We use the term HGT loosely in this study for any transfer events from non-organellar sources. Although the timing of HGT cannot be accurately calibrated in most cases, it can be inferred based on gene distribution in the recipient lineage. If the acquired gene is found in most taxa of a major lineage, it is likely that the gene was acquired prior to the divergence of the lineage. Given the paucity of sequence data from representatives of many major eukaryotic groups and the lack of consensus on eukaryotic phylogeny 47, identification of ancient HGT often becomes more difficult as phylogenetic depth increases.

A major issue related to the role of HGT in macroevolution is the scale of ancient HGT. Our analyses identified 37 anciently acquired genes in plants that account for 1.42% (37/2,605) of all generated gene trees (Table 1; Additional data file 1). It should be cautioned that HGT identification is affected by many factors, in particular taxonomic sampling, method of analysis, complications arising from IGT, and lineage-specific gains or losses (see 374849 for more discussions). For studies based on phylogenetic approaches, long-branch attraction arising from biased sequence data is also a particular concern 5051. Additionally, if the α-proteobacterial or the cyanobacterial nature of IGT-derived genes has been erased, due to either frequent HGT among prokaryotes or the loss of phylogenetic signal over time, these genes will not be properly identified and may be mistaken as HGT-derived. It should also be noted that this study is based on the genome analyses of the red alga Cyanidioschyzon, which inhabits an extreme environment in acidic hot springs and maintains a streamlined genome 41. Some anciently acquired genes might have been lost from the Cyanidioschyzon genome, but are retained in other red algal species. This could potentially underestimate the HGT frequency in plants. With the rapid accumulation of sequence data, in particular those from other red algae and under-represented eukaryotic groups, a broader taxonomic sampling will be possible and the number of anciently acquired genes identified in the plant lineage will likely change. Therefore, the data presented in this study should only be interpreted as our current understanding of the scale of ancient HGT, rather than an exhaustive list of all anciently acquired genes in plants.

Despite the difficulties in HGT identification, the multiple introductions of the same gene from various prokaryotic sources (for example, 2-methylthioadenine synthetase; Figure 1) suggest that HGT is a continuous and dynamic process. Given that phylogenetic signal tends to become obscure over time and that eukaryote-to-eukaryote transfer, which has been recorded in multiple studies 5253, is largely not covered in this study, it is possible that the identified genes in our analyses represent only the tip of an iceberg for the overall scope of ancient HGT in eukaryotes. In particular, during early eukaryotic evolution when the ancestral nucleocytoplasmic lineage emerged from prokaryotes (either by a split from archaea or by fusion of archaeal and bacterial partners) and began to diverge into extant groups, these early eukaryotes might bear more biochemical and physiological similarities to their prokaryotic relatives. Because HGT tends to occur among organisms of similar biological and ecological characters 54, the barriers to interdomain gene transfer during early eukaryotic evolution might not be as significant as observed today. Therefore, although our data suggest that HGT indeed existed in early plant evolution, many other anciently acquired genes in plants might have escaped our detection because of the limitations of current phylogenetic approaches. These genes might have shaped the genome composition of the recipient lineages and may also be, in part, responsible for the lack of resolution of relationships among major eukaryotic groups 4047.

A significant insight from prokaryotic genome analyses is the role of HGT in microbial adaptation. By acquiring ready-to-use genes from other sources, HGT avoids a slow process of gene generation and might confer to the recipient organisms immediate abilities to explore new resources and niches 555657. This may be crucial for organisms inhabiting shifting environments, where acquisition of beneficial genes from local communities is necessary for recipient organisms to avoid extinction or to optimize their adaptation. Therefore, lineage continuity and ecological stability can be achieved by increasing the genetic repertoire through recruitment of foreign genes.

An acquired gene may be novel to the recipient or homologous to an endogenous copy. In the latter case, the newly acquired homolog may be retained (for example, 2-methylthioadenine synthetase; Figure 1) and the acquisition of an additional gene copy will provide opportunities for functional differentiation and enriches the genetic repertoire of the recipient. Although all acquired genes affect genome composition and evolution, only those that potentially provide new functions will most likely induce biochemical or phenotypic changes, and consequently adaptation in recipient organisms. Some anciently acquired novel genes identified in our analyses appear to be critical for plant development or adaptation. For example, the gene encoding topoisomerase VI beta subunit (TOP6B) in plants was likely acquired from a crenarchaeote 37. TOP6B in green plants is required for endoreplication, a process of DNA amplification without cell division and a mechanism to increase cell size in plants. Top6b mutants display extreme dwarf phenotypes (about 20% the height of wild types), chloroplast degradation, and early senescence 585960.

