Horizontal gene transfer (HGT) is the nonsexual transmission of genetic material across species boundaries 1 2 . HGT is well known in bacteria, where HGT often results in adaptive gains of novel genes and traits 3 4 5 . There are fewer well-documented cases of HGT among eukaryotes 6 and the large majority of these cases appear to result in short-lived, nonfunctional sequences 6 7 8 . Consequently, the long-term evolutionary impact of HGT in multicellular eukaryotes remains largely unknown. Several cases of HGT are known or suspected in plants 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 , most involving mitochondrial sequences, and/or parasitic plants 13 14 15 17 18 19 20 23 24 25 . Parasitic plants form direct haustorial connections with their host plants and are capable of obtaining a wide range of macromolecules from their hosts, including viruses 26 , gene silencing signals 27 , and messenger RNAs 28 . Consequently, parasites may have many opportunities for HGT events and an increased likelihood that some of these result in functional, and potentially adaptive, gene transfers. Two recent reports by Yoshida et al 19 and Xi et al 25 were the first indications that nuclear protein coding sequences, likely obtained from their respective host species, could be integrated into the genomes of parasitic plants by HGT. These were important advances, but they provided few clues as to the long term impact of HGT, how the transgenes evolve, and how they may function. We hypothesized that systematic analysis of genome-scale datasets from parasitic plants could lead to evidence for acquisition and long-term maintenance of functional gene sequences in plants that had been acquired via HGT.

Albumin 1 genes are known only from a subset of species in the legume family (Leguminosae) of angiosperms where they encode seed storage proteins and insect toxins 29 30 . The albumin 1 proteins in legumes are 112 to 154 amino acids in length and rich in cysteine residues. They form a unique protein structure known as a KNOTTIN, which has three disulfide bonds and is characterized by a “disulfide through disulfide knot” 31 . The KNOTTINs are famous for their intruguing “disulfide through disulfide knot” structure and have been extensively studied in various fields, most of which are related with potentials in drug design 32 33 34 35 36 37 . Albumin 1 genes may have originated early in the diversification of papilionoid legumes 29 30 , but multiple homologous gene copies have been found only in species that are members of the more derived “Millettioid s.l.” and “Hologalegina” clades 38 .

Orobanche s. l., often known by the common name “broomrape,” includes 150-170 obligate parasitic plant species in the family Orobanchaceae. Growing evidence supports the segregation of broomrapes into four genera 39 : Aphyllon (syn. Orobanche sect. Gymnocaulis), Myzorrhiza (syn. Orobanche sect. M.), Phelipanche (syn. Orobanche sect. Trionychon), Orobanche s. str. (syn. Orobanche sect. O.). Most broomrape species have a narrow host spectrum and grow exclusively on perennial eudicot host plants 40 , with members of the Leguminosae, Solanaceae, and Asteraceae among the more common hosts 41 . As a member of order Lamiales, Orobanchaceae is phylogenetically well-separated from host members in these lineages, particularly legume hosts in the rosid order Fabales (Additional file 1: Figure S1; 42 ). A few broomrape species (e.g., P. aegyptiaca, P. ramosa, O. cernua, O. crenata, and O. minor) have become devastating pests of important crop plants, affecting their growth and resource allocation and imparting significant losses in yield 43 . P. aegyptiaca, the focal species in this study, has a broad host range that includes members of the eudicot families Apiaceae, Asteraceae, Brassicaceae, Cucurbitaceae, Leguminosae, and Solanaceae.

Here we show that a gene encoding albumin 1 KNOTTIN-like protein, closely related to the albumin 1 genes, only known from papilionoid legumes, serving dual roles in food storage and as insect toxins, was found in Phelipanche aegyptiaca and related parasitic species of family Orobanchaceae, and was likely acquired by a Phelipanche ancestor via HGT from a legume host based on phylogenetic analyses. According to genomic sequences from nine related parasite species, 3D protein structure simulation tests, and evolutionary constraint analyses, the broomrape xenogene we identified here retains the intron structure, six highly conserved cysteine residues necessary to form a KNOTTIN protein, and displays levels of purifying selection like those seen in legumes. The albumin 1 xenogene has evolved through >150 speciation events over ca. 16 million years, forming a small family of differentially expressed genes that may confer novel functions in the parasites.

