INTERCHANGE: A new strategy for horizontal transfer identification at the whole-genome scale using short read sequencing data

The inference of HTs is usually based on the use of three criteria [29,30]: i) high sequence similarity between evolutionary divergent species (HS); ii) phylogenetic incongruence between the evolutionary history of the species and that of the transferred sequence (PI); iii) patchy distribution of the transferred sequence in the phylogeny of the species (PD). In other words, no homologous sequence of the transferred DNA is found in the closest relative of the recipient species. There are several alternative hypotheses that could explain high sequence conservation, phylogenetic incongruence and patchy distribution. For instance, high similarity in genetic sequence between divergent species could originate from intense selective pressure or from introgression events. Phylogenetic incongruity could potentially result unrecognized paralogy or incomplete lineage sorting. Finally, patchy distribution might be attributable to the stochastic loss or retention of genes within specific evolutionary lineages. It is critical to underscore that the likelihood of these alternative hypotheses being accurate increase as the species involved in the HT exhibit closer evolutionary distance. Conversely, the integrated application of these criteria, particularly in scenarios involving HTs between evolutionary highly distant species, significantly strengthens the plausibility of HT hypothesis.

Existing methods for detecting HT, which rely on one or two of the previously mentioned three criteria, require prior genome assembly and annotation. This poses a significant challenge to HT studies among wild non-model species lacking accessible references genomes and annotations. We have therefore developed a new pipeline to identify regions of high conservation, potentially indicative of HT events, between two or several genomes. This approach that utilizes raw short-read sequencing data and accounts for all three HT identification criteria. Briefly, INTERCHANGE (i) first identifies similar reads derived from conserved genomic loci between the studied species using a k-mer approach (ii) assembles these reads into scaffolds (iii) annotate the scaffolds (iv) and test for high sequence similarity (HS) by comparing the sequence identity between conserved scaffolds with that of orthologous genes. Those HT candidates are then manually tested for the PI and PD criteria. The main steps of this pipeline are shown in Fig 1 and described in details in the method section.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The different steps of the INTERCHANGE pipeline of horizontal transfer identification from unassembled and unannotated genomes.

Steps 1 to 8 are completely automatic steps 9 and 10 are semi-automatic. Step1: Identification of identical k-mers using Tallymer [54]. Steps 2 & 3 Identification of reads derived from conserved regions & Removal of Tandem repeats using PRINSEQ lite [55]. Reads sharing at least 50% of identical k-mers are considered as homologous reads. Step 4: homologous reads are extracted and assembled for each pair of species using SPAdes [56]. Step 5: Scaffolds annotation using multiple protein and TEs database: CDDdelta, Repbase, mitochondrial, chloroplast, and ribosomal (TIGR) gene database. Step 6: Identification of homologous scaffolds using reciprocal best hit (RBH). Step 7: Identification of high sequence similarity threshold based on the distribution of orthologous BUSCO gene identities according to the following formula: high similarity threshold (HS) = Q3+(IQR/2); where Q3 is the third quartile, IQR is the interquartile range (Q3-Q1). Step 8: Testing for HS criteria. Step 9: Phylogenetic incongruence criteria. Step 10: testing the Patchy distribution (PD) of transferred sequence. For details see Method section.

https://doi.org/10.1371/journal.pgen.1010964.g001

INTERCHANGE validation using simulated horizontal transfers

To test the accuracy of INTERCHANGE, we simulated HT events involving both genes and transposable elements (TEs) of various classes and superfamilies between three plant models: Arabidopsis thaliana, Oryza sativa and Brachypodium distachyon (Fig 2, see method for details). These species were chosen because they have high-quality, well-annotated reference genomes and present a contrasting evolutionary divergence times. A. thaliana and O. sativa share a last common ancestor 160 million years ago, while the two Poaceae species, O sativa and B. distachyon, diverged from each other less than 46 million years ago (http://www.timetree.org/). HTs were simulated between these species with contrasting divergence times in both directions, using each species as a donor and a recipient (Fig 2A). Using the assembled reference genomes, we inserted 100 genes and 100 TEs from A. thaliana into the O. sativa reference genome, and 100 genes and 100 TEs from O. sativa into the A. thaliana reference genome, resulting in 400 simulated HT events between these two species (see methods section). Before inserting these sequences into the recipient genomes, random artificial mutations were introduced to simulate both ancient and recent HTs as described in the method section. Next, Illumina short reads were simulated for each genome harboring these in silico HTs. The same HT simulation was carried out between O. sativa and B. distachyon.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Simulation of horizontal transfer (HT) between A. thaliana and O. sativa and between O. sativa and B. distachyon.

