Design and preparation of CcmK constructs

Bearing in mind that purification tags could impact solubility and/or assembly behavior, short tetra-histidine (His₄) peptides were placed at either N- or C-terminal sides of all four CcmK isoforms from Syn6803. A tobacco etch virus (TEV) protease recognition site was included between protein sequence and tags to facilitate the preparation of untagged versions, when necessary (S1 Fig). We will refer to such constructs as ^HT-K and K^-TH to indicate tagging emplacement at either the N- or the C-terminus, respectively, and ^-K or K^- to refer to the corresponding TEV-proteolysis products. Expression of the different constructs in E coli and solubility were monitored by SDS-PAGE (S2A and S2B Fig) and compared with untagged proteins (^UnK, S2C Fig), which only differed from wild-type sequence by the presence of Ala replacing the second residue and by an extra Ala at the C-terminus. The four isoforms were well expressed, with two exceptions (^HT-K1 and K4^-TH) for which bands were fainter. After cellular lysis, CcmK2 and CcmK4 remained soluble irrespective of the presence and position of tags. Multiple bands were apparent for the K4^-TH construct, suggesting sensitivity to cellular proteases, something confirmed by the presence of multiple peaks detected by mass spectrometry (MS) on purified samples (see below). In the case of CcmK1, N-ter tags rendered the protein insoluble. In contrast to the other three isoforms, none of the tested constructs and conditions permitted expression of a soluble form of CcmK3. This was likely related to folding/oligomerization defects, since untagged ^UnK3 also expressed well but remained fully insoluble after cell lysis (S2C Fig). This result is in agreement with the lack of reports on in vitro studies for this isoform.

The five soluble proteins, namely K1^-TH, ^HT-K2, K2^-TH, ^HT-K4 and K4^-TH, were purified by immobilized metal ion affinity chromatography (IMAC) and used for subsequent studies.

Characterization of CcmK isoforms in solution

The oligomerization behavior of purified constructs was first verified by size-exclusion chromatography (SEC). CcmK1 and CcmK4 isoforms eluted at volumes expected for hexamers, with molecular masses (MW) of 65 kDa for K1^-TH, 72 kDa for ^HT-K4 and 65kDa for K4^-TH, as estimated by comparison with calibrated protein standards. On the other hand, K2^-TH elution volumes matched to a 100–110 kDa protein, suggesting equilibrium between hexamer and dodecamer species in solution, in agreement with previous reports [35, 41].

Monomer association was next investigated by electrospray ionization native-MS (n-MS), a suitable approach for the characterization in solution of protein-ligand(-protein) interactions and the investigation of complex assembly phenomena, such as viral capsid formation [42]. Clusters of peaks in the 3500–5000 m/z range were detected as most predominant under soft ionization conditions for all CcmK proteins, which were pretreated with TEV protease to increase sample homogeneity (Fig 1 and S3 Fig). These signals corresponded to multiply charged ions whose mass is consistent with CcmK hexamers (S1 Table). These species were slightly heavier than masses calculated considering monomer data measured under more stringent conditions or from MS/MS spectra (see below). These deviations are caused by water molecules and salts that remain bound to proteins under soft ion-desolvation conditions required to preserve protein complexes. Tandem MS/MS experiments further confirmed the stoichiometry of the detected CcmK complexes. Discrete peaks with a single charge state were selected (indicated with asterisk in Fig 1 and S3 Fig) and subjected to collision-induced dissociation (CID). A common dissociation pattern was observed for all CcmK. Namely, peaks corresponding to monomers and pentamers appeared in the m/z region below or above, respectively, the parent ion peak. Sharper CID product ion peaks allowed us to attain more accurate mass measurements, as a consequence of dissociation of the residual solvent and salt adducts. Measured masses for monomers matched values calculated from amino acid sequences (S1 Table). Values with K1^-, K2^- and K4^- indicated loss of the first Met residue (131.04 Da).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Oligomerization of CcmK1-4 and potential assembling intermediates with CcmK2 of Syn6803.

Positive-ion mode native ESI-MS spectra from CcmK isoforms are presented on top panels: A, Syn6803 K1^–; B, Syn6803 –K2; C, Syn6803 –K4. The spectra are dominated by signals from multiply charged ions of CcmK hexamers. In addition, potential assembling intermediates with higher oligomerization state were noticed in experiments with−K2 (B). With all proteins, tags were removed by TEV protease treatments prior to spraying. Bottom panels present collisional activation data collected on hexamer precursor ions selected from top spectra (indicated with an asterisk). An asymmetric charge partitioning is noticed, hexamers dissociation resulting in monomer and pentamer species. Species m/z values and charges are indicated for major peaks. Cartoons schematically illustrate the stoichiometries of detected protein complexes and subcomplexes. Molecular weights of neutral species estimated from data measured for species with different charges are compiled in S1 Table.

https://doi.org/10.1371/journal.pone.0185109.g001

Remarkably, n-MS data for the two CcmK2 constructs revealed the presence of a cluster of peaks in the 4700–6300 m/z range. The MW of these species was estimated to range between 239–275 and 345–367 kDa for K2^- and ^-K2, respectively. These values are considerably higher than the expected MW for dodecamers, which could not be detected for any of the two constructs (S1 Table). Taking into account the likely association of small solution molecules, these species roughly matched to 21–24 or 28–30 combined monomers, respectively. Despite not always fitting to integer combination of hexamers, the fact that such species were only detected for the two CcmK2 constructs suggested the existence of potential CcmK2 pre-assembled intermediates in solution. Ion mobility spectrometry (IMS) experiments were therefore performed to evaluate whether the shape of these objects fitted better to sheet-like intermediates or to globular aggregates. Collision cross-sections (Ω) determined for selected K2^- and ^-K2 species were slightly higher than values measured (and published) for globular proteins of similar MW (S4 Fig). Although this difference could be considered evidence for small layered assemblies, the absence of a continuous distribution of oligomers spreading up to higher m/z values, where Ω differences between sheet-like viral capsid proteins and globular structures are more notorious, limited the interpretation of CCS data.