The albumin 1 transcript was first identified as a HGT candidate in the transcriptome of P. aegyptiaca (cultured and grown on Arabidopsis and tobacco) using a BLAST-based 44 bioinformatic screen (details in Material and Methods). Albumin 1 transcripts were then searched further, using BLASTX, against the NCBI nr database and the PlantGDB database 45 . Top hits were seen (Additional file 2: Figure S2) to Medicago truncatula albumin 1 sequences, with expected values of 5e-51 and 1e-48. Additional BLAST, including Hidden Markov Model (HMM)-based psi-BLAST searches with the sequence from P. aegyptiaca were performed to attempt to detect homologs in three other members of Orobanchaceae with large transcriptome datasets (two parasites, Striga hermonthica and Triphysaria versicolor, and the nonparasitic Lindenbergia philippensis 46 ) (Parasitic Plant Genome Project, PPGP 47 ). Several large public databases, including Phytozome 48 , PlantGDB, and SOL Genomics Network 49 , were also searched. After searching 34 sequenced genomes and transcriptomes of 274 additional plant species, albumin 1 homologs were detected only in legumes and the transcriptome libraries of P. aegyptiaca.

Having identified the albumin 1 sequence in the P. aegyptiaca transcriptome, genomic sequences encoding albumin 1 were then obtained from P. aegyptiaca and eight additional broomrape species, including P. schultzii, P. ramosa, P. mutelli, P. nana, and Orobanche hederae, O. minor, O. cernua and O. ballotae. The nucleotide sequence and inferred gene structures of the albumin 1 genes in broomrape species (Figure 1; Additional file 3: Figure S3, Additional file 4: Figure S4, Additional file 5: Figure S5) are closely comparable, with inferred protein alignments 57.3-58.3% identical and 72.7%-74.3% similar (= identity + conservative substitutions) in ungapped regions between the legume and parasite proteins. Two albumin 1 genes were identified in Phelipanche species, and are identified here as copy_12653 and copy_75797, or albumin1-1 and albumin1-2, respectively. An intron disrupts the coding region at the same position in both genes and the intron sequences are similar but contain a number of insertion and deletion mutations. Only one albumin 1 gene was detected from Orobanche species. Although the intron length in albumin 1 genes of Phelipanche and legume species is not well conserved, several critical intron features are shared (Additional file 5: Figure S5). First, the starting position of the intron in both the P. aegyptiaca and M. truncatula sequences are the same, and the first nine base pairs are identical. Second, the introns have characteristic splicing sites at their 5’ and 3’ ends; 5’ ends often have GT/GU and 3’ ends often have AG, and these motifs are found in both M. truncatula and Phelipanche albumin 1 introns (Figure 1 and Additional file 5: Figure S5). Albumin1 gene sequences from Phelipanche were also searched with BLASTn against the NCBI nt database in order to search for high frequency repeats and mobile elements, but no such features were identified.

Phylogenetic analysis (Figure 2) of all known plant albumin 1 sequences showed a strongly supported clade containing all of the albumin 1 sequences from broomrapes (Maximum likelihood (ML) boostrap 98, Bayesian inference (BI) Posterior probabilities (PP) 0.99) nested deeply within the IRLC (Inverted Repeat-lacking clade) of papilionoid legumes 50 . Among legumes, the next most closely related sequences (ML bootstrap 100, BI PP 0.99) are from Onobrychis argentea and Onobrychis viciifolia. Because the node supporting the position of the broomrape clade (ML bootstrap 79, BI PP 0.99) within the papilionoid legumes is relatively weakly supported, we also tested the hypothesis that the broomrape clade of albumin 1 sequences falls outside the larger clade of legumes represented in this analysis (i.e., at a position sister to the Millettioid and Hologalegina clades). This hypothesis was rejected (Shimodaira-Hasegawa test and Kishino-Hasegawa test, using Tree-Puzzle version 5.2, Log L = -4482.60) relative to the maximum likelihood position as indicated in this tree. Two albumin 1 genes are resolved as sister clades in Phelipanche species, which are in turn resolved as sister to the single gene obtained from Orobanche species. Gene structures supported a similar conclusion (Figure 1).