a. 200 HT events were simulated in each direction (green arrows), comprising genes and TEs with equal proportion. b. INTERCHANGE results using short reads of genomes harboring simulated HTs. Y-axis indicate to the total number of HTs (scaffolds) identified by INTERCHANGE and X-axis represent filters based on scaffold size. The color codes are provided in the figure legend.

https://doi.org/10.1371/journal.pgen.1010964.g002

The INTERCHANGE pipeline was applied to these species using simulated short read sequencing data from the artificially modified genomes that incorporated simulated HT events. INTERCHANGE was able to identify 90% (361 out of 400) of simulated HT events between A. thaliana and O. sativa (Fig 2B) (HS = 80%). 10% of simulated HTs were false negatives (39 out of 400) and only one single scaffold candidate was a false positive. Among the identified HTs, 52% were genes and 48% were TEs. For the pair O. sativa and B. distachyon, INTERCHANGE identified 73% (292/400) of simulated HTs (53% genes and 47% TEs) with a false negative rate of 27% (109 out of 400) and 743 false positives. 85% of these false positives correspond to scaffolds smaller than 500 bp and correspond to highly conserved genes with a sequence similarity higher than the HS threshold (i.e HS = 87%). False negatives correspond to simulated HTs with lower sequence similarity than the threshold. This increase in false negative rate and false positive rate compared to A. thaliana and O. sativa can be explained by the smaller evolutionary distance between the two Poaceae species. As shown in Fig 2B, when only scaffolds with longer size are considered however, the number of false positives decreases significantly, with a smaller decrease in true positives and without affecting the relative proportion of detected genes and TEs. For instance, by limiting the candidates to those with a scaffold size longer than 1 kbp, the number of false positives decreased by 93% (from 743 to 52) while the number of true positives decreased by only 17% (from 292 to 241) (Fig 2B). This simulation clearly demonstrate that INTERCHANGE is able to efficiently identify simulated HTs directly from the raw reads without any detection bias towards genes or TEs. Additionally, the sensitivity of INTERCHANGE increases as the evolutionary distance between the species involved in HTs increases.

INTERCHANGE validation using real data allows detection of new horizontal transfer events

To further validate INTERCHANGE pipeline, we applied it to five distant plant genomes for which several HT have been previously reported [9]: grapevine (Vitis vinifera), peach (Prunus persica), poplar (Populus trichocarpa), date palm (Phoenix dactylifera) and clementine (Citrus clementina). These highly divergent species have experienced 6 HTs of LTR-retrotransposons named BO1, BO2, BO3, BO4, BO7 and BC1 (BO: HT between plant orders; BC: HT between plant classes) [9]. In this previous study, the identification of these HTs was done through a comparative genomic analysis using assembled and previously annotated genomes. Here, we used the unassembled short reads of the same species (S1 Table) to test whether INTERCHANGE could detect the previously reported HTs using a minimum scaffold size filter of 1 kbp to reduce the number of false positives as suggested by our HT simulation.

A total of 10 whole genome comparisons were performed between the 5 species and 31 HT candidates were identified using INTERCHANGE, of which 30 correspond to LTR retrotransposons (29 Copia and 1 Gypsy) and one single gene (Elongation factor 1) (S2 Table). In addition to the HS criteria used by INTERCHANGE, we tested for the PI and PD criteria (see Method). The PD criterion was only tested if the HS and PI criteria were satisfied. Nine HT candidates met both the HS, PI and PD criteria, while for the other 22, only one criterion was met leading to their rejection for further analysis (S2 Table). Four among the six known HTs (BO1, BO3, BO4 and BO6) were identified by our new strategy (see Fig 3). BO2 and BC1 were not detected by INTERCHANGE because they did not pass the 1kbp scaffold size filter (S2 Table). Remarkably, INTERCHANGE detected an additional five HTs that were not previously identified by El Baidouri et al. (2014). This includes two HTs between grapevine and date palm (HT1, HT2), one HT between grapevine and poplar (HT3) and two HTs between poplar and peach (HT4, HT5). Interestingly, as with the previously identified HTs, these new HTs correspond to LTR-retrotransposons (LTR-RTs) from the Copia superfamily. Strikingly, the LTR-retrotransposons identified by INTERCHANGE as having been transferred between grapevine and date palm (HT1 and HT2) have a high degree of sequence identity at 91% and 95.5%, respectively, suggesting a more recent transfer than the previously identified transfer (86% for BC1) (S2 Table). All newly identified HTs were also found in the reference genomes of species involved in HTs, providing further evidence of the reliability of the INTERCHANGE pipeline.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. HTs identified by INTERCHANGE using real data.