CcmK2 and CcmK4 assemble as flat sheets on mica

Assembly of ^HT-K2, K2^-TH and ^HT-K4, K4^-TH on mica supports was next investigated by atomic force microscopy (AFM). The influence of pH (from 5.0 to 8.0, 0.5-unit intervals), buffer composition (phosphate, MES or Tris) and protein concentration (20–120 μg/mL) was systematically evaluated. Extended islands of sheets were obtained when 0.4 μM concentrations of K2^-TH were deposited in phosphate (Fig 2A) or MES buffers at pH 6.0. Cross-sectional analysis proved the formation patches made of single protein layers (average depth of 4.1 ± 0.1 nm). Sheets were flat and seemed to derive from hexamers adopting a single orientation with regard to the mica plane. However, islands displaying striped motifs were also noticed in rare instances (arrow, Fig 2A), providing evidence for some degree of assembly plasticity for this isoform. Strips were 12–15 nm wide, with top plane lifted by about 1 nm above the lowest strip, which reached a similar height as other flat patches. Flat sheets were also imaged with ^HT-K2. Assemblies persisted over a wider pH range, spreading from 5.5 to 7.0 (Fig 2B). The result was sharply influenced by protein concentration, small increases resulting in shift from no protein adsorption to fully covered mica. Difficulties to sense mica level impeded accurate estimations of layer thickness with ^HT-K2 (4.8 ± 0.2 nm for data of Fig 2B). Intercalated patches reminiscent of assemblies bound to mica with inverse orientation were occasionally noticed, reaching 0.5 to 1 nm higher levels than the first level layer.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. CcmK2 and CcmK4 assembling revealed by AFM.

Images were recorded after absorption on mica of indicated amounts of proteins in phosphate buffers: 120 ng of K2^-TH at pH 6.0 (panel A), 60 ng of ^HT-K2 at pH 7.5 (B) and 120 ng of ^HT-K4 at either pH 7.5 (C) or pH 5.5 (E). Stripped patches (arrow) were occasionally observed with K2^-TH. Panel D results from applying FFT on image C. The black arrow indicates a periodicity spot at 6.9 nm. In C and E, ^HT-K4 islands are surrounded by connected stretches of protein positioned at higher level. F, image obtained after inverse-FFT using periodicity spots generated from a cropped section of image from panel E (dashed square). Image sizes were five times the dimension of scale bars, 120x120 nm for panel F.

https://doi.org/10.1371/journal.pone.0185109.g002

Similar experiments were carried out with ^HT-K4 and K4^-TH proteins. Well resolved images recorded for ^HT-K4 confirmed the formation of flat assemblies under a variety of conditions, with pH spreading from 5.5 to 7.5 (Fig 2C–2F). Likewise CcmK2, layer heights could not be accurately measured due to difficulties to prevent full covering of mica. An underestimated value of 3.1 ± 0.8 nm was calculated from an image recorded at pH 6.0. Fast Fourier transform (FFT) operations on images acquired at pH 7.5 revealed the occurrence of 6.9–7.1 nm periodicities (Fig 2D), matching values of 2D layer lattices from most of published 3D structures (approx. 70 Ǻ) [23, 35], but slightly longer than 6.7 nm values reported between hexamers interacting in a planar geometry within the HO BMC shell [26], or in crystals of tightly packed CcmK2 and CsoS1A proteins [32, 33]. Much like in EM studies ¹³, we could not detect patterns with strips of inversely oriented hexamers characterized in a CcmK4 crystal structure [35]. Less compact sheets formed at pH 5.5 (7.0 and 8.0 nm periodicities, depending on lattice direction), hexamers being held together seemingly through vertex-vertex contacts (Fig 2E and 2F). This organization contrasts with edge-to-edge interactions that must occur at pH 7.5, according to lattice periodicities similar to published crystal and EM data. Only at pH 5.5, central depressions at the center of hexamers were noticed, suggestive of assemblies deposited on mica with the opposite orientation. At all pHs, stretches of proteins were found decorating island edges, protruding in average 4 to 5 nm above the first layer level (Fig 2C and 2E).

Unfortunately, experimental conditions compatible with assembly of K4^-TH could not be found, something that we attributed to the proteolytic sensitivity of this construct that was confirmed in n-MS experiments (see S3B Fig).

Honeycomb-like supra-hexameric organization of CcmK1 assemblies

The propensity of K1^TH to self-organize on mica was next inspected. As for CcmK2 and CcmK4, experimental conditions were systematically screened. Sheets formed under varied conditions, yet the best resolved images were recorded in the pH 6.5–7.0 interval. Remarkably, K1^TH displayed a supra-hexameric level of organization resulting in honeycomb-like motifs (Fig 3A and 3B), contrasting with data presented for CcmK2 and CcmK4. Depending on the conditions, these motifs occurred within isolated islands (Fig 3A) or formed continuous assemblies almost fully covering the imaged surface (Fig 3B). These mosaics were detected not only growing directly on top of the mica support (Fig 3A) but also as stacked layers positioned above similarly honeycomb-structured material (Fig 3A) or even over a first layer of protein not displaying these peculiar motifs (S5E Fig). Heights of 4.8 ± 0.2 nm and 8.3 ± 0.3 nm with regard to mica were estimated for the first and second layers from Fig 3A, respectively. Values of ca. 7.2 nm were estimated from Fig 3B, despite difficulties to sense mica level at such high surface occupancy.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Supra-structural organization of CcmK1 from Syn sp. PCC6803.

