Introduction

Horizontal gene transfer (HGT) is widespread in prokaryotes. Comparative and phylogenetic analyses of eukaryotic genomes show that a considerable number of eukaryotic genes also result from HGT. However, mechanisms of HGT in eukaryotic organisms are poorly understood in comparison to gene transfer among the Prokaryota. Evidence of gene transfer from bacteria to the nuclei of multi-cellular eukaryotes is rare (Richards et al. 2006; Acuna et al. 2012). The close contacts frequently found in different kinds of symbiosis could promote HGT between species (Gao et al. 2014). One of the best studied examples of natural HGT from bacteria to plants is gene transfer from Agrobacterium sp. to plants. Agrobacterium-mediated HGT relies on a highly specific DNA transfer mechanism and the introduction of T-DNA genes with eucaryotic promoter sequences, which can be expressed in a wide variety of plants.

Plants with cellular T-DNA (cT-DNA) sequences can be considered as natural genetically modified organisms (natural GMO’s). Species of the genera Nicotiana (White et al. 1983; Intrieri and Buiatti 2001; Chen et al. 2014, 2018), Linaria (Matveeva et al. 2012, 2018; Pavlova et al. 2013), and Ipomoea (Kyndt et al. 2015) contain different types of cT-DNA structures. In Nicotiana, six cT-DNA types have been described (TA, TB, TC, TD, TE, gT), with some species having received several cT-DNAs in succession. We have argued that Agrobacterium-mediated HGT could create new species (Chen and Otten 2017). cT-DNA expression can confer new properties on natural GMO’s, as shown by opine synthesis in roots of Nicotiana tabacum (Chen et al. 2016). A search for additional natural transformants in the large number of available sequence data would provide a much better idea about the frequency of Agrobacterium-induced HGT, on the type and variation of cT-DNA structures, and on cT-DNA-induced changes in natural transformants.

Most cT-DNA sequences identified so far seem to originate from A. rhizogenes, with genes already described in various A. rhizogenes strains. However, cT-DNAs may also contain previously unknown T-DNA genes, or unusual combinations of them. N. tomentosiformis contains genes distantly related to the orf14 and agropine synthase (ags) genes, and typical A. tumefaciens and A. vitis genes resembling octopine synthase (ocs), vitopine synthase (vis), C-protein-like, and 6b. It also contains a large, previously unknown T-DNA gene, orf511 (Chen et al. 2014; Chen and Otten 2017), with unknown function. In Ipomoea, a gene distantly related to rolB and rolC was found (Kyndt et al. 2015). These findings suggest that Agrobacterium T-DNA diversity is greater than expected. The TBLASTN algorithm, which compares a protein query against a translated nucleotide sequence database, allows the detection of highly diverged gene sequences, which are undetectable at the DNA level. The aim of the present work was to search for new examples of HGT from agrobacteria to plants, using T-DNA-encoded proteins as queries against sequenced plant genomes and transcriptomes.

Results

Identification of dicot cT-DNA sequences in the Whole Genome Shotgun database

To date (April 12, 2019), data on the genomes of 275 species of dicotyledonous plants, including tobacco and sweet potato, are available in the NCBI database (O’Leary et al. 2016). This database was searched with selected T-DNA-encoded protein sequences (“Materials and methods”, Table 1). Apart from the earlier studied tobacco and sweet potato genomes (the Linaria genome is not published), we found homologues of proteins encoded by agrobacterial T-DNA genes in 23 species belonging to 17 genera, 12 families and 10 orders (Fig. 1). The data on their T-DNA-like sequences are summarized in Table 2. As a control for Agrobacterium contamination, we searched for vir sequences, which are located outside the T-DNA. No intact vir-like genes were found in the genome data of most species listed in Table 2. In Euphorbia esula, a VirB1-like sequence was found (30% identical to YP_001967531.1). This segment is 99% identical to WP_125244063.1 from Aquabacterium sp. contig PJAD010111136.1. Unexpectedly, the entire contig (56 kb) shows strong identity to various Aquabacterium sequences. Conversely, Euphorbia esula contains sus-like and orf18-like T-DNA sequences, but the fully sequenced Aquabacterium sp. does not. Therefore, the Euphorbia VirB1-like sequence is most likely due to contamination of Euphorbia DNA with Aquabacterium DNA. In Parasponia andersonii, virH1, virH2, and virF homologs were found, close to the PaT-DNA3 region, and associated with plant sequences, suggesting abnormal co-transfer of these vir genes with the T-DNA (see below).

