Modular assembly of transposable element arrays by microsatellite targeting in the guayule and rice genomes

Background Guayule (Parthenium argentatum A. Gray) is a rubber-producing desert shrub native to Mexico and the United States. Guayule represents an alternative to Hevea brasiliensis as a source for commercial natural rubber. The efficient application of modern molecular/genetic tools to guayule improvement requires characterization of its genome. Results The 1.6 Gb guayule genome was sequenced, assembled and annotated. The final 1.5 Gb assembly, while fragmented (N50 = 22 kb), maps > 95% of the shotgun reads and is essentially complete. Approximately 40,000 transcribed, protein encoding genes were annotated on the assembly. Further characterization of this genome revealed 15 families of small, microsatellite-associated, transposable elements (TEs) with unexpected chromosomal distribution profiles. These SaTar (Satellite Targeted) elements, which are non-autonomous Mu-like elements (MULEs), were frequently observed in multimeric linear arrays of unrelated individual elements within which no individual element is interrupted by another. This uniformly non-nested TE multimer architecture has not been previously described in either eukaryotic or prokaryotic genomes. Five families of similarly distributed non-autonomous MULEs (microsatellite associated, modularly assembled) were characterized in the rice genome. Families of TEs with similar structures and distribution profiles were identified in sorghum and citrus. Conclusion The sequencing and assembly of the guayule genome provides a foundation for application of current crop improvement technologies to this plant. In addition, characterization of this genome revealed SaTar elements with distribution profiles unique among TEs. Satar targeting appears based on an alternative MULE recombination mechanism with the potential to impact gene evolution. Electronic supplementary material The online version of this article (10.1186/s12864-018-4653-6) contains supplementary material, which is available to authorized users.


Background
Guayule, a species in the Compositae (Asteraceae) family, is a perennial shrub native to the Chihuahuan Desert of North America that represents a potential commercial source of natural rubber [1]. There have been the numerous unsuccessful past efforts to develop guayule as a crop [2], however, both a narrow germplasm pool and the innate genetic complexity of guayule limited past breeding improvement efforts [1]. The importance of guayule, and other plants, as alternatives to Hevea brasiliensis as natural rubber sources can be appreciated when the economics and supply of Hevea rubber are considered. Seventy percent of worldwide rubber production (12.3 million metric tons in 2015) is utilized in tire fabrication [3], and natural rubber represents one of the largest corporate purchases made by this industry. The natural rubber supply is subject to both considerable price volatility and a number of issues associated with security of supply. For example, the Hevea tree is susceptible to South American Leaf Blight (SALB) caused by the endemic fungus Microcyclus ulei. SALB terminated commercial Hevea cultivation in South America in the early 1900's and remains a major threat to rubber production [4]. Thus, identification and development of alternative sources of natural rubber are important to the tire industry. The sequencing and annotation of the 1.6 Gb [5,6] nuclear genome of a diploid guayule was undertaken to facilitate the application of current molecular and breeding tools to guayule improvement.
The genome annotation process, which follows sequence acquisition and assembly, involves identification of encoded gene and structural components, essentially converting a compiled DNA sequence into an informationrich tool with broad applicability. The identification/ characterization of repeat sequences is a requisite first step for efficient and accurate genome annotation [7]. In most chromosomal domains, the repeated DNA sequences represent largely transposable elements (TEs) of two classes. The class I retrotransposons are mobilized through an RNA intermediate and make up large percentages of plant genomes [8]. The class II transposons mobilize through a DNA intermediates, and exist both in autonomous (TE encodes requisite transposase proteins) and non-autonomous (transposase function supplied in trans) forms [9]. Both TE types have non-random distribution profiles on plant chromosomes, with the class II TEs associated with gene-rich chromosomal regions [9]. The Mu-like element (MULE) superfamily of class II TEs represents a large and diverse set of autonomous and non-autonomous elements that can make up a significant percentage of plant genomes [10,11]. The MULE TEs, particularly the Pack-MULEs, have been proposed as important mediators of plant gene evolution [12][13][14][15].
In the course of characterizing repeated sequences in initial guayule genome assemblies, a number of short, unrelated repetitive DNA sequences associated with microsatellites were identified. Microsatellites, or simple sequence repeats (SSRs), are regions of tandemly repeated short (1-6 bp) DNA repeats that are common features in genomes [16]. The guayule microsatelliteassociated elements were identified as microsatellitetargeted non-autonomous MULE elements by both structural and sequence similarities to guayule autonomous MULEs. While specific microsatellite-associated class II TEs [17][18][19], including autonomous MULE elements [20,21], have been identified in plants, the elements in guayule had unique and unexpected features.

