LTRtype, an Efficient Tool to Characterize Structurally Complex LTR Retrotransposons and Nested Insertions on Genomes.

The amplification and recombination of long terminal repeat (LTR) retrotransposons have proven to determine the size, organization, function, and evolution of most host genomes, especially very large plant genomes. However, the limitation of tools for an efficient discovery of structural complexity of LTR retrotransposons and the nested insertions is a great challenge to confront ever-growing amount of genomic sequences for many organisms. Here we developed a novel software, called as LTRtype, to characterize different types of structurally complex LTR retrotransposon elements as well as nested events. This system is capable of rapidly scanning large-scale genomic sequences and appropriately characterizing the five complex types of LTR retrotransposon elements. After testing on the Arabidopsis thaliana genome, we found that this program is able to properly annotate a large number of structurally complex elements as well as the nested insertions. Thus, LTRtype can be employed as an automatic and efficient tool that will help to reconstruct the evolutionary history of LTR retrotransposons and better understand the evolution of host genomes. LTRtype is publicly available at: http://www.plantkingdomgdb.com/LTRtype/index.html.

Introduction

It has long been recognized that transposable elements compose an important fraction of most eukaryote genomes. Transposable elements are usually classified into three groups by transposition mechanisms, known as LTR retrotransposons, non-LTR retrotransposons, and DNA transposons. Among them, LTR retrotransposons are particularly prevalent in most plant genomes, where they appear to be the major determinant of the tremendous variation in genome size (Wessler, 2006; Wicker et al., 2007; Wendel et al., 2016). They have been found to serve as a major contributor to large plant genomes (Flavell, 1986; SanMiguel et al., 1998; Kumar and Bennetzen, 1999; Meyers et al., 2001; Nystedt et al., 2013). LTR retrotransposons are a class of mobile genetic elements containing two identical or similar long terminal repeats (LTRs) and one internal region (IN) between them, which are transposed through the reverse transcription of an RNA template via “copy-and-paste” in the genome. For decades it has been recognized that evolutionary dynamics of these elements actively act as a motivating force to drive the genome evolution. LTR retrotransposons can largely make contributions to the variation of genome size through the amplification and recombination, and likewise bring about genomic structural variation and organization. For instance, LTR retrotransposons determine the growth of genome size through retrotransposon amplification (SanMiguel et al., 1998; Bennetzen, 2002; Kellogg and Bennetzen, 2004), and they also cause genome size reduction through unequal homologous recombination and illegitimate recombination (Devos et al., 2002; Vitte and Panaud, 2003; Ma et al., 2004; Bennetzen et al., 2005). Taking the Arabidopsis thaliana genome for example, rapid loss was meanwhile revealed to counteract genome expansion through recombination despite a large number of recent amplification of LTR retrotransposons (Devos et al., 2002).

Extraordinary quantities of TEs, particularly the LTR retrotransposons, have greatly hampered the genome assembly and annotation. Thus far, a multitude of computational tools have been developed for the detection of LTR retrotransposons in the rapidly emerging genomic sequences. Common structural features of LTR retrotransposon elements make it possible to de novo detect novel LTR retrotransposon families having low sequence homology to known queries or families with a typical structure. With this regard, several programs were specifically designed for the ab initio computer discovery of LTR retrotransposons. LTR_STRUC (McCarthy and McDonald, 2003) is the best known of these programs, which has broadly been applied to numerous genomes, such as the Glycine max (Schmutz et al., 2010), Mus musculus (McCarthy and McDonald, 2004), Oryza sativa (Tian et al., 2009), and Pan troglodytes (Polavarapu et al., 2006). Furthermore, the other programs, including LTR_par (Kalyanaraman and Aluru, 2006), LTR_Rho (Rho et al., 2007), LTR_FINDER (Xu and Wang, 2007), LTR_harvest (Ellinghaus et al., 2008), and LTRdigest (Steinbiss et al., 2009), were developed for the de novo prediction of LTRs, and these programs consider further features of LTR retrotransposons in post processing steps to enhance the quality or sensitivity of the predictions. Most extensively implemented approaches of LTR retrotransposon identification were based on similarity searches against a target genome. These tools, such as REPuter (Kurtz et al., 2001), RECON (Bao and Eddy, 2002), RAP (Campagna et al., 2005), PILER (Edgar and Myers, 2005), RepeatMasker[1] and LTR Annotator (You et al., 2015), are able to detect repeat sequences in the genome, but they incorporated almost no defragmentation and have definitely resulted in the overestimated number of LTR retrotransposons. Limitations of such sort of approaches are apparent, since they are able to find elements in the database without difficulty but they may scarcely predict repeat elements with distantly divergent sequences. All the above-mentioned annotation tools may limitedly apply to detecting full-length elements (LTR-IN-LTR) and their fragmented elements, but they are unable to reconstruct structurally complex elements as well as their nested cluster.

