ABSTRACT
Planctomycetes are distinguished from other Bacteria by compartmentalization of cells via internal membranes, interpretation of which has been subject to recent debate regarding potential relations to Gram-negative cell structure. In our interpretation of the available data, the planctomycete Gemmata obscuriglobus contains a nuclear body compartment, and thus possesses a type of cell organization with parallels to the eukaryote nucleus. Here we show that pore-like structures occur in internal membranes of G.obscuriglobus and that they have elements structurally similar to eukaryote nuclear pores, including a basket, ring-spoke structure, and eight-fold rotational symmetry. Bioinformatic analysis of proteomic data reveals that some of the G. obscuriglobus proteins associated with pore-containing membranes possess structural domains found in eukaryote nuclear pore complexes. Moreover, immuno-gold labelling demonstrates localization of one such protein, containing a β-propeller domain, specifically to the G. obscuriglobus pore-like structures. Finding bacterial pores within internal cell membranes and with structural similarities to eukaryote nuclear pore complexes raises the dual possibilities of either hitherto undetected homology or stunning evolutionary convergence.
INTRODUCTION
A nucleus surrounded by a double membrane envelope is a universal characteristic of eukaryote cells [1] and is thought to be universally absent from the prokaryote domains Bacteria and Archaea. The nucleus is accompanied by a complex apparatus for transport of macromolecules, including a multi-protein nuclear pore complex embedded in the nuclear envelope, and a soluble transport system [2]. The nuclear pore complex and many of its component proteins appear universal among eukaryotes, spanning yeast, trypanosomes and vertebrates [3], and the Last Eukaryotic Common Ancestor already possessed a complex version of the nuclear pore complex, nuclear envelope and connected endomembrane system [4,5,6]. Gemmata obscuriglobus, a member of the bacterial phylum Planctomycetes, possesses compartments including the nuclear body containing DNA and ribosomes, riboplasm containing ribosomes but no DNA, and the paryphoplasm, a ribosome-free compartment [7,8]. By whole-cell tomography and conventional transmission electron microscopy (TEM), we have previously established that the riboplasm and nucleoid compartments of G. obscuriglobus cells are bounded by membranes [7,8,9]. Confocal fluorescence micrographs of cells where the nuclear region has been stained with DiOC6 membrane stain and DAPI DNA stain are also consistent with the membrane-bounded nature of the DNA in this organism [10]. An earlier study of G. obscuriglobus internal membranes differs in its conclusions from those of ours, proposing that only one invaginated membrane exists in such cells and that there is no membrane enclosure of the Gemmata chromosome [11], and instead a tubulovescular model for internal membranes has been proposed [12]. Our tomography analysis of G.obscuriglobus cells demonstrated that the internal membranes do not display continuity with the cytoplasmic membrane apposed to the cell wall [7]. Such a cell plan implies specialized internal membrane(s) distinct from the cytoplasmic membrane and would also require some form of transport system (e.g. pore structures) for macromolecules passing between the internal compartments and the rest of the cytoplasm. This hypothesis is consistent with the recent finding of confinement of translation to non-nucleoid regions of G. obscuriglobus cells [13]. A corollary of our study of G. obscuriglobus internal compartments was that several different types of membranes might be isolatable from lysed cells, and we have confirmed this concept here. There has been extensive debate regarding the evolutionary significance of compartmentation in G. obscuriglobus [14,15,16]. However, such discussions have been limited by lack of knowledge about Gemmata membrane composition and the structure of internal membranes in particular. Recently, components of cell walls characteristic for Gram-negative bacteria such as peptidoglycan and lipopolysaccharide have been found in Gemmata obscuriglobus [17,18] correlating with other data on occurrence of peptidoglycan in Planctomyces limnophilus [17]and an anammox planctomycete species[19]. The exact location of these components within planctomycete is yet unknown, but the results suggest a potential for planctomycete cell plan to relate more closely to Gram-negative cell wall and structure than previously thought as outlined in published hypotheses[20,21]. The implications of these results for interpretation of planctomycete internal membranes and their evolutionary significance are not yet clear. Here we present evidence that some of the internal membranes of G. obscuriglobus possess pores with complex structure. Moreover we identify proteins specific to these membranes, some of which possess structural domains also found in eukaryote nucleoporins. The evolutionary implications of these results are considered, both from the perspective of common ancestry with the eukaryote nuclear pore complex, and from the viewpoint of convergent evolution.
RESULTS AND DISCUSSION
Planctomycetes possess pores in internal membranes
Pore-like structures (termed ‘pores’ throughout the remainder of the text) in the internal membranes of the planctomycete bacterium Gemmata obscuriglobus can be observed in transmission electron microscopy images of thin or thick sections of whole cells (Fig 1A, 1B, S1 Fig). When thin sections of cells prepared either via high-pressure freezing or cryosubstitution without high-pressure are examined, favourable planes of section reveal pore-like structures along sectioned nuclear envelopes. In Fig 1B a circular complex (arrowhead) appears in a gap between folded regions of the nuclear envelope membranes (arrows). Appearance of the pore is comparable to that of sectioned eukaryote yeast pores prepared by high-pressure freezing [22]. The diameter ca. 35 nm of the circular complex in Fig 1B is consistent with the diameter of pore structures in negatively-stained preparations (see below). One ring of the pore complex analogous to that of the nuclear and cytoplasmic rings of the eukaryote nuclear pore may be visible en face by a favourable tangential plane of section (Fig 1B). The position of such a pore within the membrane in a whole cell can be seen in S1A Fig, and the details of ring and central plug and spoke-like structures within such pores seen in face are also evident in S1B Fig.
When cells of G. obscuriglobus are gently lysed, internal membrane can be released through the broken cell walls (Fig 1C), the overall shape of which is consistent with a sphere or compressed sphere (Fig 1C and 1D). The released membranes are covered with circular ring-like structures, possessing a dense circular centre (Fig 1E).