AFM images were recorded after absorption on mica of 100 ng of K1^-TH 6803 (A,B), 65 ng of K1^-H 6803 (C), 12 ng of ^H-K2 7942 (D), 12 ng of K2^-H 7942 (E), 500 ng of untagged K1^- 6803 (F) or 50 ng of K1^-ncTK 6803 (G). Experiments were performed in either MES pH 6.5 (D), NaPi pH 6.5 (B, C, E, F, G) or TRIS pH 7.0 (A). Arrows in panel A indicate structures of size and thickness comparable to individual polygonal tiles. For panels A and F, solutions were supplemented with 0.5 mM guanosine-5′-tetraphosphate (GtetraP) and 5 mM MgCl₂, respectively. Cross-sectional analysis graphs indicate material thicknesses traced along discontinuous white lines on images. Image sizes are 5 times those of indicated scale bars.

https://doi.org/10.1371/journal.pone.0185109.g003

In spite of an evident heterogeneity of size and shape, average motifs might be described as polygons, approximately 25 to 35 nm large. A mean surface of 720 ± 90 nm² was estimated from visually-selected representative cases. Since the largest face of a CcmK1 hexamer covers approximately 45 nm² in the crystal structure[33], honeycomb tiles should contain in average at least 16 hexamers, depending on the curvature. Such structures were limited by edges that protruded above their central depression by roughly 1.2 to 1.5 nm. These values are well below CcmK hexamer thickness of about 3.5 nm. Thus, the contours cannot be explained as stacks of protein positioned on top of a flat plane of CcmK1. Heterogeneity was accompanied by variability of cross-section profiles (shown in Fig 3). It is thus not possible to conclude anything about the form of these patches, which might be either V- or round-shaped, or a mixture of the two. For some polygons, the edges progressively rose over the full path from the center (approx. 15 nm). In contrast, other motifs were basically flat and edges protruded sharply, over the size of a single hexamer (7 nm). These two situations would roughly correspond to 5°-12° bending angles distributed between a few hexamers within the same section of round-shaped patches, or 10°-24° at sharp edges of V-shaped patches. Noteworthy are spots of size and thickness comparable to individual polygonal tiles that were regularly detected under different set of conditions (e.g. arrows in Fig 3A), especially at low protein concentrations. However, the formation of honeycombed mosaics only occurred when negatively charged phosphate/phosphate esters or sulfonates were present, either as additives or buffer, during mica adsorption. Other additives assayed, including sulfate and bicarbonate salts or divalent cations, did not promote the formation of polygon-tiled assemblies (S5 Fig).

Overall, these data suggest that CcmK1 of Syn6803 might play a role different from that ascribed to CcmK isoforms for CB assembly. Such behavior was not unveiled in 3D and EM structures, which pointed to similar (canonical) flat arrangements for the three isoforms, the single exception being the striped pattern of CcmK4 in 3D crystals[35]. We therefore questioned whether this apparent discrepancy for CcmK1 could derive from alterations caused by studied protein constructs or by different experimental constraints. These possibilities were tackled by studying K1^-H, the construct used in previous structural investigations [23, 33]. Assemblies of purified K1^-H formed under similar experimental conditions than with K1^-TH, and more importantly giving rise to the same supra-structural motifs (Fig 3C). Polygons were heterogeneous, yet of similar size than those produced by K1^-TH (average 700 ± 100 nm² area). Surrounding edges protruded ca. 1 nm above the central depression. Assembled islands were positioned on average 4.3 ± 0.1 nm above mica, closer than values measured for K1^-TH islands (Fig 3A). The difference could arise from the presence of a 9 residue-longer linker preceding oligohistidine tags in K1^-TH, as compared to K1^-H (S1 Fig), and argues in favor of an assembly model with tags mediating attachment to mica.

Two other CcmK1 variants were studied to shed light on the role of tags. First, untagged K1^- was prepared by TEV protease treatment of K1^-TH. The resulting protein failed to assemble on mica, irrespective of protein amounts (60 to 500 ng), buffer composition (phosphate, MES, Tris) or pH (6.0 to 7.0, 0.5-unit intervals). Assembled islands could finally be produced at 8 times higher protein loads than with tagged proteins, and only in the presence of magnesium ions, conditions that resembled those applied in the AFM study of HO BMC-H [39]. Assembled islands of K1^- were flat, without apparent higher order motifs (Fig 3F). Heights above mica plane were estimated 3.7 ± 0.2 nm, similar to 3.5 nm values reported for untagged HO BMC-H [39], and in agreement with a closer proximity to the mica plane than with tagged K1^-TH or K1^-H. The second variant differed from K1^-TH by the replacement of His₄ by Lys₃ tags, which in virtue of their positive charge should still ensure attachment to anionic mica. This protein K1^-ncTK was obtained by TEV treatment of a purified CcmK1-ncTEV-Lys₃-TEV-His₄ construct (K1^-ncTK-TH), with ncT standing for “non-cleavable TEV peptide” (S1 Fig). Assembly of K1^-ncTK on mica occurred under similar conditions than with K1^-TH, and most importantly giving rise to comparable polygonal mosaics (Fig 3G). Differing from K1^-TH and K1^-H, however, arrangements reaching two slightly different heights were noticed. Polygone edges also protruded to higher levels (up to 2 nm) above their center than K1^-TH or K1^-H (cross-section profile of Fig 3G).