Table 1 Query sequences for searching proteins encoded by T-DNA and vir genes
Full size table
Fig. 1
figure 1

Number of species with sequenced genomes in different plant orders of Eudicots and number of naturally transgenic species among them. Relation of orders is after APG IV (Angiosperm Phylogeny Group 2016). Orders with naturally transgenic plants are in bold. The numbers in parentheses denote the number of species with sequenced genomes and the number of species with cT-DNA. *Solanales and Lamiales include data on natural transformants described earlier. **Cucurbitales and Cornales only show T-DNA-like sequences in the TSA transcriptome database

Full size image
Table 2 cT-DNA genes detected by analysis of WGS database
Full size table

cT-DNA-positive contigs were analyzed in greater detail. Unfortunately, the quality of genome assembly was found to be variable among different species. In some cases (Eugenia uniflora, Euphorbia esula, and Silene latifolia), various contigs were shorter than 1000 base pairs, making it difficult to identify sequences surrounding the cT-DNA fragments, and to identify the original insertion sites. Nevertheless, the presence of homologous fragments in related species or subspecies strongly indicated their integration into the genome of the ancestral form. T-DNA-like sequences, present in groups of related species, were identified within the genera Arachis, Cuscuta and Juglans, in two subspecies of Humulus lupulus, in two subspecies of Silene latifolia, and in two related genera from the family Cannabaceae: Parasponia and Trema. In the following part we will describe the new natural transformants in detail, starting with the simplest cT-DNAs, which carry only opine or plast genes.

cT-DNAs with only opine genes

Several plant genera contain cT-DNA sequences which carry only opine genes. We will first discuss the well-known genera Arachis and Juglans, and then present the remaining ones.

Cultivated peanut (Arachis hypogaea) is an allotetraploid species whose ancestral genome is most likely derived from the A-genome species, A. duranensis, and the B-genome species, A. ipaensis (Kochert et al. 1996). Arachis monticola is a close relative of the domesticated peanut. It is the only other tetraploid species besides A. hypogaea in the genus Arachis. A. monticola might be an immediate wild ancestor of cultivated peanut (Guillermo et al. 2007), or a weedy form, segregated from cultivated peanuts (Pattee et al. 1998). A. duranensis contains two copies of a cucumopine synthase (cus)-like gene, one complete, the other truncated. The common parts are 99% identical. The full size cus homologs from A. duranensis cultivars P1263393 (Table 2) and V14167 are 99% identical. The partially deleted copies differ in length and localization (upstream or downstream) relative to the intact gene. Probably, the A ancestor carried two full-size genes, which diverged over time. A. ipaensis contains a strongly rearranged cus-like gene, showing a deletion, a replacement of part of the gene, and insertion of a large DNA fragment (0.5 Mb). Analysis of plant sequences next to the cT-DNA suggests that they result from different integration events. Apart from a cus gene, A. ipaensis also contains a mannopine synthase 2′ (mas2′) homolog. Both parents’ cT-DNAs are present in the peanut crop genome. However, the cT-DNA derived from A. duranensis contains only one full-size copy of the cus-like gene (Supplementary Fig. 1). The cT-DNA from A. hypogaea cultivar Shitouqi (CP030990.1) is 99% identical to that of cultivar Tifrunner (Table 2).

Juglans cathayensis (Chinese walnut), J. manshurica (Manchurian walnut), and J. sigillata (iron walnut) also contain cT-DNA sequences with intact, succinamopine synthase (sus)-like opine genes. These species are closely related (Stanford et al. 2000; Dong et al. 2017), and their cT-DNAs are 99% similar. J. regia is closely related to J. sigillata, but their exact relationships are not yet fully established. Interestingly, cT-DNA sequences were found in J. sigillata, but not in J. regia. The phylogenetically more distant species J. microcarpa, and J. hindsii do not contain cT-DNA (Supplementary Fig. 2).

Apart from Arachis and Juglans, homologs of opine genes were also found in Euphorbia esula (green spurge), Nissolia schottii (yellowhoods), Humulus lupulus (hops, used in beer production), Eutrema yunnanense (japanese horseradish or wasabi, used for seasoning), Psidium guajava (goyave), Silene latifolia (white campion), Dianthus caryophyllus (carnation), Cuscuta australis, and Cuscuta campestris (dodder). Humulus, Eutrema, Silene, Dianthus and Cuscuta contain intact opine genes. The opine genes of N. schottii and P. guajava are interrupted by stop codons. In Euphorbia esula, five contigs with sus-like sequences were found. The three longest sequences are mentioned in Table 2. However, none of the contigs covers the gene completely. No stop codons were detected within these fragments. Overlapping parts are 97–99% similar. Additional studies are required to determine the precise copy numbers and extent of these genes.

