Introduction

The term genetically modified (GM) organism represents an organism ‘in which the genetic material has been altered in a way that does not occur naturally by mating and/or natural recombination’1. Plant individuals developed through use of transgenic techniques have, therefore, been regarded as GM organisms, partly due to absence of clear evidence for natural occurrence of such genetic modification2,3. The consequent stringent regulation has been a hurdle for commercialisation of transgenic technology-derived products, and equivalent restrictions are likely to be imposed on emerging gene-edited crops, particularly those with transgenes, despite the presence of scientific evidence indicating that potential risks of such GM products on human health and ecology do not substantially differ from those of non-GM counterparts2,3,4,5.

Horizontal gene transfer (HGT) is a known evolutionary mechanism in both prokaryotes and eukaryotes, permitting acquisition of adaptation novelty from taxonomically distant species6. Such events in multi-cellar eukaryotes are, however, less frequently documented, and identification of novel candidates from eukaryote genomes is required for understanding molecular basis of the mechanism, and the resulting effect on evolution in eukaryote lineages7. Angiosperms (flowering plants) represent multi-cellar plant spices, which relatively recently diverged [around 200 million years ago (MYA)] from other land plant species in the evolutionary time8. Following to the appearance, diversification of the lineages occurred during the last 100–125 million years, and the taxa is currently composed of around 300,000 spices, dominating both natural and cultivated landscapes9. Due to the importance as primary food source for human and animals, angiosperms have been major targets for genetics and genomics, especially for the crop breeding purposes10. The recent advances in DNA sequencing and computational technologies facilitated whole genome and transcriptome sequencing of angiosperm species, despite their expanded genome sizes, typically from a couple hundred mega to over ten giga base pairs in length, and complexity of transcriptome in them11,12.

Investigation of horizontally transferred genes has been commonly performed through a comparison of DNA sequences from taxonomically distant species, and a subsequent phylogenetic analysis, for identification of DNA sequences with an unusually high identity between the two distant lineages13. In plants, mitochondrial genes (sequences) transferred from other plants have been relatively frequently reported14. The genus Amborella represents a basal lineage of angiosperms, and, in amborella (Amborella trichopoda Baill.), extensive ancestral horizontal transfer of mitochondrial DNA genes was inferred15. The amborella mitochondrial genome was completely sequenced, to confirm the presence of an unusually large number of genes transferred from green alga, moss and angiosperm mitochondria. Due to the scale of integration, a mitochondria fusion-based model was proposed for the extensive HGT case16. Another well-characterised example in flowering plants is the genes transferred through Agrobacterium-based conjugation. The Agrobacterium-based event was originally identified in tree tobacco (Nicotiana glauca Graham, Solanaceae family)17. From the tree tobacco genome, a sequence showing a similarity to the Agrobacterium Ri plasmid was found, and multiple gene(-like) sequences were located on the Agrobacterium-derived insertion. Similar Agrobacterium-derived sequences have been identified from 15 Nicotiana species, suggesting multiple HTG events from Agrobacterium during the recent evolutionary time18. Agrobacterium-originated genes have also been identified in common toadflax (Linaria vulgaris, Plantaginaceae family) and sweet potato (Ipomoea batatas (L.) Lam, Convolvulaceae family)19,20. Recent studies have revealed potentially frequent nuclear gene exchanges between angiosperms, especially between host and parasitic species21. From purple witchweed (Striga hermonthica (Delile) Benth., Eudicots clade), which may infest plants in the Monocot clade, the ShContig9483 gene was identified, showing an unusually high DNA sequence similarity to sorghum (Sorghum bicolor L.) and rice (Oryza sativa L.) genes. At the 3′ end of the ShContig9483 sequence, consecutive adenine (A) nucleotides were found, suggesting a possibility of mRNA-mediated HGT from a Monocot species22. Through a systematic analysis for parasitic plants of the Orobanchaceae family, a total of 42 genes were identified as HGT candidates between parasitic and host plants, in addition to previously reported candidates21. Due to conservation of the intron positions between putative decedents of gene donor and recipient, a proportion of the candidates were suggested to have been integrated through a transformation-like mechanism. Among the 42 HGT candidates, 9 sequences were annotated as plant defence-related genes, suggesting that those genes may have contributed to pathogen resistance and/or parasite-host interaction.