The amino acid sequence alignments of albumin 1 from legumes to P. aegyptiaca show conservation of all cysteine residues essential for disulfide bond formation in albumin 1 proteins (Figure 3A). We investigated whether the predicted albumin 1 proteins from parasites maintain the characteristic KNOTTIN structures found in the legume albumin 1 proteins using Knoter1d 31 53 . Simulated 3D structures show that the Phelipanche albumin 1 proteins form a characteristic KNOTTIN structure with three-disulfide bonds and a “disulfide through disulfide knot”. KNOTTIN protein structures are also predicted in all of the other full-length albumin 1 genes in Phelipanche species. Knoter1d assigned scores greater than 35 to each Phelipanche albumin 1 sequence; a score greater than 20 in this analysis passes the Knoter1d criteria for identification as an albumin 1 structure. The predicted 3D structures for P_aegyptiaca_Albumin1-1 (Figure 3B) and P_aegyptiaca_Albumin1-2 (Figure 3C) are very similar to the insect toxic albumin 1 protein from M. truncatula. Albumin 2, a non-KNOTTIN legume protein, has no discernable homology with the albumin 1 protein in legumes (Figure 3E).

Having found that the horizontally acquired albumin1 genes were present in related species of broomrapes we then asked if the genes are evolving under purifying selection indicative of a functional protein coding sequence. dN (nonsynonymous substitutions), dS (synonymous substitutions) and dN/dS were calculated for all three lineages of the broomrape albumin 1 clade (= albumin 1 in Orobanche, albumin1-1 and albumin1-2 in Phelipanche) and for the albumin 1 sequences from three closely related legume sequences; Astragalus monspessulanus, Onobrychis argentea and Onobrychis viciifolia. Synonymous substitutions in the albumin 1 genes (dS) outnumber non-synonymous substitutions (dN) by at least 3:1 in most lineages (Figure 4), and dN/dS, reflecting the level of purifying selection, is similar in broomrapes to the value estimated for closely related albumin1 sequences from legumes. All cysteine residues were also identified as evolving under purifying selection, suggesting that the horizontally acquired albumin 1 genes in broomrapes are functional (Bayes factors ranging from 3.04 to 27.52.)

Having observed evidence for selection for structural conservation, we investigated whether these genes exhibit transcription profiles that suggest a new or unique pattern of expression in parasites. Normalized expression levels of both albumin 1 genes in P. aegyptiaca were estimated as reads per kilobase per million reads (RPKM) for eight libraries representing major stages of belowground and aboveground parasite development (Figure 5, Additional file 6: Table S1, Additional file 7: Table S2). Both genes displayed lowest expression levels at stage 3 (haustorial attachment stage) and highest at stage 6 (above-ground tissues). Transcripts were particularly abundant at stage 6.2 (reproductive), more than 1000x higher than the haustorial stage.

Biogeographic overlap and common feeding interactions between diverse broomrapes and temperate papilionoid legumes increase the likelihood that the HGT event occurred in a common ancestor of the parasites that was in direct contact with legume host plants. An alternative (and less parsimonious) explanation is that another organism or virus that co-occurred in the same habitats as the ancestral lineages served as a “stepping stone” for a two- or more step transfer. However, this is not supported in our searches of the sequence databases. Based on fossil-calibrated age estimates of legume lineages 50 51 , we estimate that this horizontal acquisition occurred in an ancestral broomrape that lived in the Miocene epoch, about 16 Mya. Both parasite and their legume host groups have northern temperate distributions, and their lineages likely overlapped in the past as they do now, providing a minimal requirement for a horizontal gene transfer to occur. Another possibility, however unlikely, is that albumin1 was a more recent acquisition that underwent strong convergence at the protein level with this legume lineage. However the branch lengths we observed in the phylogeny (Figure 2) were not unusually long in our DNA-based phylogeny, and given the large collection of related sequences we obtained from other broomrape species, we have reduced any tendency the Orobanche/Phelipanche lineage may have had to connect by chance to a deep branch. Thus, the convergence hypothesis is not supported. Because the breadth of Phelipanche and Orobanche species we have sampled spans the deepest branches of broomrape diversity 55 , the albumin gene can be inferred to have survived through an extended evolutionary radiation of at least 150 species 55 56 57 or more, if the number of now-extinct broomrape species could be estimated.