Lines represent the HT events identified from genome short read sequencing data. In green, HTs that were identified in a previous study [9] using reference genome and detected by INTERCHANGE from short reads. In gray, HTs missed by INTERCHANGE. In red, new HTs only identified by INTERCHANGE.

https://doi.org/10.1371/journal.pgen.1010964.g003

To further assess the capability of INTERCHANGE in identifying transferred genes, we focused on the recently reported case of Glutathione S-transferase gene (GST) transfer from the fungi species Epichloe aotearoae to the cereal plant Thinopyrum elongatum (S1 Table) [17]. As Illumina reads for E. aotearoae were unavailable in public database, we opted for a closely related species harboring the GST gene and possessing the required data (Epichloe clarkii). INTERCHANGE was executed using Illumina reads from T. elongatum and E. clarkii. This analysis successfully detected one significant HT candidate, corresponding to the GST gene homologs of E. aotearoae in the T. elongatum genome (see S1 Data). Moreover, we performed simulations of Illumina reads from E. aotearoae, utilizing the E. aetearoe reference assembled genome (NCBI: ASM72985v1). A similar analysis was then carried out using INTERCHANGE, which identified again one candidate that corresponded to the GST gene transfer. Together, this comprehensive evaluation effectively illustrates the capacity of INTERCHANGE to detect both horizontally transferred genes and TEs.

Characterization of horizontal transfers between wild non-model species from a natural ecosystem

To deepen our understanding of the biotic interactions that can facilitate HTs in natural ecosystems, and to ascertain whether certain genes or TEs are more prone to HT than others, we selected the Massane forest as our study site. This protected reserve, located in southern France, is a unique European location that earned a UNESCO World Heritage Site designation in July 2021. It is mainly composed of beech trees (Fagus sylvatica) located between 600 and 1158 m of altitude and covering 336 hectares in total. We sampled 17 different species from this ecosystem including 14 plant species and three fungi (see Table 1). The plant species selected include six tree species, four lianas/climbing plants, two herbaceous, one bramble, and one shrub and are all non-parasitic species. The three fungi species were collected from tree trunks or dead wood and include two saprophytic and one parasitic species. The selection of these species was based on a combination of biological and technical criteria, including: (i) the nature of their biotic interactions, including those with close relationships such as parasitism or physical proximity, and those with no known close interactions; (ii) genome size, with the aim of obtaining sufficient sequencing coverage for the detection of HTs. The selected species have a genome size smaller than 3 Gbp; (iii) abundance in the Massane forest, to facilitate collection; and (iv) taxonomic diversity to optimize phylogenetic representation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Species sampled in the Massane forest and whose genome has been sequenced using Illumina short-read sequencing.

https://doi.org/10.1371/journal.pgen.1010964.t001

The genomes of the selected species were sequenced using Illumina short-read technology with 20X coverage (see Method and S3 Table). Using INTERCHANGE, we performed 136 whole-genome pairwise comparisons to identify highly similar regions between these species that may have originated from HTs. INTERCHANGE detected 68 HT candidates comprising 46 TEs and 22 genes and involving 8 out of the 17 studied species (see S4 Table). In order to avoid redundancy of candidates due to the presence of multiple paralogs, clustering was performed using SiLiX [31], resulting in 48 HT clusters (see S4 Table). To test the PI criteria, HT candidates where aligned to 400 plant genomes using Blastn (See S5 Table). Phylogenetic trees of the transferred TEs were constructed and compared to the phylogenetic trees of species (see Method Step 9). In total, of the 48 HT candidates (22 genes and 25 TEs) that met the HS criteria only 12 TE candidates also met the PI criteria (S2–S13 Figs, S2 Data). 11 of the 12 TE candidates belong to the LTR Copia superfamily (named MaCo01 to MaCo12, for Massane Copia) and 1 element belongs to the Gypsy superfamily (MaGy01, for Massane Gypsy). We further checked the presence/absence of these TEs in the genome 400 plant genomes to test for the PD criteria. This analysis clearly showed that these elements have a patchy distribution, thus confirming the occurrence of HT (see S14–S24 Figs). Notably, both the phylogenetic trees and the patchy distribution of these transferred TEs in 400 plant species point to other possible HT of the same elements between multiple plant species (S8 Table). This is further supported by the utilization of RANGER-DTL software [32], which tests for transfer, duplication, and loss scenarios, strongly suggesting that these elements may have undergone multiple horizontal transfers during their evolution (S4 Data).

