GenomeDelta: detecting recent transposable element invasions without repeat library

Abstract

We present GenomeDelta, a novel tool for identifying sample-specific sequences, such as recent transposable element (TE) invasions, without requiring a repeat library. GenomeDelta compares high-quality assemblies with short-read data to detect sequences absent from the short reads. It is applicable to both model and non-model organisms and can identify recent TE invasions, spatially heterogeneous sequences, viral insertions, and hotizontal gene transfers. GenomeDelta was validated with simulated and real data and used to discover three recent TE invasions in Drosophila melanogaster and a novel TE with geographic variation in Zymoseptoria tritici.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-024-03459-5.

Background

Transposable elements (TEs) are short DNA sequences capable of increasing their copy numbers within a host genome. They are common in many organisms and often make up a large part of their genome [1, 2]. While some TEs may confer benefits to hosts [3, 4], the majority of TE insertions are likely neutral or deleterious [5, 6].

Consequently, host genomes have evolved elaborate defense mechanisms, frequently involving small RNAs [7]. TEs can evade host silencing through horizontal transfer (HT), i.e., the transmission to naive species that lack the TE [8–11]. HT is a common phenomenon in prokaryotes [12], but recent studies suggest that HT (especially of TEs) is also prevalent in eukaryotic organisms [8, 13].

The evidence for HT of TEs has typically been indirect. Such evidence includes a patchy distribution of the TE among closely related species or a high similarity between the TE of the donor and recipient species, which is frequently quantified by the synonymous divergence of the TE [14, 15].

However, for TEs that have spread very recently, direct evidence of the recent invasion can be obtained, for example, when the sequence of the TE is absent in older samples but present in more recently collected ones. To illustrate, the P-element invaded D. melanogaster populations between 1950 and 1980 [16, 17]. Consequently, sequences with similarity to the P-element are absent in natural D. melanogaster strains collected before 1950 but present in strains collected after 1980. It is possible that such recent invasions are common. In D. melanogaster, an important model organism for studying TE dynamics, 11 different TE families invaded natural populations during the last 200 years [9, 16–22]. It is feasible that other organisms might also experience a high rate of recent invasions.

To obtain direct evidence for recent TE invasion, it is necessary to compare the sequencing data from old and recent samples. Recent samples can be collected from natural populations, whereas old samples may be derived from several sources, including old laboratory strains, genomes of historical specimens from museums, or ancient DNA extracted from archeological remains [20, 23, 24]. The number of species with sequencing data from historical specimens is rapidly increasing [25] (for example, Anopheles sp. [26], Apis mellifera [27], Columba livia [28], and Canis lupus [29]). Therefore, it is in principle feasible to discover direct evidence for recent TE invasions in an increasing number of species. However, discovering recent invasions with existing approaches typically requires comparing the copy numbers of known TE families in old and young samples (e.g., [9, 20]). These approaches thus require prior knowledge of the sequences of the TEs, i.e., a repeat library. Generating repeat libraries is notoriously difficult, requiring extensive manual curation [30–32]. This issue is further compounded by the fact that even for the few species for which a high-quality repeat library is available, the library may be incomplete and not contain the sequences of TEs that have spread very recently. For example, the high-quality repeat library of D. melanogaster [33] lacks the sequence of the retrotransposon Spoink, which spread in natural populations between 1983 and 1993. This is because the reference strain used for generating the repeat library was likely collected prior to that period. This is part of the reasons why the Spoink invasion was only recently discovered, several years after the invasion [21].

The development of an approach that enables identification of recent TE invasions independent of repeat libraries would represent a substantial conceptual advance in the field. For this reason, we developed GenomeDelta. GenomeDelta is based on the idea that recent invasions will lead to sequences that are present in recently collected samples (i.e., after the invasion) but absent in old samples (i.e., before the invasion). As input, GenomeDelta requires a high-quality assembly (ideally a long-read assembly) of the recently collected sample and short-read data of the old sample. GenomeDelta then identifies sequences that are present in the assembly but absent in the short-read data. As this approach does not require prior knowledge about the sequences, it allows to comprehensively identify sample-specific sequences (e.g., TEs that invaded recently) in model and non-model organism. Importantly, GenomeDelta is not designed to detect copy number differences among samples or differences in the insertion sites. Apart from finding recent TE invasions, GenomeDelta may also be used to detect sequences showing a geographically heterogeneous distribution, such as the TE Styx in Z. tritici [34], recent endogenous virus insertions [35], and recent lateral gene transfer [36]. We thoroughly validated our novel tool with simulated and real data. We also provide a detailed manual and a walkthrough. Finally, we show that GenomeDelta can be used to gain novel biological insights. With GenomeDelta, we discovered three novel TE invasion in D. melanogaster in the last three decades and a novel TE (Rosetta) with a spatially heterogeneous distribution in Z. tritici.