Although TE-nest (Kronmiller and Wise, 2008, 2013) and REannotate (Pereira, 2008) can identify nested events of transposable elements, they fail to adequately locate the structurally complex LTR retrotransposons. Until now, little has been known regarding genome-wide patterns of structurally complex retrotransposons and their contributions to the modification of genomes. The difficulty increases without a doubt to efficiently annotate such elements due to the recombination, DNA losses and nested events in a genome. Single Molecule Real Time (SMRT) Sequencing technologies are able to generate long reads that are essential to complete plant reference genomes that will provide unprecedented opportunities to obtain in-depth knowledge on evolutionary behaviors of retrotransposons and explore their contributions to the genome structure, function and evolution on the strength of large-scale genome information. Thus, it is urgently needed to develop efficient tools to genome-wide characterize various types of structurally complex retrotransposon elements and better annotate genomes in rapidly deposited large-scale genomic sequences.

Here we developed a novel software program, named LTRtype, for the purpose of proficient discovery of diverse types of the structurally complex LTR retrotransposons (Figure ) and nested insertions (Figure ) in large quantities of genomic sequences. In addition to an effective solution of fragmented retrotransposon sequences from BLAST searches, this program has incorporated a combination of rapid algorithms, which may not limit to LTR pairs but indeed is able to identify retrotransposon elements with three or more LTRs. These elements include: (1) normal full-length elements (LTR-IN-LTR); (2) solo-LTR elements; (3) complex elements with three or more LTRs (e.g., LTR-IN-LTR-IN-LTR); (4) fragmented elements of the above-mentioned structural types; and (5) nested insertions among these different LTR retrotransposon elements. The program LTRtype reported here is able to proficiently mine increasingly sequenced genomes by using multithreading technologies, and thus provides a convenient and friendly service for users.

Structurally diverse types of long terminal repeat (LTR) retrotransposons identified in the genome. Triangles in blue and yellow indicate the two different target site duplications (TSDs), rectangles in brown represent LTRs, dotted line rectangles represent fragmented LTRs, black thick lines denote internal regions, and arrows show the direction of DNA strand.

Typical nested structural types of LTR retrotransposons in the genome. (A) normal full-length elements nested by solo-LTRs, normal full-length elements, or complex elements with three LTRs; (B) solo-LTR elements nested by solo-LTRs, normal full-length elements, or complex elements with three LTRs; (C) complex elements with three LTRs nested by solo-LTRs, normal full-length elements, or complex elements with three LTRs. Triangles in blue and yellow indicate the two different TSDs, rectangles in brown represent LTRs, dotted line rectangles represent fragmented LTRs, black thick lines denote internal regions, and arrows show the direction of DNA strand.

Materials and Methods

Flowchart of LTRtype

The first step of LTRtype in the flowchart (as shown in Figure ) is to collect all full-length LTR retrotransposons and construct library files in FASTA format. There are two major methods to retrieve the full-length LTR retrotransposons. One is to straightforwardly download from the known repeat databases, e.g., Repbase: http://www.girinst.org/; PlantGDB: http://www.plantgdb.org; Plant repeats: http://www.tigr.org/tdb/e2k1/plant.repeats/, while the other is to mine target genomes by using authorized tools of LTR retrotransposons, such as LTR_STRUC (McCarthy and McDonald, 2003), LTR_FINDER (Xu and Wang, 2007), LTR_harvest (Ellinghaus et al., 2008), and so on. Then, the library will be merged by their overlapping lengths and sequence similarities.

Flowchart of the characterization of LTR retrotransposon elements in the program LTRtype.

The second step is to align the LTR and IN libraries with target genomes using RepeatMasker and preprocess the aligned fragments by eliminating the overlapped elements.

The third step is to merge adjacent fragments into longer or more intact LTR or IN sequences that belong to the same family, as RepeatMasker usually align a complete LTR or IN fragment into several fragments that may be separately or partially overlapped. The yielded datasets become candidate fragments for further analyses.

The fourth step is to link candidate fragments by discriminating trees according to their physical positions. If something can be connected one another, then we construct the linked fragments and take else as inserted elements or unconnected fragments. Note that inserted elements should be linked again until nothing could be connected.

The last step is to create a linked list and categorize different structural types of LTR retrotransposon elements.