Wet-lab validation of the horizontally transferred LTR-retrotransposons

To rule out potential contamination between the investigated species, despite all the precautions taken during sampling (see Methods), we performed PCR and sanger sequencing to check the presence of the identified HTs in the genome of the 8 species involved in HTs. To do this, we re-sampled two additional individuals from each of the 8 species and extracted their DNA. For each transferred LTR-RT, a set of two primer pairs was designed to amplify different regions of each element (S6 Table and S26 and S27 Figs). In the two individuals of the species involved in the HTs, the transferred LTR-RTs were successfully amplified. For one candidate (MaCo11), only one of the two primer sets results in PCR amplicon (S26 Fig). One PCR amplicon from each transferred LTR was selected for sanger sequencing. Multiple sequence alignment of the PCR product sequences and the sequences of the elements detected using INTERCHANGE, as well as the construction of a phylogenetic tree, validate the occurrences of these HTs (S3 Data). We also verified the presence of the transferred LTR-RT in the genomes of the donor/recipient species for which a reference genome is available. These are A. glutinosa, F. sylvatica, F. excelsior and P. avium. All the transferred LTR-RTs implicating one of these species where identified in their respective reference genome. Additionally, we sampled and sequenced the genome of two H. helix individuals (Ivy A: ∼4Gb; N50 = 14.4 kbp and Ivy B: ∼4Gb; N50 = 14.6 kbp) and one F. sylvatica (7.4 Gb; N50 = 20 kbp) using Nanopore sequencing (see Method section). We were also able to unambiguously identify the transferred LTR-RT involving these two species in different Nanopore reads corresponding to different paralogs (see S28 and S29 Figs). Taken together, these results clearly refute the possibility that the HTs identified in this study are the result of contamination.

Lianas are hotspots of horizontal transfers

Among the 12 HTs that we identified, none involved a saprophytic or parasitic fungi. As shown in Fig 3, these HTs occurred between 8 out of the 17 studied species. The species involved in these transfers are essentially trees and climbing plants. Indeed, 5 of the 6 analyzed tree species have experienced at least one HT event. These are, in decreasing order of HT frequency: Fraxinus excelsior (6 HTs), Fagus sylvatica (5 HTs), Alnus glutinosa (2 HTs), Acer monspessulanum (2 HTs), Prunus avium (1 HT). For the climbers, two species among the five analyzed have undergone HTs, namely Dioscorea communis (5 HTs) and Hedera helix (2 HTs). These HTs were identified between phylogenetically distant species that do not belong to the same plant class. In particular, the five HTs involving D. communis (Fig 4), which is a monocot species, occurred with eudicot species that diverged over 150 million years ago. Interestingly, most HTs involving D. communis (4/5 HTs) and H. helix (2/2 HTs) occurred with tree species which may suggest that the close physical relationship between lianas and trees may be a facilitator of HTs between those plants. Additionally, we found that some species pairs underwent multiple independent HTs of different LTR families such as the ones that occurred between D. communis and F. excelsior (2 HTs) and between F. excelsior and F. sylvatica (2 HTs). The direction of the HTs could not be determined, although the patchy distribution is clearly shown for all transferred LTR-RTs (S14–S25 Figs), as this will require further sampling and sequencing of additional plant genomes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The phylogenetic tree of the 17 analyzed Massane species.

The curves represent the identified HTs and link the involved species. Blue and red curves represent Gypsy and Copia HTs, respectively. The asterisks indicate multiple HTs. The horizontal scale represents the divergence time in million years (source: timetree.org). The figure was generated using scripts from Zhang et al. [12]. Correspondence of species names: Ace: Acer monspessulanum, Aln: Alnus glutinosa, Bry: Bryonia dioica, Dio: Dioscorea communis, Fag: Fagus sylvatica, Fra: Fraxinus excelsior, Fom: Fomes fomentarius, Gen: Genista pilosa, Hed: Hedera helix, Her: Hericium clathroides, Lon: Lonicera periclymenum, Ple: Pleurotus ostreatus, Pru: Prunus avium, Rub: Rubus ulmifolius, Sal: Salvia sp, Sen: Senecio inaequidens, Sor: Sorbus aria.

https://doi.org/10.1371/journal.pgen.1010964.g004

Sequence identity between the transferred LTR-RTs varies from 89 to 97% (S4 Table), corresponding to an age of transfer between 3.8 and 1.15 million years (Mya) (using the molecular clock rate of Ma and Bennetzen, 2004) [33]. This indicates that these HTs occurred millions years ago and are therefore ancients. This is also supported by the PCR analysis on different individuals that indicate that these HTs are likely to be fixed in populations of these species. Additionally, as shown earlier, for the species for which the reference genome is available (F. sylvatica, F. excelsior, P. avium and A. glutinosa) the transferred elements were found in their respective genomes pointing to ancient HT events. From the available data, it cannot be determined where these HTs took place and whether they occurred in the Massane forest, despite its ancient origin. It is important to note however that all species involved in these HTs are native to European and Mediterranean regions and that their respective geographic distributions overlap, indicating that they have been in contact for a long period of time, thus facilitating the occurrence of HTs.