Plant material
All genomic and transcriptome sequences were derived from a single diploid guayule plant, accession W6 429 developed from a selection obtained in a 1942 Durango, Mexico collection expedition [1,5,6].
Illumina (300 bp paired-end) shotgun sequencing libraries were made using the Kapa Biosystems protocol: High-Throughput NGS LibraryPreparation Technical Guide for Illumina TM platforms (KR0427 -V1.12) and sequenced with MiSeq ® Reagent Kit v3 Reagents on an Illumina MiSeq. The Roche 454 sequencing libraries were prepared using the Rapid Library Preparation Manual for GSFLX+ and GS Junior + Series (May 2011) and sequenced with Roche 454 GSFLX +. The Illumina matepair libraries with insert sizes of 500, 700 and 1 kb were made following the Nextera® Mate Pair Library Illumina HiSeq 2500.

Meraculous and CLC Genomics Workbench genome assemblies
Meraculous is a whole genome assembler for next generation sequencing data geared for large genomes [23]. The guayule genome was first assembled using the Mercaculous assembler with the Illumina reads (shotgun and Nextera). This assembly encompassed 1.5 Gb (N 50 = 22Kb, 260 k Scaffolds) and mapped > 95% of the shotgun reads. CLC Genomics Workbench assembly (version 8.5.1, in conjunction with scaffolding using SSPACE version 2.0 [24]) employed the Illumina and Roche 454 reads and provided improved representation of simple sequence repeat domains. This assembly encompassed 0.9 GB (N 50 = 28Kb, 59 k Scaffolds) and retained > 95% of the annotated genes.

Computational de novo identification of SaTar elements
Software designed to identify potential Satellite Targeted (SaTar) elements in any genome was employed. Entered scaffolds are scanned for repeated sequences (250-800 bp in length, ≥ 20% GC content) defined by TA microsatellite domains (≥ 12 bp in length). The program utilizes MISA to identify microsatellite domains (http://pgrc.ipkgatersleben.de/misa/misa.html). Software and Users Guide available at: http://probes.pw.usda.gov/Guayule.

Sequencing, assembly and annotation of the guayule genome
The genes annotated from a single artificial 16 Mb assembly (approximately 1% of the genome) composed of the 33 longest Meraculous scaffolds is shown in Fig. 1a. The locations of the 505 annotated genes are indicated, a gene density consistent with predictions based on the overall genome ( Table 1). The observed non-genic domains in this assembly are composed largely of retrotransposon and unclassified interspersed repetitive sequences (Table 1). However, no evidence for recent, widespread, amplification of specific Class I or Class II transposon/ retrotransposon families was found in either the guayule assemblies or the collected reads. Syntenic analysis of the genes encoded on individual scaffolds was consistent with two rounds of genome-wide duplication in the last 40 MY [29].
This Meraculous assembly allows characterization of particular genes of interest for guayule improvement, including those involved in rubber biosynthesis. For example, Fig. 2 shows the structures of genes encoding proteins associated with a selected portion on the mevalonate sourced rubber biosynthetic pathway. In general the Meraculous assembly contains the expected isoforms/structures of these genes [30,31], and encode mRNAs specifically associated with guayule rubber biosynthesis (SRPP-1, Fig. 2) [32].