AFM data for K1^-TH, K1^-H or K1^-ncTK were not well enough resolved to define unambiguously the contour of assembly units. Efforts to enhance image definition by combining series of successive scans covering the same region failed. Comparison of five aligned images taken at 15 min intervals evidenced continuous size and shape alterations of polygonal patches, indicative of dynamic fluctuations of the assemblies (see below), or alternatively of perturbations induced by the AFM tip. With the same intention, we attempted the imaging of 2D sheets after cross-linking reactions with glutaraldehyde. Unfortunately, the low protein coverage detected on mica indicated disruption of the assemblies by the dialdehyde.

Assembly dynamics

The dynamics of K1^-TH assembly was next investigated by high speed (HS-)AFM. This approach was applied recently to characterize how flat 2D sheets form from individual HO BMC-H that are mostly recruited to the edges of assembled islands [39]. Two experiments were mounted, addressing either the characterization of assembly formation or the dynamics of pre-formed suprahexagonal motifs. To monitor the assembly process from its beginning, the protein was injected once the AFM probe already engaged in contact with the mica surface. Frames were recorded every 4 sec for 1 hour period. In that manner, the formation of the polygonal tapestry made of K1^-TH could be monitored (S1 Movie and S6 Fig). Despite the fact that image quality was limited, recorded data demonstrated that the assembly formation on mica is a dynamic process. Continuous events of attachment and detachment of individual proteins occurred during the first minutes (S6 Fig, frames 1 to 4). This was followed by the appearance of small seeds (frames 5 to 8), some anchored to the surface, and the later emergence of patches reminiscent of individual curved polygonal structures (frames 9 to 12). Later stages revealed a tendency for the curved domains to serve as template and boost the emergence of new neighbor curved structures, the ensemble finally coalescing into honeycomb mosaics like those presented in Fig 3A–3C (frames 13 to 16). Data indicated tendency for the HS-AFM data also demonstrated that polygonal motifs are continuously remodeled both in size and shape.

For the second experiment, a polygonal tapestry of K1^-TH was allowed to form prior to scanning the surface at 10 sec intervals for 20 minutes. Comparison of aligned frames demonstrated that most active regions for sheet remodeling are assembly edges, as proven by higher standard deviations measured at such emplacements when comparing the series of aligned images (S2 Movie and S7 Fig). Structural fluctuations were also noticed at edges of honeycomb polygonal motifs, although to a lesser extent than was observed for the island edges. Overall, these data demonstrated the suitability and weakly invasive character of AFM approaches. Visualization of attachment and detachment events of oligohistidine tagged CcmK1 on mica supports gave additional credibility to the significance of the structural motifs characterized herein.

Honeycomb supra-hexameric organization of CcmK1 disrupted by single point mutations

Residues K25, R28, D49 and R80 are estimated by theoretical algorithms to be major contributors to inter-hexamer interactions. Their role in promoting high-order assembly was therefore investigated. Residues were mutated either individually or in combinations of two (with the exception of the K25A/R28A, which was not attempted). Unfortunately, although well expressed, none of the proteins incorporating the K25A mutation could be recovered in IMAC-purified fractions (arrows, S8 Fig), contrasting with a 10-fold increased solubility over WT protein, reported for the corresponding amino acid exchange variant K26A of a PduA [43]. Since substitutions of charged residues were expected to shift the isoelectric point of hexamers and thereby could influence attachment to mica, AFM studies were carried out with each variant at pH values ranging from 5.5–7.0 (0.5-unit intervals). The supra-structural assemblies characterized for K1^-TH (and K1^-H) were disrupted in R28A or R80A single mutants and in the R28A/R80A double mutant (S9 Fig), which remarkably continued to assemble. The three mutants gave rise to islands of flat sheets, resembling those observed with CcmK2 and CcmK4. In contrast, the D49A mutant displayed a trend to produce higher-order assemblies reminiscent of honeycomb polygonal motifs described above for WT protein, yet with much more irregular sizes and shapes (S9B Fig). Finally, the double mutant D49A/R80A did not result in any identifiable organized pattern on mica (S9E Fig).

Residues on the second α helix and at the C-ter extension contribute to CcmK1 curvature

Only seven residues and the presence of a short C-terminal extension distinguish CcmK1 from CcmK2 of Syn6803 (see S10A Fig). Comparison of the two protein structures indicates that, apart from the flexible C-terminal extension, the most diverging region localizes at the C-terminus of the second α-helix where three consecutive diverging residues accumulate (S10B Fig). The calculated root mean square deviation (RMSD) of coordinates for the 24 backbone atom (ba) of residues 61 to 66 is 1.55 Å. For comparison, RMSD of 0.51 Å were measured for the remaining 344 ba in between residues 3 and 94 of the two proteins. Similar differences are found when comparing the structure of CcmK1 Syn6803 and those of CcmK2 from Synechococcus elongatus PCC 7942 (hereafter Syn7942) (1.45 vs. 0.33 Å) or CcmK4 Syn6803 (1.52 vs. 0.69 Å).

The possible implication of this α2 short helical segment and of the C-terminal extension in the formation of assemblies displaying supra-hexameric motifs was first examined in AFM experiments with CcmK2 Syn7942. N-ter His4-tagged (^H-K2 Syn7942) and C-ter His6-tagged (K2^-H Syn7942) versions were purified and their organization on mica inspected. Both proteins showed a clear propensity to form flat 2D sheets (Fig 3D and 3E), in agreement with sequence and structural similarity to CcmK2 Syn6803. Assembled patches of ^H-K2 Syn7942 displayed two orientations with regard to mica surface, whereas a single orientation was observed for K2^-H Syn7942. Average layer thickness over mica measured for the latter was 4.2 ± 0.2 nm, and approximately 2.4 and 3.3 nm for the two ^H-K2 orientations, although these values were undoubtedly underestimated bearing in mind the difficulties to sense mica. Supra-hexameric organizations were not detected in any of the conditions screened, which covered variable protein concentrations, buffers, and pHs ranging from 6.0 to 8.0.