We have also demonstrated the presence of pores on the internal membranes of intact G. obscuriglobus cells using the freeze-fracture technique (Fig 2A and 2B). In freeze-fracture replicas, it is possible to unambiguously identify the membrane surfaces displaying pores on a single major internal membrane-bounded organelle, since the position of this membrane-bounded structure is clearly visible within the cross-fractured cells. The pores, circular structures, appear on one outer surface or fracture surface of the organelle envelope, which clearly overlies another fractured membrane of this envelope, indicated by the splits visible in the envelope. The complexity of the organelle membranes has been observed in Gemmata in tomographic and thin-sectioning studies [7,8,10]. In enlarged view (Fig 2A insets, Fig 2C) each pore represents a circular complex in which an outer ring surrounds a centre which usually appears darker than the outer ring to a degree depending on the metal shadow angle relative to the pore aspect. The size of pores in the freeze-fracture replica in Fig 2C are ca. 32 – 45 nm, which is consistent with dimensions obtained from negative staining but such dimensions would be expected to be more variable due to metal shadow used to prepare the replicas. It is of interest that we obtained similar freeze-fracture replica images of pore-like structures with central core and ring structure on internal membranes in another Gemmata strain, CJuql4 (ACM5157) (S2 Fig), closely related to the general Gemmata cluster phylogenetically, and a member of the same roughly genus-level phylogenetic group with high bootstrap support in a phylogenetic tree of planctomycetes [23]. These pore-like structures have a central core and ring structure and are 23-38 nm in diameter. Cells of this additional Gemmata group strain have been shown to possess an internal membrane-bounded nucleoid-containing compartment in thin sections, similar to that of Gemmata obscuriglobus [23]. These results therefore reinforce the conclusions from Gemmata obscuriglobus regarding the significance of pores as structures associated with the Gemmata group-characteristic internal membrane envelopes often surrounding the nucleoid region. A membrane-bounded region surrounding the nucleoid represents a type of structure known only from the Gemmata group strains so far among either planctomycetes or among any other domain Bacteria.
In negatively-stained membrane fragments released from cells lysed by sonication, the pores consist of a thin inner ring immediately surrounding the electron-dense pore centre, and an outer thicker ring distinguishable from the inner ring (Fig 3A). Assuming uniform distribution, there are ca. 87 such pores per µm2 of such a membrane sheet. At higher magnification the pores of G. obscuriglobus appear to be composed of subunits with internal “plug” (Fig 3B and 3C). In membrane fragments released by mechanical lysis and negatively stained with uranyl acetate, the outer diameter of the pore based on the outer ring diameter was calculated as 33.5 ± 2 nm, the diameter of the inner ring as 17.5 ± 2 nm, and that of the electron-dense pore centre (equivalent to an ‘inner pore’ or ‘central plug’) as 9.5 ± 0.8 nm (S3 Fig).
After treating of the membranes released from lysed cells and isolated (as ‘fraction 3’, see below) via density gradient fractionation with detergent, aggregates of individual pores could be detected (S4 Fig) with only degraded membrane between them. Individual pores in such preparations show a central dense core surrounded by a light ring and in some cases material projecting from the outer rim of the ring possibly representing spokes normally connecting inner to outer ring in intact pore complexes.
In addition to these pores, two smaller classes of pores are also found. They may be seen most clearly in membrane fragments released via sonication (Fig 3A). The larger of these two classes consists of rings that are 14.5 ± 2 nm in diameter with an inner dense centre of 5 ± 1 nm wide, while the smaller class consists of pores 6± 0.9 nm wide and which also possess a dense centre and can appear in clusters. Neither of these smaller pore types seem comparable in structure to eukaryote nuclear pore structures in the sense of possession of both inner and outer rings as well as an inner dense centre. It is also possible that at least the 14.5 nm diameter class is the result of a reverse view (‘basket’ side view) of only part of the larger pore structures seen because of a folded membrane, since they appear most clearly in examinations of negatively stained membrane sheets rather than via other EM preparation methods. If true additional pore types, they suggest either specialization in pore function in these membranes or some stages in assembly for the largest pores. It is the largest pore type we have seen which is in any case most relevant to the argument of this paper since it is the class displaying most complex structure and is the class comparable with that of eukaryotic nuclear pores.
Thus, we have determined here from studies of whole cells and membranes released from lysed whole cells that at least one type of internal membrane contains pores. This has been demonstrated by three distinct electron microscope preparative techniques – thin-sectioning and freeze-fracture replica technique of the whole cryosubstituted cells, and negative staining of membranes released from lysed cells. If pores are genuine structures, one would predict that several different methods should be consistent e.g although negatively stained internal membranes released from lysis display the clearest examples of pores, methods such as freeze-fracture and thin-sectioning should also reveal entities with comparable structure on internal membranes in whole cells, and this is indeed the case. If such structures are significant functionally, one would expect they would be associated with a specific type of internal membrane and be able to find differences in protein composition of that distinct type of internal membrane.
To confirm the association of pores with a specific type of internal membrane and to investigate the composition of such membranes further we fractionated the internal membranes and applied proteomics and EM methods including immunoelectron microscopy to localize a specific protein of potential relevance to pore structure.
There are three types of membranes in G.obscuriglobus cells, one of which contains pores
To purify membranes for structural and proteome studies we applied a two-step density gradient fractionation technique (S5 Fig). The discontinuous and subsequent continuous gradient fractionations aimed to separate the membranes at the highest possible level. This procedure resulted in appearance of three distinct membrane types (S6 – S8 Fig). Only one of the fractions (fraction 3, consisting of characteristic ‘canoes’), displayed pores on the membrane surfaces when examined via TEM after negative staining (S7 Fig). The large pores are present at high density (ca. 200 / µm2 of membrane sheet) and similar in size and structure to the complex pores seen in unfractionated membrane released from lysed cells (e.g. Fig. 3A). We were able to apply clear markers for two of these fractions via an antibody against a beta-propeller-containing protein (a protein exclusive to fraction 3, see below and S9A Fig), and via a specific antibody against G. obscuriglobus clathrin-like membrane coat (MC) protein gp4978 (shown in a previous study to react specifically with membrane vesicles associated with endocytosis-like protein uptake in G. obscuriglobus [24]. The anti-beta-propeller-containing protein reacted only with fraction 3 but not with fractions 2 or 6, and the anti-MC protein antibody demonstrated reactivity only against fraction 2 (S9B Fig).
The gradient fractionation technique demonstrates that membranes related to distinct internal structures of G.obscuriglobus can be separated, and the probing with antibodies clearly shows that the fractions do not contain significant amount of cross-contamination.
Structural analyses of the pores reveal their similarity to the nuclear pores of eukaryotes
Electron tomography of the fraction 3 membranes shows that, topologically, these pores are membrane insertions (Fig 4A and 4B, S1 Video and S2 Video). By analysing digital slices through the tomogram, we see continuous membrane envelope followed by a pore structure and then followed again by continuous membrane (Fig 4B). Such pores display a projection on one side and in some slices the plug could be seen (Fig 4A and 4B). Cryo-EM analyses (S10 Fig) and a modified Markham rotation analysis of the electron micrographs from frozen-hydrated pore-containing membranes (Fig 4C and 4D) suggests an eight-fold symmetry for the organization of these pores, indicating that these complexes are likely to be modular constructions. It should be noted that we have used the same technique here as applied in the first demonstration of eight-fold symmetry in nuclear pores from invertebrate and vertebrate animal species[25]. Eight-fold rotation gave strongest reinforcement e.g. relative to 7-fold rotation. Markham rotation cannot by itself prove 8-fold symmetry of pore structure, but taking both the Markham analysis and the clear radial and octagonal symmetry visible in the original micrographs into account, on current evidence, 8-fold symmetry is the most likely possibility.