In Nissolia schottii, three contigs contain mikimopine synthase (mis)-like sequences. The similarity of the sequences surrounding the first two copies of the mis-like gene (Table 2) is about 80%. One mis copy is full-length, the other truncated, and 100% identical to the complete copy. The third copy is located in a different region, and 92% identical to the first and second copy. vis-like sequences from Humulus will be considered below, in comparison with genes from other Cannabaceae.

In the case of Eutrema, an intact ocs-like gene was found in E. yunnanense, but no T-DNA-like sequences were detected in E. salsugineum or E. heterophyllum. Psidium contains a full-size mannopine synthase 1′ (mas1′)-like sequence and a mas2′-like fragment. Silene is represented in the database with three genotypes belonging to two subspecies. They contain a large amount of cus-like genes or gene fragments with small sequence differences. One contig of S. latifolia subsp. alba (LHUT01012347.1) contains two copies of the cus gene in direct orientation, one full-size, the second with deletions. The relatively poor quality of the genome assembly may cause an artifical increase in cT-DNA copies. To clarify the situation about these multiple cus-like copies, molecular studies are required. Dianthus caryophyllus carries an intact cus-like gene. It is interesting to note, that carnation plants with blue flowers are a well-known example of a man-made GMO ornamental plant (Tanaka et al. 2009). The two Cuscuta species carry highly similar mis sequences (99% identity), surrounded by similar plant sequences. This indicates a common and recent origin.

cT-DNAs with only plast genes

Two species contain only plast-like genes (Otten 2018) on their cT-DNAs, Eugenia and Vaccinium. Eugenia uniflora (pitanga or Suriname cherry) (Nascimento e Santos et al. 2015) is a plant of the Myrtaceae family. It is native to tropical South America’s east coast, but is also grown in the West Indies and Florida (Morton 1987). Eugenia uniflora contains rolB-like and e-like sequences; both are truncated and contain stop codons. Unfortunately, the assembly quality of this genome is not very good. Therefore, the structure and localization of the cT-DNA are unclear. Vaccinium macrocarpon (American cranberry) belongs to the Ericaceae family, and forms a symbiotic relationship with ericoid mycorrhizal fungi. It has an single, apparently intact plast gene. Interestingly, the predicted protein sequence is more related to Plast proteins from the fungus Laccaria bicolor than to agrobacterial Plast sequences. This group also contains a protein from the bacterium Ensifer sp. YR511 (Fig. 2). In the following paragraphs we will present natural transformants with more complex cT-DNA structures, carrying both opine and plast genes.

Fig. 2
figure 2

Vaccinium Plast protein clusters with fungal Plast proteins. Molecular phylogenetic analysis of Plast proteins from Agrobacterium, Fungi and Vaccinium was done with the Maximum Likelihood method. The tree with the highest log likelihood (− 3418.91) is shown. The percentage of trees in which the associated taxa clustered together is shown next to the branches. The tree is drawn to scale, with branch lengths measured in number of substitutions per site. The analysis involved 16 amino acid sequences. All positions containing gaps and missing data were eliminated. There were a total of 144 positions in the final dataset. Evolutionary analyses were conducted in MEGA7 (Kumar et al. 2016)

Full size image

cT-DNAs with plast and opine genes in Cannabaceae

cT-DNAs were detected in three genera from the Cannabaceae family: Parasponia, Trema and Humulus. Whereas Humulus only contains opine gene homologs, Parasponia and Trema contain both plast and opine gene homologs and have complex multiple cT-DNA structures. Prunus andersonii contains no less than nine cT-DNAs, each being part of a long contig (Fig. 3). All PaT-DNA regions, except PaT-DNA9, are imperfect inverted repeats. Based on the degree of divergence between the repeats (indicated between brackets), we propose that PaT-DNA1 (78%), 4 (78%) and 8 (78%) were the first to enter the plant genome, followed by PaT-DNA5 (80%), 7 (82%), 6 (85%), 3 (88%) and 2 (89%). PaT-DNA9 cannot be dated by this method. The PaT-DNA3 and PaT-DNA4 sequences display large gaps, which may carry additional cT-DNA sequences.