Several other novel genes are functionally related to the biogenesis and development of plastids. These include genes acquired from different bacterial groups. For example, MGDG synthases are responsible for the generation of MGDG, a major lipid component of plant photosynthetic tissues. MGDG synthases appear to be encoded by a single-copy gene in red and green algae, but three copies exist in Arabidopsis and they are further classified into two types (type A, including MGD1, and type B, including MGD2 and MGD3). In Arabidopsis, MGD1 is localized in the inner membrane of chloroplasts and it is responsible for the majority of MGDG biosynthesis. No mgd1 null mutants are found in Arabidopsis, suggesting that MGD1 is essential for chloroplast development and plant growth 61. In contrast, MGD2 and MGD3 are highly expressed in non-photosynthetic tissues and likely provide an alternative route for MGDG biosynthesis under phosphate starvation conditions 616263. Therefore, ancient HGT, gene duplication and subsequent functional differentiation provide a mechanism for specialized MGDG production in different tissues and growing conditions. As another example, knocking down the expression of the chlamydiae-related ATS1 and ATS2 in Arabidopsis will lead to small, pale-yellow plants, suggesting that the chloroplast development has been seriously impeded 64.

Not all acquired genes may bring new biochemical functions to the recipient organism. The acquired gene may displace the existing homolog and, if they are functionally equivalent, the impact of gene transfer on the adaptation of the recipient may be limited. Such homolog displacement may be considered selectively neutral 6566, though their contributions to genome evolution should not be ignored.

Although the role of HGT in eukaryotic evolution is gaining increasing appreciation, there are very few studies available on the number of acquired genes resulting from homolog displacement without introducing new functions. According to the gene transfer ratchet mechanism proposed by Doolittle 67, homolog displacement might be pervasive in unicellular eukaryotes and bacterial genes, either intracellularly or horizontally derived, may gradually replace all endogenous copies over time. Although our analyses only address anciently acquired genes prior to the split of red algae and green plants, homolog displacement indeed appears to be frequent compared to the acquisition of genes with novel functions. For example, at least three genes encoding organellar aminoacyl-tRNA synthetases (that is, leuRS, tyrRS, and ileRS) were likely acquired from other prokaryotic sources (Table 1; Additional data file 1). These aminoacyl-tRNA synthetases are often shared by both mitochondria and plastids 68, suggesting that both plastidic and mitochondrial aminoacyl-tRNA synthetases might have been frequently displaced in plant evolution.

It should be noted that the displacement of aminoacyl-tRNA synthetases is relatively easy to identify because these genes have low substitution rates and they are universally present in all organisms 3869707172. Many other cases of homolog displacement may not be as easily detected because of complications arising from possible independent gene losses/gains or lack of phylogenetic information retained in the acquired gene 3765. In our analyses, homologs for most identified genes can be found in multiple extant cyanobacteria. Given the cyanobacterial origin of plastids, a cyanobacterial copy of these genes might have existed when the plastids were first established; therefore, an IGT event and subsequent displacement of the original plastidic genes by later non-cyanobacterial homologs cannot be excluded, though such a scenario is highly unlikely to have occurred to all these genes. Overall, our data show that many acquired genes may have resulted from homolog displacement without introducing new functions, suggesting that the number of acquired genes does not predict the role of HGT in the adaptation of recipient organisms. It is unclear whether such a gene displacement pattern also exists in non-photosynthetic eukaryotes.

Plastids are the key evolutionary novelty that defines photosynthetic eukaryotes. Aside from photosynthesis, some other important biochemical activities, including biosyntheses of fatty acids and isoprenoids, are also carried out in plastids. Intriguingly, over 78% (29/37) of the anciently acquired genes identified in our analyses are either predicted or experimentally determined to be related to the biogenesis and functionality of plastids (Table 1); these include genes possessing novel functions and those resulting from homolog displacement. Because of the extremophilic lifestyle of Cyanidioschyzon and its streamlined genome, some acquired genes related to non-photosynthetic activities might have been eliminated from the genome. It remains to be investigated whether such a high density of acquired genes that are functionally related to plastids also exists in other photosynthetic eukaryotes, including mixotrophs and those inhabiting broader niches. Nevertheless, given the total number of these plastid-related genes identified in our analyses, it appears that concerted gene recruitment from multiple sources or selective retention of the acquired genes occurred to optimize the functionality of plastids during early plant evolution. The observation that some independently acquired bacterial genes are functionally related to plastids has also been reported in the chlorarachniophyte Bigelowiella natans, which contains plastids derived from a secondary endosymbiont 21.