To further uncover the participation of the diverging α2 helix stretch and/or of the last C-ter residues in generating CcmK1 Syn6803 honeycomb mosaics, two new K1^-TH constructs were prepared: i) AAN K1^-TH: the triple mutant with N⁶³I⁶⁴R⁶⁵ replaced by the corresponding AAN residues of CcmK2 Syn6803; ii) Δ9 K1^-TH: protein with the last 9-residues deleted (blue bar, S10A Fig). The two proteins were purified and studied on mica. Protein AAN K1^-TH resulted in flat sheets placed at 3.9 ± 0.3 nm above the mica plane (Fig 4A), when assembly was allowed to occur in the presence of phosphate or phosphoesters. Something similar was noticed with Δ9 K1^-TH, the protein tiling at about 3.4 ± 0.1 nm heights in the presence of MES or NaPi buffers (Fig 4B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Residues of the second a helix and the C-ter extension of CcmK1 from Syn sp. PCC6803 are required for formation of polygonal mosaics.

AFM images were recorded after absorption on mica of 100 ng of A⁶³AN K1^-TH 6803 (A), 100 ng of D9 K1^-TH 6803 (B) and 45 ng of either N⁶³IR K2^-TH 6803 (C), or 9Cter K2^-TH 6803 (D), or N⁶³IR/9Cter-K2^-TH 6803 (E). Experiments were performed in either MES pH 6.5 (B, D, C) or NaPi pH 6.5 (A, E). For panel E, the solution was supplemented with 0.5 mM GtetraPi. Cross-sectional analysis graphs indicate material thicknesses traced along discontinuous white lines on images. Image sizes are 5 times those of indicated scale bars.

https://doi.org/10.1371/journal.pone.0185109.g004

Three CcmK2 Syn6803 variants were engineered, purified and assayed with the intention to further confirm the implication of the α2 helix stretch and the C-terminal extension to the peculiar assembly behavior of CcmK1: i) NIR K2^-TH: the triple mutant with A⁶³A⁶⁴N⁶⁵ replaced by the corresponding NIR residues of CcmK1 Syn6803; ii) 9Cter K2^-TH: CcmK2 protein incorporating the last 9-residues of CcmK1; iii) NIR/9Cter K2^-TH, which combined the two modifications of CcmK2. The first two proteins gave rise to flat assembled patches (Fig 4C and 4D), often contoured by aggregated material. In agreement with the observations described above, supra-hexameric polygonal motifs only formed when the two elements were integrated in CcmK2 (Fig 4E). Overall, these data convincingly evidenced the combined participation of the two structural elements to the formation of honeycomb mosaics observed with wild-type CcmK1.

In silico simulations of CcmK assembly

CcmK1 assembly behavior was investigated by atomistic molecular dynamics (MD). Simulations were launched on ensembles of three interacting hexamers, initially positioned as observed in the 3D crystal structure (PDB code 3BN4). The original structure included sulfate molecules modeled at the positively charged site found at the twofold axis of symmetry, where two hexamers meet [33]. Taking into consideration the requirement for phosphate/sulfonate molecules to attain curved assemblies in our experiments, the importance of this site was inspected as well. Thus, interfacial sulfates were either removed (w/o Pi), or replaced by inorganic phosphate (Pi) or methyl-triphosphate molecules (MePi₃) for MD simulations. These simulations clearly evidenced weak or no preference for flat structures under any condition. Thus, interhexamer tilting and bending angles measured for intermediate structures throughout MD trajectories deviated rapidly from values measured for the 3D crystal structure (Fig 5A and 5B, S3–S5 Movies). Most notable distortions were noticed for the interhexamer bending angle with CcmK1:MePi₃, values shifting on average by 17° with regard to the crystal structure (Fig 5B, S3 and S4 Movies). Moreover, a clear tendency of hexamers to curve towards the same side was revealed with this ligand (convex face towards the interior of the curved assembly), in contrast with data from simulations with Pi or w/o Pi assemblies (S5 Movie), which indicated bending in both directions and average values close to the starting planar situation. A closer inspection of engaged interactions suggested that ionic contacts between R66 side-chain and phosphates of MePi₃ could be a determinant in holding bent structures (S4 Movie). Interhexamer tilting was also manifested, yet average values remained close to starting values, the value shifting by about 8° in simulations with sandwiched Pi molecules. Concerning interhexamer distances, average value decreased with MePi₃ by about 0.4 Ǻ with regard to the 3D structure, whereas the value augmented slightly by 1.1 and 1.6 Ǻ for Pi and w/o Pi structures, respectively. Contrasting with CcmK1 simulations, higher conformational rigidity was revealed by similar MD simulations on a dimer of a B. subtilis operon repressor of pyrimidine biosynthesis (PyrR) [44] (Fig 4B, S6 Movie).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Theoretical simulations of CcmK assembly.