Fig. 3
figure 3

cT-DNAs in Parasponia andersonii (Pa1 to Pa9) and Trema orientalis (To1 and To2). Dark brown arrows: plant genes, green arrows: T-DNA-like sequences, thin green arrows: cT-DNA inverted repeats. Orange arrows with numbers: genes found by database annotation. Thin blue and red arrows show different direct repeats

Full size image

PaT-DNA1 contains orf3, orf8, rolB, and orf14 homologs on each of the two arms. PaT-DNA2 contains sus and orf14 homologs on the two arms of its inverted repeat, the left part of PaT-DNA2 also contains a rolB homolog. The right arm orf14-like sequence is shorter than the left arm one. In addition to these T-DNA-like sequences, PaT-DNA2 contains two IS630-like transposase genes from Alphaproteobacteria, a complete gene on the left arm, and a partial one on the right arm. Bacterial insertion sequences have been detected in T-DNAs from various Agrobacteria (Machida et al. 1984; Otten et al. 1992; Fournier et al. 1993). The PaT-DNA2 IS630-like sequences were most likely part of the original T-DNA insert.

PaT-DNA3 contains c′-gene, rolC, orf13, sus, and agrocinopine synthase (acs) homologs on its left part. The right part is smaller and contains a sus-like sequence. About 2.5 kb to the right, a plant retrotransposon sequence is found, followed by a 3 kb region with Agrobacterium virF-, virH1- and virH2-like sequences, and a plant sequence with unknown function. Thus, the PaT-DNA3 insert is closely linked to a Ti plasmid fragment from outside the T-DNA borders. Another Parasponia contig, JXTB01000142.1 (850 kb), also contains a bacterial, non-T-DNA pTi sequence (492–5522), 69% identical to several fragments of pTi_Tun151 (KY000068). These code for an ABC transporter permease, a phosphodiesterase, and an arabinose phosphate phosphatase. This pTi-like sequence is connected to a plant sequence beyond nucleotide 5522. No T-DNA-like sequences were found on JXTB01000142.1.

The PaT-DNA4 left and right arm contain orf2-, orf3n-, orf8-, rolB-, and rolC-like sequences, and two different sus-like sequences (sus-like1 and sus-like2). The orf3n copy on the right is interrupted by a 34 kb plant sequence (910411–944678). Apart from its inverted repeat, PaT-DNA4 also shows several direct repeats (Fig. 3). The 6 kb long direct repeat A1 (892861–898343) is 99% identical to A2 (1899807–905285). The A repeats contain smaller, direct repeats (C2 and C3 for A1, C4 and C5 for A2, 94% identity), an isolated copy (C1) is found on the left part of PaT-DNA4. Repeat B2 (886476-887492), localized between the two cT-DNA arms, is 82% identical to B1 (880331–881415). B3 (893372–894466) is inverted with respect to B1 and B2.

PaT-DNA5 also has a complex structure, in which rolC-, orf2-like, orf3-like, and orf8-like sequences alternate with plant genes of unknown function, one coding for DNA polymerase, and one coding for an Ig-like domain-containing protein. In addition, PaT-DNA5 carries two IS66-like fragments (unrelated to IS630). Fragments A1, B1, C1 and A2, B2, C2 form an inverted repeat with 80% identity between the two arms, but do not seem to contain T-DNA-like genes.

PaT-DNA6 contains orf14- and sus-like genes on both arms of its inverted repeat. IS630-like sequences (one full-size, the other partial) are found between the orf14- and sus-like sequences and belong to the initial T-DNA structure. The PaT-DNA6 IS630-like sequences are different from the PaT-DNA2 IS630-like sequences (50% protein identity). An acs-like sequence and an additional sus-like gene are localized between the two arms, the latter has 63% and 66% identity to the sus-like genes of the left and right arm. PaT-DNA7 is a shortened version of PaT-DNA6, and is surrounded by similar plant sequences. It contains a sus-like gene surrounded by partial IS630-like elements in opposite orientation, 82% identical to each other. These sequences are surrounded by two inverted orf14-like sequences and an additional, partial orf14-like sequence.

PaT-DNA8 carries an unusual plast gene which encodes a protein with weak homology to protein D and other Plast proteins, one orf14-like gene, and two copies of a sus-like gene. It also carries a remnant of an orf511-like gene. An orf511 gene has sofar only been found in the N. tomentosiformis TD cT-DNA (Chen et al. 2014), its function is unknown. PaT-DNA9 contains an intact vis-like sequence.

Another naturally transgenic species from the Cannabaceae family is Trema orientalis, closely related to Parasponia. cT-DNA sequences were found in two contigs. ToT-DNA1 is similar to PaT-DNA8, the average identity of their matching fragments being 84%. ToT-DNA1 is organized as an imperfect inverted repeat, containing two copies of a truncated sus-like gene, and two copies of an orf14-like and d-like gene. Average similarity between the two T-DNA arms is 85%. ToT-DNA2 is 95% identical to PaT-DNA9 and contains a vis-like sequence. Most likely, the insertion events which gave rise to ToT-DNA1/PaT-DNA8 and ToT-DNA2/PaT-DNA9 predate the Trema/Parasponia separation. Nearly all cT-DNA genes of Parasponia and Trema are degenerated. However, the left PaT-DNA2 sus-like gene and the vis-like genes of ToT-DNA2 and PaT-DNA9 are intact, and may be functional. Additional studies are needed to confirm this.