3-D reconstructions of the pores inserted into the membranes confirms a basket-like structure, with struts connecting the pore to a distal ring, and projecting from the part of the pore inserted in membrane (Fig 5 and S3 Video).
In summary, the pores embedded in the internal membranes of G.obscuriglobus display startling similarities to the nuclear pore complex of eukaryotes [26,27,28,29]. For structural comparison of the Gemmata pore with a eukaryotic pore model [2] see S11 Fig. The pores are comparable to the appearance and distribution of nuclear pores on negatively stained isolated nuclear membranes of yeast, plant and mammalian liver cells [30,31,32]. Those similarities include apparent eight-fold symmetry of its components, the presence of a basket structure projecting asymmetrically from the membrane plane, and the presence of at least two rings within that plane. These structurally complex pores of G.obscuriglobus are considerably smaller than nuclear pores on isolated nuclear membranes of eukaryotic cells. At ca. 35 nm in frozen-hydrated membranes, they are only approximately a third the diameter of characterized nuclear pores [28]. The frozen-hydrated NPCs of yeast are 96 nm wide [33], the Xenopus or yeast nuclear pore complexes in thin-sectioned cells are ca. 120 nm and 103 nm wide respectively [22,34], negatively stained NPCs of Xenopus are 107 nm wide [35], and detergent-released negatively stained are 133 nm wide [36], and finally intact Dictyostelium slime mould NPCs studied by cryo-electron tomography are 125 nm wide [26]. However, despite the difference in size of the Gemmata pores compared to those of eukaryote nuclear envelopes, there are clear analogies in structure, since the eukaryote pores also have a central plug and a central ring-like assembly composed of spokes sandwiched between a cytoplasmic ring and nuclear ring[22], and in negatively stained detergent-treated nuclear envelopes of Xenopus oocytes pores are visible as rings containing a central plug connected to the ring by spokes [27].
Our structural studies of the Gemmata pores are summarised in a deduced model (Fig 6). From Cryo-EM (Fig 4C and 4D) the pores from the “basket side” have two rings of ca 20 nm (inner ring) and 35 nm (outer ring) in diameter. The pores from this side display an octagonal symmetry, consistent with rotational folding analysis of cryo-electron micrographs from frozen-hydrated TEM of pore-containing membranes (Fig 4C and 4D). Within the central core region is a central ‘plug’ of ca. 10 nm in diameter (Fig 4C and 4D). At the higher magnification the central plug is seen as a structure connected to the inner side of the pore (Fig 3C). From the opposite side the pores also have two rings. As from transmission electron micrographs of the whole compartments lysed by grinding in liquid N2 (Fig 1C and 1E), Pt/C-shadowed replicas of whole cells (Fig 2C), and 3D reconstructions of the pores (Fig 5A and 5E) the outer ring is ca. 25 nm and the inner ring is ca. 20 nm in diameter. As calculated from the isolated membranes used for tomography studies (Fig 4B) the maximum distance from the top side of the pore to the end of the basket is 25 nm.
Whole membrane proteome analyses reveal distinct protein content for three types of internal membranes and the cell wall
Proteomics was applied to cross-compare proteins from the fractions obtained via gradient centrifugations (see above) in order to distinguish proteins unique to the pore-containing membranes. Protein composition appeared different for pore-containing membranes (fraction 3) compared to those with no visible pores (fractions 2 and 6) (Fig 7A). 342 membrane and membrane-associated proteins were identified from all fractions examined (Fig 7B, S1 Table), 46% of which are annotated as hypothetical proteins. In non-pore-containing membranes (fractions 2 and 6), many constituents of the respiratory chain, ABC transporters and secretion system components were identified. The recently described MC-like vesicle-associated protein [37] was found exclusively as a constituent of fraction 2 (S1 Table). Fraction 2 appears enriched with vesicles, which are present at high number within the paryphoplasm. Some vesicles may be formed during endocytosis and are derived from the cytoplasmic membrane [24,38]. Proteins such as glycosyl transferases, a number of dehydrogenases, including the NADH-dependent dehydrogenase, and the periplasmic solute-binding protein, were restricted to fraction 6, which suggests that this fraction is enriched for the cytoplasmic membranes, and the other two fractions do not contain detectable amount of cytoplasmic membrane debris. ATP synthase was found only in this cytoplasmic membrane fraction (fraction 6) and in the vesicle-enriched membrane fraction (fraction 2), and not in the pore-containing membrane fraction (fraction 3). This indicates that the function of the pore-containing membranes is probably not that of electron transport or energy generation. In the pore-containing membranes (fraction 3), we identified 128 proteins, 39 (30.5%) of which were unique to this fraction (S2 Table). Common proteins found by proteomics analysis in all three fractions (S1 Table) are predicted pilins and predicted ribosomal proteins, which we believe are the result of contamination since actual pili structures were detected in small number by electron microscopy in all three fractions, and heavy ribosomal subunits which might be distributed along the entire gradient. In our proteomic analysis regarding significant proteins correlated with presence of pore structures, we only consider proteins exclusive to fraction 3.
It should be noted that the pores described here structurally resemble crateriform structures, a characteristic signature structure of planctomycete surfaces [39] and isolated cell walls[40] including those of G. obscuriglobus[41]. These crateriform structures could be seen on the surface of whole cells as circular regions and in the cell slices used for TEM as pits protruding the cell wall and cytoplasmic membrane, confirming published data on G. obscuriglobus [42]. It might be proposed that the pores described here represent in fact the crateriform structures as seen in purified cell walls (Fig S12). However, these crateriform structures when negatively stained display a dense centre but only one surrounding ring, so apparently differ from the more complex internal membrane pores. The membrane fractions we used as starting material for the gradient fractionations should not have contained significant amounts of wall, as they are effectively eliminated during preparations of the membranes for initiating the gradient fractionations; the walls are denser than any of the membranes and are pelleted and thus separated at relatively low centrifugation speed. Any walls remaining after this step are pelleted during fractionation in any case. The walls also cannot be lysed in the buffers usually used for dissolving the membranes for Western blot analysis, so their proteins would not be detected in isolated gradient-fractionated membranes. Compared to the membranes, the walls are highly resistant to boiling in 10% SDS, and such boiling is used for purification of the walls from the membranes during wall isolation. In a separate experiment we have isolated the walls via boiling the G.obscuriglobus cells with 10% SDS and analysed their protein content. The cell walls revealed some proteins homologous to proteins which have in previous studies been shown to be characteristic of the wall of planctomycetes [43,44]. Via proteomics we have identified the major constituents of the walls, the so-called YTV-proteins (Gobs_U38067, GobsU_28375, and GobsU_21360). The bacterial cell wall marker peptidoglycan has now been reported to comprise at least part of the wall composition in G. obscuriglobus and some other planctomycetes [17,19], and lipopolysaccharide has been reported in Gemmata obscuriglobus [18] but proteins appear to comprise significant proportions of wall in G. obscuriglobus as well as other planctomycetes [40,41,45]. However, no planctomycete cell wall/surface protein homologs were found via proteomics in any of the three membrane fractions isolated from fractionation of lysed cells, reinforcing evidence from analysis of enzyme markers that all of these three fractions are from membranes other than cell wall/cell surface structures [42]. In addition, sectioned whole cells immuno-gold-labelled with a fraction 3 specific antibody (see below) were shown to label only internal membranes, and there was no labelling of walls or other surface structures. Thus we conclude that the pores in fraction 3 membranes are not wall/surface crateriform structures either on the criterion of location within the cell or on a criterion of protein composition. The pores represent structures embedded into internal membranes and do not represent crateriform structures or any other wall components. This implies potential performance by such internal membranes of functions such as transport of material between internal membrane-bounded cell compartments.