No nuclear gene transferred from non-plant species into angiosperms has, however, been reported until recently23. In a previous systematic study for genomes of representative angiosperms [Arabidopsis thaliana (L.), rice, sorghum and poplar (Populus trichocarpa Torr. & A.Gray ex. Hook.)], no evidence for an HGT event from the fungal linage was found, in contrast to identification of two and three candidates in moss and lycophyte lineages, respectively13. It had, therefore, been believed that such an event hardly or never happened in the angiosperm lineage. The β-1,6-glucanase-like gene (LpBGNL) of perennial ryegrass (Lolium perenne L.; Poeae tribe, Loliinae sub-tribes) was the first documented candidate for an ancestral nuclear gene transfer event from non-plant species into the angiosperm lineage. A high sequence similarity (90%) to the corresponding gene of Epichloë species inferred that the gene was transferred from an ancestral species of the Epichloë taxon, which composes a sister group to the ergot fungi (the genus Claviceps)24. As asexual Epichloë species may establish symbiotic relationships between the Loliinae and Dactylidinae species, close physical contact may have facilitated the HGT. More recently, a glutathione S-transferase gene (Fhb7) was identified from a distant relative of wheat, Thinopyrum elongatum25,26. This gene showed up to 97% DNA sequence similarity to the corresponding genes in Epichloë species, suggesting another transfer event from the Epichloë lineage. Due to contribution to fungal-pathogen resistance, especially against Fusarium, the Fhb7 gene is expected to be valuable for wheat breeding.

In the current report, identification of additional HGT candidates in cool-season grass species is described. The genes are likely to have been transferred from the Epichloë lineage, and the identification of the non-plant species-originated genes in fully-sequenced crop species evidences that modern biotechnology-like genetic modification has naturally happened in food crop species. Characterisation of such 'natural' adaptation transgenes may increase public acceptance of advanced genetic improvement technologies.

Results

An Illumina short-read sequencing library was prepared using short RNA from an endophyte-absent (E−) perennial ryegrass individual. To catalogue expressed non-polyadenylated [poly(A)] RNA from perennial ryegrass, the sequencing library was analysed on the Illumina MiSeq platform. A single Illumina MiSeq run generated a total of 8,216,014 reads. Subsequently, a dataset of unique short RNA reads, including 1,424,274 sequences, was generated (Supplementary Information 1). To identify perennial ryegrass genes potentially derived from the fungal taxa, a BLAST search of the perennial ryegrass (E−) transcriptome shotgun assembly (TSA) and unique short RNA read datasets was performed against the Epichloë festucae transcriptome23,27. Totals of 88 and 121 sequence similarity hits were identified from the TSA and unique short RNA read datasets (Supplementary Information 2 and 3). A manual examination was performed using the NCBI BLAST function, to eliminate sequences derived from microbiome, those related to highly conserved genes between the two taxa, such as actin and ubiquitin genes, and low confident matches. The manual examination initially excluded 116 hits due to low sequence similarity. A total of 76 hits were related to highly conserved genes between eukaryotes, and 12 hits were concluded to be contribution from plant-related microbiome. After exclusion of those hits, 5 were found to be putative HGT-related hits, and those hits were corresponding to two E. festucae genes, and the β-1,6-glucanase gene (Supplementary Information 4). From the TSA data, a sequence [unique identifier (UI): ID_150936_C1449060_17.0] showing a relatively high similarity (85%) to an E. festucae gene (UI: EfM3.066060.partial-2.mRNA-1) was identified. As the predicted product of the Epichloë gene showed a sequence similarity to a fungal transcriptional regulatory (FTR) protein (Genbank UI: CRL18938; identity: 48%, e-value: 2e−150) of Penicillium camemberti (Ascomycota, Trichocomaceae), the perennial ryegrass gene was designed LpFTRL (L. perenne FTR-like). The other candidate was identified from the non-polyadenylated RNA unique reads (Sequence UI: 734684-1), and designated LpDUF3632, due to a similarity (96%) to the E. festucae DUF3632 (domain of unknown function 3632)-like gene (UI: EfM3.028800.mRNA-1). Sequence contigs containing LpFTRL and LpDUF3632 from shotgun whole genome sequencing assembly data of the perennial ryegrass Impact04 genotype (E−) were identified23. The contigs were deposited into the NCBI GenBank (UIs: MG680924 for LpFTRL and MG680923 for LpDUF3632), of which sizes were 17.2 kb and 6.1 kb in length, respectively. Although in the E. festucae (Strain: Fl1) genome contigs, flanking genes to the FTRL and DUF3632 sequences were identified within distances of 1–2 kb and 3–5 kb, respectively, no such flanking gene was predicted from the perennial ryegrass contigs. Instead, plant repetitive element-like sequences were identified in both contigs. A putative intron was identified for LpDUF3632, while no intron was predicted for LpFTRL.