Structure of guayule gSaTars
In the course of characterizing the repetitive DNA content in the guayule genomic scaffolds, in particular characterization of non-autonomous Class II TEs, an unexpected pattern emerged. When scaffolds were probed with libraries of small (≤1.7 kbp) terminal inverted repeat elements, clusters of the elements were frequently observed. These clusters had two unexpected properties. First, they were largely associated with microsatellite domains such that individual elements within the cluster are flanked by simple sequence repeats, often (TA) n . Second, within the clusters individual elements were modularly assembled, in no case did one microsatellite-flanked element interrupt another. This architecture suggested that the observed modular assembly resulted from the specific targeting of these sequences to microsatellite domains [17,18,20,21]. For this reason, these elements are referred to as SaTars, or Satellite Targeted, transposable elements.
The guayule SaTars, or gSaTars, described here range in size from approximately 400 to 1700 bp, depending on the individual family ( Table 2, Additional file 1). As shown in Fig. 3, they have several structural properties common to non-autonomous class II TEs. The gSaTars are of appropriate size and are defined by imperfect terminal inverted repeat (TIR) domains ( Table 2). Eleven of the fifteen gSaTar families contain individual members not localized to microsatellites and flanked by 10 bp target site duplications (TSDs) (Fig. 3, Table 2). In contrast to standard non-autonomous Class II TEs however, a subset of the gSaTars appear preferentially targeted to microsatellite domains ( Table 2). Probing (BLAST) the genome with either form (TSD or microsatellite targeted) While the gSaTar elements predominantly localized to (TA) n microsatellite domains, in guayule the type and complexity of the microsatellite domains flanking gSaTar elements are highly variable (Fig. 4, for example). Among the gSaTar families, at least 78% of the elements are flanked by either microsatellite domains or TSDs ( Table 2). The remaining elements, i.e. those with neither architecture, appear to represent loci in which terminal and/or flanking sequences have been altered by local deletions and/or rearrangements.
In Fig. 4, individual members of four gSaTar families flanked by either TSDs or microsatellite domains are aligned. In cases where the gSaTar elements are flanked by identical microsatellite domains, both structures are consistent with derivation from a straightforward "cut and paste" transpositional mechanism resulting in a 10 bp TSDs. However, as indicated in Fig. 4, the microsatellite sequences immediately flanking the elements are often inconsistent with this mechanism.
The relative frequency of the different gSaTar families in the genome are highly variable, and can be estimated   by analysis of their distribution on the 16 Mb (1% of genome) artificial assembly (Fig. 1a). This assembly contains 215 gSaTar elements, 32% of which are linked to other gSaTars (Additional file 2) indicating the presence of over 20,000 total elements constituting approximately 1% of the guayule genome.