Results

GenomeDelta

We developed GenomeDelta (GD) to identify genomic sequences that are present in one sample (P presence) and absent in another sample (A absence). Given the two sets of genomic sequences "P" and "A", GenomeDelta aims to identify the set of sample-specific sequences $P-A$. Note that GenomeDelta does not compute the set $A-P$.

As input, GenomeDelta requires a high-quality assembly for sample P and short-read data for sample A. GenomeDelta is based on the idea that sequences that are present in P and absent in A (i.e., $P-A$) can be identified as coverage gaps when short reads of A are aligned to an assembly of P (Fig. 1A).

Fig. 1.

Overview of GenomeDelta. A Recent TE invasions will lead to coverage gaps when reads of a sample collected before the invasion (H10, collected in 1815) are aligned to the assembly of a sample collected after the invasion (TEN015, collected in 2015). The coverage gap in this example is due to the retrotransposon Spoink, which invaded D. melanogaster populations between 1983 and 1993 [21]. B Workflow of GenomeDelta. Reads (FASTQ) of a sample A (absence) are aligned to an assembly (FASTA) of another sample P (presence). The sequences of coverage gaps are extracted, similar sequences are clustered, a multiple sequence alignment is performed, and consensus sequences are generated. Finally, the sequences that are present in P but absent in A are reported. C Overview of the input and output of GenomeDelta. As output, two fasta-files are generated, one with the consensus sequences of repetitive elements and one with the sequences of non-repetitive elements. D Graphical output generated by GenomeDelta, providing an intuitive overview of the sample-specific repetitive sequences. The length, the copy number, and the coverage bias (see text) are shown for each identified repetitive sequence

One major field of application for GenomeDelta is the identification of novel TE invasions. TE invasions add novel sequences to the genome that are present in samples collected after the invasion (P) but absent in samples collected before the invasion (A). Another major use-case of GenomeDelta is the identification of sequences (TEs) that are present in one geographic region (P) but absent in another (A). For example, KoRV (koala retrovirus) insertions are present in the genomes of koalas sampled from the North but not in all the koalas from the South of Australia [37]. In summary, GenomeDelta may be used to identify sequences (repetitive or non-repetitive) showing a spatial or temporal heterogeneous presence/absence pattern.

To identify the sample-specific sequences ($P-A$), GenomeDelta aligns the sequencing reads of A to the assembly of P and computes the coverage. Next, GenomeDelta identifies coverage gaps, extracts the sequences of the gaps, groups them by sequence similarity (e.g., the different insertions of a TE family), performs a multiple sequence alignment (MSA), and reports the consensus sequences (Fig. 1B). Separate results are reported for repetitive and non-repetitive sample-specific sequences (Fig. 1C). Finally, GenomeDelta estimates the reliability for each of these sample-specific sequences ($P-A$) by computing a coverage bias score. The coverage bias is estimated as $bias=2\ast \frac{f}{g+f}-1$ where f is the coverage in the regions flanking the coverage gap (10,000 bp in each direction) and g the average genomic coverage. The bias ranges from $-1$ to 1, where 0 indicates an unbiased coverage and $-1$ and 1 a highly biased coverage (either highly decreased or increased; Fig. S1). As output, the sample specific-sequences are provided as two fasta-files, one for the consensus sequences of repetitive elements and one for the non-repetitive sequences. Additionally, a bed file with the genomic coordinates and the coverage bias of each coverage gap is reported. To provide an intuitive graphical overview, GenomeDelta also generates a summary plot, showing for each sample-specific repetitive sequence the copy number, the length, and the coverage bias (Fig. 1D).

GenomeDelta can be easily installed with conda ([38]). A detailed manual and a walkthrough with data from D. melanogaster are available. GenomeDelta is distributed under the Open Software License v.2.1.