A-B, All-atom molecular dynamics of an ensemble of three CcmK1 Syn6803 hexamers originally positioned as in the published crystal structure (PDB code 3BN4). A, Ribbon representation of the simulated system (excluding explicit water and salt molecules). The position of atoms selected for measurement of tilting and bending angles are indicated with red and blue spheres, respectively (only shown for a single couple of hexamers for clarity). Tilting is defined by main-chain alpha carbon (Cα) atoms from residues 23 and 17 from interfacial monomers in each hexamer (illustrated by blue arrow). Bending angle values were based on the positions of Cα atoms from residues 26 and 37 from interfacial monomers (black arrow). B, Plot of tilting and bending angles during all-atom MD simulations (30 ns), measured at 0.25 ns intervals (black squares). Three different starting models were investigated, depending on the absence (w/oPi) or presence of Pi or MePi₃ ligands replacing crystallographic sulfates at the interface between hexamers. Shown are deviations from approximate 180° measured for the original crystal structure (empty circles). Thick traces represent mean values over the MD run. Similar data measured during a similar 30 ns simulation with a PyrR dimer (PDB code 1A4X) are presented for comparison. Here, tilting is defined by Ca atoms from residues 100, 104 of chain A, and 105, 131 of chain B, whereas bending angle is read from Cα atoms from residues 104, 143 of chain A and 143, 103 of chain B. Angles measured from the starting crystal structure are presented as empty squares. C, Lateral views of CcmK1 structure (left), and its representation in the coarse-grained model (middle). The white sphere indicates hexamer orientation. Two hexamers assembled with a nonzero curvature are illustrated below, indicating the preferred angle ν of interaction (bottom). D, snapshot of a coarse-grained simulation run with parameters ε₀ = 8.3 k_BT, ν = 1.4 rad, ε_ads = 2.5 k_BT, α = 0.1 and hexamer concentration 0.0125 σ^-3. The solid support plane appears blue and only hexamers near the surface are plotted. The simulations show the presence of stable patches of 15–20 hexagons with raised borders that resemble structures formed during the assembly of Ccmk1 with phosphate/sulfonate anions. E, snapshot of a coarse-grained simulation run with parameters ε₀ = 6.7 k_BT, ν = 1.45 rad, ε_ads = 10 k_BT, α = 0.2 and hexamer concentration 0.0125 σ^-3, showing the formation of a canonically flat assembly, that occurs even with non-flat preferred curvatures at weak bending stiffness and high adsorption strengths.

https://doi.org/10.1371/journal.pone.0185109.g005

The formation of larger assembled patches was next simulated using a minimal coarse-grained model. This is a well-established tool for investigating complex assembly processes [31]. Although structure-function relationships remain inaccessible, these models permit to parse biophysical constraints that might influence the final fate of such complex processes. As remarkable demonstration, Perlmutter et al. recently defined energetic regimes and conditions that lead to empty or cargo-filled BMC [31]. In our model, hexamers were represented as nearly spherical beads interacting with a simple potential that included: a short-range repulsion term that prevents hexamer collapse; an attraction term driving the assembly; bending and torsion terms accounting for the preferred angle of interaction between hexamers and the energetic cost of bending and twisting them from the optimal angle (more details are provided as M&M and ref [45]). Based on this model, Brownian dynamics simulations successfully reproduced the formation of assembled patches of CcmK1. Depending on selected relationships of parameter values, either flat patches or curved assemblies formed (Fig 5C and 5D and S7 Movie). Curved patches reminiscent of those found in honeycomb-like structures, composed by 15–20 assembly units, formed when the preferred interhexamer interaction angle was not flat. Despite the fact that assembly could occur in solution or from surface-attached hexamers, the survival of curved patches on the flat support required the imposition of weak adhesion forces, supporting our experimental working choice of mediating contacts with mica through oligohistidine tags.

Cloning, expression, and protein purification

Full-length CcmK1 (UniProt entry P72760), CcmK2 (P72761), CcmK3 (P73406) and CcmK4 (P73407) genes from Synechocystis sp. PCC6803 (codon-optimized for expression in E. coli) and from CcmK1 (Q03511) from Synechococcus elongatus PCC7942 (natural sequence), the triple A⁶³AN mutant and the Δ9 C-ter variant of K1^-TH Syn6803, as well as all mutants/variants of K2^-TH Syn6803 were synthesized (Genecust and Twist Bioscience) with N-ter or C-ter tag extensions (showed in S1 Fig). Sequences were cloned in either a pET-15b vector using Xba1/Xho1 or Nco1/Xho1 restriction sites or in pET-26b using Nde1/Xho1 (as indicated in S1 List). Single and double amino acid exchange variants of interfacial residues were prepared from reactions combining two mutagenic primers (for K25A + R80A or for R28A + D49A), following protocols supplied with the QuikChange Lightning multisite site-directed mutagenesis kit (Agilent Technologies). The importance of these residues to interhexamer interactions was predicted by the ANCHOR theoretical algorithm (http://structure.pitt.edu/anchor). Clonal DNA was purified and verified by sequencing. Full DNA and primer sequences are provided as supplementary data (S1 List). For subsequent studies, E. coli BL21 (DE3) cells were transformed following standard protocols.

