1, CG15455_inv1, CG31762_inv1, CG32139_inv1, CG33529_inv1), we identified the surrounding orthologous regions from 10 other Drosophila species using BLAST via the DroSpeGe website 53. For each of the 15 regions we performed the sliding-window BLAST protocol described above, testing the D. melanogaster sequence and the D. pseudoobscura sequence independently against sequence from every other species. The presence of the standard (D. melanogaster-like) or inverted (D. pseudoobscura-like) sequence arrangement was recorded in each case.

Orthologous sequences were extracted from the following assemblies: dsim_wu050602 (D. simulans), dsec_br051028 (D. sechellia), dyak_caf051213 (D. yakuba), dere_caf051209 (D. erecta), dana_caf051209 (D. ananassae), dper_br051028 (D. persimilis), dmoj_caf051209 (D. mojavensis), dwil_caf060213 (D. willistoni), dvir_caf051209 (D. virilis), and dgri_caf051209 (D. grimshawi). The D. simulans and D. yakuba assemblies were provided by the Genome Sequencing Center, Washington University 54. The D. erecta, D. ananassae, D. mojavensis, D. virilis, and D. grimshawi assemblies were provided by Agencourt Bioscience Corporation 55. The D. sechellia and D. persimilis assemblies were provided by the Broad Institute 56. The D. willistoni assembly was provided by the J. Craig Venter Institute 57.