Validation

We thoroughly validated GenomeDelta with simulated and real data. For validations with simulated data, we used a chromosome sequence of D. melanogaster (chromosome arm 2R) as template and inserted 25 copies of a randomly generated sequence with a length of 5000 bp into this template. We thus obtain an artificial sequence with (P) and without TEs (i.e., the template, A). Next, we generated artificial short reads for the sequence without TEs (A). We simulated artificial reads yielding different coverages and coverage distributions (uniform coverage and random position of reads; Table 1). We also used Gargammel [39] to simulate reads mimicking properties of ancient DNA (i.e., fragmentation, cytosine deamination, bacterial contamination [24]). Finally, we plugged the artificial reads into GenomeDelta to identify the sample-specific sequences. To evaluate the performance of GenomeDelta, we computed the true positive rate, the false positive rate, the length of the identified consensus sequence, and the similarity between the observed consensus sequence and the simulated sequence (Table 1). The 25 artificial insertions were detected in all scenarios. The obtained consensus sequence was 100% identical to the simulated sequence, and the length of the consensus sequence was close to the simulated 5000 bp. With a low coverage, especially when properties of ancient DNA were simulated, the false positive rate was elevated and the length of the consensus sequence deviated from the expected one (i.e., 5000; Table 1). However, with a coverage of $\ge 5$, no false positives were found and the observed length was close to the expected one (Table 1). To test the performance of GenomeDelta with short sequences, we performed an additional validation with a sequence of length 1000 bp. With such a short sequence, GenomeDelta identified more false positive sequences when the coverage was low (i.e., 1x, Table S1). Finally, we simulated reads with different lengths and sequencing error rates. The error rate of the reads and the read length had little impact on the performance of GenomeDelta (Tables S2, S3).

Table 1.

Validation of GenomeDelta with simulated data. We introduced 25 copies of an artificial TE with a size of 5000 bp into a template sequence and tested if the TE sequence was accurately identified with GenomeDelta. As input, we used the template with the TEs (P) and artificial reads simulated from the template without TEs (A). We simulated different read coverages using either a uniform or a heterogeneous coverage (random position of reads). We also simulated properties of ancient DNA with Gargammel. We evaluated the number of true positive insertions (TP), the number of false positive insertions ($F{P}_r$, $F{P}_{\mathit{nr}}$: repetitive or non-repetitive), the length of the reported consensus sequence [len. (bp)] as well as the sequence similarity between the reported consensus sequence and the simulated insertion [sim (%)]

Method	Coverage	TP	$F{P}_{\mathit{nr}}$	$F{P}_r$	len. (bp)	sim (%)
uniform	1	25	8	0	5000	100
uniform	5	25	0	0	4998	100
uniform	10	25	0	0	4998	100
random	1	25	607	2	5251	100
random	5	25	0	0	4998	100
random	10	25	0	0	4997	100
gargammel	1	25	2410	17	5435	100
gargammel	5	25	0	0	5000	100
gargammel	10	25	0	0	4997	100

Our validations with simulated data suggest that GenomeDelta accurately identifies sample specific-sequences with a length $>1000$ bp. Furthermore, the short-read data (of sample A) ought to have a minimum coverage of 5.

Next, we validated GenomeDelta with real genomic data from an insect (the fruit fly, D. melanogaster) and a mammal (the koala, Phascolarctos cinereus).

Based on various approaches, ranging from phenotyping (hybrid dysgenesis) to whole genome sequencing, previous works showed that seven TEs spread in D. melanogaster populations between 1800 and 1980 [9, 16–20, 40]. These works showed that Blood, 412, Opus spread between 1850 and 1930, Tirant around 1935 followed by the I-element, Hobo, and the P-element (Fig. 2A). We tested whether GenomeDelta identifies these known invasions when short-read data of a strain collected around 1815 (H10) are aligned to the assembly of a strain collected around 1975 (Pi2) [41, 42]. Note that the short-read data are derived from a strain kept for $\approx$200 years in a museum, thus the sequencing reads are fragmented with an average size of around 50 bp [41]. GenomeDelta detected 27 repetitive sequences that are specific to Pi2 (Fig. 2B). A blast search of the consensus sequences against a TE library [33, 43] revealed that the 7 TE that invaded D. melanogaster recently are the most notable outliers in terms of length, copy numbers, and coverage bias (Fig. 2B). Note that several fragments are reported for some TEs (Blood, I-element, Tirant; Fig. 2B). This can be explained by the fact that degraded fragments of these TEs, likely the remnants of ancient invasions, are present in all genomes, including the genome of the strain sampled at 1815 [9, 20]. Reads derived from these fragments may be mapped on the new TE insertions, thus fragmenting the coverage gap generated by the new insertion. The other sample-specific sequences were less promising, as, for example, seen by the high coverage bias. Closer inspection revealed that many of these sequences were located in telomeric regions where few reads aligned in Pi2 (Table S4). GenomeDelta thus identified the 7 TEs that invaded D. melanogaster between 1815 and 1975 based on sequencing reads from historical specimens [41].