Thus, we have demonstrated here that the three membrane types separated by density gradient centrifugation are distinct from each other in their composition of specific proteins. Continuity of cytoplasmic membrane with internal Gemmata membranes has been proposed as part of the concept of the planctomycete cell plan as essentially one of a classical Gram-negative cell [11,46]. However, our data concerning protein composition of isolated membrane fractions does not support this concept, but rather is consistent with a concept that genuine internal compartmentalisation exists within the Gemmata planctomycete cell, where the compartments are separated from each other by different types of membranes. Our data is also consistent with distinction of cytoplaamic membrane from two different types of internal membranes.
Bioinformatic analyses of the membrane proteome
We characterised the identified set of proteins using a range of bioinformatics-based analyses (S Text and S3 – S6 Tables). Initial blast analyses indicated that many of the proteins showed little or no similarity to proteins outside G. obscuriglobus. We therefore performed Blast cluster analysis (VisBLAST) to establish whether any proteins in the pore-containing membranes exhibit sequence similarities to one another. We also performed profile-based (phmmer) screens to search for more distant similarities, and ran structural predictions using Phyre2 [47] to assess similarity of proteins to known folds. Sequence and structural analyses of our membrane proteomics data revealed the presence of a number of bacterial transmembrane proteins, including outer membrane efflux proteins, translocons and porins (S Text), underscoring the bacterial nature of these membranes. However, none of these were unique to pore-containing fraction 3, so their origin as contaminants from cell structures or components other than membranes of fraction 3 is possible, and they cannot be implied as characterizing any specific fraction 3 membranes. In addition, published structural data are not consistent with any of these transmembrane or ‘outer membrane’ protein homologs forming pores with dimensions and attributes similar to the pores we observe in G.obscuriglobus.
Cluster analyses performed on all proteins identified through proteomics revealed two groups of proteins from fraction 3 with substantial sequence similarity (S13 and S14 Fig). Phyre2 models generated for one of these clusters identified a conserved C-terminal beta-propeller fold for 8 of the 11 proteins making up this cluster (Fig 7C; S15 – S18 Fig, S4 Table). Beta-propeller folds are found in protein constituents of eukaryotic nuclear pore complexes and coated vesicles, and it has been proposed that the eukaryotic nuclear pore and coatomer complexes evolved from a suite of membrane-curving proteins with common structural elements [48]. We therefore searched for evidence of other folds associated with eukaryotic nuclear pore complexes. Most notably, we identified two fraction 3 proteins (ZP_02735673 and ZP_02736511) that model well (>95% confidence, Phyre2) against clathrin adaptor core proteins, exhibiting an alpha-solenoid architecture (S Text, S18 Fig, S3 Table).
The presence of protein folds characteristic of the eukaryote nuclear pore complex [49] is intriguing in light of our deduced pore model, since in the eukaryote nuclear pore complex, beta-propeller-and alpha solenoid (stacked alpha-helices)-containing proteins act as scaffold proteins [50,51]. Beta-propeller proteins form vertices in a lattice-like model of the NPC and have special sequence-independent protein—protein interaction functions [52] while stacked alpha-helices of other scaffold nucleoporins are central to the lattice model interactions, forming edges of the NPC scaffold lattice [50] and are also structurally related to soluble proteins significant to nucleocytoplasmic transport through nuclear pores [53]. While it is remarkable that structural prediction yields folds known from the eukaryote nuclear pore complex, neither β-propeller folds nor alpha-solenoids are unique to the eukaryote nuclear pore complex or endomembrane system, and examples of both folds are known from both Bacteria and Archaea [54]. For all G. obscuriglobus proteins carrying folds also found in eukaryote nuclear pore proteins, we find no evidence of substantive sequence similarity with eukaryote counterparts. Evidence of such similarity might be expected if recent horizontal gene transfer from eukaryotes was their origin. We therefore conclude these genes are not the result of recent transfer from eukaryotes. Our data instead indicate these are genes of bacterial origin.
Immuno-gold labelling confirms association of the pores with internal membranes
One of the proteins identified in fraction 3 exhibiting a β-propeller fold (ZP_02736670), was selected for antibody generation with the aim of using the antibody to immunolocalize the protein. This antibody (ab6670) showed high specificity as established by Western Blot, and reacted specifically with fraction 3, which comprises pore-containing membranes (S9B Fig). It was used for immunolocalization experiments to assess localization of the pore-containing membranes within the cells. On whole sectioned cells, gold particles were observed exclusively at membranes within the cell cytoplasm and internal to both cytoplasmic membrane and paryphoplasm (ribosome-less cytoplasm). The antibody labelled membranes comprising the nuclear body envelope and membranes associated with riboplasm (ribosome-containing cytoplasm) (Fig 8 and S19 Fig). Consistent with this result, the antibody also recognised pores in the purified membranes from fraction 3 (Fig 9 and S19 Fig) Membranes from other fractions, without such pores, were not labelled when tested. Such labelling is consistent with the presence of the ZP_02736670 β-propeller fold protein within pore complexes found via proteomics exclusively in fraction 3 (the origin of the protein data from which the peptide antibody was prepared).