Genomic distribution of gSaTar elements
The distribution of gSaTar elements along the guayule scaffolds is strikingly different from MITEs and other non-autonomous Class II TEs. MITEs are associated with gene-rich domains in the genome [33,34]. Specific targeting of MITE insertion events to elements of the same, and different, MITE families results in clusters of these elements [35]. The gSaTar elements are also found in clusters on the guayule scaffolds, but with frequency and architecture strikingly different from described in MITEs. In terms of frequency, the gSaTar elements are localized to clusters more frequently than observed with the most commonly associated rice MITE elements ( Table 2) [35]. And in contrast to the rice-MITE clusters where insertion of MITEs into existing elements is more commonly observed than adjacent insertion [35], the gSaTar clusters are always assembled modularly, with intact individual elements linked via microsatellite domains. Examples of gSaTar clusters are shown in Fig. 5 (selected cluster domain sequences in Additional file 3). In these clusters, elements from the different families are assembled by adjacent insertions flanked by microsatellites. Within the guayule gSaTar clusters internal deletions are observed, some of which remove the intervening satellite sequence in addition to gSaTar element ends resulting fusion of elements within the cluster ( Fig. 5h and i, for example). However, in cataloging over 300 gSaTar clusters, not a single case of one gSaTar element interrupting another has been observed, supporting a mobilization mechanism distinct from that involved in the expected non-autonomous Class II TE transposition. Analysis of one of the gSaTar clusters (Scaffold23207) revealed a large, low copy, TA-satellite-linked, repetitive DNA at one end (Additional file 4a). Further characterization indicated that this sequence represents a terminal fragment of a larger defective autonomous gSaTar-MULE1 (AgS-MULE1) Mu-like transposon [10], an apparently full length copy of which occurs on scaffold15495 (Additional file 4b). This 5.3 kb MULE element is defined by 400 bp TIRs. A BLAST search of the guayule CLC assembly with the autonomous MULE TIRs returned 5 hits, and in all cases the terminal TIRs were flanked by microsatellite domains. Five additional Autonomous gSaTar-MULEs have been identified in the guayule assemblies (Additional file 1). As indicated (Additional file 5), the TIRs defining four of these families are similar to those defining previously described gSaTar families (Additional file 1). Both the definition of gSaTar elements by long TIRs similar to the AgS-MULE elements described above, and the identification of individual members of eleven of the fifteen families flanked by 10 bp TSDs in complex sequence rather than microsatellite domains [36], indicate that the gSaTar elements described here represent non-autonomous MULEs that have been specifically targeted to microsatellite domains.

