Manually curated library of transposable elements from Aedes albopictus

TE discovery for manual curation in Ae. albopictus.

For the construction of the MCTE library, TE consensi were obtained from the AalbF2 genome assembly (Aalbo_primary.1; (Palatini et al. 2020)) via three distinct annotation algorithms: EDTA-2.0.0 (Ou et al. 2019), RepeatModeler-2.0.2a (Flynn et al. 2020), and MITE-Tracker (Crescente et al. 2018). The resulting TE consensi were combined into a single multifasta file, along with a subset of sequences from RepBase 25.08 (Jurka et al. 2005; Bao, Kojima, and Kohany 2015) (including sequences from Anopheles gambiae, Drosophila melanogaster, Invrep -Invertebrata-, and Invsub -Invertebrata subfamilies-). RepeatMasker hits shorter than 80 bps were removed because of the likelihood of spurious results. After masking the genome, the OneCodeToFindThemAll pipeline (Bailly-Bechet, Haudry, and Lerat 2014) was applied to reconstruct TE copies by merging close-by RepeatMasker hits via the following command:

 build_dictionary.pl --rm Aalb.out --unknown --fuzzy > dico_fuzzy.txt \ one_code_to_find_them_all.pl --rm Aalb.out --ltr dico_fuzzy.txt --fasta --flanking 100 --strict --unknown --insert 80

A custom script was then used to parse the output of one code to find them all (Bailly-Bechet, Haudry, and Lerat 2014) and extract a fasta file containing all the TE copies sequences (https://github.com/Tcvalenzuela/Manual-versus-automatic-annotation-of-transposable-elements/octfta_to_fasta.sh).

To group TE copies by families, RepeatMasker-identified insertions were clustered via cd-hit (Huang et al. 2010; Li and Godzik 2006) with parameters set to match the 80-80-80 rule as much as possible: identity of 80% or greater along more than 80% of the sequence in sequences longer than 80 bp as described in Flutre (2011), as follows: cd-hit est -i copies.fasta -o consensi.fasta -c 0.8 -G 0 -aS 0.8 -M 90000 -d 0.

After several test runs, to increase the high copy number families, we selected only clusters with more than seven insertions. Consensi were called from the remaining clusters with the Refiner tool from RepeatModeler2 (Flynn et al. 2020) after downsampling 500 sequences from the largest clusters. To reduce redundancy among family consensi, these were in turn clustered with cd-hit-est and consensi called again with Refiner (Flynn et al. 2020). The process was repeated until no redundancy was detected in the library (total number of iterations: 8). In the end, 23,009 family consensi were obtained at this step.

To increase the likelihood of obtaining full-length TE models, all 23,009 consensi were extended via the following method. First, the putative TEs were screened for low complexity and simple repeat sequences via TRF (Benson 1999). To gather sequences for extension, the filtered sequences were used as a library for RepeatMasker (v4.1.5) on the AalbF2 genome assembly. An alignment per family is generated in stockholm (.stk) format, followed by extension, using somewhat relaxed parameters for extension into the flanking sequence based on the chromosomal locations present in the .stk files. Following extension, additional identifying information might have been obtained, such as long terminal repeats (to allow for endogenous retroviruses identification), terminal inverted repeats (for DNA family identification), or polyA tails (for LINE/SINE identification). Therefore, RepeatClassifier (a utility part of RepeatModeler) is run to take this additional information into account.

The extended consensi was used as a reference in a second RepeatMasker run, this time performed on AalbF3, a deduplicated version of F2 plus optical mapping reference genome assembly GCA_018104305.1 (Boyle et al. 2021). The output of RepeatMasker was used to sort the list of consensi by the total number of full-length insertions, where a full-length insertion was defined as at least 90% of the length of the consensus, with a maximum nucleotide divergence of 20% (Flutre et al. 2011). Using the custom script https://github.com/Tcvalenzuela/Manual-versus-automatic-annotation-of-transposable-elements/detectFullSize.py, the list of consensi was finally sorted by full-length frequency, and manual curation was performed starting from the most frequent consensi. Elements with fewer than three insertions were not considered for manual curation and were not part of the final library either.

Manual curation of TEs from Ae. albopictus

The relevant information necessary for manual curations, such as the absolute frequency of insertions, pre- and post-extension length, insertion frequency in each genome assembly (AalbF3 and AalbF5 versions), ratio of extended consensus length to original consensus length, extended consensus coverage, extended consensus median coverage, insertion frequency, and full-length insertion frequency, was computed for all the consensi. The frequency of full-length insertion was used as a priority list for manual curation.

 Manual curation was conducted on the TE consensus of the 800 most frequent full-length insertions, resulting in a final TE library containing 497 TEs. First, from the insertion coordinates from RepeatMasker and using the custom script https://github.com/Tcvalenzuela/Manual-versus-automatic-annotation-of-transposable-elements/GetMultipleAln.sh, each insertion was extended to 2,000 bp on both flanks and extracted, and the 100 longest insertions were clustered together via ClustalO (Sievers and Higgins 2014) for examination of the characteristic component of the respective category of TEs. Additionally, each TE consensus was examined via a set of manual curation identification tools, such as TE-Aid (Goubert et al. 2022), RepeatClassifier (Flynn et al. 2020), and alignment visualisation via Aliview (Larsson 2014), together with databases as Repbase (Jurka et al. 2005; Kohany et al. 2006; Kapitonov and Jurka 2008) and CDD protein domains (Lu et al. 2020; Marchler-Bauer et al. 2015), following the annotation recommendations from Goubert et al. (2022).