Fig. 2.

Validation of GenomeDelta with real data. A Overview of the invasion history of D. melanogaster until 1975 as revealed by previous works. B Overview of the sequences identified by GenomeDelta when a sample collected around 1815 (H10) is compared to a strain collected in 1975 (Pi2). Note that the identified sequences correspond to the recent invaders. C Previous work shows that KoRV insertions are present in the genomes of koalas sampled in the North but not in the South of Australia. D Overview of sequences identified by GenomeDelta when short reads from a southern koala (FRIS-F-SAWS22) are aligned to an assembly of a northern koala (GCA_002099425.1). Note that the best candidate sequence (i.e., coverage bias close to zero) corresponds to KoRV

We next evaluated the influence of the coverage on the performance of GenomeDelta with real data. The sample used to identify the 7 invasions, H10, yielded a coverage of 33x [41]. We subsampled the number of reads of H10 to coverages of 10x, 5x, and 1x, and again used GenomeDelta to identify sequences specific to Pi2 (Fig. S2). With a coverage of 10x, GenomeDelta identified 5 out of the 7 TEs (Blood and Tirant were missing). With a coverage of 5x, we identified the same 5 out of the 7 TE, but additionally Hobo and Opus were fragmented into two clusters. With a coverage of 1x, just one short fragment of the I-element was found (Fig. S2). In agreement with the validation with artificial data, the subsampling of real data (from historical specimens) suggest that a coverage $>5$ should be used with GenomeDelta.

We next tested the performance of GenomeDelta with a mammal, i.e., the koala (Phascolarctos cinereus). Insertions of the koala retrovirus (KoRV) have been found in the genomes of koalas sampled from the North of Australia but not in koalas sampled from the South [37]. KoRV may be at the transition stage between an exogenous element (e.g., virus) and a vertically transmitted endogenous element (i.e., a transposable element [37]). We tested whether GenomeDelta manages to identify KoRV, by comparing short-read data from a southern koala (FRIS-F-SAWS22 [44]) to the assembly of a northern koala (i.e., the reference genome GCA_002099425.1, which is based on a koala from the Sunshine coast; Fig. 2C). GenomeDelta identified several sample-specific repetitive sequences (Fig. 2D). The most promising sequence (low coverage bias, high copy number, and substantial length) corresponds to KoRV (Fig. 2D). Compared to D. melanogaster, we find more false positive sequences (57 in koalas and 11 in D. melanogaster; Fig. 2B, D). This can likely be explained by the much larger genome size of koalas as compared to D. melanogaster (3500 Mb in koalas and 200 Mb in D. melanogaster [45, 46]). As a control, we also compared two koalas from a northern population with GenomeDelta (Sunshine-Coast-M-79817 and the reference genome) and did not find any sequences matching with KoRV (Fig. S3). In agreement with previous work, GenomeDelta thus identified the integration of KoRV in the genomes of northern koalas but not in southern koalas [37].

In summary, we thoroughly validated GenomeDelta with artificial and real data. We also showed that GenomeDelta may be used with historical DNA and that a minimum coverage of 5 should be targeted.