SaTar elements in the rice genome
In order to determine if the SaTar mobilization pathway is a general feature of plant genomes, a specific search algorithm for identification of repetitive DNA sequences flanked by microsatellite domains was developed (Methods section). The complete genomic assemblies of rice (Oryza sativa (japonica) v7_JGI, [37,38]) was evaluated employing this algorithm. Five families of rice SaTars (rSaTars) were returned by the algorithm (Table 3). While the SaTar elements are considerably less frequent in rice as compared to the guayule genome (Table 3, Fig. 1b), the rSaTar and gSatar elements contain a number of common features. Similar to the gSaTars, the rSaTar families range in size from 380 to 1900 bp. Two resources were employed in the classification of these elements, the RiTE rice transposable element database [36] and a comprehensive listing of rice MULE elements [39]. Four of these families were identified as non-autonomous MULE or Mutator-MITE elements, the most abundant rice TE type (Additional file 6). By comparison to previously mapped rice MULE elements [39], rSaTar1, 3 and 5 were further characterized as non-Pack-MULEs. While the diverse MULE element families in the rice genome are largely flanked by 9 bp TSDs [39], only the rSaTars in Table 2 were flagged as microsatellite associated by the search algorithm. The final element, rSaTar2, was previously identified as small MITE-like mobile elements targeted to (TA) n microsatellites (Micron) in rice [17]. However, as shown in Additional file 7, the TIR domains of the rSaTar2/Micron elements are highly similar to the TIRs of an autonomous Mu-like transposon found on the chromosome one of the Oryza sativa (japonica) genome, indicating that this element is also a non-autonomous rice MULE.
These rSaTar elements share common features with the guayule gSaTars described above (Table 2). First, the rSaTar1, 2 and 4 families are defined by TIRs. Second, the elements were frequently associated with microsatellite domains (Table 3). Similar to the four of the gSaTar families, the rSaTar1, rSaTar2, rSaTar4 and rSaTar5 families had no members flanked by TSDs. In contrast, among the rSaTar3 elements, this structure was dominant. In addition, the rSaTar elements were frequently clustered (Fig. 6, Additional file 8). For example, of the 36 rSaTar1 elements in the rice genome, 6 are linked to rSaTar2 elements via microsatellite domains (Fig. 6). Of the rSaTar2 elements, 27 are linked to rSaTar4 elements (Additional file 8). Finally, in none of the clusters, containing either a single or multiple rSaTar types, does one rSaTar element interrupt another.
In addition to these non-autonomous rSaTar elements, four defective autonomous MULEs which appear to have been mobilized as rSaTars (Autonomous rSaTar-MULEs, ArS-MULEs) have been identified in the rice genome (Additional file 9). Again similar to the guayule architecture, other rSaTar elments are found in the satellite domains flanking these autonomous TEs (Additional files 9 and 10). Of the 12 ArS-MULE1 elements, 6 are linked to rSaTar2 elements.
In addition to the Oryza sativa ssp. japonica assembly, the availability of a draft genome from the second cultivated subspecies or varietal group, Oryza sativa ssp. indica (assembly ASM465v1 [40]) facilitates identification of potential rSaTar element insertion sites. These groups diverged relatively recently (approximately 0.44 MYA [41]). Of the 51 rSaTar2 sites on japonica chromosome 1, 9 are conserved on the indica assembly (Additional file 11). In another 20 cases, however, these japonica insertion sites are represented by unoccupied TA microsatellite domains of variable length in the indica assembly (Fig. 7, Additional file 12). The remainder of the japonica rSaTar2 insertion domains are either deleted or unassembled in the indica genome. A similar relative rSaTar3 insertion into an empty satellite flanking an rSaTar1 element on japonica chromosome 3 is shown in Fig. 8. The presence of both conserved and non-conserved rSaTar2 loci in japonica and indica indicates that the SaTar targeting system was active before and after the divergence of these two subspecies/varietal groups.
Three families of sorghum SaTars (sSaTars) were identified employing the algorithm described above (Additional file 13). All three sSaTar elements are small (233-373 bp) MITE-like TIR elements. Similar to rSa-Tar1and 2, no sSaTar-like elements defined by TSDs were found in the sorghum genome. As indicated in Additional file 13, there are a total of 1261 of the sSaTar elements on the sorghum genome, leading to on approximately one element every 600 kb (1261 elements on 730 mb). However, the clustering of these elements is  Table 1). The algorithm employed to identify SaTar elements in rice and sorghum failed to return elements from the citrus genome (Cclementina_182_v1) [43]. However, a manual search of sequences flanking TA satellite domains resulted in identification of three families of small citrus SaTars (cSaTars) (Additional file 15). The cSaTar elements are similar to the guayule, rice and sorghum elements of (definition by TIRs, clustering on chromosomes, modular assembly). In addition, while the majority of these elements are flanked by microsatellite domains, each family contains members flanked by 9 or 10 bp TSDs (Additional file 15). However, the cSaTar elements appear relatively ancient as compared to the gSaTar and rSaTar elements. Examples of cSaTar clusters are shown in Additional file 16. Finally, in no cases did the sSaTar or the cSaTar elements within clusters interrupt each other.

Discussion
The successful development of guayule as a commercial source of natural rubber is dependent upon improvement of both its agronomic properties and rubber yield.
To this end, the sequencing, assembly and annotation of the guayule genome was undertaken to promote the application of current molecular and breeding tools improvement efforts. While fragmented (260 k Scaffolds), the Meraculous assembly reported here is essentially complete allowing, for identification and utilization of specific genes and gene components, as well as contributing to the development of molecular breeding tools (such as genotyping-by-sequencing [5]).

