Our first goal was to test the widely but a priori accepted hypothesis that the NPC is homologous in all extant eukaryotes by investigating the distribution of homologs of the metazoan NPC and NPCa proteins across eukaryotic lineages. We retrieved the sequences of 65 metazoan NPC and NPCa proteins and searched for their homologs in all eukaryotic phyla for which sequences are available in current databases, such as fungi, green plants, Rhodophytes, Conosa, and Diplomonads (Table 1; Additional data file 1).

Two different phyletic patterns are expected depending on: whether the NPC was a very recent evolutionary innovation and the outcome of independent evolutionary processes in different eukaryotic lineages; or whether it originated before the last eukaryotic common ancestor (LECA 3). In the first case, very few metazoan NPC and NPCa proteins would have homologs in all eukaryotic lineages; and in the second case, the vast majority of metazoan NPC and NPCa proteins would have homologs in all eukaryotic lineages 26.

Retrieving homologs for NPC and NPCa proteins was unexpectedly difficult, despite the apparent structural conservation of the NPC between fungi and metazoa 8. The ability to identify and successfully retrieve homologs by BLAST and PSI-BLAST approaches is notably dependent on the evolutionary rates of sequences. For example, attempts to retrieve a rapidly evolving Arabidopsis thaliana sequence using a slowly evolving Homo sapiens sequence, or vice versa, may be unsuccessful if these homologous sequences have evolved beyond recognition. To overcome this limitation, we multiplied the seeds for our BLAST searches. Interestingly, we observed that 40 of the 65 NPC and NPCa proteins studied were present in at least the fungal, animal and plant lineages (Table 1). Furthermore, mining of protist EST databases, notably of stramenopiles, expanded this taxonomical distribution (Table 1), revealing that 48 of the 65 proteins under study were present in bikonts (the grouping of plants and all protists excepted conosa 27) and in unikonts (the grouping of opisthokonts: metazoa and fungi, and conosa). Among these 48 proteins, 27 of the 33 components of the NPC (Table 1; Figure 1) and 16 of the 17 proteins involved in nucleocytoplasmic transport were conserved in unikonts and bikonts against only four of the 14 proteins associated with the nuclear envelope (Lbr, Narf, Rfbp and Man1; Table 1). Thus, we did not observe any of the outcomes of the two a priori models, but we obtained an intermediate picture, in which most but not all of the metazoan NPC and NPCa proteins have homologs in other eukaryotic lineages. A unique and ancient origin of the NPC and, by extension, of the nuclear compartment itself would be favored because similar patterns of distribution would be better explained by an inheritance from the LECA than by multiple convergent recruitments. This claim would be strengthened if phylogenies of these eukaryotic ubiquitous proteins are all in agreement with the eukaryotic tree 26. Indeed, phylogenetic analyses of these proteins led to trees in which the relationships between the eukaryotic lineages were generally well preserved; most of the trees displaying apparent phylogenetic oddities could be easily rationalized by reconstruction artifacts due to heterogeneity of evolutionary rates (not shown).

Interestingly, the ubiquitous homologs are broadly located on the NPC structure (Figure 1), suggesting that a large fraction of the genes for NPC components originated once, prior to the LECA (27 of the 33 nucleoporins have homologs in unikonts and bikonts), and that the LECA likely had a complex nucleoplasmic transport system (16 of the 17 proteins have homologs in unikonts and bikonts) and possibly a large and modern-type nucleus.

We reckon that one has to be cautious when making conclusions about the lack of homologs in some lineages, such as conosa, for which no complete genome was available when we conducted this study (Table 1; Figure 1). This reduced our ability to shed light on several steps of NPC evolution. In organisms with complete genome sequences available, such as metazoa, fungi, and green plants, an absence may be interpreted as either a true loss, but also as the outcome of evolution beyond recognition. For example, the absence of a metazoan and fungal Nup214/Nup159p homolog in green plants (despite the presence of the homolog of its partner Nup88/Nup82p) may well reflect a true loss of this gene in the green plant lineage or an innovation in the opisthokont lineage (metazoa and fungi). If this absence is proven to be true, it could suggest some limited structural reorganization of the NPC. However, this apparent absence could also simply reflect a fast evolutionary rate for this protein in green plants or in opisthokonts, or both.