Evolutionary implications of the findings
Our results imply two alternative possibilities, both equally profound. One is that the architectural similarity of pores in Gemmata bacteria and eukaryotes may be the result of a shared deep evolutionary origin, so that they are homologous and related by vertical inheritance. This would be consistent with past analyses that have placed Planctomycetes among the deepest branching phyla within domain Bacteria [55], and more recent analyses support the view that other members of the PVC (Planctomycetes, Verrucomicrobia and Chlamydiae) superphylum also carry compartmentalized cell plans [56,57]. In our view, the lack of detectable sequence similarity precludes homology resultant from recent horizontal gene transfer from a eukaryotic source. More generally, the lack of an unambiguous signal for common descent, together with the notable differences in size, weakens the case for common descent of the G. obscuriglobus pores and eukaryotic nuclear pores. An alternative interpretation to deep shared evolutionary origin of pores and pore complex proteins is that the similarities we observe between the G. obscuriglobus pore complex and the eukaryote nuclear pore complex are the result of convergence spanning all the way from gross architecture down to the recruitment of individual protein components with shared domain architecture. Under this interpretation the G. obscuriglobus pore complex is an analog of the eukaryote nuclear pore complex rather than a homologous structure derived by descent from a common ancestor. In other words, similar solutions to macromolecular transport across internal membranes and between cell compartments may have evolved independently, once in eukaryotes and once in planctomycete bacteria, via the parallel co-option of protein folds present in both bacteria and eukaryotes. The structures we report support the view that these solutions are architecturally similar, and in this regard, further characterization of the G. obscuriglobus pore will be critical for understanding the evolution of internal cell structure in this bacterium. The pore structures we have found are not straightforward to explain within the context of recent exciting results regarding Gram-negative bacteria cell wall components in planctomycetes, but there is no evidence from our data consistent with a concept that pore-containing membranes are outer membrane or other cell wall components, or that the pores contain any wall-specific proteins. Such pore structures are however consistent with the known unique internal cell structure features of planctomycetes, and consistent with distinctive planctomycete cell organization and functional properties. Significantly, when considered alongside the recent discoveries of endocytosis [24] and compartmentalized transcription and translation [13], the parallels between the cell biology of Gemmata obscuriglobus and that of eukaryotes are nothing short of remarkable, whether due to hidden homology or an analogous reuse of a similar set of protein folds. Detailed analysis of the composition of the internal bacterial pores will no doubt enable the evolutionary origin of these structures to be definitively established.
MATERIALS AND METHODS
Cell lysis and electron microscopy of nuclear body envelopes
In the case of mechanical lysis experiments, Gemmata obscuriglobus was grown on M1 agar for 6 days at 28 ˚C, and growth harvested by washing into filtered Milli-Q water. For mechanical lysis via grinding with alumina, suspensions were mixed with alumina powder (Sigma Alumina Type A-5) in an eppendorf tube and a plastic pestle manually rotated within the tube for ca. 60 sec to lyse cells. The supernatant from this homogenate after allowing particles to settle was either harvested directly for negative staining or this supernatant purified by centrifugation using a microfuge at 20160 g. Negative staining was performed by mixing a homogenate drop on a pioliform–coated specimen support grid with 1% uranyl acetate containing 0.4% sucrose. In the case of preparations that were sonicated, G. obscuriglobus culture was grown on M1 medium for 8 days at 28˚C. Cells were suspended in 1 ml of sterile Milli-Q water, and sonicated in a Branson Sonifier 250 at amplitude output level 1 for ten thirty second intervals, separated by 30 sec rests. The resultant suspension was pelleted in a benchtop centrifuge for five mins at 20160 g, and the pellet was resuspended in 50µL sterile Milli-Q H2O. Cells from these suspensions were negatively stained using 2% ammonium molybdate pH6.5 (mixture of 5µl each of suspension and stain was prepared on a carbon-and pioloform-coated specimen support grid followed by removal of excess fluid with filter paper and air-drying).
Cryo-electron microscopy
4 µL aliquots of density-gradient-purified membrane fractions from sonication-lysed G. obscuriglobus cells were deposited onto holey carbon films on hexagonal 200 mesh copper grids (‘C-flat’, Protochips, NC) in the humidified chamber of a commercial thin-film vitrification apparatus (CP3, Gatan, Pleasanton, CA). An additional 4 µL of colloidal gold (nominal size 10-12 nm) was added to the membrane fraction and the grid was blotted from both sides for 3.5 seconds before automatic propulsion into liquefied ethane. Grids were transferred under liquid nitrogen to a cryo-sample holder (Model 914, Gatan) for transfer to the microscope and observation at a stable temperature of approximately −175 Celsius. A JEM-1400 transmission electron microscope (JEOL, Japan) fitted with a high-contrast polepiece and LaB6 cathode, and operating at an accelerating voltage of 120 kV was used for data acquisition. Micrographs were recorded at a nominal defocus value of −10 µm and an electron dose of approximately 4 000 electrons/nm2/micrograph. Detector noise was reduced by means of a median filter with a radius of 1 pixel.
Electron tomography from thick sections
For electron tomography of isolated membrane samples, 300nm thick sections were cut using Leica EM UC6 ultramicrotome (Leica Microsystems, Austria). The membranes were cryofixed as described for thin-sectioning. Dual-axis tilt-series data was collected at 39000x magnification on an FEI Tecnai F30 (FEG)TEM (FEI Company, the Netherlands) operating at 300kV, over a tilt range of +/−66º at 1º increments for the a-axis and 2º increments for the b-axis, using SerialEM software (The Boulder Lab for 3D Electron Microscopy, USA).
Membrane fractionations
Cells were collected from two to three M1-agar plates, washed once, and resuspended in 500 μL of bt-DMSO buffer [10mMbis-Tris (pH 7.5), 0.1mMMgCl2, and 20% DMSO] supplemented with 10 μL of protease inhibitor mix (Protease Inhibitor Mixture Set 3; Merck), 10 μg of DNase, and 10 μg of RNase. Cells were then sonicated using a Branson Sonifier 250, and unbroken cells were spun down in a microfuge at 5,000 × g for 10 min. Supernatant was centrifuged at 100,000 × g in a Beckman Coulter tabletop ultracentrifuge (Optima TLX). The pellet was resuspended in 500 μL of bt-DMSO buffer, loaded onto a five-step sucrose gradient, and centrifuged in an SW60 rotor on a Beckman Coulter L8-60M ultracentrifuge at 215,000 × g for 4 h. Visible pink or white membrane fractions were collected via puncturing the side of the tube and to remove sucrose, centrifuged at 100,000 × g in a Beckman Coulter tabletop ultracentrifuge (Optima TLX), and the pellets were resuspended in ≈500 μL of bt-DMSO buffer. The material was loaded onto a 20–60% or 30-70% sucrose/bt-DMSO continuous gradient. After centrifugation using SW60 rotor, for 16 h at 215,000 × g, ≈400 μL fractions were collected from a puncture at the bottom of the tube, To concentrate and remove sucrose, fractions (in all cases containing a visible band) were diluted with bt-DMSO buffer and centrifuged at 100,000 × g in a Beckman Coulter tabletop ultracentrifuge (Optima TLX). The pellets were resuspended in 5mM Tris, pH7.5 and immediately used for electron microscopy experiments or proteomics. Each fractionation was performed twice using separate culture batches of cells, and 2-3 technical replicates from each fractionation were used in proteomics experiments.