Within the Cannabaceae family, Humulus contains a vis-like gene, like Parasponia and Trema, but Cannabis sativa does not. In two subspecies of Humulus, the vis-like gene is present in three copies. Two of them form a direct repeat, the third one is located in another contig. The vis-like genes of Parasponia, Trema and Humulus are highly similar (Supplementary Fig. 3). However, in Humulus the surrounding sequences differ from those in Parasponia and Trema. This indicates independent acquisition of the gene by Humulus and the Parasponia/Trema ancestor. The 6854-8448 fragment of Humulus contig BBPC01047313.1 is similar to Parasponia PaT-DNA7, the intact vis-like ORF (7250-8323) is located within this fragment. The 6854-8448 fragment most likely delimits the cT-DNA insert. The high similarity of the cT-DNA sequences in these species may be due to recent transformation by similar Agrobacterium strains.

cT-DNAs with plast and opine genes in Azadirachta indica, Quillaja saponaria, and Camellia sinensis

Azadirachta indica, or neem tree, has been used in folk medicine in India for over 2000 years (Kausik et al. 2002). Four Azadirachta indica contigs contain cT-DNA sequences. They are relatively short, but also contain plant sequences, demonstrating integration of T-DNA into the plant genome. These contigs contain sequences similar to orf8, orf14, and cus. Because the contigs are small, it is not possible to tell whether they are part of the same cT-DNA or located on different cT-DNAs. To elucidate the fine structure of the cT-DNA in Azadirachta indica, additional experiments are required.

Quillaja saponaria (soap bark tree) is a medicinal plant native to South America (Muravieva 1983). It contains rolB-like, orf13-like, e-like, and sus-like sequences. The e-like sequence is associated with a bacterial IS3-like sequence. The rolB-like gene on contig PVLG01028938.1 is situated at the right border of the contig, and therefore partial. A TBLASTN search with RolB did not yield additional Quillaja contigs, indicating that coverage may be incomplete. The quality of the assembly does not allow us to draw conclusions about the structure of the Quillaja cT-DNA(s).

Camellia sinensis is a species of evergreen shrub or small tree, and has been used for thousands of years to make tea. Its genome has been sequenced (Wei et al. 2018) and our search showed that it contains homologs of rolB, sus, and two acs-like genes. These sequences are located in the same contig, and organized as an imperfect inverted repeat of 5.3 kb, the common parts are 90% identical. No intact ORFs are found on this fragment.

Summarizing the analysis of the plant genomes, we note that opine genes are more common and better preserved. Most of the plast genes acquired stop codons and probably lost their function. The presence of intact Agrobacterium-derived ORFs in several natural transformants suggests that these genes may be expressed. This led us to analyze the TSA database to search for expressed cT-DNA genes.

Identification of dicot cT-DNA sequences in the Transcriptome Shotgun Assembly database

The Transcriptome Shotgun Assembly (TSA) database was searched for cT-DNA-like sequences with the BLASTX option, using the T-DNA protein query sequences (Table 1), and with BLASTN using the nucleotide sequences of natural transformants listed in Table 2. It should be noted that the lack of TSA sequences for a given species does not mean a lack of such sequences in the genome, because transcripts could be missing in the libraries if expression is limited to certain tissues or stages. For this reason, the results of this section should not be viewed as definitive, but as starting material for future research. At this stage, it is not possible to match the data of the fully sequenced genomes and the transcriptomes. Within some genera, some species have only sequenced genomes, while others have only sequenced transcriptomes. The TSA library was found to contain transcript sequences from some of the cT-DNA genes of the above-mentioned species, or from closely related species, these are summarized in Table 3.

Table 3 Expressed cT-DNA genes
Full size table

The TSA database analysis yielded representatives of seven additional dicot genera with cT-DNAs (Table 4). These are Aeschynomene evenia (shrubby jointvetch), Salix purpurea (purple willow), Luffa aegyptiaca (sponge gourd), Cyclocarya paliurus (sweet tea), Citrus maxima (pomelo), and Diospyros lotus (Caucasian persimmon, one of the oldest cultivated plants). Among these candidates, Diospyros lotus is especially worth mentioning. Its transcriptome contains opine genes and plast genes. They show similarity to acs, cus, sus, orf2, orf3n, orf8, orf13, orf14, orf17n and orf511 sequences. Some of these are combined into longer sequences, which may result from abnormal read-through transcription. Table 5 shows the identity percentages of extended areas among contigs. The TSA data require verification by additional methods, since they cannot exclude contamination with Agrobacterium DNA, and do not provide information on the location and structure of the inserts. The TSA analysis confirmed the predominance of opine genes, already noticed for the WGS data.

