An Atlas of Plant Transposable Elements

Data source

All genomes (Supplementary Material 1, Extended data) were downloaded from the Ensembl Plants¹⁹ database, version 41 (57 genomes) and 45 (plus 10 new genomes).

Annotation of transposable elements

We used similarity-based methods and de novo techniques to build a collection of putative transposable elements, based on the SPTEdb pipeline.²¹ We refined, extended and increased steps in order to produce a novel annotation (Figure 1). Our reformulated steps (details in Supplementary Material 2, Extended data) guarantee a comprehensive knowledgebase of these TEs.

Figure 1. Steps in transposable elements identification.

Dataset: Genome assemblies were downloaded from Ensembl Plants. Identification: 1A) RepeatScout was used to search for putative repetitive sequences and further classification by PASTEClassifier, resulting in a library. 1B) RepeatModeler was also used to find a consensus of TEs sequences. 2) RepeatMasker was run with Repbase library and libraries from RepeatModeler and RepeatScout. 3) For Class II - Subclass 2 TEs, we also used HelitronScanner and MITE-Hunter. 4) In order to find LTR and Non-LTR retrotransposons, we used LTR_retriever and MGEScan-non-LTR, respectively. Filter: A cut-off filter was applied to remove low complexities, simple repeats and other nomenclatures that were not classified into TEs. Annotation: In result of the pipeline, we have a Transposable Element annotation for each genome analyzed.

RepeatScout was performed separately; the output was unified in a library to be labeled by PASTEClassifier³² and later combined into a final annotation. To automate the pipeline, an in-house framework in Perl language was developed for each software output to be uniformized, described in steps 1 to 4. A main script in Bash starts the process of automatization using Perl scripts. All steps were supervised by researchers, carefully checked, and the output was manually verified at each step for each genome. Records classified as low complexity, simple repeat and other nomenclature not related to Class I or Class II TEs were discarded.

Due to the extensive genome sizes of Triticum aestivum (14,5 Gb), Triticum dicoccum (10,4 Gb) and Triticum turgidum (10,4 Gb), we adapted our pipeline for their analysis, based on the approach of Jamilloux et al.²⁴ For these species, we applied our pipeline on chromosome 1 (which is the longest pseudomolecule), as the large genomes were eventually duplicated into new copies, increasing the number of these same repeats in the genome, and did not significantly impact discoveries related to new or different TEs families.²⁴

TE evidence score

To test the reliability of our TE annotation pipeline, we scored sequences that had duplicated annotation in the same loci (Figure 2). We developed a statistical metric (labeled as TE-Score, shown in each record as the ninth column for each genome annotation file) that identify and ponder sequences types that have been identified by the programs. The TE-Score is a metric (0 to 1) that is given by

Figure 2. The TE Score: the average amount of sequence identification made by programs in all genomes.

Correlation analysis

To test for correlations between genome size and transposable elements percentage by genome in base pairs, we first normalized using log10, and then we applied the Pearson Correlation Coefficient in SPSS version 25.

Web implementation

APTE is hosted at the Universidade Tecnológica Federal do Paraná (Cornélio Procópio, PR, Brazil). It uses Debian 11 as operating system, Apache 2 as web server, PHP 5.6 as web programming language. We also used Zend Framework 2, which implements model, view, controller (MVC), a methodology for web development that can be expanded for any future additional functionality. On the front-end, we used HyperText Markup Language 5 (HTML5), Cascading Style Sheet 3 (CSS3) and JavaScript to perform dynamic functions that provide user-friendly navigation. A built-in genome browser (JBrowse, version 1.14.1) is available to visualize and download the data as well.

