RepeatFiller newly identifies megabases of aligning repetitive sequences and improves annotations of conserved non-exonic elements

Abstract

Background: Transposons and other repetitive sequences make up a large part of complex genomes. Repetitive sequences can be co-opted into a variety of functions and thus provide a source for evolutionary novelty. However, comprehensively detecting ancestral repeats that align between species is difficult because considering all repeat-overlapping seeds in alignment methods that rely on the seed-and-extend heuristic results in prohibitively high runtimes.

Results: Here, we show that ignoring repeat-overlapping alignment seeds when aligning entire genomes misses numerous alignments between repetitive elements. We present a tool, RepeatFiller, that improves genome alignments by incorporating previously undetected local alignments between repetitive sequences. By applying RepeatFiller to genome alignments between human and 20 other representative mammals, we uncover between 22 and 84 Mb of previously undetected alignments that mostly overlap transposable elements. We further show that the increased alignment coverage improves the annotation of conserved non-exonic elements, both by discovering numerous novel transposon-derived elements that evolve under constraint and by removing thousands of elements that are not under constraint in placental mammals.

Conclusions: RepeatFiller contributes to comprehensively aligning repetitive genomic regions, which facilitates studying transposon co-option and genome evolution. Source code: https://github.com/hillerlab/GenomeAlignmentTools.

Keywords: conserved non-exonic elements; genome alignments; transposons.

Figure 1:

Missed repeat-overlapping alignments and concept of RepeatFiller. Illustration of RepeatFiller. Focusing on unaligning regions in a reference and query genome that are flanked by up- and downstream aligning blocks, RepeatFiller performs a second round of local alignment considering also repeat-overlapping seeds. Newly found local alignments (red boxes) are inserted into the context of other aligning blocks (grey boxes). Unaligning regions that are larger than a user-defined threshold are not considered because the chance of aligning non-orthologous repeats is increased.

Figure 2:

RepeatFiller adds several megabases of aligning transposable elements to existing mammalian genome alignments. (A) Phylogenetic tree of human and 20 non-human mammals whose genomes we aligned to the human genome. The amount of new alignments detected by RepeatFiller is shown in megabases and in percent relative to the human genome. Bar charts provide a breakdown of newly added aligning sequences into overlap with transposons, simple repeats, and non-repetitive sequence. (B) Application of RepeatFiller to fragmented mammalian assemblies still adds a substantial amount of new alignments.

Figure 3:

RepeatFiller also detects additional alignments for non-mammalian genomes. The figure shows how many new alignments were detected by applying RepeatFiller to pairwise alignments of birds, reptiles, and drosophilids. Both the amount (in megabases) of new alignments and the percent of the reference genome additionally aligned are shown. Bar charts show which portion of newly added alignments overlap repetitive sequences.

Figure 4:

Examples of newly identified CNEs near MEIS3. UCSC genome browser [45] screenshot shows an ∼11 kb genomic region overlapping the gene MEIS3, a homeobox transcription factor that is required for hindbrain development. Visualization of the 2 multiple genome alignments (without RepeatFiller at the top, with RepeatFiller below; boxes representing align regions with darker colors indicating a higher alignment identity) shows that RepeatFiller adds several aligning sequences, some of which evolve under evolutionary constraint and thus are CNEs (red boxes) only detected in the RepeatFiller-subjected alignment. The RepeatMasker annotation shows that these newly identified CNEs overlap transposons. The zoom-in shows the 21-mammal alignment of one of the newly identified CNEs, which overlaps a DNA transposon. While this genomic region did not align to any mammal before applying RepeatFiller, our tool identified a well-aligning sequence for 17 non-human mammals (red font). A dot represents a base that is identical to the human base, insertions are marked by vertical orange lines, and unaligning regions are showed as double lines.

Figure 5:

Examples of newly identified CNEs upstream of AUTS2. UCSC genome browser screenshot shows a ∼1.5 Mb genomic region around AUTS2, a transcriptional regulator required for neurodevelopment. CNEs only detected in the RepeatFiller-subjected multiple alignment are marked as red tick marks. The zoom-in shows the 21-mammal alignment of one of the newly identified CNEs. While only the rhesus macaque sequence aligned to human before applying RepeatFiller, our tool identifies a well-aligning sequence for all 19 other mammals (red font). A dot represents a base that is identical to the human base. The RepeatMasker annotation (bottom) shows that this newly identified CNE overlaps a DNA transposon.

Figure 6:

Additional alignments found with RepeatFiller reveal absence of conservation in the genomic regions that were erroneously classified as conserved before. (A, B) UCSC genome browser screenshots showing 2 examples of genomic regions that were only classified as constrained in a multiple genome alignment generated without applying RepeatFiller. Dots in these alignments represent bases that are identical to the human base, insertions are marked by vertical orange lines, and unaligning regions are shown as double lines. The alignments show that the sequences of species added by RepeatFiller (red font) exhibit a number of substitutions. This explains why these regions were not classified as constrained anymore, despite adding more aligning sequences. Note that in (B) only the sequence of the rhesus macaque was aligned before applying RepeatFiller. Sequences in both (A) and (B) overlap long interspersed nuclear element transposons (LINEs). (C) Difference in evolutionary constraint in non-exonic alignment columns that are only classified as constrained in either alignment. For each alignment position, we used GERP++ to compute the estimated number of substitutions rejected by purifying selection (RS). The difference in RS between alignments with and without RepeatFiller is visualized as a violin plot overlaid with a white box plot (box spans the first to third quartile and indicates the median). This shows that almost all non-exonic bases that were only detected as constrained in the alignment with RepeatFiller (orange background) have a positive RS difference, indicating that the newly aligning sequences added by RepeatFiller largely evolve under evolutionary constraint. In contrast, non-exonic bases only detected as constrained in the alignment without RepeatFiller (grey background) often have slightly negative RS differences, indicating that many of the newly added sequences do not evolve under constraint. The 2 distributions are significantly different (P < E−16, 2-sided Wilcoxon rank sum test).