Auto-Link by Discriminating Tree

Discriminating tree is a method to discriminate whether the aligned fragments should be connected that belong to the same LTR retrotransposon element in the genome. The discriminating tree is classified into the four branches: LTR-LTR, IN-IN, LTR-IN, and IN-LTR (Figure ). There are three output states of the discriminating tree, that is, 1 is linked, 0 is not linked, and 2 stands for inserted. Taking the linked IN-IN for example (Figure ), the three separated fragments are aligned to adjacent coordinates in the rice genome. The fragment a with coordinates of 1,001–1,500 in the genome corresponds to coordinates of 1–500 in the Osr1_IN, the fragment b with coordinates of 1,501–1,800 corresponds to coordinates of 1–300 in the Osr2_LTR, and the fragment c with coordinates of 1,801–2,300 corresponds to coordinates of 501–1,000 in the Osr1_IN. Here, the known length of Osr1_IN is 1,000 bp. Then the state of link (a, b) = 2, and link (a, c) = 1, that is, the fragments a and c is connected as a single element, and fragment b is regarded as an inserted element.

Auto-link by discriminating tree. (A) Discriminating tree; (B) Link the aligned fragments by taking the linked IN-IN for example. Triangles in yellow indicate TSDs, rectangles in brown represent LTRs, and black thick lines denote internal regions.

Create the Dynamic Link List and Identify Different Structural Elements

It is feasible to create a series of dynamic link lists according to the link states of the aligned fragments. Here, a dynamic link list stands for one linked block domain in the genome, and constructs one structural type of LTR retrotransposon element. There are seven adjacent aligned fragments in the genome, and these link states are individually given as follows: link (a, b) = 1, link (b, c) = 2, link (b, d) = 2, link (b, e) = 1, link (c, d) = 1, link (d, e) = 0, link (e, f) = 1, and link (f, g) = 0. Then, they may construct three dynamic link lists: (a-b-e-f), (c-d), (g), and form three LTR retrotransposon elements (Figure ).

The creation of the dynamic linked list. a, b, c, d, e, f, and g Show the aligned fragments of LTR retrotransposons.

Comparisons with LTRtype, REannotate, and TEnest

We downloaded the sequences of maize LTR retrotransposons from Repbase version 18.11 (Jurka et al., 2005), and built the library file (Supplementary Library S1), including a total of 302 full-length elements; they were further subdivided into two sequence libraries containing 302 pairs of LTRs and INs, respectively. Maize genome sequences of Chromosome 10 (gb: CM000786.2) were downloaded from NCBI. LTR retrotransposons of maize were mined by running LTRtype (default), REannotate (-n -t -d = 10k -s = 10k -c), and TEnest (default), respectively. Because default parameters are loose, we adjusted them when running REannotate.

Mining and Analyzing LTR Retrotransposons in the A. thaliana Genome

The sequences of A. thaliana LTR retrotransposons were downloaded from Repbase version 18.11 to build the LTR retrotransposon library files including a total of 146 full-length elements (Supplementary Library S2). This library file thus contained 146 corresponding pairs of LTRs and IN, respectively. The A. thaliana genome sequences (TAIR10) were downloaded from http://www.arabidopsis.org/. We identified different types of structurally complex LTR retrotransposons and characterized their nested insertions LTR retrotransposon on the A. thaliana genome by running LTRtype (default).

Nucleotide sequence divergence among pairs of intra-element LTRs was used as a molecular clock, as they are identical at the time of insertion. The ages of full-length LTR retrotransposons were then determined by comparing their 5′ and 3′ LTRs (SanMiguel et al., 1998). MEGA5 (Tamura et al., 2011) was employed to calculate the number of transition and transversion mutations. Insertion dates were estimated using the Kimura two-parameter method (Kimura, 1980). The average mutation rate of 1.3 × 10-8 substitutions per synonymous site per year (Ma and Bennetzen, 2004) and 7 × 10-9 substitutions per synonymous site per year (Ossowski et al., 2010) were applied to estimate insertion times of the LTR retrotransposons in the A. thaliana genome. The time (T) since element insertion was estimated using the formula: T = K/2r.

Results and Discussion

Hardware Requirements

The hardware requirements vary with genome sizes. Both Intel and AMD x64 architectures are supported. Note that there are no specific requirements including CPU, memory and disk space to run the LTRtype. Taking rice genome (∼400 Mb) for an example, it is needed to have ∼500 Mb RAM, 4+ cores, and ∼1 Gb disk space.

User Input

LTRtype provides some default parameter settings to make users convenient, but the four parameters (-P,-p -i, and -d) must be set by users themselves. -P follows the path of RepeatMasker; -p follows the path of blast searches; -i follows the LTR retrotransposon library file (FASTA format). Reference LTR names may not have the gap and ‘| ’, but they must be suffixed with the string ‘_LTR.’ Likewise, reference IN names must be suffixed with the string ‘_IN’; -d follows the genome sequence file (FASTA format). In order to facilitate the view output, the sequence names do not contain the gap and ‘|’.

Usage: perl LTRtype.pl [options]

Options:

–P