Protein synthesis was induced with 0.2 mM IPTG in LB when cells attained mid log phase (OD = 0.6–0.8) at 37°C. Expression was allowed to proceed for 3–4 hours at 37°C before cells were harvested and stored at -20°C. Frozen cells were resuspended in 1/10^th of culturing volume of lysis buffer (20 mM Tris-HCl, 300 mM NaCl, 10 mM imidazol, pH 7.8), supplemented with DNase I (5 μg/mL final conc) and lysozyme (0.05 mg/mL). After incubation at room temperature with gentle agitation for 10 minutes, cell were lysed at 4°C in the presence of PMSF (1 mM final conc.) by 4 cycles of 30 sec sonication at 30% power, with intermediate pauses of 1 min without sonication (VibraCell 72434, Bioblock Scientific). Insoluble material was removed by centrifugation for 20 min at 20.000 x g (4°C). Soluble fractions were loaded on cobalt-loaded TALON Superflow metal affinity resin (Clontech) conditioned at 4 to 10°C. After thoroughly washing with Sol A (20 mM NaPi, 300 mM NaCl, 10 mM imidazol, pH 7.8), elution was effected with Sol B (300 mM imidazol in Sol A). EDTA (5 mM final conc.) was added immediately after elution. Proteins were buffer-exchanged against Sol C (10 mM HEPES, 300 mM NaCl, pH 7.5) by 3–4 steps 10-fold dilution/concentration steps in Vivaspin Turbo 15, 10 kDa MWCO devices, before concentrating the protein to 1–2 mg/mL. Protein concentrations were calculated from 280 nm absorption readings, using theoretical extinction coefficients estimated from protein sequences with ExPASy ProtParam tool (http://web.expasy.org/protparam/).

TEV-treated untagged proteins

Untagged proteins were obtained after removal of oligohistidine tags by incubation of ^HT-K, K^-TH or K^-ncTK-TH proteins (0.5 mg/mL final conc.) in 50 mM Tris pH 8.0 / 300 mM NaCl /2 mM DTT/ 1 mM EDTA, with turbo TEV protease (GenWay, 5 μg/mL). Reactions proceeded at 25°C overnight, followed by buffer-exchange at 4°C against Sol C (exactly as described above). Proteins were concentrated to 1–2 mg/mL.

Size-exclusion chromatography

Protein sizes were estimated by SEC using a Beckman Ultraspherogel SEC2000 column (7.5 x 300 mm) mounted on a Waters 2690 HPLC separation module. Protein solution (10–20 μL) was injected at 1 mL/min flowrate after equilibration of the column with 20 mM Tris-HCl, 300 mM NaCl at pH 7. Elution was monitored with a Waters 996 Photodiode Array Detector. Elution volumes (280 nm absorption) were used to estimate protein MW by comparison to next calibration standards run under identical conditions: Ferritin (440 kDa), Aldolase (158 kDa), Conalbumin (75 kDa), Ovalbumin (43 kDa) and Ribonuclease (13.7 kDa).

Native-mass spectrometry

Prior to measurements, TEV-treated proteins were buffer-exchanged against 150 mM aqueous ammonium acetate at pH 8 using Amicon Ultra-0.5 mL centrifugal filters (MWCO = 10 kDa; Millipore). MS measurements were performed on a commercial electrospray ionization hybrid quadrupole time-of-flight (ESI-Q-ToF) mass spectrometer (Q-TOF Ultima API, Micromass, Manchester, U.K.) equipped with a 32k quadrupole and a high-pressure collision cell (MS Vision, Almere, Netherlands). All experiments were carried out in the positive ion mode. A commercial nano-ESI source was used at ambient temperature. Typically, 3–5 μL of sample were loaded into a Au/Pd-coated glass capillary emitter (1 μm outlet inner diameter; Thermo Scientifc, Madison, WI, USA). To generate electrospray, a voltage of 1.9–2.3 kV was applied to the emitter and a backing pressure of 0.5 bar was used to assist sample flow. In order to assure gentle transmission of the ions from atmospheric pressure to vacuum, the pressures in the first pumping compartment were increased to 3.5–4.0 mbar. The pressure inside the collision cell was set to 1*10⁻² mbar. Several tuning parameters including the cone voltage, RF lens 1 offset, offset 1, and the collision energy offset were optimized to obtain efficient ion transfer, good signal intensity and resolution. The quadrupole transmission profile was adjusted for the desired m/z range. Mass spectra were recorded in a m/z 50–8000 window with a scan time of 2 s and an interscan delay of 0.1 s. Spectra were recorded using the MassLynx 4.0 software (Waters, Manchester, UK) and baseline-corrected, normalized, and smoothed using MATLAB R2015a (MathWorks, Natick, MA, USA). Fifty individual scans were typically combined to produce a mass spectrum. For calibration, CsI clusters formed by electrospraying an aqueous CsI solution (40 mg/ml) were used. The recorded spectra were averaged (50 scans), smoothed with a moving average algorithm (width of ±3 steps) and centroid spectra were generated at 80% peak height. The m/z axis was calibrated by fitting a polynomial function.

For tandem mass spectrometry experiments, precursor ions were isolated in the quadrupole mass analyzer and accelerated into an argon-filled linear hexapole collision cell. Various collision energy offsets were applied upstream of the collision cell, with argon at a pressure of 3.0 x 10⁻² mbar.

Drift times for CcmK2 constructs and protein calibrants were measured by ion mobility mass spectrometry experiments performed on a hybrid quadrupole-IMS-TOF instrument (Synapt G2S HDMS, Waters, Manchester, UK). For calibration, the following standard proteins were used: concanavalin A, bovine β-lactoglobulin, bovine serum albumin, alcohol dehydrogenase, transthyretin, as described in reference [62]. Nitrogen was used as a buffer gas in the ion mobility cell operated at a nominal pressure of 1.65 mbar. Typically, ion mobility spectra were acquired for 2 minutes. The source temperature was kept at 30°C, and the pressure in the first pumping stage was increased to 3.0 mbar. Drift times of the folded calibrant protein ions were extracted and used to produce a calibration curve for collision cross section. Ω values for CcmK2 hexamers as well as for species detected at higher m/z were compared to sheet-like structures of norovirus and HBV that were determined by Heck and coworkers under the same conditions (S4 Fig) [42].