A further in silico examination was performed to eliminate a possibility of DNA sequencing artifacts or contribution from microbiome. Through a BLASTN search against nucleotide sequences catalogued in the NCBI GenBank database, relatively high sequence similarity matches against LpFTRL were identified from Tausch's goatgrass (Aegilops tauschii L., UI: XM_020322891.1) and barley (UI: AK375773.1), followed by ascomycota species, while only fungal genes showed a significant sequence similarity to LpDUF3632 (Table 1). From publicly available barley and hexaploid wheat genome sequences, putative orthologues for LpFTRL were identified in barley chromosomes 1H and 7H (Ensembl UI: HORVU1Hr1G009870 and HORVU7Hr1G108080, respectively), while the corresponding sequences were found from chromosome 7 of each sub-genome of hexaploid wheat (TraesCS7A02G564400, TraesCS7B02G375200, and TraesCS7D02G460600). Significant matches to LpFTRL was also identified from the NCBI short read archive (SRA) data of cereal rye (Secale cereal L.) chromosome 7 (SRA UI: ERX140518). The LpDUF3632 sequence was subjected to a BLASTN search against NCBI SRA data from cool-season grass species. Significant sequence matches were found from Italian ryegrass (L. multiflorum L.) and tall fescue (Festuca arundinacea L.), but not from orchard grass (Dactylis glomerata L.) or Antarctic hairgrass (Deschampsia Antarctica É.Desv.) (Fig. 1 and Supplementary Information 5). Significant matches to LpFTRL were found from SRA data of cool-season grass species, except for transcriptome shotgun sequencing SRA of harding grass (Phalaris aquatica L.). As a control experiment, LpBGNL, the perennial ryegrass architecture candidate genes, LpABCG5 and 6 (GenBank UIs: JN051254.1 and JN051255.1), and an Epichloë-specific gene, makes caterpillars floppy (mcf)-like gene (EfMCF, GenBank UI: KJ502561.1) were sought to find significantly matching sequences for LpBGNL, and LpABCG5 and 6, but no sequence for the mcf-like gene from each SRA dataset28,29. Sequences corresponding to the two candidate genes were sought in genomes of other fully-sequenced angiosperm species, Brachypodium [B. distachyon (L.) P. Beauv.], rice, sorghum, Zea Mays (L.) and A. thaliana, and no significant match was, however, obtained. To confirm the gene presence/absence status, a PCR-based assay was performed. Presence of the DUF3632-like sequence in Loliinae species [perennial ryegrass, darnel (L. temulentum L.), tall fescue, and sheep fescue (F. ovina L.)] was confirmed, while no PCR amplification was observed from orchard grass and harding grass. Presence of FTRL-like sequence was confirmed in all tested Poeae species, including harding grass (Fig. 1, and Supplementary Information 6).

Table 1 Result of the DNA sequence similarity search for LpFTRL and LpDUF3632. ‘GPUK’ denotes for genome project at University of Kentucky. ‘N.A.’ stands for ‘not applicable’.
Full size table
Figure 1
figure 1

Presence/absence status of the LpFTRL and LpDUF3632-like sequences. The divergent point of plant species from other species (fungi and animals) is indicated with the filled hexagon (> 1,000 MYA), and the filled circle indicate the divergent point (32–39 MYA) of species belonging to the Triticeae and Poeae tribes from other Poaceae species. The filled diamond indicates the divergent point (59 MYA) of Epichloë species from Claviceps species. The triangles represent species belonging to the Loliinae or Poa subtribe, or Epichloë or Claviceps genera. The presence/absence status of putative homologues (orthologues) is shown on the right side of species/subtribe/genius name, in witch ‘In silico’ and ‘PCR’ denote the database and PCR-based screening results, respectively. The plus( +), minus( −), and ‘N/A’ denote ‘presence’, ‘absence’, and ‘not analysed’, respectively. Although results of database-based screening for LpFTRL-like sequences in harding grass and one of Poa species were negative, indicated with an asterisk(*), the PCR-based assay indicated presence of orthologues in harding grass. Further details about database-based screening can be found in Supplementary Information 5. The putative periods of the ancient HGT events for FTRL, DUF3632, and BGNL are shown with red arrows.