Copia LTR-retrotransposons are the most frequently horizontally transferred elements in the investigated plant species

Despite the fact that our approach of HT identification does not focus on specific types of sequences such as TEs or genes, unlike all other approaches, the HTs identified in the Massane forest involve only LTR-retrotransposons. Some might argue that INTERCHANGE has a bias toward LTR-retrotransposons transfer detection, but our simulations have shown that this is not the case. Further characterization of the protein-coding genes of these transferred LTR-retrotransposons shows that 11 out of 12 belong to the Copia lineages (i.e. MaCo01 to MaCo11) and one belongs to the Gypsy lineage (MaGy01). This result is consistent with what we observed in our previous work, where Copia were more frequently transferred than Gypsy (28 Copia vs 7 Gypsy) [9]. However, in order to ascertain this, it is essential to check whether this is not due to an overrepresentation of the Copia superfamily among other LTR-RTs in the surveyed genomes. For this purpose, we estimated the relative frequencies of Copia and Gypsy in the 8 species involved in HTs by aligning their raw genomic reads to a collection of reference protein sequences [34] using Diamond Blastx [35] (See Method). As shown in Fig 5A, Copia elements were more prevalent than Gypsy elements in 6 out of the 8 species, equally abundant in D. communis and less prevalent in H. helix. On average, Copia were 1.4 times more prevalent than Gypsy. However, this cannot explain that 11 out of the 12 identified HTs belong to the Copia clade.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. The relative abundance of LTR-retrotransposon superfamilies in species that have experienced HTs.

a) Relative frequency of Copia and Gypsy in the studied species involved in HTs. In blue: Copia frequency, in yellow: Gypsy frequency b) Phylogenetic tree of transferred Copia detected in this study using the RT domain. In bold, the consensus sequence of the reference Copia lineages. Maco1 to 11 correspond to horizontally transferred elements identified between the plant species from the Massane. BO1 to BO8, BG1 to BG and BC1 correspond to Copia elements identified in our previous study. Correspondence of species names as in Fig 3C) Copia lineages relative frequencies in species involved in HT.

https://doi.org/10.1371/journal.pgen.1010964.g005

The transferred Copia LTR-retrotransposons belongs to Ale and Ivana lineages

We then investigated whether some Copia lineages have a greater propensity to transfer than others. To this end, we extracted from the Rexdb database the reverse transcriptase (RT) protein sequences of 17 Copia reference clades described in the literature [34] as well as those of the 11 transferred Copia (MaCo01 to MaCo11) identified in this study (21/22 paralogs) (see Method). Furthermore, we also extracted the RT domain of the 28 Copia previously identified as horizontally transferred between several plant species [9]. The constructed phylogenetic tree shows that for the previously described HTs, 21 Copia (75%) belong to only two lineages Ale (13) and Ivana (8) where the 7 remaining HTs belong to different lineages such as Tork, TAR or Bianca (Fig 5B). This trend was even more pronounced for the Copia elements identified in the Massane forest. In fact, all transferred elements belong only to these two lineages: Ivana (6 /11) and Ale (5/11) (Fig 5B).

These results suggest that these two lineages are more prone to HTs compared to other Copia lineages in the studied species. In order to check whether this observation could be due to an overrepresentation of these two Copia lineages in the analyzed plant genomes, we estimated the frequency of all known Copia clades in the 8 plant species involved in the identified HTs (see Method). This analysis shows big disparities in Copia lineage frequencies in those species with no particular conserved trend. In five of the eight species involved in HTs (A. glutinosa, D. communis, F. sylvatica, P. avium, R. ulmifolius) the Ale lineage was the predominant Copia lineage ranging from 8.6% in A. monspessulanum to 37.8% in P. avium (Fig 5B). On average, Ale accounted for 24% of the Copia elements in these genomes, followed by SIRE lineage (16.6%). Meanwhile, the percentage of the Ivana lineage varies from 3.3% in A. monspessulanum to 17% in D. communis with an average of around 6.5%. These results show that the strong bias observed in transferred elements belonging to Ale and Ivana lineages cannot be explained by the relative abundance of these lineages in the genomes of the species involved in the HTs.