Computational resources

To run the pipeline described in Figure 1, we used three platforms: (i) to runRepeatModeler,^25,26 RepeatScout,^26,27 RepeatMasker,²⁷ LTR_retriever,²⁸ MITE-Hunter²⁹ and HelitronScanner³⁰; (ii) to perform MGEScan-non-LTR³¹ and PASTEClassifier³²; and (iii) to unify and filter outputs to the main annotation. The hardware utilized were (i) Xeon E7540 2.00 GHz 256GB memory, Xeon E5-2620v3 2.40 GHz 64GB memory, 2x Intel i7-3820 3.60Ghz 32GB memory and Intel i7-3820 3.60Ghz 64GB memory, (ii) Intel i7-3820 3.60Ghz 64GB memory, and (iii) Intel Core 2 Duo 2.4 GHz 8GB memory, a total of 30 physical cores and 456 GB of memory. In order to present a scale of time elapsed to measure, filter and standardize the results, we estimate that for the A. thaliana genome, the time needed to get the final annotation was ~18 hours, using all resources mentioned, including post-processing scripts (detailed on our website).

Overview of TE portion

We retrieved a total of 49,802,023 TE records from 67 plant genomes, representing a total of 47,992,091,043 (~47,62%) base pairs (bp) of the total genomic space. This information is distributed in ~57,36% (28,565,034) TEs organized into class, order and superfamily. In addition, ~42,64% (21,236,989) elements could not be assigned to any type of known TE and they were labelled as unknown. They likely represent chimeric and/or partial elements for which we were not able to perform the full classification. For known TEs, we identified that ~62,85% were retrotransposons, and ~37,15% were DNA transposons. All assigned classifications of TEs identified along the 67 genomes are shown in Figure 3. The distribution of TEs in the analyzed genomes are somewhat similar (Figure 4), especially in genomes that have a shorter phylogenetic distance (e.g., Oryza spp, Triticum spp). However, even close-related genomes exhibit uneven TEs distribution (e.g., Arabidopsis spp). Two main hypotheses might explain the variation of TE content: (a) different evolutionary stories, since the two major genome duplication events are shared by all seed plants (epsilon) and flowering plants (gamma), followed by the lineage-specific duplication events,²⁰ and (b) specific pressures to maintain, expand and purge TEs in each lineage.

Figure 3. Class, order and superfamilies identified among the 67 plant genomes used in this study.

Figure 4. Overview of Class I and Class II composition of TEs in each genome organized in a phylogenetic tree.

We have noted that our approach permitted better TE annotations in genomes assembled at chromosome scale, and we also observed that the amount of TEs is generally related to the genome size, since larger genomes have higher occurrences of TEs (Figure 5). However, for incompletely and draft-assembled genomes, it tends to decrease the number of TEs, once the assembly into small parts (scaffolds or contigs) may impact the genome assembly quality, collapsing repeated contigs (mostly TE- derived) and interfering with the proper identification of these TEs.

Figure 5. Correlation between genome size and TE content.

On the left, the bar chart in blue, the genome size (in Gb), and, in green, the transposable elements distribution in analyzed genomes (in percentage). On the right, we normalized, in base pair, genome size and TE using log(10) and then we correlated (Pearson) the genome size by transposable elements. r and p-value are shown in the top-left of each chart. A) Using all the 67 annotated genomes; B) For all genomes with recent WGD (Whole Genome Duplication) events, blue circles; C) Excluding genomes that experienced recent WGD, red circles.

TE database comparison

To compare the results of the identification performed and to ensure the reliability (details in Supplementary Material 1, Extended data) of our approach, we used SPTEdb²¹ annotation data of the genome Populus trichocarpa (black cottonwood), which is explored in Table 1. The second comparison of TE annotations was performed for the Glycine max (soybean) genome, in which we used SoyTEdb²² to compare the amount vs. type of TEs, shown in Table 1. A third comparison used data from GrTEdb²³ to explore the amount of TEs in Gossypium raimondii (cotton), available in Table 1.

Underlying data

All data underlying the Plant TE Atlas is available in the portal http://apte.cp.utfpr.edu.br/.

Extended data

Zenodo: Datasets from An Atlas of Plant Transposable Elements, https://doi.org/10.5281/zenodo.5672122.³³https://doi.org/10.5281/zenodo.5574528

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero “No rights reserved” data waiver (CC0 1.0 Public domain dedication).

Analysis code available at: https://github.com/alerpaschoal/apte_pipeline/

Archived code at time of publication: https://doi.org/10.5281/zenodo.5672122

License: CC0