Full size image

A phylogenetic analysis was performed using the FTRL and DUF3632 sequences from plant and fungal taxa. Close relationships of plant FTRL and DUF3632-like sequences between corresponding Epichloë sequences were revealed (Fig. 2). An LpDUF3632-derived DNA-based marker was developed, and a genetic linkage analysis was performed, using the p150/112 reference linkage mapping population of perennial ryegrass (Supplementary Information 7). The marker was assigned into perennial ryegrass linkage group (LG) 3 (Fig. 3), which corresponds to the chromosome 3 of Triticeae species.

Figure 2
figure 2

Phylogenetic tree of Epichloë-Triticeae/Poeae HGT candidates and corresponding fungal genes. The phylograms for LpFTRL (a) and LpDUF3632 (b) generated based on predicted amino acid sequences. Sequences from plant species are shown in red. For LpFTRL, the sequence from Trichoderma was selected as an outer group to obtain a root of the phylogenic tree (a). For LpDUF3632, amino acid sequences from Tolypocladium and Colletotrichum were selected as an outer group. A DUF3632-like gene was found from Periglandula ipomoeae, which is shown with an asterisk(*) (b). For fungus sequences, strain and sequence identifiers of the Genome Project at the University of Kentucky are shown in brackets. For plant sequences, NCBI GenBank UI is shown in brackets.

Full size image
Figure 3
figure 3

Genetic linkage mapping of LpDUF3632. The LpDUF3632-related DNA marker is shown in red. Genetic distance is shown on the left side of the linkage map.

Full size image

A gene expression analysis was performed, based on a read count approach of short read sequencing data. From the E− perennial ryegrass transcriptome data23, a leaf tip-specific expression pattern of LpFTRL was proposed (Fig. 4a and Supplementary Information 8). Although the counts per million reads (CPM) were relatively low compared with those of LpBGNL, LpDUF3632 was ubiquitously expressed (Fig. 4b). The SRA data from endophyte-infected (E+) and E− perennial ryegrass plants, deposited by Massey University, were also subjected to the analysis. The SRA data did not include a leaf tip sample, and the expression level of LpFTRL was below the limit of detection in most tested tissues, regardless of the E+/E− status. Expression of the corresponding Epichloë gene (EfM3.066060) was not also observed in most tested tissues of the E+ plants (Fig. 4c and Supplementary Information 9). Expression of LpDUF3632 was observed in all tested tissues, and the expression patterns were similar between E+ and E− plants. Expression of the Epichloë DUF3632 gene (EfM3.028800) was detected from all tested tissues of the E+ plant, with a relatively high expression level in sheath (Fig. 4d). From short-read transcriptome sequencing data of the E+ and E− perennial ryegrass seeds/young seedlings23, an increment of LpDUF3632 expression level was observed in E+ seeds after a germination-treatment, although the expression levels of LpFTRL, EfM3.066060, and EfM3.028800 stayed low throughout the tested 10 days (Fig. 4e,f, and Supplementary Information 10).

Figure 4
figure 4

The expression levels of HGT candidates and corresponding fungal genes in perennial ryegrass. (a, b) The expression levels of LpFTRL and LpDUF3632 in each tissues of the perennial ryegrass Impact04 genotype. (c, d) The expression levels of the HGT candidates and corresponding fungal genes in E+ and E− perennial ryegrass genotypes. (e, f) The time course gene expression analysis for the HGT candidates and corresponding fungal genes in perennial ryegrass seeds and young seedlings. The x-axis shows tissue types (a–d) or period after a germination treatment(e, f), and normalised read count numbers [counts per million (CPM)] are indicated on the y-axis. The green and red bars or lines show the expression levels of the HGT candidates in E− and E+ plants, respectively, and the blue bars or lines show the expression levels of corresponding fungal genes in E+ plants. Note that the scales of the y-axes are not uniformed.