Novel biological insights

Finally, we provide two examples illustrating how GenomeDelta can be used to generate novel biological insights. First, we identified three novel TE invasions in D. melanogaster with GenomeDelta (Fig. 2A; [22]). We used GenomeDelta to align reads from a sample collected in the early 1800s (H10) to the assembly of a strain collected in 2016 (TOM007). As expected, we found all the previously described TE invasions that occurred between 1810 and 1975 (Fig. 2A, B) and Spoink, which spread between 1983 and 1993 [21]. Surprisingly, GenomeDelta also discovered three novel TE invasions in D. melanogaster: Micropia-like element (MLE), Souslik, and Transib1 (Fig. 3A). The coverage gaps caused by each of these TE families are shown in Fig. S6. These TEs are also present, at least partially, in a repeat library generated from the long-read assemblies of recently collected D. melanogaster strains ([47]; for a cross-reference of the TE names between this study and the study of Rech et al. see Table S5). We are describing these novel invasions in detail in a separate work [22]. Here, we use these three novel invasions to showcase how candidates identified with GenomeDelta may be further validated and investigated. We will thus start with the raw results provided by GenomeDelta. As a first step, we investigated the coverage bias. Since the three novel sequences have a low bias (i.e., close to zero), they may be considered promising candidates (Fig. 3A blue). Next, a blast search revealed that these sequences correspond to three different annotated TEs, a “Micropia-like” element (MLE) described in D. melanogaster (GenBank: MN418888.1), Transib1 described in D. melanogaster and D. simulans, and to Souslik identified in D. simulans (GenBank: BK008880.1). For some TEs, such as Blood, the consensus sequence identified by GenomeDelta may be fragmented or incomplete (Fig. 3). To obtain the complete consensus sequence of the novel TEs, we extracted the sequence of each insertion together with the flanking sequences (3000 bp) using bedtools. Next, we performed a multiple sequence alignment of these sequences (TE + flanking region) with MUSCLE and constructed a novel consensus sequence using the GenomeDelta script “MSA2consensus.py,” which employs a simple majority rule for generating consensus sequences. While the consensus sequences of MLE and Souslik remained largely unchanged, the length of the consensus sequence of Transib1 increased from 1323 to 3030 bp. This shows that the initial consensus sequence of Transib1 extracted by GenomeDelta was incomplete. This finding can be attributed to the presence of degraded fragments of Transib1, likely remnants of past invasions, in all D. melanogaster strains [22, 48]. Reads from ancient Transib1 fragments may be misaligned to the recent Transib1 insertions, thus interfering with GenomeDelta’s identification of coverage gaps. Next, we identified the LTR sequence of MLE and Souslik and the inverted repeat sequences of Transib1 using BLAST. Based on long-read assemblies, we confirmed the presence of these three TEs in strains collected around 2015 and their absence in strains collected before 1975 [22]. We inferred the exact timing of these three invasions using 585 D. melanogaster samples collected during the last 200 years, revealing that MLE, Souslik and Transib1 invaded D. melanogaster around 1985, 2005, and 2013, respectively [22]. Finally, we aimed to identify the species that acted as donor of the horizontal transfer triggering the invasions, by analyzing the genomes of 1400 arthropod species [22]. The most likely donor species for MLE is a Drosophila species of the willistoni group, while Souslik and Transib1 were likely donated from the sister species D. simulans [22]. By identifying three recent TE invasions in D. melanogaster, we demonstrated that GenomeDelta can be used to generate novel insights about sequences showing a temporally heterogeneous distribution [22].

Fig. 3.

GenomeDelta can be used to generate novel biological insights. A GenomeDelta identifies three novel TE invasions in D. melanogaster (blue). Short-read data of a strain collected in the early 1800s (A: H10) were aligned to the assembly of a strain collected in 2016 (P: TOM007). Previously identified TE invasions are in black. B GenomeDelta identifies a novel TE (Rosetta) with a spatially heterogeneous distribution in Z. tritici. Short-read data of a sample from Iran (A: SRR5194593) were aligned to the assembly of a sample from the Netherlands (P: GCA000219625.1). C Copy number of Rosetta in Z. tritici samples collected from different geographic regions. As controls, a TE with a known spatially heterogeneous distribution (Styx) and a TE present in all strains (Neptunus) are shown. TE copy numbers were estimated with DeviaTE