Analysis of structural symmetry
To assess symmetry of structure within pore-like structures within a pore-containing membrane fraction purified from membranes released from lysed G. obscuriglobus cells, Markham rotation [58](Markham et al., 1963) was performed according to the modifications suggested by Friedman [59](Friedman, 1970) by assuming any possible symmetry up to and including 8-fold symmetry. Pores in isolated pore-containing membranes from fraction 3 from the membrane fractionation described above were imaged en face by cryo-electron microscopy and were extracted from images and rotated in silico by set increments corresponding to the assumed symmetry. For example, reinforcement of possible 8-fold symmetry was assessed by superimposition of the corresponding 45.0 degree rotations, 7-fold by 51.4 degrees and so on. The procedure was conducted on raw data only rather than the bandpass-filtered structures. ImageJ was used for all feature extraction, rotation and summation.
Preparation of material used for Markham rotation: Isolated membranes from fraction 3 of the density gradient fractionation of membranes from lysed Gemmata cells were processed for cryo-electron microscopy and symmetry analysis via Markham rotation. Briefly, isolated membrane sheets were vitrified by rapid immersion in liquid ethane prior to mounting the samples in a cryo-sample holder (Model 914, Gatan, Pleasanton, CA) and imaging the frozen-hydrated specimens in a JEM-1400 transmission electron microscope (JEOL Ltd, Japan) equipped with a charge-coupled device detector (Gatan) and low-dose exposure conditions (< 4,000 electrons/nm2) to avoid radiation damage. The accelerating voltage was 120 kV and the nominal magnification of 15,000 corresponded to a pixel size of 0.69 nm at the detector. A minority of pores appeared to be markedly ellipsoid in projection. This was found to be the result of a tilted or folded membrane, which was taken into account by manual tilting of the specimen. Note that this differs from slight deviations in circularity that probably represent the respective functional states of transport-competent pores[60].This tilt was always less than 10 degrees, indicating that the untilted membrane sheets were mounted approximately orthogonal to the beam.
Freeze-fracture electron microscopy of whole cells
Cells from a 14-day culture of G. obscuriglobus ACM 2246 were grown on soil extract agar medium. Cells were harvested directly without chemical fixation into 20% (v/v) aqueous glycerol as cryoprotectant prior to freezing in liquid Freon 22. Fracturing was performed using a Balzers BAF 301 apparatus fitted with a complementary fracturing device, at −115 ˚C and 10−7 torr (1 torr = 133 Pa). Replicas were produced using platinum/carbon and stabilized with a layer of carbon.
Cryosubstitution and thin-sectioning
For cells cryofixed by plunging into liquid propane, cells of G. obscuriglobus were cryofixed using a Reichert-Jung KF80 cryofixation system. Cryosubstitution was performed with 2% osmium tetroxide in molecular-sieve-dried acetone at −79 ˚C (dry ice/acetone bath) for 50 hr. The temperature was increased to −20 ˚C over 14 hr. Specimens were brought to room temperature and then washed in acetone. Specimens were embedded in Epon resin, then sectioned and stained on pioloform-covered specimen support grids with aqueous uranyl acetate and lead citrate. For cells cryofixed by high-pressure freezing, cells of G. obscuriglobus were cryofixed by first mixing with 2% agarose and placing the suspension between hexadecene-soaked planchettes, then frozen using a BAL-TEC HPM 010 High-Pressure Freezer. Frozen cells at −160 ˚C were warmed to −85 in 1.9 hrs at 4 C/hr, stored at −85 ˚C for 52 hrs and then raised to −20 ˚C over 11 hrs in a Leica EM AFS cryosubstitution apparatus. Cells were then embedded in Epon resin and sectioned and stained as above.
Electron microscopy
For experiments other than those involving tomography, specimens were examined using a JEOL 1010 transmission electron microscope operated at 80 kV.
Proteomics
Before proteomics, pellets of membrane fractions in Tris buffer were dissolved using Laemmli buffer, protein concentration was measured using BCA Protein assay (ThermoFisher Scientific) and 20 µg of each suspension was loaded onto PAA gels. The proteins, separated by 1-D SDS-PAGE electrophoresis, were cut out from the 8-12% PAA gels for mass spectroscopy. Gel slices were destained with 50% ACN in 50 mM ammonium bicarbonate (ABC) followed by dehydration in 100% acetonitrile (ACN). Trypsin (80 ng) in 50 mM ABC was added and gel slices rehydrated at 4 °C for 10 min, followed by incubation at 37 °C overnight. Peptides were extracted three times with 50 ul of 50% ACN / 0.1% formic acid, followed by clean up with a ZipTip (Millipore). Peptides were separated using reversed-phase chromatography on a Shimadzu Prominence nanoLC system. Using a flow rate of 30 µl/min, samples were desalted on an Agilent C18 trap (0.3 x 5 mm, 5 µm) for 3 min, followed by separation on a Vydac Everest C18 (300 A, 5 µm, 150 mm x 150 µm) column at a flow rate of 1 µl/min. A gradient of 10-60% buffer B over 30 min where buffer A = 1 % ACN / 0.1% FA and buffer B = 80% ACN / 0.1% FA was used to separate peptides. Eluted peptides were directly analysed on a TripleTof 5600 instrument (ABSciex) using a Nanospray III interface. Gas and voltage settings were adjusted as required. MS TOF scan across m/z 350-1800 was performed for 0.5 sec followed by information dependent acquisition of the top 20 peptides across m/z 40-1800 (0.05 sec per spectra). Data were converted to mgf format and searched in MASCOT accessed via the Australian Proteomics Computational Facility and searched against the LudwigNR database, limited to ‘other bacteria’, using trypsin as enzyme, 2 mis-cleavages, MS tolerance of 0.5 Da and MS/MS tolerance of 0.2 Da. Oxidation (met, variable) and carbamidomethylation (cys, fixed) modifications were also included.
The MS analyses performed were strictly qualitative, not quantitative, and therefore no normalisation of protein amount prior to trypsin digested was performed. A nominal amount (eg 80 ng) trypsin is added per gel slice, as it is not feasible to determine the amount of protein in each gel band processed for MS. This is typical practice in the proteomics field. Samples were ziptipped after digestion, prior to MS, in part to desalt/concentrate the samples, but also to ensure the LCMS system was not overloaded (ziptips have a limited loading capacity (5ug)).