Full size image

Nucleotide substitution ratio was calculated for the horizontally transferred genes. The synonymous substitution rate (Ks/million years) of the plant FTRL sequences was not substantially different from those from fungi sequences (Table 2). Although some variation was observed in the non-synonymous substitution rates (Ka/million years), the rate from the plant sequences was between those from the Claviceps-Epichloë and E. gansuensis-other Epichloë combinations. The Ka/Ks ratio for the plant FTRL sequences were 0.376, which was close to that of the E. gansuensis-other Epichloë combination. A comparison with E. festucae DUF3632 indicated a similar Ka/Ks ratio (0.31) for LpDUF3632 (Supplementary Information 11). Comparison of codon usage of LpFTRL orthologues from diploid plant species and corresponding Epichloë sequences revealed a similar usage ratio for each amino acid, except for the termination codons (Fig. 5). Only ‘TGA’ was used in the plant genes as termination codon, the ratio of ‘TGA’ was only 38.5% in the Epichloë sequences.

Table 2 Ka and Ks rates for LpFTRL orthologues and corresponding Epichloë and Claviceps genes. The ratios calculated from the Claviceps and Epichloë, Claviceps and Plants, Triticeae and Poeae (plants), and E. gansuensis and the rest of Epichloë combinations are shown as Claviceps-Epichloë, Claviceps-Plants, Triticeae—Poeae, and E. gansuensis-other Epichloë, respectively. As Epichloë and Claviceps were diverged around 58.8 MYA followed by horizontal transfer of the FTRL sequence into plant species after 39 MYA, the comparison of the Claviceps and plants combination include the diversification period in the Epichloë lineage (from 58.8 MYA to the HGT event), while the Triticeae and Poeae combination merely represent diversification between plant species.
Full size table
Figure 5
figure 5

Visualised codon usage ratio for 18 amino acids and the termination signal. The y-axis shows the ratio, and ‘B’, ‘P’, ‘E’, and ‘C’ on the x-axis denote for codon usages in ‘barley (Hordeum vulgare subsp. vulgare) 1,490 coding sequences (CDSs)’, ‘plant FTRL genes’, ‘Epichloë FTRL genes’, and ‘Claviceps purpurea 37 CDSs’, respectively. Each color (dark blue, orange, grey, yellow, light blue or green) represents the ratio of each codon, of which DNA sequence is shown on the right side of data from the C. purpurea 37 CDSs. The usage information of barley and C. purpurea CDSs was obtained from the Codon Usage Database (https://www.kazusa.or.jp/codon/), and shown in the figure as codon usage reference in plant and fungal species.

Full size image

Pearl barley, wheat plain flour, rolled oat, and rice flour products were obtained from retail shops, and DNA was extracted from the retail products. PCR primers for the LpFTRL sequence within a conserved region between plant and Epichloë species were designed. Through PCR, the sequences corresponding to LpFTRL were amplified from barley, wheat, oat, and E. festuca gnomic DNA (gDNA) templates, but not from rice, confirming the presence of the fungus-originated gene in Triticeae and Poeae cereal species (Fig. 6, Supplementary Information 12). A control experiment with fungi-specific primers confirmed absence of gDNA of Epichloë and Claviceps species, especially the rye ergot fungus, C. purpurea, in the retail products24,29.

Figure 6
figure 6

PCR-based detection of LpFTRL-related gene from retail products. (a) Retail products of barley (i), wheat (ii), oat (iii), and rice (iv). (b) PCR assay results were visualised on an agarose gel, using the primers designed for LpFTRL. The size (bp) of PCR amplicons is indicated on the right side of each image. As a control experiment, PCR primers for the florigen candidate gene (FT in wheat and barley, and Hd3a in rice) were used. For demonstration of absence of Epichloë and Claviceps species in the retail products, PCR primers specific to the fungal species (EfMCF_F and R, and C.purpurea_D0288F and D0289R) were used. The gDNA samples from the perennial ryegrass genotype Impact04 and E. festucae were used, as positive controls for amplification with the plant and fungus-specific primers, respectively. With the PCR primers specific to Claviceps species, amplification from E. festucae gDNA template was observed, presumably due to sequence similarity between Epichloë and Claviceps species. ‘NTC’ stands for ‘no DNA template control’.