Interestingly, eight proteins (Pom121, Gp210, and the lamina-associated proteins Emerin, Otefin, Lamina A/C, Lamina B1 and B2, Lap1 and Lap2) were found only in metazoa, whereas five proteins (Pom152, Pom34, Ndc1, Nup1p and Nup2p) appeared as fungi specific (Table 1). Could this reflect lineage-specific innovations? In metazoa, Pom121 and Gp210 are involved in the anchoring of the NPC to the nuclear membrane 5. The lack of apparent homologs of these genes in fungi indicates that they likely have an analogous anchoring system. Indeed, structural analyses have shown that three analogous proteins (Pom152, Pom34, and Ndc1) that do not display any sequence similarity with Pom121 and Gp210 perform this function in fungi 6. These observations favor the hypothesis of a lineage-specific innovation with non-homologous replacement, followed by loss of the ancestral anchoring system in one of the two lineages. Additional information about the NPC anchoring structure in other opisthokonts, and in conosa (for which no homologs of those genes have been detected) may help to determine in which lineage (fungal or metazoan) this replacement occurred. A similar hypothesis could be formulated for the metazoan-specific nucleoporins Nup153 and Nup50. Structural analyses revealed that fungi possess analogues of Nup153 and Nup50 called Nup1p and Nup2p, respectively 5. As plants harbor a candidate homolog of Nup50, a replacement of these proteins may have occurred specifically in fungi. An alternative explanation would be that they have evolved beyond recognition. Further investigations of structural data, especially from protists and plants, will be required to further test these hypotheses.

To understand the evolution of NPC protein sequences, we compared evolutionary rates: between markers for all the species (Figure 3); between markers for three given lineages independently (Figures 4 and 5); and within lineages (Figure 6). We produced a very conservative estimate because we considered only the 22 datasets composed of unambiguously aligned sequences having multiple representatives in green plants, fungi, and/or metazoan groups (the datasets used are available in Additional data file 2). Other markers presented too little sequence conservation and/or too limited taxonomic samples in the three lineages analyzed. We show that these 22 ubiquitous proteins present important differences in their rates of evolution (Figure 3a). For instance, some proteins (Nup160 or RanGAP1) displayed on average six times more substitutions than others (Lap2) (Figure 3a). The position within the NPC structure did not explain these differences in evolutionary rates as proteins evolving at either rapid or average rates are uniformly distributed across the NPC and found in almost all of the NPC subcomplexes (Figure 3b). However, such a global average rate of evolution, because it is estimated for all species altogether, is not the most accurate way to describe the evolution of protein sequences, which might be lineage-dependent. Thus, we estimated the evolutionary rates in fungi, metazoa, and plants separately (Figures 4 and 5). This analysis revealed that the markers were not homogeneously slowly or rapidly evolving. In fact, they evolved at different rates in the different lineages, without any general rule and without any obvious correlation with their structural location (Figures 4 and 5). For instance, Nup93 and Nup54 evolved at average rates in metazoa and in fungi, but slowly in plants (Figures 4 and 5). Some markers such as RanGAP1 are slowly evolving in the green plants and in metazoa but evolving at an average rate in fungi, while Importin is slowly evolving in fungi but rapidly evolving in plants and at an average rate in metazoa (Figures 4 and 5). Rae1 protein displays slowly evolving evolutionary rates within fungi and metazoa and average evolving evolutionary rates in plants; Nup133 and Nup160 evolve at average rates within metazoa but very rapidly in fungi, and so on. Evolutionary rates were also sometimes heterogeneous within a given lineage. For instance, Rae1 evolves faster than average in Drosophila melanogaster but slower than average in Mus musculus and H. sapiens (Figure 6).