The horizontally transferred Copia are active after their transfer but show low transpositional activity in both donor and receiver species

To better understand the dynamics of the transferred elements, we estimated their copy number in both donor and receiver species using unassembled raw genome sequencing reads. To this end, we used the coverage of single-copy BUSCO genes as a standard to normalize the observed coverage of each transferred element (see Methods). This analysis show that copy number of the transferred elements varies from single copy to 28 copies for the Gypsy MaGy01 with an average of 4.3 copy per species (see S7 Table). For the HTs where we could identify the direction of HT, we can observe that some Copia elements exist as single copies (3/8), indicating that these elements likely didn’t transpose after their transfer. The remaining elements show, on the contrary, a transpositional activity in the host recipient genomes that results in several copies. However, this post-transfer transpositional activity appears to be low, with only 2 to 5 copies observed for each transferred LTR-RT. It is interesting to note that this low transpositional activity is also observed in the donor species, suggesting that it is an inherent feature of the transferred Copia element.

INTERCHANGE a new pipeline for HT characterization at the whole-genome scale using raw sequencing reads

In this study we investigated for the first time HTs between wild non-model species within a natural ecosystem. We sequenced the whole genome of 17 species including trees, climbing plants and fungi and characterized HTs directly form raw sequencing reads thanks to INTERCHANGE pipeline. Using this tool, we were able to report new HT events in previously studied species that had not been identified using previous methods [9]. In this study, we demonstrated the utility of INTERCHANGE for genome-wide screening of HT events between non-model species, using both simulated and real datasets. Importantly, our approach obviates the need for costly and time-consuming genome assembly and annotation, which are often major bottlenecks for conducting such studies.

As indicated by our evaluation of INTERCHANGE some of the known or simulated HT even did not pass the 1 kbp scaffold size filter. A decrease in scaffold size may allow characterization of these events, but this will be at the expense of specificity. Other parameters may also impact the specificity or sensitivity of INTERCHANGE. For example, the smaller the k-mer size, the greater the sensitivity and vice versa, but this will lead to an increase in the computational time needed to perform all the possible pairwise comparisons. The different parameters of INTERCHANGE can be modified by the user which allows great flexibility. However, we shall point out that INTERCHANGE can only detect relatively recent HTs because high sequence divergence between older transferred sequences will not satisfy the HS criteria. Despite these limitations, our results show that this pipeline is very efficient at detecting gene and TEs HT events at the whole-genome scale using unassembled sequencing reads and is therefore a tool of choice for future studies of HT in natural ecosystems. INTERCHANGE could also be used to identify conserved sequences such as homologous genes, TEs or other types of sequences from unassembled genomes, which could be very useful for comparative genomics studies.

No plant-fungus horizontal transfer was identified at the Massane forest

Using the INTERCHANGE pipeline, we were able to identify 12 HTs implicating 8 plant species. We did not identify any transfer involving fungi even though the three studied species are saprophytic or parasitic and known to proliferate on tree trunks or dead wood. It is broadly accepted that close relationships such as endosymbiosis or parasitism are favorable for HTs in eukaryotes [19,21,24,36–38]. The absence of plant-fungus HT in this study may suggest that such events are rare or too old to be detected [39].

Liana-tree interactions: a possible new route of horizontal transfer between non-parasitic plants?

Our results also show that the two climbing plants, the common ivy (H. helix) and black bryony (D. communis) have experienced several HTs events predominantly with trees. These findings are in agreement with our previous study that showed a higher frequency of HTs between grapevines and several tree species [9]. To date, no hypothesis has been put forward to explain this higher frequency of HT in grapevines and whether it is due to an inherent genetic trait or to its particular ecological lifestyle. In fact, similarly to common ivy and black bryony, wild grapevine use trees as support for growth which could explain the high HT frequency observed in this species. A recent study on four different and closely related Vitis species seems to confirm this trend [40]. Using comparative genomics, the authors identified dozens of HTs between these four closely related vine species and mainly trees belonging to highly divergent taxa, although they did not highlight the greater frequency of HTs between grapevine and trees. In light of our findings, we hypothesize that liana-tree interactions may favor HTs between non-parasitic plants and could be considered as route by which HTs occur frequently in nature. For the three other climbing plant genomes analyzed in this study, we did not find any HT. Therefore, the question of whether some climbing plants are more prone to HTs than others remains unanswered at this point. This needs to be tested on a larger sampling of liana species. It is also possible that the tree-to-tree HTs that we identified between beech (F. sylvatica), ash (F. excelsior) or alder (A. glutinosa) could be mediated by other, yet not sequenced, climbing plant species.