Full size image

Discussion

In the current study, identification of fungus-originated genes from crop species has been reported. The possibility of DNA sequencing artifacts or contribution from microbiome has been excluded through a rigorous approach. The identification of plant repetitive element-like sequences in the perennial ryegrass genome contigs (GenBank UIs: MG680924 and MG680923) suggested that LpFTRL and LpDUF3632 are located in the plant genome. A BLASTN search for LpFTRL indicated presence of putative orthologues in Tausch's goatgrass, barley, wheat, and cereal rye. Similarly, LpDUF3632-like sequences were identified from SRA data of Italian ryegrass and tall fescue. However, no similarity hit was obtained from Brachypodium, rice, sorghum, Zea Mays (L.) and A. thaliana. The genome sequences and SRA data described above were generated at divergent institutes and organisations, and the lineage specificity of the genes, hence, suggested ancestral HGT, rather than that the two sequences were derived from DNA sequencing artifacts, mis-annotation or contributions of microorganisms.

Although the in slico screening suggested absence of an FTRL-related sequence in harding grass transcriptome, PCR amplification was observed from the harding grass gDNA template (Fig. 1). The tissue-specific expression pattern of LpFTRL suggested a possibility that the corresponding sequence is present in the harding grass genome, and the expression level was, however, below the limit of detection in the leaf tissues, which were used for transcriptome sequencing. Therefore, it is likely that the FTRL-related sequence is also present in harding grass.

The phylogenetic analysis suggested a possibility that HGT events occurred after diversification of ancestral Epichloë species from the Claviceps lineage around 59 million years ago (MYA)29. Molecular taxonomy and the estimated divergence ages suggested that the two genes were transferred from the fungus lineage into plant species, but not vice versa. Based on the divergence age of the Brachypodieae tribe from other Pooideae species, the FTRL gene was predicted to have been transferred into plants after 32–39 MYA30. A formation of a cluster of the plant FTRL sequences suggested a single transfer event, rather than multiple events. It is, therefore, possible that the fungus gene was transferred into the chromosome 7 of a common ancestor of Triticeae and Poeae species, followed by gene duplication in barley genome. The DUF3632-like sequence is likely to have been transferred into a common ancestor of Loliinae species, independently of the FTR and β-glucanase(-like) gene, as the DNA-based marker for LpBGNL has been assigned into perennial ryegrass LG 523. The phylogenetic tree of the DUF3632(-like) genes suggested a possibility that the ancestral gene transfer event occurred after diversification of modern Epichloë species (7.2 MYA)29. These results indicated a possibility of at least three independent HGT events since 39 MYA. Compared with these three events, it is possible that the Fhb7 gene in T. elongatum was transferred more recently. The Fhb7 gene is likely to have been transferred after T. elongatum diverged from wheat, which was estimated to have been 4.77–4.96 MYA26. Comparison of genomic sequences including DUF3632-like genes suggested a transformation-like ancestral HGT event, due to conservation of the position of the putative intron between plant and Epichloë species. The transformation-like events have also been suggested for LpBGNL and Fhb723,26.

Identification of LpFTRL facilitated a unique opportunity to compare nucleotide substitution rates between fungal and plant lineages, using the sequences diverged only after 59 MYA (the Claviceps-Epichloë divergence period). Interestingly, the Ks/million years ratios of the plant FTRL sequences and counterparts in fungi were not substantially different (Table 2). The Ka/Ks ratios for the plant FTRL (0.376) and corresponding E. gansuensis-other Epichloë sequences (0.3619) were relatively close, suggesting that both plant and fungi genes have been subjected to selection pressures. The Ka/million years and Ks/million years rates from the E. gansuensis-other Epichloë combination were higher than the other combinations, which could be partly attributed to the shortest divergence period (7.2 million years) in the tested combinations. The Ks and Ka values for LpFTRL and LpDUF3632 proposed that plant FTRL and DUF3632 genes retain functionalities at the protein level, and the genes have contributed to natural adaptation of the plant species. The intron sequences of LpDUF3632 were much less conserved than the exon sequences, likely due to lower levels of the pressure, which implies that the selective pressures have been an important factor for retaining of sequence similarity to the corresponding Epichloë sequences. This hypothesis suggests a possibility of occurrence of additional unrevealed ancient HGT events, of which transferred genes (or sequences) have, however, been eliminated from plant genomes during the subsequent evolutionary process, due to lack of adequate levels of the pressure. The close physical contacts between the host plants and symbiotic and/or pathogenic fungi could have facilitated the relatively frequent transfer events21. As both Epichloë and Claviceps species infect plant reproductive tissues24,29, the presence close to germ cells might have been an important factor of the multiple HGT events31.