Antibodies
A polyclonal antibody (designated as Anti-Protein 6670) was raised by GenScript Inc. (Piscataway, NJ, USA). The antigen used for immunization was VPVTDDTRKEPTETC, derived from the translated protein ZP_02736670. Immunogen was a Peptide-KLH conjugate, and host strain was New Zealand rabbit. The antibody was affinity purified and stored in Phosphate Buffered Saline (PBS, pH 7.4) with 0.02% Sodium Azide at −20°C. Membrane preparations of G. obscuriglobus were resolved by SDS/PAGE (10%) and the specificity of the antibody was tested via western blotting at 1:2000 dilution (S10 Fig). HRP-conjugated goat anti-rabbit antibody (Cell Signalling Technology, Australia, 1:5000 dilution) was used as secondary antibody. Detection was done by using ECL Western Blotting System (GE Healthcare Life Science). The ‘anti-MC protein’ antibody against the Gemmata obscuriglobus protein gp4978, a clathrin heavy chain-like membrane coat (MC) protein already shown to be present on internal membrane vesicles of Gemmata obscuriglobus associated with endocytosis-like protein uptake in this species[24]. This antibody to gp4978 was a rabbit polyclonal antibody raised against recombinantly expressed and purified gp4978 protein identified as a eukaryotic MC coatomer protein (National Center for Biotechnology Information reference sequence: ZP_02732338.1; see [37]).
Immuno-gold labelling
Ultrathin sections of high-pressure frozen and cryosubstituted intact G. obscuriglobus cells or density gradient-purified fraction 3 membranes on formvar-carbon-coated copper grids were floated onto drops of Block solution containing 0.2% (wt/vol) fish skin gelatin, 0.2% (wt/vol) BSA, 200 mM glycine, and 1× PBS on a sheet of Parafilm, and treated for 1 min at 150 W in a Biowave microwave oven. The grids were then transferred onto 8 µL of primary antibody, diluted 1:25 in blocking buffer, and treated in the microwave at 150 W, for 2 min with microwave on, 2 min off, and 2 min on. The grids were then washed on drops of Block solution three times and treated in the microwave at 150 W each time for 1 min before being placed on 8 µL of goat anti-mouse IgG Fc (γ)-specific antibody conjugated with 10 nm gold (British Biocell International, catalog no. EM GAM10) diluted 1:50 in Block solution and treated in the microwave at 150 W, for 2 min with microwave on, 2 min off, and 2 min on. Then grids were washed three times in 1× PBS, each time being treated for 1 min each in the microwave at 150 W, and four times in water for 1 min each in the microwave at 150 W, and examined via transmission electron microscopy.
LEGENDS FOR SUPPLEMENTARY VIDEOS AND FIGURES
S1 Video: Electron tomography of the fraction 3 membranes containing pores
Curving sheets comprising isolated membranes from fraction 3 (isolated via density gradient fractionation from lysed G. obscuriglobus cells) can be seen in section. As tomogram slices of the spirally coiled sheets are passed through during the movie’s course (in effect passing through successive and different planes of a thick section), they can be seen to be interrupted periodically by non-membranous pore structures some regions of which project from one side of each of the membranes which comprise a coil. If one membrane sheet is examined at different points of the movie, several pores can be identified as slices of the membrane sheet are passed through successively. Bar, 50 nm.
S2 Video: 3D reconstruction of an internal pore-containing membrane of Gemmata obscuriglobus
The electron tomography membrane reconstruction shown here is derived from a representative of the fraction 3 membranes in the thick section tilt-series.
S3 Video: 3D reconstruction of a pore embedded in the internal pore-containing membrane of Gemmata obscuriglobus.
The reconstruction is derived from a pore in the membrane shown in S1 Video and S2 Video.
Bioinformatics analyses
We performed a number of bioinformatics analyses on the 128 unique proteins identified through proteomics as belonging to the pore-containing membrane fraction (fraction 3). We first searched for similarity to known proteins in the non-redundant protein database (downloaded from NCBI) using BLASTP and PHMMER.
BLASTP (2) reported 112 proteins with significant homologs (E<0.001) (S3 Table). Of these, 33 are in the membrane fraction 3 (84.6% of all fraction 3 proteins). Of the significant hits, 25 top hits are to non-identical sequences in G. obscuriglobus, 18 are to Planctomyces, and 16 to S. acidiphila. One sequence found no hits at all (ZP_02735547). Almost half of all hits are to hypothetical (50), unnamed (5) or probable (2) proteins. Among hits with assigned function, 6 are flagellar, 12 ribosomal, 6 are membrane-related, and 4 are efflux proteins.
PHMMER (from the HMMER package, ver. 3.0(1)), returned significant hits for 115 proteins (E<0.001) (S3 Table), of which 21 are to G. obscuriglobus, 20 are to Planctomyces, and 21 are to S. acidiphila. The functional distribution is similar to the BLAST results: A large fraction are hypothetical (51), unnamed (7) and probable (3), while annotated functions include flagellar (6), ribosomal (13), efflux (4) and membrane (5) hits. Notably, PHMMER and BLASTP find the exact same top hits for 64/128 proteins. 17 are non-identical G. obscuriglobus proteins, 11 are from Planctomyces, and 14 from S. acidiphila.
As BLASTP and PHMMER screens both identified hits to other proteins coded in the G. obscuriglobus genome, we examined similarity among proteins from the pore-containing fraction. We performed all against all BLAST (2) followed by a Markov clustering on the results (3) as implemented in VisBLAST (4) with default parameters E<0.001 and i-value=2.0. This yielded two large clusters (both containing 11 proteins), and a number of small triplet and doublet clusters (Fig S13). The vast majority of proteins were singletons (91 proteins; S3 Table), and the clustering was robust at higher E-value cutoffs (up to E=10) and i-values (from 1.2 to 5). One of these large clusters contained 7 proteins that were unique to fraction 3 with the remaining four proteins being divided between fractions (2,3) and (2,3,6). The second large cluster contained no proteins that were unique to fraction 3. We performed the same clustering on the full set of all 512 proteins and the same clusters were recreated. VisBLAST was used to cluster proteins based on sequence similarity. An E-value cut-off of 0.001 was used together with an i-value of 2. Interestingly, the cluster containing 8 proteins unique to fraction 3 did not change at all in this larger analysis, which indicates that it is indeed membrane-specific. The other large cluster expanded with proteins belonging to fractions (2,6), (6) and especially fraction (2). Other large clusters were found within the set of 512 proteins, however none contained fraction 3 proteins (Fig S14).
Transmembrane helix structure potential was predicted for all 128 fraction 3 sequences using TMHMM2 (5). Of these, 42 showed significant signal of one or more transmembrane helices. Seven of these are unique to fraction 3 (S3 Table). We also looked for evidence of coiled coils using Paircoil2 (6). We report 15 proteins with significant signal of a coiled coil structure (S3 Table).