Imaging with AFM

Proteins were 10–50 fold diluted in different solutions, adjusted to pHs 5.0 to 8.0, and containing or not commercially available additives (Sigma-Aldrich) at 0.5 mM final concentrations (as indicated in figure legends): Sol AFM1: 10 mM NaPi, 300 mM NaCl; Sol AFM2: 10 mM MES:NaOH, 300 mM NaCl; Sol AFM3: 10 mM Tris:HCl, 300 mM NaCl. Two μL of these solutions were then dispensed onto freshly-cleaved mica and proteins allowed to adsorb for 15–30 min. Samples were imaged after dilution with 150 μl of the same buffer. For conventional AFM imaging, a Multimode 8 AFM (Bruker), equipped with a 160-μm scanner (J-scanner) and oxide-sharpened Si₃N₄ cantilevers (k = 0.09 N·m⁻¹, Olympus) was operated in contact mode in buffer at ambient temperature and pressure. Minimal loading forces of approximately 300 pN were applied during AFM imaging, at scan frequencies of 3–4 Hz using optimized feedback parameters. Images were acquired at 512 x 512 pixel resolution.

HS-AFM images were captured at between 30 and 60 Hz in buffer in AC mode using a NanoWizard ULTRA speed A (JPK) equipped with an ULTRA Speed 2.8 μm scanner and ‘Ultra-Short Cantilever’ USC-0.3 MHz probes (0.3 N·m⁻¹, Nano World), or in tapping mode using a Dimension Fastscan AFM (Bruker) equipped with 30 μm Icon scanner and ScanAsyst fluid plus probes (0.7 N·m⁻¹, Bruker).

Standard image analysis and treatments were initially performed using NanoScope Analysis software (Bruker). When necessary, AFM images were processed with 0 to 3^th order plane fitting and 0 to 3^rd order flattening to reduce XY tilt. HS-AFM frames were aligned using ImageJ scripts (mostly StackReg). To analyze protein dynamics from consecutive series of AFM images, images were aligned as stacks, and standard deviations evaluated using the Z-project command.

Atomistic molecular dynamics simulations

Three hexamers were assembled using CcmK1 structure (PDB ID 3BN4) and applying symmetry operations within Swiss-PdbViewer (http://www.expasy.org/spdbv/)[63]. Interhexamer sulfates were either deleted, or replaced by phosphate, or MePi₃ ligands, which were parameterized using the PRODRG server (http://davapc1.bioch.dundee.ac.uk/cgi-bin/prodrg)[64]. In the case of PyrR protein, the crystallized dimer was used (1AX4). After a first energy minimization within YASARA, the CcmK1 ensemble or the PyrR dimer were hydrated within a cubic cell with dimensions extending by 20 nm around protein atoms, which was filled with explicit solvent. Periodic boundary conditions were applied and the AMBER14 force field was employed. The cut-off for the Lennard-Jones potential and the short range electrostatics was 8 Ǻ. Long-range electrostatics were calculated using the Particle Mesh Ewald (PME) method with a grid spacing <0.1 nm, 4th order PME-spline, and PME tolerance of 10⁻⁵ for the direct space sum. Force field parameters for protein and phosphate atoms followed general AMBER force field atom type assignments. YASARA’s pKa utility was used to assign pKa values at pH 7.0. The simulation cell was neutralized with NaCl (0.9% final concentration) by iteratively placing sodium and chlorine ions at the coordinates with the lowest electrostatic potential. The entire system was energy-minimized using steepest descent minimization, in order to remove conformational stress, followed by a simulated annealing minimization until convergence (<0.05 kJ/mol/200 steps). Simulations were run at 298 K, with integration time steps for intra-molecular and inter-molecular forces of 1 fs and 2 fs, respectively. After equilibration, simulations were continued for 15 ns. This was followed by an identical second simulation, with attribution of random initial atomic velocities. Intermediate structures were saved every 250 ps.

Dihedral angle analysis and generation of figures was done with Pymol (https://www.pymol.org/).

Coarse-grained simulations

Brownian Dynamics simulations of a coarse-grained model were implemented to analyze the assembly of CcmK hexamers. In the model, a CcmK hexamer is represented at very low resolution as a nearly spherical bead of diameter σ, described in terms of its position r_i and orientation Ω_i. The interaction between hexamers is modeled using the same potential as in Ref. [45], which depends on three main parameters: the binding energy between hexamers, ϵ₀; the preferred angle of the inter-hexamer interaction, ν; and the local bending stiffness dictated by α. The interaction between an hexamer and the substrate was modeled using a potential where z is the vertical distance to the substrate, σ_s = σ/2 is the equilibrium distance, ϵ_s is the adhesion strength and θ_i is the polar angle of the orientation of hexamer i. The orientational dependence of this potential is introduced to mimic the asymmetry between the concave and convex faces of the CcmK proteins and the presence of the tags.

We worked using reduced units in terms of the diameter of hexamers σ, their diffusion coefficient D, and the binding energy ε₀. The differences between CcmK proteins are modelled using different interaction parameters. In these reduced units, typical parameters used in the simulation are: binding stiffness, α = 0.1, preferred angle between hexamers, ν = 1.4 radians and torsion constant, k_t = 1.5. Brownian Dynamics simulations were implemented using a stochastic Euler’s algorithm with a timestep Δt = 10⁻⁵σ²/D. In the simulations, N hexamers are placed initially at random positions and orientations inside a cubic box with periodic boundary conditions and their dynamics is monitored for typically 4 10⁸ timesteps (corresponding to ca. 0.8 ms). Snapshots and movies of the simulation were prepared with VMD (http://www.ks.uiuc.edu/Research/vmd/).