Because the introns of Phelipanche albumin 1 xenogenes maintain critical splicing sites and share the same starting positions and first nine base pairs with the known M. truncatula albumin 1 intron, it is likely that the HGT event in broomrapes involved transfer of a genomic sequence rather than a separate cDNA. Following the transfer, albumin 1 genes in broomrape species have evolved under purifying selection consistent with what is observed in related legume albumin 1 genes. This observation, as well as the stage-specific transcription patterns, conserved cysteine residues and predicted 3D KNOTTIN protein structures, strongly suggest that albumin 1 genes encode functional proteins in broomrape species, and could potentially serve a function similar to its role in legumes, providing a large pool of sulfur storage and exhibiting toxicity to insect herbivores in certain legumes 29 30 . A recent report involves panicoid grass species with C3 or C4 photosynthetic pathways. Evidence was presented that nuclear genes were horizontally transferred between panicoid species and were subsequently adapted into the existing pathways with the effect of advancing the extent of C4 photosynthesis in some lineages 21 . These results indicate that HGT may promote the sharing of adaptive traits among related species. In comparison, the albumin example described here shows how a completely novel and highly specialized trait has been acquired at an ancestral stage from a distantly related donor species and maintained by the recipient lineage throughout an extended period of evolutionary history.

The albumin 1 genes in P. aegyptiaca are highly transcribed in most of the developmental stages we examined. Transcripts are more abundant in reproductive tissue, and lowest in the young haustorium (stage 3), which represents the earliest point in our tissue sampling where the parasite is in direct contact with the host plant. This suggests that the novel gene in P. aegyptiaca is probably not encoding a protein that is playing a direct role in the process of haustorial formation, and that albumin 1 expression is down-regulated as the parasite devotes energy to the essential process of establishing host vascular connections. It is also possible that the low expression in the haustorial stage could help the parasite avoid detection or minimize a negative impact on the health of the host plant during early stages of parasite contact and feeding.

Several other parasitic lineages, including members of Cuscuta (Convolvulaceae), Cassytha (Lauraceae), Apodanthaceae, Hydnoraceae, and the order Santalales, regularly feed upon legumes 58 and therefore might also have had opportunities to acquire albumin 1 sequences through HGT. Large transcriptome datasets are currently available for only two of these, the generalist parasite Cuscuta pentagona (Convolvulaceae) and for the legume specialist feeder Pilostyles thurberi (Apodanthaceae) 17 . Both of these parasites, and other species in these genera, feed widely on legumes. No homolog of albumin 1 was detected in BLAST searches of the Pilostyles transcriptome in the 1KP dataset 59 . However, albumin 1 sequences were detected in the same dataset and in two additional transcriptome libraries from Cuscuta pentagona (J. Westwood, unpublished data but publicly available through 1KP Blast database). Phylogenetic analysis nests the Cuscuta sequences well within Leguminosae, but on an independent branch from the broomrape sequences (Additional file 8: Figure S6), suggesting that these transcripts in Cuscuta represent a different HGT event into Cuscuta from a lineage of papilionoid legumes that was different from the source of the broomrape albumin 1 xenogene. The putative Cuscuta albumin 1 similarly encodes a protein predicted to have KNOTTIN structure (Knoter1d score: 33 to 35). No other albumin 1 sequences were identified elsewhere in searches of REFSEQ or publicly available plant transcriptome datasets.

Because of their extensive, intimate contacts with host plant tissues, and the wide range of materials that are commonly transmitted across haustorial connections 27 28 46 60 61 62 63 , parasitic plants play an important role as recipients and donors for HGT in plants 13 14 15 17 19 24 . As parasitic plants increasingly become the targets for genome-scale analyses, it should become possible to estimate the frequency and likely mechanisms of HGT events between parasites and hosts involving albumin 1 and other genes, the likelihood of more complex stepping-stone models, and how often HGT leads to long-term maintenance of new genes and novel traits.

The authors declare that they have no competing interests.

Conception and design of PPGP transcriptome study (JHW, CWD, MPT, JIY); conception and design of HGT study (YZ, CWD); Phelipanche and Orobanche plants, DNAs, PCR, cloning, and chromosome walking (MF-A, JHW); plants, RNAs, and libraries for transcriptome sequencing (MF-A; LAH, PER, MD), legume DNAs (MFW), data analysis and presentation (YZ, EKW, YJ, NJW, MF-A, CWD); wrote manuscript (YZ and CWD, with contributions from all of the authors). All authors read and approved the final manuscript.