The ‘natural’ transgene was detected from a range of grain products. Cultivation of barley, rye, and wheat, including emmer and einkorn wheat, began over 10,000 years ago, and domestication of oat was completed by around 3000 years ago32,33. Compared with those, GM technologies and products have substantially shorter histories, shorter than 50 years since the first report of genetic engineered bacteria, and 25 years since the first approval of GM commercial food products34,35. The public concerns have been one of major factors involved in the regulatory decisions36,37. The identification of the ‘natural’ transgenes in staple species with long cultivation histories, however, may increase public acceptance of GM and gene-edited crop products19. Lateral acquisitions of nuclear genes, especially from taxonomically distant species, have been rarely documented, and it has been believed that such ‘natural’ genetic modifications never happened in flowering plants13. With the previously reported ‘natural’ Agrobacterium-mediated conjugation(-like) events in tobacco and sweet potato19, the identification of transformation(-like) events questions the traditional scientific perception. Occurrence and frequency of HGT may have been largely depending on the closeness and frequencies of physical contacts of the gene-donor, typically microbiome and parasitic plants, to germ cells of the recipient plants21,31.

Materials and methods

Non-poly(A) tailed RNA sequencing

Fresh young leaves of an individual of perennial ryegrass cultivar Trojan were harvested. Total RNA including small molecules was extracted using the RNeasy Plant mini kit (QIAGEN) following the modified protocol of related products (Purification of miRNA from animal cells using the RNeasy Plus Mini Kit and RNeasy MinElute Cleanup Kit; https://www.qiagen.com/nl/resources/download.aspx?id=4a299ac7-4a55-4ae0-aaca-f3ae8f66d9c1&lang=en). A sequencing library was prepared using the Small RNA sequencing library preparation kit (New England Biolabs; NEB), and a fraction containing non-coding RNA molecules (140–150 bp in length, including sequencing adapters) was purified using the BluePippin platform (Sage Science). The products were characterised on the 2200 TapeStation instrument using the D1000 kit (Agilent). The library was loaded on the MiSeq platform (Illumina), following the manufacture’s instruction, and a sequencing analysis was performed with the MiSeq Reagent Kit v3 (150-cycle) kit (Illumina). The outcome data were processed with the FastX-tool-kit package (https://hannonlab.cshl.edu/fastx_toolkit/), to remove adapter sequence and generate the unique read dataset.

BLAST-based screening

The transcriptome dataset of E. festucae (file name: M3 transcript sequences) was obtained from the website of Kentucky university (https://www.endophyte.uky.edu). The TSA (GFSR01000001-GFSR01044773) and non-poly(A) tailed RNA unique read datasets of perennial ryegrass were prepared for the DNA sequence homology search. The homology search was performed with the megablast function of the BLAST + package (https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE_TYPE=BlastDocs&DOC_TYPE=Download), and the threshold E value was set at 1 e−10. The resulting data was imported into the Microsoft Excel software for the manual examination. The manual examination was performed using the NCBI BLAST tools (https://blast.ncbi.nlm.nih.gov/Blast.cgi).

The homologues sequence in representative flowering plant species were performed on the Ensembl Plants website (https://plants.ensembl.org/index.html). The LpFTRL sequence was subjected to BLAST-based search against cereal rye SRA dataset (BioProject UI: PRJEB3229). The LpDUF3632 sequence were subjected to the search against the following SRA datasets; Italian ryegrass (SRA UIs: SRX1604870 and SRX1604871), tall fescue (SRX1056957), orchard grass (ERX1842528 and SRX738187), Antarctic hairgrass (SRX465632), Poa species (SRX745831, SRX745855, and SRX745858), and harding grass (SRX669405).