Guayule gSaTar elements are MULE TEs
The gSaTar elements were identified in the guayule genome as families of microsatellite-defined, non-autonomous MULE elements with a unique and unexpected chromosomal distribution profile. The MULE designation is based upon structural features of the gSaTar elements, characterization of insertion sites and TIR sequence similarity to autonomous guayule MULE transposons.
The gSaTar elements are similar to MULE elements in terms of size and structure (defined by TIRs) (Fig. 3, Table 2) [10]. Previously described non-autonomous MULE elements are flanked by 8-11 bp TSDs generated during the "cut and paste" transpositional process [15]. Most of the non-autonomous gSatar families described here (Table 2) have individual members flanked by 10 bp TSDs, indicating that these elements are capable of mobilization through a similar pathway. The designation of non-autonomous gSaTar elements as MULEs is also supported by TIR sequence similarity to autonomous, microsatellite-targeted, AgS-MULE elements that are similarly targeted to microsatellite domains (Additional files 1, 4 and 5). This sequence similarity indicates that the non-autonomous gSaTars are mobilized by AgS-MULE transposase activities supplied in trans.
However, previously described autonomous and nonautonomous MULE elements do not display the sequencespecific (microsatellite domain) targeting exhibited by the gSaTars and AgS-MULEs. In many cases examination of the structures flanking microsatellite targeted gSaTar elements reveals a lack TSDs, these insertion sites are thus inconsistent with insertion via the conventional MULE "cut and paste" mechanism (Fig. 4). The potential utilization of two distinct insertion mechanisms within gSaTar families suggests the possibility element subtypes mobilized via independent pathways. However, probing (BLAST) the genome with either form (TSD or microsatellite targeted) returns equally elements with both types of insertion events, indicating that the two distinct insertion profiles do not appear to be associated with specific gSaTar subtypes. Rather, the gSaTar elements appear to be non-autonomous MULE elements that interact with, or are substrates of, an ancillary system that results in targeted insertion of the mobilized element into microsatellite domains.
The microsatellite targeting of gSaTar and AgS-MULE elements commonly results in clustering into mixedfamily multimers, within which individual elements are present in ordered, continuous arrays of uninterrupted elements. This structure feature was validated through PCR amplification of genomic regions containing gSaTar elements in guayule (Additional file 17). The expected nested architecture within these TE multimers, with older component elements interrupted by those more recently inserted [35,44], was not observed in any of the gSatar-or AgS-MULE-containing clusters.
Given approximately 20,000 total members of the gSa-Tar families described here, a random distribution would result in approximately 1 element every 80 kb. However, the observed distribution is strikingly non-random. Approximately 30% of the gSaTar elements from these families are linked directly to another gSaTar element. This value (30%) represents a minimum level of association as our identification of gSaTar families is incomplete, and in many cases these elements are located on the scaffold ends where potential association would not be detected (gSaTar elements are present at approximately 12% of scaffold ends). Given random insertion, the four linked elements shown in Additional file 3c would be expected to be distributed over approximately 320 kb, not the observed 2.3 kb.
Both the structure and distribution of the gSaTar and AgS-MULE elements suggest that while they share features with MULEs and Class II TEs in general, and can in cases be mobilized similarly, gSaTar transposition involves distinct and/or additional mobilization components. As such, they appear to serve as substrates in a novel mechanism that results in specific targeting to microsatellite domains and modular assembly of diverse elements at specific chromosomal loci.

SaTar elements in other plant genomes
To evaluate whether similar SaTars existed in other plant species, rice (Oryza sativa Japonica) was selected because of its well defined genome [37], the abundance and diversity of MULE elements [39], and the availability of genomic sequence from Oryza sativa Indica [40], a closely related subspecies/varietal group. The rice genome was analyzed employing a specific SaTar search algorithm (repetitive DNA sequences flanked by (TA) n microsatellite domains) and five rSaTar element families were returned (Table 3). The rSaTar elements were similar in size, structure and genomic distribution to the elements from guayule (Fig. 6, Table 3, Additional file 8). These elements could be classified as non-autonomous MULEs by sequence comparison (BLAST) to previously characterized TEs [36,39], or by identification of an autonomous microsatellite targeted MULE with similar TIRs (Additional file 7). In addition, ancient autonomous, microsatellite-associated rice ArS-MULE elements present in rSaTar clusters were identified (Additional file 9).
Finally, SaTar families in sorghum (sSaTars, Additional files 13 and 14) and citrus (cSaTars, Additional files 15 and 16) were identified, suggesting that microsatellitetargeting dependent MULE stacking is a general feature of plant genomes. The relative similarity of SaTar family members varies considerably in the different species, for example the individual members of the cSaTar families have much lower similarity than observed among the gSaTars, indicating that SaTar targeting is evolutionarily intermittent.