Table 4 New cT-DNAs in the genomes of dicots, based on TSA data
Full size table
Table 5 Sequence identity (%) between fragments of Diospyros lotus contigs and genes, located in overlapping fragments
Full size table

Search for monocot cT-DNA sequences in the WGS and TSA databases

As of June 2019, 73 monocot sequences were available in the WGS database. We searched these sequences in the same way as for dicots. T-DNA-like sequences were found in Dioscorea alata (greater yam), but not in D. rotundata (white Guinea yam). Greater yam is an important and geographically widely distributed staple food (Cormier et al. 2019). Although sweet potato (Ipomoea batatas) is also called yam, it is unrelated to Dioscorea alata. Dioscorea alata contigs CZHE02045212.1 (3009 nt) and CZHE02050078.1 (2550 nt) are 96% similar. CZHE02045212.1 potentially encodes an intact Cus-like protein, 85% identical to WP_071201425.1 from Agrobacterium vitis, CZHE02050078.1 encodes a truncated Cus-like protein. Another Dioscorea species, Dioscorea bulbifera (potato yam) was among the first monocot species to be transformed (Schäfer et al. 1987).

The TSA database contains 132 monocot sequences. We found cus-like sequences in Musa acuminata AAA group (dessert banana), in accession numbers JV331205.1 (157–1071), JV353951.1 (3–554), and JV360234.1 (8–316), with identity values to WP_071201425.1 of 64%, 57%, and 63% respectively. JV331205.1 is intact, the two other ones are truncated. Banana is one of the oldest cultivated plants, found on all continents.

Discussion

In 2012, we reported the identification of a new natural transgenic plant, Linaria vulgaris (toadflax) by analyzing more than a hundred species of dicotyledonous plants, using PCR primers designed from typical T-DNA gene sequences (Matveeva et al. 2012). The current study shows that a search for HGT sequences by bioinformatic methods is an order of magnitude more efficient. This is not surprising, since such an approach allows detection of highly diverged sequences, not possible with PCR primers. Our analysis of the WGS and TSA databases (in all, 631 dicot and 205 monocot species) revealed numerous examples of naturally transgenic plants. New T-DNA-like dicot sequences were found in representatives of 39 (23 + 16) species, 24 (17 + 7) genera, 17 (12 + 5) families, and 12 (10 + 2) orders. Previously, six species from these databases (N. tabacum, N. tomentosiformis, N. otophora, N. noctiflora, Ipomoea batatas and I. trifida) were already found to be transgenic. Thus, our data indicate that about 7% of the sequenced dicot species are naturally transformed. With an estimated number of 175,000–200,000 dicot species, this yields a minimum of 10,000 naturally transformed dicot species. Out of 205 available WGS and TSA monocot sequences, only those of Dioscorea alata and Musa acuminata contain T-DNA-like sequences.

Many of the cT-DNA sequences appear to be relatively small T-DNA fragments. They may result from incomplete transfer of T-DNA, or from partial cT-DNA deletion subsequent to the initial insertion event. In some cases, these fragments may have been amplified. Partial deletions of cT-DNA sequences are known for various cultivars of N. tabacum (Chen et al. 2014), and a cT-DNA duplication was detected in N. otophora (Chen et al. 2018). Many cT-DNA structures are inverted repeats (Chen et al. 2014), with typical LB-associated sequences at both ends. PaT-DNA1, 4, 6, and 7 from Parasponia are organized in the same way. Such structures are consistent with the proposed mechanism of T-DNA integration by polymerase theta (PolQ), which can link two T-DNA molecules with their 3′ ends (LB ends) to plant DNA breaks (van Kregten et al. 2016).

Apart from Agrobacterium and plants, various Fungi (Mohajjel-Shoja et al. 2011) and Rhizobium species (Chen et al. 2014; Chen 2016) also contain genes encoding Plast-like sequences. In the case of Fungi, these sequences could result from Agrobacterium transformation. In the case of the Rhizobia, these bacteria do not contain pTi/pRi sequences, and their Plast-like sequences cluster separately from the Agrobacterium ones (Chen 2016). Thus, their origin remains unclear. In most plant cases described here, the T-DNA-like sequences were phylogenetically closer to Agrobacterium sequences than to sequences from other taxons. However, in two species of the genus Vaccinium, a cT-DNA plast-like gene was found which is closer to fungal (Laccaria bicolor) and bacterial (Ensifer sp.) plast sequences. We have proposed (Mohajjel-Shoja et al. 2011; Chen and Otten 2017) that fungal plast genes are derived from transformation with an unknown Agrobacterium strain, because of their rare and patchy distribution among fungal groups. A similar Agrobacterium strain could then also be the source of the Vaccinium plast-like genes. Additional unusual types of T-DNA-like genes include a c’-like gene in PaT-DNA3, a d-like gene in PaT-DNA8 and a highly diverged acs-like gene in PaT-DNA6. The corresponding Agrobacterium strains remain to be identified.