We next tested whether GenomeDelta can also be used to generate novel insights about sequences showing a spatially heterogeneous distribution. We utilized the publicly available genomic resources for the important crop pathogen Zymnoseptoria tritici (see Additional files 1 and 2), which causes septoria leaf blotch, one of the most important diseases of wheat [49]. Z. tritici is native to the Middle East, but spread to all continents between 500 and 200 years ago [50]. Elegant recent work revealed that several TEs (Styx, Juno, Deimos, and Fatima) show a spatially heterogeneous distribution in Z. tritici, where, for example, the TE Styx is present in populations from Europe but not in the Middle East [50] (see also Fig. 3C). The authors attributed this heterogeneous distribution of the TEs to spatial differences in the efficiency of the genomic host defenses against TEs (repeat-induced point mutations [51]). Using GenomeDelta, we aligned short reads from a Middle East sample (SRR5194593) to the assemblies of 15 Z. tritici strains collected from diverse continents. As expected, a comparison between the sample from the Middle East and Australia identified TEs that were previously shown to have a geographic heterogeneous distribution (Styx, Juno, Deimos, and Fatima, Fig. S4; [50]). Interestingly, by comparing a sample from the Middle East to a sample from the Netherlands, GenomeDelta revealed a novel TE with a heterogeneous distribution, i.e., Rosetta (Fig. 3B). Although Rosetta has been previously annotated in Z. tritici, a spatially heterogeneous composition has not been described before [50, 52]. To further investigate the abundance of Rosetta in samples from different geographic regions, we used DeviaTE [53]. DeviaTE estimates the copy numbers of TEs by normalizing the coverage of TEs to the coverage of single copy genes. We found that Rosetta is most prevalent in Oceania and Europe but largely absent in Africa, the Middle East, North and South America (Fig. 3C). As control, we also analyzed the abundance of Styx with DeviaTE. In agreement with previous work, Styx was found in Europe and the Americas but not in Africa and Oceania (Fig. 3C). As a further control, we included a TE (Neptunus) present in all populations. A high number of Neptunus insertions can be found in all analyzed samples (Fig. 3C). GenomeDelta thus identified Rosetta as a novel TE with a spatially heterogeneous distribution in Z. tritici. Since the distribution of Styx and Rosetta varies among regions (e.g., Rosetta but not Styx is present in Oceania), it is questionable whether differences in the efficiency of the genomic defense as proposed previously [50] can account for the spatially heterogeneous distribution of the TEs. Differences in the host defense ought to affect all TEs equally. An alternative hypothesis might be that Styx and Rosetta recently spread in Z. tritici following a horizontal transfer. The differences in distribution of Styx and Rosetta can then simply be explained by different geographic origins of the horizontal transfer that triggered the invasions or by differences in invasion routes caused by stochastic migration events transmitting the TE between the populations.

In summary, we showed that GenomeDelta can be used to gain novel biological insight into spatially (Z. tritici) and temporally (D. melanogaster) heterogeneous distributions of TEs.

Discussion

In this work, we presented GenomeDelta, a tool designed to detect genomic sequences that are present in one sample but absent in another one. We thoroughly validated GenomeDelta with both artificial and real data and showed that GenomeDelta can be used to generate novel biological insights by detecting recent TE invasions in D. melanogaster and identifying a TE with a geographically heterogeneous distribution in Z. tritici (Fig. 3).

As major advantage, GenomeDelta can be used to identify sample-specific sequences without prior knowledge about the sequences. Such sample specific sequences are of biological interest and could be generated due to varying processes, ranging from horizontal transfer, to differences in the host defense against foreign DNA and to locally restricted negative selection against some sequences. One important use-case for finding sample specific sequences is the identification of recent TE invasions, where a TE is present in recent but absent in older samples. Previous approaches for finding such recent TE invasions required a comprehensive repeat library, which enables estimating copy number differences of repeats among samples of interest. Creating such repeat libraries is notoriously difficult and requires substantial manual curation [30–32]. Even more problematic is that a single repeat library for a species may not be sufficient, as some TEs may not be present in the library. For example, several TEs that invaded D. melanogaster recently (Spoink, MLE, and Souslik) are not present in the high-quality repeat library of D. melanogaster (these TEs spread after the strain used for generating the repeat library was collected [22, 33]). A comprehensive identification of all TE invasions would thus require a "pan repeat library" for a species, i.e., a library comprising the repeat sequences of a large number of strains sampled at different times from diverse geographic regions. Generating up-to-date repeat-libraries is thus a major bottleneck with conventional approaches for finding recent TE invasions. The identification of sample-specific sequences with GenomeDelta, independent of a repeat library, is thus a major conceptual advance that enables identifying all recent TE invasions in model and non-model organisms.