This phenomenon of concerted gene recruitment for the origin and optimization of key evolutionary novelties of the recipient also exists in other eukaryotic groups. In the protozoan group diplomonads, about half (7/15) of the acquired genes are related to the anaerobic lifestyle of the organisms. These genes were interpreted to have been acquired from various organisms, including other eukaryotes, and might be responsible for the lifestyle transition from aerobes to anaerobes in diplomonads 24. Another example is related to ciliates that live in the rumen of herbivorous animals. In this case, over 140 genes were transferred from diverse bacterial groups to rumen ciliates, the vast majority of which are related to degradation of carbohydrates derived from plant cell walls 30. A third example is the evolution of nucleotide biosynthesis in the apicomplexan parasite Cryptosporidium, where two independently acquired genes, one each from γ- and ε-proteobacteria, and likely two other plant-like genes facilitated the establishment of salvage nucleotide biosynthetic pathways 3673, allowing the parasite to obtain nucleotides from their hosts. Therefore, concerted recruitment or selective retention of foreign genes apparently is not a unique phenomenon in the origin and optimization of evolutionary novelties of unicellular eukaryotes. In the case of plants, ancient endosymbioses and HGT events in concert drove the establishment of plastids. In the cases of diplomonads, rumen ciliates and Cryptosporidium parasites, multiple independent HGTs from other organisms contributed to the major lifestyle transitions in the recipient organisms. In all these cases, the origin of evolutionary novelties may be viewed as a result of gene sharing with other organisms.

Although the current data suggest that HGT events are frequent in unicellular eukaryotes 21242630, how and to what degree they have affected the evolution of the recipients remain largely unclear. An interesting observation from the studies of HGT in eukaryotes is that the vast majority of well-documented cases involve prokaryotes as donors 263031. Given the ubiquitous distribution of prokaryotes and their greater species and metabolic diversity, the gene pool of prokaryotes conceivably was significantly larger than that of eukaryotes, in particular during early eukaryotic evolution. Therefore, it is interesting to speculate whether early eukaryotes continuously obtained genes from a larger prokaryotic gene pool 67, either individually or occasionally in large chunks, through HGT events in response to the environment, as we have now observed in many prokaryotes and unicellular eukaryotes. Such changes in genetic background and biochemical system would likely induce shifts in ecology, physiology, morphology or other traits of the recipient lineage. Concerted gene recruitment in plants, diplomonads, rumen ciliates, Cryptosporidium parasites and possibly many other organisms suggests that independently acquired genes are able to generate and optimize key evolutionary novelties in recipient organisms. Whether such ancient gene recruitment events and the novelties they generated were ultimately responsible for the emergence and adaptive radiation of some major eukaryotic groups warrants further investigations.

Protein sequences for the red alga Cyanidioschyzon merolae were obtained from the Cyanidioschyzon Genome Project 4274. Expressed sequence tag (EST) sequences were obtained from TBestDB 75 and the NCBI EST database. All other sequences were from the NCBI protein sequence database.