Horizontal transfers in plants mainly involve low copy number LTR-retrotransposons belonging to the Ivana and Ale lineages of the Copia superfamily

Our study reveals that LTR-retrotransposons are the only genetic elements that experienced HTs in the studied plant species, which confirms earlier reports, but remains without mechanistic explanation. In fact, successful HT requires three key steps, namely the "excision" of genetic material in the form of DNA or RNA molecules from the donor genome, its transport to the recipient species and finally its integration into the target genome. Due to their transposition life style, LTR-retrotransposons are able to generate extrachromosomal double stranded DNA encapsidated in the VLP (Virus Like Particule) and accumulating in the cytoplasm of the cells [41,42]. They also have the ability to integrate into the host genome using the integrase (IN) [43,44]. LTR-retrotransposons may therefore be more likely to achieve successful HT, given their ability to generate double stranded DNA encapsidated in the VLP and because of their ability to integrate the host genome. Although both Copia and Gypsy elements can produce VLPs, Copia appears to be more prone to horizontal transfer than Gypsy elements. If Copia and Gypsy superfamilies differ mainly in the order of the IN and RT domains, there are some genomic and transpositional features specific to each of these superfamilies. For instance, Copia elements are abundant in gene-rich euchromatic regions while Gypsy elements are mainly located in Heterochromatic and pericentromeric regions [45]. Copia are also generally activated in response to environmental stress as it has been shown for many plant species [46]. It is therefore possible that Copia elements, because of their presence in transcriptionally active regions of the genome and because of their responsiveness to environmental stresses could facilitate their HT.

Intriguingly, Copia elements that have been horizontally transferred between plant genomes belong mainly to the Ivana and Ale lineage and are low copy numbers. The reasons why Ale and Ivana clade appear to be more prone to HT compared to the other Copia lineages remain unknown, as there are no well-known common specific features of these two clades that clearly differentiate them from others [47]. A recent population genomics study in Arabidopsis arenosa showed that Copia elements, particularly Ale and Ivana respond to temperature and irradiance [48]. It is also interesting to note that the TEs shown to be currently active in A. thaliana namely EVADE [49] and ONSEN [50], also belong to the Ale and Ivana clade, respectively, and the latter is active in response to heat stress [50]. As for the transferred Copia, these two families also have a low copy number: two copies for EVADE and eight copies for ONSEN. When a TE family reaches high copy numbers, it tends to be silenced by the production of small interfering RNAs and the epigenetic machinery depositing DNA methylation [51]. The silenced TEs would not be candidates for HT. If this holds true, the question of the presence and survival of Gypsy families in eukaryotic genomes remains to be explained by other mechanisms. Considering our study and previous ones on HT in plants, the propensity of Copia elements and in particular Ale and Ivana lineages to transfer horizontally cannot be explained. Further studies are needed to elucidate the reasons for the remarkable ability of low copy number Copia to transfer horizontally in plants.

Materiel

Illumina genome sequencing

DNA from each sample was extracted using the CTAB2X [52] and the quality of the DNA was estimated by Nanodrop (Thermo Scientific) and Qubit (Invitrogen) quantification. DNA libraries and sequencing was outsourced to Novogene company using the Hiseq 2000 and Novaseq 6000 platforms. Briefly, a total amount of 1μg DNA per sample was used as input material for the DNA libraries. Sequencing libraries were generated using NEBNext DNA Library Prep Kit following manufacturer’s recommendations and indices were added to each sample. The genomic DNA was randomly fragmented to a size of 350bp by shearing, then DNA fragments were end polished, A-tailed, and ligated with the NEBNext adapter for Illumina sequencing, and further PCR enriched by P5 and indexed P7 oligos. The PCR products were purified (AMPure XP system) and resulted libraries were analyzed for size distribution by Agilent 2100 Bioanalyzer and quantified using real-time PCR. Paired-end sequencing was performed using a coverage of 20X and a read length of 150bp for each sample.

Nanopore genome sequencing

High-molecular-weight genomic DNA was extracted from 0.41 g to 0.5 g of frozen leaf tissue according to [53] and the Oxford Nanopore Technologies protocol (February 2019). Briefly, after lysis of cell membranes with Carlson buffer, proteins were removed with chloroform. DNA was purified using Qiagen Genomic-tip 100 columns following the manufacturer’s instructions. A selection of fragments > 10 Kb was performed using AMPure XP beads. DNA quantification was performed by Nanodrop (Thermo Scientific) and Qubit assays (Invitrogen) and the quality was assessed on a 0.8% agarose gel. We then followed the 1D genomic DNA protocol by ligation with the SQK-LSK109 kit to prepare the 3 libraries using 3 μg, 3.9 μg, and 4.1 μg of DNA (beech, ivy A, and ivy B), respectively. We successively loaded 1.7 μg of library onto a Flowcell R9, 2.6 μg and 2.7 μg of libraries onto two Flowcell R10. We produced 7.4 Gb, and 2 times 4 Gb of fastq pass reads with N50s of 20 kb, 14.4 kb and 14.6 kb, respectively. Bascalling was performed using guppy in the high accuracy (hac) mode (https://nanoporetech.com/nanopore-sequencing-data-analysis).