GenomeDelta identifies sample-specific sequences ($P-A$) by aligning short reads (sample A) to an assembly (sample P). Our validations indicate that historical DNA can be used (sample A) and that the short-read data (sample A) should have a minimum coverage of 5. This coverage requirement of 5 may be a limitation for some projects where shallow sequencing was performed for several samples. In this case, one workaround may be to simply pool the reads of all samples with shared properties (e.g., to pool all samples from the same geographic region or the same decade).

Another limitation is that GenomeDelta solely identifies sequences with qualitative differences, i.e., being present in some samples and absent in others. It will not identify sequences having quantitative differences in copy numbers among samples.

Perhaps the major limitation of GenomeDelta is the requirement for a high-quality genome assembly (sample P). High-quality assemblies are necessary, as TE sequences not present in the assembly cannot be detected (e.g., repeat sequences may be missing in low-quality assemblies). It is therefore not possible to identify sequences specific to samples from which a high-quality assembly cannot be generated, such as historical samples, which typically yield only highly fragmented DNA [23]. However, the need for a high-quality assembly is not unique to GenomeDelta; such assemblies are also indispensable for approaches that rely on a repeat library. In fact, without a high-quality assembly, it is impossible to construct a comprehensive repeat library.

Despite these limitations, we anticipate that GenomeDelta may be used for a wide range of applications. Our primary motivation for designing GenomeDelta was to discover recent TE invasions. For example, we used GenomeDelta to discover the three most recent TE invasions in D. melanogaster (Fig. 3A; [22]). As another application, GenomeDelta might be used to identify sequences occurring in populations of some regions but not in others. We demonstrate this by using GenomeDelta to discover a novel TE with a geographically heterogeneous composition in Z. tritici (Rosetta; Fig. 3B, C). Such a geographically heterogeneous distribution of TEs might result from spatial differences in the efficiency of the genomic host defense against TEs [50]. Whether GenomeDelta is best used to search for recent invasions or geographically heterogeneous TE distribution will depend on the research question, the available data/samples, and the population structure of the investigated species. In species with a prominent population structure, there could be some geographic heterogeneity in the TE composition for an extended period, which could be identified with GenomeDelta. On the other hand, a TE may rapidly spread in all individuals of species with little population structure (e.g., Transib1 in D. melanogaster [22]). A heterogenous TE composition may thus be transient, and it may be better to use GenomeDelta to search for recent invasions. Finally, GenomeDelta could also be used to identify sample-specific non-repetitive sequences such as recent endogenous retrovirus insertions [35] and recent lateral gene transfer [36].

Conclusion

GenomeDelta identifies genomic sequences that are present in one sample and absent in another, such as recent transposable element (TE) invasions, without a repeat library. This represents a significant advantage, as constructing repeat libraries is challenging and the libraries are frequently incomplete, even for well-studied model organisms like D. melanogaster. GenomeDelta operates by identifying coverage gaps when short reads from one sample are aligned to a high-quality assembly of another sample.

The tool has a wide range of applications for studying genomic variation. For instance, it has been used to discover three recent TE invasions identified in D. melanogaster. Additionally, GenomeDelta has proven effective in identifying sequences with geographically heterogeneous distributions, exemplified by its identification of a TE with a heterogeneous distribution in Z. tritici.

GenomeDelta has been validated using both simulated and real data, including historical specimens, demonstrating its reliability and versatility for genomic studies.

Methods

Code structure

To identify sequences that are present in sample P and absent in sample A, GenomeDelta requires an assembly in FASTA format (P) and sequencing reads in FASTQ format (A).