These irregular rates of evolution, at all levels of analysis (between markers, between lineages and within a lineage) suggest multiple independent adaptations to independent constraints. Because NPC and NPCa proteins are involved in very diverse functions, the contrast between their ubiquitous distribution, their lack of sequence conservation, and their heterogeneity of evolutionary rates probably reflects a higher plasticity of sequences than for NPC structure, which could thus have become frozen very early in eukaryotic evolution.

Yet, if the evolutionary rate of NPC protein sequences is very heterogeneous, the domains detected in 43 proteins by querying the SMART database 28 were generally conserved (Additional data file 10 and Figure 7); 7 out of 43 of the proteins tested presented no domain organization. We found no loss or gain of domains for 23 of the remaining proteins over NPC evolution in four organism representatives of three majors phyla, metazoa, fungi and green plants. Only 12 proteins displayed less than 90% of identical domains between plants, fungi and metazoa, and only half (Narf, Nup214, Luma, Ranbp7, Ranbp8, p30 and Nup35) showed a significant change. For example, Narf has either lost an iron-only hydrogenase domain in H. sapiens and Schizosaccharomyces pombe or gained it in D. melanogaster and A. thaliana. Conversely, other proteins (Aladin, Nup43, Rae1, RanGAP1 and Seh1) show variation only in the number of repeated domains. For example, if we take H. sapiens as a reference, Aladin seems to have gained two WD domains in S. pombe, and one in D. melanogaster, and to have lost two such domains in A. thaliana.

This strong domain conservation for NPC proteins all over the NPC structure and despite the multiple changes in the rest of the sequence illustrates the strength of the structural constraints acting on NPC and NPCa proteins, probably since LECA.

Thus, while the presence of NPC and NPCa proteins seems to be necessary, most of their sequences can be highly adapted and plastic. These differential evolutionary constraints between sequences and NPC structure are an example of tinkering in eukaryotic evolution, a trick to overcome the frozen structural evolution (that is, the structure and complexes in interaction are preserved, but the sequences of their components vary). Thus, while the global structure of the NPC seems mostly preserved and rigid, it is also strikingly flexible outside the preserved domains, enough to accommodate multiple different functions and to interact with an indefinite number of partners.

The age of the NPC structure - as ancient as LECA - raises the question of its origin. The possibility of a pre-LECA NPC deserves consideration. Indeed, a structure comparable to a nucleus (membranes surrounding and isolating the DNA from the rest of the cytoplasm) has been observed in some members of the Planctomycetales, possibly one of the most ancient bacterial phyla 2930. However, available data concerning the nature, the composition, the structure, and the function(s) of these nuclear-like structures in Planctomycetales have not yet established whether they were homologous to the eukaryotic nucleus. Importantly, some Methanogens (Archaea) also display intriguing inner membranes 3132. Could these structures in prokaryotes and eukaryotes have a common origin or did they appear independently in the three domains of life? Moreover, could viruses have played an important role in the origin of the nucleus and of the NPC as sometimes suggested 33?

To address this, we tested whether some phylogenetic connections between the eukaryotic NPC components and some putative prokaryote and viral homologs may be proposed. This may provide some answers, even though the absence of a convincing rooting of the Tree of Life does not allow any obvious temporal polarization 34. For instance, if homologs of the NPC genes were found in prokaryotes, and in particular in Planctomycetales, this could be an argument in favor of a very ancient origin of the genes constituting the NPC (before the separation of the three domains), consistent with a very ancient origin of the nucleus itself. On the other hand, if no prokaryotic homologs are found, the hypothesis of a strictly eukaryotic construction of the NPC (and nucleus) might be most parsimonious.