PCR and Sanger sequencing

We utilized the Taq DNA Polymerase 2x Master Mix RED to carry out our PCR amplification. The mix was prepared for each sample by adding 7.5 μl of Taq 2x Master Mix, 6.4 μl of H2O, 0.6 μl of the oligo F and R mix (10μM each), and 0.5 μl of DNA at 10ng/μl. The PCR program we employed consisted of a lid temperature of 98°C, followed by 34 cycles of denaturation at 95°C for 30 seconds, annealing at 60°C for 30 seconds, and extension at 72°C for 2 minutes. This was followed by a final extension at 72°C for 5 minutes and a cooling step at 4°C for 10 minutes. Finally, the samples were electrophoresed on a 1% agarose gel containing TAE 0.5X for 25 minutes at 135 volts, along with a 1kb+ size marker, to visualize the PCR products. PCR products have been purified and sent to Eurofinsgenomics for sequencing using the LightRun Tube platform.

Detection of Horizontal transfer using INTERCHANGE pipeline

In silico simulation of horizontal transfer events

In this simulation, we randomly selected 100 genes and 100 TEs from each species and randomly introduced them into the reference genomes of the other species. For TEs, we selected different copies belonging to the major classes (Class I and Class II) and to various TEs superfamilies (LTRs, LINEs, MuDR, hAT, Mutator, Helitrons, etc.) with an equivalent proportion when possible. Before inserting these genes and TEs into the recipient genome, we artificially introduced mutations to create sequence divergence, simulating both recent and ancient HTs. The mutated sequences had a sequence divergence ranging from 80% to 100% identity compared to the original copies in the donor species. Wgsim tool (https://github.com/lh3/wgsim) was used to simulate 150 bp length paired-end reads with 20X coverage from the donor and recipient genomes carrying the in silico HTs using default parameters.

Estimation of LTR-retrotransposons copy numbers in unassembled genomes

To estimate the copy number of each retrotransposon in the species involved in the transfer, we calculated the number of mapped reads on each transferred retrotransposon compared to the numbers of mapped reads on single-copy genes. Total reads for each species were mapped onto the transferred LTR-retrotransposons. For each LTR-retrotransposon, we calculated the coverage at each nucleotide of the element. The median coverage was taken as a proxy for the coverage of the element in the genome. The same strategy was adopted to estimate the coverage of the BUSCO genes of the studied species. We then used the following formula to estimate the total copy number of each transferred LTR-RTs using Illumina reads: Copy number = (MCT / MCB), where MCT is the median coverage of LTR-RTs and MCB: median coverage of BUSCO genes. To test whether this approach is an appropriate method to estimate copy number using genomic raw reads, we compared the copy number estimated from unassembled genomes with that obtained from assembled reference genomes in species for which the latter is available (S5 Table). Copy numbers estimated from unassembled genomes and those obtained by Blastn against reference genomes are highly correlated, validating our approach (Pearson correlation; R = 0.982, p-value = 4.699E-10).

Phylogenetic tree of Copia lineages

We extract the RT (reverse transcriptase) domain of the transferred Copia elements in both donor and receiver species (22 paralogs corresponding to the 11 Copia families). For 60% of the paralogs (13/22), the RT domain was assembled using our automatic INTERCHANGE pipeline. For the others, the RT domain was lacking. We then manually reassemble the lacking RT domains using raw Illumina reads of the corresponding species. For species for which the reference genome is available (F. sylvatica, F. exclesior, P. avium and A. glutinosa), we realigned the raw reads to the reference genomes and used the reference elements as a guide for manual assembly. Alternatively, we used homologs from other closely related plant species to guide manual assembly. Using this strategy, we obtained for most Copia paralogs involved in HTs nearly the complete elements with the corresponding RT domain (21/22).

Frequency estimates of the different Copia and Gypsy lineages

To estimate the relative frequency of Copia and Gypsy in the sequenced genomes, we aligned the raw genome reads of each species to a collection of protein sequences corresponding to the different known Copia and Gypsy lineages from the RexDB [34] database by Diamond Blastx (evalue 1e-5) [35]. The number of aligned reads on each superfamily and on each lineage was reported to the total number of aligned reads to estimate their relative frequency.

Plant sampling statement

The authors declare that the samples taken in the framework of this study comply with local and national legislation and that an authorization has been granted.