GenomeDelta proceeds in several steps relying on widely used tools. First, GenomeDelta aligns the reads from A to P using bwa mem (v 0.7.17, [54]). Note that it is important to use an algorithm that performs a local alignment of the reads (such as bwa mem), otherwise the boundaries of the coverage gaps may be inferred less accurately. The mapped reads are sorted with samtools (v 1.17, [55]), converted to bam files, and the coverage is computed with samtools depth. Next coverage gaps, i.e., regions with zero coverage (default threshold), are identified. To allow for some spurious mapping of reads, GenomeDelta enables merging adjacent coverage gaps separated by a maximum distance of d (Fig. S5). All coverage gaps having a minimum size (customizable; per default 1000 bp) are extracted into a fasta-file with bedtools, and a bias score is computed for each gap. The score is computed as $bias=2\ast \frac{f}{g+f}-1$ where f is the coverage in the 10 kb regions flanking the gap (on both sides) and g the average coverage of the genome (Fig. S1). Bedtools is used to compute the coverage in the regions flanking the gaps (v 2.30.0, [56]). The bias score ranges from $-1$ to 1, with 0 indicating an unbiased coverage in the flanking regions (i.e., flanking regions have the same coverage as the genomic average). A sample-specific repetitive sequence will lead to several coverage gaps, where each gap correspond to one insertion of the repetitive sequence (e.g., the dispersed insertions of a TE family). GenomeDelta derives a single consensus sequence for each repetitive sequence. To do this, GenomeDelta, extracts the sequence of each coverage gap, clusters them based on a similarity search with BLAST (v 2.6.0, [57]) and a Python script (blast2clusters.py). We use a minimum of 3 sequences per cluster. For each cluster, a multiple sequence alignment (MSA) is generated with MUSCLE (v 3.8.1551, [58]) and the consensus sequence is derived using a Python script “MSA2consensus.py.” The coverage bias of each cluster is calculated as the median of the individual biases of the clustered sequences.

As main output GenomeDelta provides two fasta files: (1) GD-candidates.fasta contains the consensus sequences of the repetitive clusters and (2) GD-non-repetitive.fasta has the sequences of the non-repetitive coverage gaps. Furthermore, a bed file is generated, which includes the genomic coordinates and the coverage bias of each coverage gap. Additionally, GenomeDelta allows to access all generated intermediate files in the output folder. Finally, GenomeDelta provides a plot summarizing the properties of the repetitive clusters, i.e., the number of sequences in the cluster (e.g., the copy numbers of a TE family), the average length of the sequences, and the average coverage bias. The plot is computed with the R package ggplot2 (v 3.4.4, [59]).

Simulated data

To simulate artificial data, we used the chromosome arm 2R of D. melanogaster from the assembly GCA000001215.4 (strain ISO1 [60]) as reference sequence (sample A). We randomly inserted repetitive sequences into this reference sequences with SimulaTE (v 1.31, [61]). Artificial reads were generated with a Python script (create-reads.py). Artificial reads that capture properties of ancient DNA were simulated with Gargammel (v3, [39]) using 10% bacterial contamination (Wolbachia pipientis, an endosymbiont of D. melanogaster), 8% contamination with modern D. melanogaster DNA, and 82% DNA of interest. The similarity between the observed consensus sequence and the simulated one was computed with BLAST 2.6.0 [62].

Validation with real data

To validate GenomeDelta with real data, we used short reads of the strain H10 (1815, Sweden) as well as an assembly of strain Pi2 [41, 42]. We randomly subsampled the reads with Rasusa v0.8.0 [63] to obtain different coverages.

Z. tritici

We used DeviaTE (v0.3.8) [53] to estimate the copy numbers of different TEs, including Rosetta (the novel TE identified by GenomeDelta), in Z. tritici strains collected from diverse geographic regions. Short reads were aligned to the sequences of the TEs and to three single copy genes (MYCGRDRAFT-39655, MYCGRDRAFT-70396, and MYCGRDRAFT-99758) with bwa bwasw (version 0.7.17-r1188) [54]. DeviaTE estimates the copy number of a TE by normalizing the coverage of the TE by the coverage of the single copy genes [53].

Authors' contributions

R.P. and R.K. conceived the project. R.P. implemented the code and wrote the manuscript draft. R.K. finalized the manuscript text. A.H. validated the code and contributed to writing the manuscript.

Funding

This work was supported by the Austrian Science Fund (FWF) grants P35093 and P34965 to RK.

Data availability

GenomeDelta is open source and freely available at https://github.com/rpianezza/GenomeDelta. A detailed manual with installation instructions as well as a walkthrough are available. The source code of GenomeDelta as well its validation are also available in Zenodo [64].

Declarations

Ethics approval and consent to participate

Ethics approval is not applicable for this study.

Competing interests

The authors declare no competing interests. Robert Kofler was a Guest Editor for this article collection but was not involved in the editorial process of this manuscript.