Hence, we specifically looked for homologous sequences in prokaryotes and viruses, even if they were at first not retrieved when multiple extant eukaryotic seeds were used. Clearly, the large evolutionary distances between eukaryotes and prokaryotes and the heterogeneity of evolutionary rates in sequences complicate such analyses 3536. Ancestral sequences inferred using Codeml 37, software taking into account the heterogeneity of rates of evolution, for genes with sufficiently long unambiguously aligned regions provided us with additional seeds. Interestingly, BLAST searches seeded with these ancestral sequences systematically recovered previously identified eukaryotic sequences (a positive control on the quality of ancestral sequences) and sometimes retrieved new prokaryotic sequences that were otherwise undetected (Tables 1 and 2).

We found seven proteins with such additional prokaryotic homologs, leading to a total of 15 proteins with prokaryotic homologs: 9 are NPCa proteins (p30, Nurim, Importins, Ha95, Luma, Lbr, Rfbp, Ddx19 and Narf), whereas only 6 are NPC proteins (Nup37, Nup43, Seh1, Rae1, Aladin, Sec13R) (Table 2; Figure 8). All the NPC proteins with prokaryotic homologs detected have WD repeated domains, suggesting that this domain, if not convergent, may be very ancient and would have originated before the separation of the tree domains of life. Five of these NPC proteins are involved in the anchoring system (Ha95, Luma, Nurim, Lbr, Narf). Interestingly, all NPC proteins are localized on the nuclear side, except Aladin, which locates near Nup358 on the cytoplasmic face of NPCs 38. In addition, two of the NPCa proteins are involved in nucleocytoplasmic transport (Importins and Ddx19). This result is very suggestive because our phylogenetic approach was very conservative: only 31 proteins were used to infer ancestral sequences (see Materials and methods), and the evolutionary distances and the heterogeneity of the evolutionary rates are obviously larger between prokaryotes and eukaryotes than inside eukaryotes alone. This search could then be improved when additional eukaryotic sequences are known.

From these results, an exciting hypothesis may be that an ancient universal system of transmembrane transport was recruited during early eukaryotic evolution (before LECA) to form the NPC. However, too little is currently known about the function of these proteins in prokaryotes to test this hypothesis.

It was interesting to find homologs in Methanosarcinales since those Archaea could display inner membranes. Yet, the absence of undisputable homologs in Planctomycetales, even if the complete genome of Pirellula were available, does not support a relationship between their nucleus-like structure and the eukaryotic nucleus. In the detail, the taxonomical distribution of prokaryotic NPC protein homologs is intriguing (Table 2; Additional data files 3, 4, 5, 6, 7). The species harboring these proteins are mainly members of Cyanobacteria for Bacteria and Methanosarcinales for Archaea. The prokaryotic homologs of NPCa proteins are more patchily distributed than those of the NPC proteins. They are mainly present in various phyla of Bacteria such as Proteobacteria, Cyanobacteria, Green non sulfur bacteria or the Cytophagales-Flavobacteria-Bacteroides group. This patchy taxonomical distribution could be explained by multiple independent losses, the proteins being kept in some species for different purposes, but also - and more likely - by several independent gene transfers from eukaryotes to prokaryotes. For Ha95, Luma and Nurim, the hypothesis of lateral gene transfers between metazoa and prokaryotes seems the most likely explanation (see for instance the phylogenies of Luma, found only in metazoan and in Mesorhizobium loti and of Nurim, found in some Cyanobacteria plus α-Proteobacteria; Additional data files 5 and 6). These examples of transfers from eukaryotes (and sometimes specifically from metazoa) to prokaryotes suggest that NPC and NPCa proteins can be functional in a prokaryotic cellular context even in the absence of a nuclear compartment. In any case, this illustrates the plasticity, flexibility, multitasking and recruitment potential for these NPC/NPCa proteins, already suggested by their highly specific rates of evolution.