We next examined fold architecture of proteins in the membrane pore fraction using Phyre2 (7). This approach uses homology modelling to infer the structure of an amino acid sequence based on resemblance to known structures. 127 of the sequences were modelled in full, but one protein (ZP_02734818, 2558 aa) was analyzed in pieces because of its large size. The Phyre2 result for this protein is thus based on the best scoring subsequence. In the S3 Table we report the top Phyre2 structural hit for each protein. These hits are extracted automatically from the output, and we note that they may not represent the sole best hit (multiple hits often have the same highest confidence score). Based on PDB descriptions, these unfiltered results include membrane proteins (14), flagellar proteins (7) and ribosomal proteins (8), along with a number of other bacterial membrane/transport-related hits.
Comparing all structural predictions with our clustering analysis reveals a number of interesting patterns. Most significantly, cluster 1 consists of proteins modelled by Phyre2 as β-propeller-containing. This is notable given the presence of the β-propeller architecture in protein constituents of the eukaryotic nuclear pore complex (8). Of the 11 proteins in this β-propeller cluster, seven are unique to the pore-containing membrane fraction (3) (Fig S14 and S3 Table). The second large cluster (cluster 2) is dominated by pilins, and the proteins in this cluster mainly come from membrane fraction (3, 2, 6). Approximately half the structural predictions for singleton proteins showed significant structural similarity to porins and membrane proteins, ribosomal subunits, and flagellar proteins (S3 Table). The predicted triplet cluster contains flagellar proteins.
A closer investigation of the structure predictions for the 11 members of the cluster 1 showed that eight yield at least one structure prediction with a confidence >95%. In most cases, multiple predictions are made covering all parts of the sequences.
As constituency in a cluster does not establish whether all constituents share a common region of sequence similarity, we performed multiple sequence alignments across all members of both cluster 1 and cluster 2 (using MAFFT, option L-ins-i, (9). For cluster 1, all sequences displayed similarity in the C-terminal region. Fig S15 shows the alignment for the eight sequences for which we also obtained high confidence (>95%) structures with Phyre. We evaluated the full alignment using the T-Coffee CORE program (10), which shows a moderately robust 8-way alignment with a CORE-score of 69 (where 100 is perfect alignment). It is clear that the conservation is most pronounced in the C-terminal end of the sequences from around position 850 in the alignment. Indeed, if only the C-terminal part of the alignment is analysed using T-Coffee, the score increases to 81. This corresponds well with the observation that the majority of hits retrieved when searching the non-redundant protein database using both BLAST and PHMMER are also against the C-terminal ends.
With the aim of better characterising the commonalities of cluster 1, we focused on the structural predictions associated with the common C-terminal region. Note that there is some disagreement between the top hits shown in S3 Table (which was automatically generated) and those derived from the conserved C-terminal region (S4 Table). If both coverage and confidence scores generated by Phyre 2 are considered, the structures associated with the C-terminal region (S4 Table) emerge as the best hits. Some cluster 1 proteins also yield significant predictions for their N-terminal ends, but these are not in conflict with results from their respective C-termini, indicating these may be multi-domain proteins. In all 8 cases where a significant (confidence >95%) structure model is obtained, the C-terminal predictions are for β-propeller structures that overlap with the conserved C-terminal region of the sequences (Fig S15). Furthermore, Phyre2 modeled all 8 proteins to the same PDB template (2C4D), which we interpret as independent verification that these proteins share a common structural fold. These results are not due to extensive sequence similarity, as the overall sequence identity between the queries and the PDB template ranges from 13% to 19%.
Results from our structural analysis of cluster 1 are given in S4 Table, and structural models are depicted in Fig 7C (all structures are visualized using The PyMOL Molecular Graphics System, Version 1.4.1 Schrödinger, LLC).
For the second large cluster (Fig S14), we performed the same type of analysis. All can be aligned (Fig S16), 10 of 11 sequences have significant (>95% confidence) structure predictions, and in all cases the best hit in terms of both confidence and coverage (S5 Table) was modeled against PDB file 1OQW. The predicted structures are all very similar and consist of a single α-helix.
Two proteins (ZP_02735673 and ZP_02736511) show possible α-solenoid structures with stacked α-helices (Fig S18). Both are singletons in the cluster analysis, and both are present in membrane fraction 3 (the first is in both fractions 3 and 2, and the latter is unique to fraction 3). The left-hand structure in Fig S18 (ZP_02735673) models against alpha-solenoid structures in pdb with high (>99%) confidence, with models spanning >95% of the sequence. The model shown is based on 1OYZ, a hypothetical protein from E. coli, which is classified in SCOP as a member of the ARM repeat superfamily. Within the top 10 hits are structures that derive from Bacteria, Archaea and Eukaryotes, including clathrin adaptor core proteins (2VGL, 1W63). The right-hand structure in Fig S18 (ZP_02736511) contains two high confidence domain models. The N-terminal region models to the same alpha-solenoid structure as seen in the left-hand structure in Fig S18. This spans 30% of the protein sequence. In the adjacent central region, Phyre2 models a response regulator (top hit: 1ZES) with high (>99%) confidence.
A possible FG repeat-containing protein (ZP_02734840) was found in the pore-containing membrane fraction and in the total nuclear membrane proteome. With 5 FGs in the first 200 residues, this conforms to a recent definition of FG-repeat nucleoporin (11) but the C-terminal half of the protein models as a transmembrane beta-barrel protein.
We also performed a more general screen for bacterial transmembrane proteins among our structural predictions. To do this, we screened results for predicted structures of the following type: “outer membrane lipoprotein”, “outer membrane efflux protein”, “transmembrane beta barrel”, “porin”, “tolc” and “oprd/g/h/p/”. The same protein might have hits in more than one of these categories. For transmembrane beta barrels, we chose the best hit to a beta barrel spanning the membrane even if the term “transmembrane” was not used to describe that particular hit (however, for all proteins in this category at least one hit is called “transmembrane”). For all other categories we chose the best hit containing the specified keyword(s). S6 Table summarises these hits with their ID, membrane fraction, confidence score and coverage (as reported by Phyre2), and the PDB template used. For the oprx group we also list the specific class.
ACKNOWLEDGEMENTS
The authors acknowledge the facilities, and the scientific and technical assistance, of the Australian Microscopy & Microanalysis Research Facility (AMMRF) at the Centre for Microscopy and Microanalysis, The University of Queensland. We thank Dr. Vitaliya Sagulenko for help with Western-blot analyses. We thank Tony Romeo of Electron Microscope Unit (EMU), University of Sydney, for help with preparing some of the freeze-fracture replicas. JAF acknowledges assistance in developing some preliminary density gradient protocols and donation of gp4978 antibody from Jody Franke now of Creighton University, Omaha, Nebraska USA.