Anciently acquired genes in this study include those horizontally acquired prior to the split of red algae and green plants. A list of ancient HGT candidates was first generated based on phylogenomic screening of the Cyanidioschyzon genome using PhyloGenie 41 and the NCBI non-redundant protein sequence database. The vast majority of the genes on this list are predominantly identified in bacteria and archaea, and therefore are likely of prokaryotic origin. To reduce the complications arising from potential cases of IGT, we adopted an approach combining sequence comparison, phylogenetic analyses, and statistical tests. Each gene on the list was first used to search the NCBI protein sequence database. Because of the cyanobacterial origin of plastids and the α-proteobacterial origin of mitochondria, genes with cyanobacterial and plastid-containing eukaryotic homologs as top hits were considered as likely plastid-derived; those with α-proteobacterial and other eukaryotic homologs as top hits were considered as likely mitochondrion-derived. These potentially organelle-derived genes were removed from the candidate list and the remaining genes were subject to detailed phylogenetic analyses. Gene tree topologies generated through detailed phylogenetic analyses were subject to careful inspections; any genes that formed a monophyly with cyanobacterial and plastid-containing eukaryotic homologs or with proteobacterial and other eukaryotic sequences were also eliminated from further consideration. Additionally, alternative topologies representing various evolutionary scenarios for each gene were statistically evaluated based on AU tests 43. Genes for which a straightforward IGT scenario (versus IGT followed by secondary transfers) could not be rejected (p-value > 0.05) were also removed from the HGT candidate list. For a few genes, the gene tree topology may be explained by either a straightforward HGT or an IGT followed by secondary HGT events to other organisms; we prefer the scenario of straightforward HGT in these cases to that of secondary HGT, based on an assumption that chances for the same gene being repeatedly transferred among different organismal groups are relatively rare. In several other cases (for example, Figures 1 and 2d), the distribution of the subject gene may also be explained by either multiple independent HGT events or a single HGT followed by differential gene losses. In such cases, we prefer the gene loss scenario based on an assumption that independent acquisitions of the same gene, by closely related taxa, from the same donor are rare. Because identification of HGT heavily relies on an accurate organismal phylogeny and because the relationships among many major eukaryotic lineages remain unsolved 4047, HGT events among eukaryotes were not included in our analyses in most cases, except for those between photosynthetic eukaryotes where secondary or tertiary endosymbioses and subsequent gene transfer to host cells have been frequently documented 212676.

Sequences were sampled from representative groups (including major phyla of bacteria and major groups of eukaryotes) within each domain of life (bacteria, archaea, and eukaryotes). Because of the potential for sequence contaminations, eukaryotic EST sequences whose authenticity is suspicious (for example, high nucleotide sequence percent identity with bacterial homologs and/or absence of homologs from genomes of closely related taxa) were not included in the analyses. Multiple protein sequence alignments were performed using MUSCLE 77 and clustalx 78, and only unambiguously aligned sequence portions were used. Such unambiguously aligned positions were identified by cross-comparison of alignments generated using MUSCLE and clustalx, followed by manual refinement. The alignments are available in Additional data file 1. Phylogenetic analyses were performed with a maximum likelihood method using PHYML 79, a Bayesian inference method using MrBayes 80, and a distance method using the program neighbor of PHYLIP version 3.65 81 with maximum likelihood distances calculated using TREE-PUZZLE 82. All maximum likelihood calculations were based on a substitution matrix determined using ProtTest 83 and a mixed model of four gamma-distributed rate classes plus invariable sites. Maximum likelihood distances for bootstrap analyses were calculated using TREE-PUZZLE 82 and PUZZLEBOOT v1.03 (by Michael E Holder and Andrew J Roger, available on the web 84). Branch lengths and topologies of the trees depicted in all figures (Figures 1 and 2; Additional data file 1) were calculated with PHYML. For the convenience of presentation, gene trees were rooted using archaeal (or archaeal plus eukaryotic) sequences, or paralogous gene copies if ancient gene families were involved, as outgroups; otherwise, trees were rooted in a way that no top hits of the sequence similarity search were used as an outgroup. Nevertheless, all gene trees should be strictly interpreted as unrooted.

Following detailed phylogenetic analyses, alternative tree topologies for each remaining HGT candidate were assessed for their statistical confidence using Treefinder 85. In most cases, multiple constraint trees for each HGT candidate were generated using Treefinder by enforcing: monophyly of all eukaryotic sequences; monophyly of cyanobacterial, plant and other plastid-containing eukaryotic sequences; and monophyly of cyanobacterial, plant, and closely related bacterial sequences. These alternative topologies assumed that the subject gene in plants is not HGT-derived; they served as null hypotheses that all eukaryotic sequences have the same eukaryotic or mitochondrial origin or that plants acquired the subject gene from plastids, sometimes followed by secondary HGT to other bacterial groups. AU tests, which have been recommended for general tree tests 43, were performed on alternative tree topologies (non-HGT hypotheses) and the tree generated from detailed phylogenetic analyses (HGT hypothesis). In this study, topologies with a p-value < 0.05 were rejected.

Targeting signal of identified protein sequences was predicted using ChloroP 86 and TargetP 87. Additional information about protein localization in green plants was obtained from The Arabidopsis Information Resource (TAIR).