Homologous sequences of all the identified nucleoporins in vertebrates and in fungi 563940 (completed by the list of proteins published in the Nuclear Protein Database 41), of proteins involved in the NPC anchoring system 5642, and of several important protein partners in and around the nuclear envelope (Table 1) were retrieved from the National Center for Biotechnology Information 43 with the programs BLASTP, TBLASTN, and PSI-BLAST 4445. To avoid incorrect assignment for non-homologous sequences containing the phylogenetically weakly discriminant WD domains and FG repeats, we considered as homologous only those sequences with long stretches of sequence homology outside of these regions with repeats. When no homologous sequences were retrieved outside metazoa, additional searches were performed using each new sequence as a seed to complete the retrieval phase and initiate new searches. Homologous proteins were aligned with ClustalW 46 and the alignment was then manually refined with the ED program of the MUST package 47. Regions of unambiguous alignment were manually selected using the program NET from the MUST package 47. All the alignments are available upon request from CB or EB.

Eukaryotic EST databases were mined for each gene with a satisfactory phylogenetic alignment. The EST databases we used (Additional data file 8), included more species than in Mans et al. 3 because they notably contained stramenopiles. This approach is far from being ideal, however, because the absence of an EST in a given lineage does not mean that these species do not harbor the corresponding homologs in their genome. In addition, many homologs were probably not retrieved because of the limited size of the databases. Indeed, the largest database (diatoms and conosa, a group including Dictyostelium and Entamoebae species) provide the largest number of hits.

All protein alignments were used to calculate phylogenetic trees by maximum likelihood (ML), maximum parsimony (MP) and Neighbor Joining (NJ) methods with the programs PHYML version 1.0 48 (JTT+F+Γ model taking into account among-site rate variations), PMBML (JTT+PMB model) 49 and TREE-PUZZLE version 5.1 50, PAUP version 4.0 beta 51 and MUST 47.

We selected a few proteins for further in-depth phylogenetic analyses by maximum likelihood (PROML; nine user defined categories) 52 when they presented a broad taxonomic distribution and enough unambiguously aligned sites. Bootstrap values were calculated with an exact procedure (100 replicates were generated using SEQBOOT 52, and trees were inferred by an ML method with Γ distribution using PUZZLEBOOT) to estimate the robustness of phylogenetic inference.

For 22 proteins with a good alignment and a comparably broad dataset, two conservative estimates of the evolutionary distances between species were deduced from distance matrices, calculated using TREE-PUZZLE version 5.1 with a JTT model corrected by a Γ law and eight categories of rates of evolution 50. First, the average rate of evolution of a given species in reference to the whole dataset shows if a given species X was evolving slower, faster, or at an average rate relative to other species for this gene. This measure allows the identification of rapidly evolving species. Second, the relative rate of evolution compared only with species of the same lineage (when there are at least three) indicates if a given species X was evolving slower, faster, or at an average rate relative to other members of its lineage. This measure provides an insight into the heterogeneity inside a lineage, and allows one to test, for instance, if the acceleration of rates are phylogenetically consistent. These estimates were calculated by Evospeedometer 53.

The presence of domains in the sequences was investigated using the SMART server 28. This also allows, in addition to the HMMER searches of the SMART database, which is the default option, detection of outlier homologs and homologs of known structures, signal peptides, internal repeats, intrinsic protein disorders, and PFAM domains. All NPC and NPCa proteins present in at least two of the three lineages, metazoa (H. sapiens, D. melanogaster), fungi (S. pombe) and green plants (A. thaliana) were investigated.

Ancestral sequences were reconstructed for 31 proteins. Only regions of proteins with significantly long, contiguous and unambiguously aligned regions (>200 successive positions) were used (Additional data file 9). A maximum likelihood tree for each of these proteins was calculated by PMBML (JTT+PMB model), with user-defined categories. The topology of this tree was provided as an intree to CODEML 37 (WAG model, pre-estimated alpha parameter by TREE-PUZZLE version 5.1 50, for eight categories of rates of evolution), which infers the ancestral sequences for each node of the tree. Ancestral sequences were extracted from the outfile of CODEML using ancestRetrieve 54.