DNA sample preparation

Plant seeds were obtained from the Genetic Resources Unit of Institute for Biological Environmental and Rural Studies (IBERS; Aberystwyth, Wales, UK) and the South Australian Research and Development Institute (SARDI)23. The plants were germinated on a filter paper in a petri dish. gDNA was extracted from fresh leaf of each plant genotype using the DNeasy Plant mini kit (QIAGEN) following the manufacture’s instruction. gDNA was extracted from barley (pearled grains), wheat (plain flour), oat (rolled grains) and rice (flour) using the DNeasy Plant mini kit.

PCR-based screening

PCR primers were designed using the Oligo Calc tools (https://biotools.nubic.northwestern.edu/OligoCalc.html). For cross-species amplification, primers were designed to generate short DNA fragments (< 251 bp in length) within highly conserved regions of the target sequences (Supplementary Information 13). The PCR was performed on the CFX Real-Time PCR Detection Systems (Bio-Rad), with the Luna Universal qPCR Master Mix (NEB), and data analysis was performed using the CFX Manager Software (Bio-Rad). Visualisation of PCR products was performed on the 2200 TapeStation instrument or on an agarose gel (2.0% w/v) stained with SYBR Green (Thermo Fisher) through electrophoresis.

Validation of the PCR-based screening method

PCR primers were designed based on perennial ryegrass sequences to amplify larger fragments (> 900 bp), for use as DNA templates of a standard curve assay (SCA) (Supplementary Information 13). The PCR amplicons were amplified from the Impact04 genotype, using the MyTaq DNA Polymerase kit, and the amplicons were cleaned using the Monarch PCR & DNA Cleanup Kit (NEB). DNA concentration was adjusted to 1 pg/μl, and a tenfold series dilution (from 1 pg/μl to 1 × 10−4 pg/μl) was prepared. For fungus-specific primers, a fourfold series dilution (from 1 ng/μl to 1 × 4−4 ng/μl) of E. festucae gDNA was prepared. The SCA was performed on the CFX Real-Time PCR Detection Systems, with the Luna Universal qPCR Master Mix, followed by data analysis with the CFX Manager Software (Supplementary Information 14).

Genetic linkage mapping

PCR primers (LpDUF3632_SCA_F and R in Supplementary Information 13) were used for amplification of an LpDUF3632-related sequence from the C3 parental genotype of the p150/112 reference genetic mapping population23. A sequencing library for the MiSeq platform (Illumina) was prepared from the PCR amplicons, following the previously described MspJI-based method. The prepared library was characterised and quantified with the TapeStation and Qubit instruments (Thermo Fisher Scientific). The library was loaded on the MiSeq platform (Illumina), following the manufacture’s instruction, and the outcome reads were assembled with the Sequencher 5.0 program (Gene Codes), to find the single nucleotide substitutions between the C3 haplotypes. An additional reverse PCR primer (5′-AGCGTGCTTGCCAGCCTCTC-3′) was prepared and used with the LpDUF3632_SCA_F primer, for the PCR–RFLP assay. DNA fragments were amplified from each individual of the genetic mapping population and digested with the SacII restriction enzyme (NEB). Genetic linkage analysis was performed through use of the JoinMAP application.

In silico analysis

Gene structure prediction was performed using the FGENESH program of the Softberry website (https://www.softberry.com/berry.phtml) using the ‘Monocot plants’ parameter. Alignments of amino acid sequences were generated with the CLUSTALW program (https://www.genome.jp/tools/clustalw/) with the default parameters. Sequence homology search was performed with the NCBI (https://www.ncbi.nlm.nih.gov/) and PredictProtein (https://open.predictprotein.org/) websites. Amino acid and DNA sequences were prepared for phylogenetic analysis with the MEGA7 program (https://www.megasoftware.net/) (Supplementary Information 15 and 16). Figure legends generated with the program can be found in Supplementary Information 17. Non-synonymous and synonymous nucleotide substitution ratio was calculated using the MEGA7 program. For gene expression analysis, in-house transcriptome read datasets were prepared through filtering of the transcriptome sequencing reads from Impact04 tissues using Impact04 genome contigs (> 999 bp). The number of reads which contained LpFTRL or LpDUF3632 sequences (no sequence mismatch for 60 bp or longer) were counted. For the gene expression analysis with the SRA data from Massey University (NCBI BioProject UI: PRJNA292034), the NCBI BLASTn function was used, with the megablast algorithm and word size of 64.