Among the newly described naturally transgenic plant species, those containing only opine genes predominate (16 out of 23 species from the WGS database and 14 out of 16 from the TSA database). This might be due to several reasons. The first one is related to the T-DNA transfer mechanism. T-DNA transfer starts from the right T-DNA border, and is not always complete. Because opine genes like mas2′ (in A. rhizogenes strain 8196, Hansen et al. 1991), nopaline synthase (nos), cus, sus, mis, vis, and ocs are situated close to the right border, they are more likely to be transferred in case of incomplete transfer. Since opine genes are not known to favour regeneration, regeneration of such incompletely transformed cells most likely occurred spontaneously. The second possibility is that the opine genes were initially located on a T-DNA fragment together with plast genes like rolB and rolC, allowing formation of hairy roots with a high potential for regeneration. If plast genes reduced growth or fertility of the regenerant plants, they may have been lost by negative selection. Alternatively, if the initial transformation event involved the insertion of different T-DNA fragments, some with plast genes, the others with opine genes, the plast genes could have been lost by segregation. Some of the potential opine enzymes reported here, are only distantly related to known sequences and could produce new types of opines. Thus, their properties should first be extensively investigated in vitro, before searching the corresponding opines in the natural transformants. In addition, it will be necessary to determine where these genes are expressed by using reporter genes. In several N. tabacum cultivars, the TB-mas2′ gene is expressed in roots and leads to measurable amounts of opines (Chen et al. 2016).

Longer cT-DNAs with a combination of opine and plast genes, usually have an imperfect repeat structure. This was already noted for Nicotiana and Ipomoea, and could reflect a basic property of the T-DNA transfer system, or a higher stability of such repeats over longer periods.

Of particular interest are the complex cT-DNA sequences from Parasponia, Trema, Quillaja, Camellia and Azadirachta. Most of their genes are degenerated and carry stop codons. However, two opine genes of Parasponia have an intact ORF and may still encode opine synthesis, in spite of the rather large divergence between the repeats. In Parasponia, these divergence rates range from 12 to 22%. In Linaria, Nicotiana, and Ipomoea these are 8%, 1–6% and less than 1% respectively (Chen and Otten, 2017). As in the case of the TA, TB, TC and TD regions of N. tomentosiformis, the different Parasponia cT-DNA sequences are footprints of multiple transformation events, that took place well before the transformation of Linaria, Nicotiana and Ipomoea. PaT-DNA6 and 7 result from duplication of the same original insert, like the TE region of N. otophora (Chen et al. 2018), adding to the complexity of cT-DNA structure and function in natural transformants. The ancestor sequences of ToT-DNA1/PaT-DNA8 and ToT-DNA-2/PaT-DNA9 were inserted before the Parasponia-Trema divergence. PaT-DNA8 shows 22% divergence between the two arms, ToT-DNA1 only 15%. Thus, the divergence rate between the repeats was faster in Parasponia as in Trema. The insertion of ToT-DNA2/PaT-DNA9 seems to be more recent, with only 5% divergence. Parasponia andersonii and Trema orientalis are non-legume plants, which evolved nitrogen fixation capability through symbiosis with a large range of Rhizobium species (Op den Camp et al. 2012). The strong association between Parasponia/Trema and Rhizobiaceae and the close relationship between Rhizobium and Agrobacterium may have favored multiple T-DNA transformation events. In this scenario, Agrobacterium strains would have transmitted their Ri plasmids to Rhizobium strains, which then transferred T-DNA sequences to their host plants. It would be interesting to test whether some Rhizobia associated with Parasponia andersonii and Trema orientalis carry Ri plasmids, and whether such strains can lead to transformation and regeneration of their hosts. More generally, natural transgenic species may have a higher spontaneous regeneration capacity than non-transformed species.