SaTar targeting
The identification of SaTar elements in diverse plant genomic backgrounds offers the potential for an improved characterization of mobilization mechanisms. As discussed above, MULEs, as Class II TEs, are generally mobilized by a transposase-directed "cut and paste" mechanism [45] that results in TSDs, the size of which is diagnostic of the transposons superfamily. As indicated in Fig. 2, individual members of gSatar families not targeted to microsatellites are flanked by TSDs expected from Mu-superfamily elements (8-11 bp TSDs in low-copy DNA), and are not assembled into arrays of unrelated elements joined end to end. The same is true of the non-targeted rSaTar3 elements (Table 3). In contrast, the sequences flanking many of the microsatellitetargeted gSaTar elements are inconsistent with the "cut and paste" mechanism ( Fig. 4) suggesting involvement of a transpositional component distinct from other MULEs.
Given the general instability of microsatellite sequences [16], more recent SaTar insertion events would be expected to most accurately retain architectural features generated by the transposition. The TAFTA transposons in maize [20] represent the most recent events of MULE SaTar-like insertions described. In these events, empty target sites consisted of short TA microsatellite domains of (6-8 bp) which are expanded following insertion to up to (TA) 50 on each side of the inserted TAFTA element [20]. While similar apparent microsatellite expansions are easily found on comparison of relative rSaTar2 insertions in Japonica and Indica (Figs. 7 and 8, Additional file 12), the innate instability of the microsatellite domains, as well as inability to clearly differentiate rSaTar2 relative insertions and deletions, limits the utility of this data. In maize, a TA site-specific transposase was proposed to establish targeting, with expansion of the flanking microsatellite via DNA polymerase stuttering during repair of single stranded gaps [20]. However, in guayule (and other plant genomes, including rice) the SaTar-flanking microsatellite domains are often composed of diverse, non-TA, microsatellites (Fig. 4), complicating this targeting model. Rather, the targeting of SaTars to a diverse set of different microsatellite repeats suggests involvement of homologous recombination in SaTar insertion, not involved with insertion of non-SaTar MULE TEs. A separate model for the mechanism of SaTar targeting, and thus modular stacking of these TEs, would involve generation of closed circular SaTar intermediates [46] with microsatellite domains added as filler sequences between the element termini, followed by homologous recombination into an existing genomic microsatellite target domain. While this model is speculative in nature, the involvement of circular intermediates in MULE transposition has been suggested [12], and circular autonomous elements which include filler sequence have been described [47].

SaTars and gene evolution
While the actual importance of the MULE elements in the evolution of gene and genome architecture remains a topic of investigation [9], both the diversity and distribution of these elements in plant genomes suggests the potential to play an important role as mediators of plant gene evolution, including both coding and regulatory functions [12][13][14][15]. The specific targeting of these TEs to microsatellite domains, and the resulting modular assembly of mixed MULE clusters, offer improved evolutionary potential by preventing inactivation of active genes [9,21] as well as proximally, and modularly, locating sequence divergent elements with the potential to contribute coding and non-coding gene components.
The SaTar elements described here do not represent a complete catalog of these elements in any of the selected genomes. However, the described elements are sufficient for demonstrating conservation of SaTar structure, chromosomal distribution profiles and cluster architecture among these plant species. It is clear that accurate determinations of both the frequency of SaTar clustering and overall contributions to these genomes will require more thorough characterization, and that the values presented here represent minimal estimates of their overall contributions to these genomes.