PaT-DNA2, and PaT-DNA6 and 7 of Parasponia carry different IS630-like bacterial insertion elements, PaT-DNA5 shows an IS66-like sequence. In Quillaja, an IS3-like element is found close to an e-like plast gene sequence. IS elements can be easily transferred between bacteria, and their frequent insertion in the Agrobacterium genome, including the Ti plasmids, strongly contributes to Agrobacterium evolution, pTi structure and modification of T-DNA function (Otten et al. 1992). The presence of IS630-, IS66- and IS3-like sequences in natural transformants shows that Agrobacterium can transfer bacterial elements to plants in a two-step HGT process. In the first step, IS elements from other bacteria insert into a T-DNA (probably in a random way). In the second step, they are transferred to a plant as part of the T-DNA. IS elements are not expected to function in plants, as they lack plant expression signals. We could not find free IS-like elements in Parasponia, indicating that they did not transpose. These elements may have been transferred by chance, without influence on the plant, but they could have played a role in allowing efficient regeneration, by inactivating a T-DNA gene interfering with that process. Agrobacterium strain A66 for example, carries an IS66 insertion element in the auxin synthesis gene iaaH, which leads to shooty tumors (Binns et al. 1982; Machida et al. 1984). Apart from IS elements, we also found other agrobacterial sequences in Parasponia. One fragment carried vir genes and was situated close to a T-DNA, the other fragment carried Ti plasmid genes, unconnected to T-DNA sequences. These two cases represent another type of HGT, with a more or less random transfer of non-T-DNA sequences, probably through abnormal activity of the Agrobacterium DNA transfer system.

Transformed plants carrying non-T-DNA sequences like vir region DNA and vector backbone sequences have been reported by several authors (Ooms et al. 1982; Ramanathan and Veluthambi 1995; Kononov et al. 1997; Gelvin 2017). The more or less frequent occurrence of such abnormal structures may depend on the properties of the virulence genes. In the case of the natural GMOs, the original Ri plasmids and their vir genes are unknown. Since Agrobacterium can transfer large fragments of its chromosomal DNA via the T-DNA transfer system (Ulker et al. 2008), these sequences may also be found in natural GMOs. This interesting possibility has not been tested sofar.

The present findings expand the list of natural GMOs to a much larger number of plant taxa and suggest some directions for further research. One of these is the functional study of intact cT-DNA genes. Another concerns the variability and evolution of cT-DNA sequences in natural plant populations and in cultivated species. Among the new natural GMOs are several plants used for food, drinks and medicine, with large collections of accessions and cultivars already available, and rapidly increasing amounts of sequence data. Such data will provide excellent material for studies on cT-DNA evolution. Based on our results, it is clear that throughout their history, almost all human cultures have encountered natural transformants, which they adopted for food, drinks, medicine or decorative purposes.

Materials and methods

Identification of cT-DNA

In order to detect new cT-DNA sequences, we performed a 4-step blast search. In the first step, representative protein sequences of A. rhizogenes oncogenes, their homologs from Ipomoea and Nicotiana plants, from the fungus Laccaria bicolor and protein sequences of opine genes of different strains of Agrobacterium sp. (Table 1) were recovered, and used as queries to search the National Center for Biotechnology Information (NCBI) Whole-Genome Shotgun (WGS) contigs of all plant genomes sequenced to date, using the TBLASTN algorithm. In the second step, Vir protein sequences (Table 1) were used to search for possible Agrobacterium contaminations in those genomes, where T-DNA-like sequences were detected. In case homologs of vir genes were detected, the surrounding sequences were studied. When plant genes were found to be linked to vir genes, the hypothesis of contamination was rejected. In the third step, contigs that potentially encoded T-DNA-like protein sequences with identity levels 30% or higher, were analyzed further. They were used as queries in BLASTX to detect the closest protein homologs and to identify proteins encoded by plant genes surrounding the cT-DNA. The resulting cT-DNA maps (based on sequence similarities) were mapped to annotated sequences from the same plant species, wherever possible. The Vector NTI AdvanceTM software was used to build the combined maps. In the fourth step, the TSA database was used to search for expressed new cT-DNA genes, using sequences described in the third step as a query. In addition to this, the TSA database was used to search for cT-DNA transcripts, as described for step one.

Phylogenetic analysis of cT-DNA sequences

Phylogenetic analysis of cT-DNA sequences was done in MEGA 7.0 (Kumar et al. 2016). Evolutionary history was inferred by using the Maximum Likelihood method based on the Tamura-Nei model (Tamura and Nei 1993). The bootstrap consensus tree inferred from 500 replicates is taken to represent the evolutionary history of the taxa analyzed (Felsenstein 1985). Initial tree(s) for the heuristic search were obtained automatically by applying Neighbor-Join and BioNJ algorithms to a matrix of pairwise distances estimated using the Maximum Composite Likelihood (MCL) approach, and then selecting the topology with superior log likelihood value.