Hexameric assembly of the bifunctional methylerythritol 2,4-cyclodiphosphate synthase and protein-protein associations in the deoxy-xylulose-dependent pathway of isoprenoid precursor biosynthesis.

The bifunctional methylerythritol 4-phosphate cytidylyltransferase methylerythritol 2,4-cyclodiphosphate synthase (IspDF) is unusual in that it catalyzes nonconsecutive reactions in the 1-deoxy-D-xylulose 5-phosphate (DOXP) pathway of isoprenoid precursor biosynthesis. The crystal structure of IspDF from the bacterial pathogen Campylobacter jejuni reveals an elongated hexamer with D3 symmetry compatible with the dimeric 2C-methyl-D-erythritol-4-phosphate cytidylyltransferase and trimeric 2C-methyl-D-erythritol-2,4-cyclodiphosphate synthase monofunctional enzymes. Complex formation of IspDF with 4-diphosphocytidyl-2C-methyl-D-erythritol kinase (IspE), the intervening enzyme activity in the pathway, has been observed in solution for the enzymes from C. jejuni and Agrobacterium tumefaciens. The monofunctional enzymes (2C-methyl-D-erythritol-4-phosphate cytidylyltransferase, IspE, and 2C-methyl-D-erythritol-2,4-cyclodiphosphate synthase) involved in the DOXP biosynthetic pathway of Escherichia coli also show physical associations. We propose that complex formation of the three enzymes at the core of the DOXP pathway can produce an assembly localizing 18 catalytic centers for the early stages of isoprenoid biosynthesis.

Life depends on isoprenoids such as sterols, dolichols, triterpenes, ubiquinone, and plastoquinone and components of macromolecules such as the prenyl groups of prenylated proteins and isopentenylated tRNAs (1,2). This large family of natural products contributes to numerous and varied biological functions including electron transport processes in respiration and photosynthesis, hormone-based signaling, the regulation of transcription, and posttranslational processes that control lipid biosynthesis, meiosis, apoptosis, protein cleavage, and degradation. In addition, isoprenoids fulfill an important role as a structural component of cell and organelle membranes.
Enzymes of the DOXP pathway present attractive targets for the development of broad spectrum antimicrobial drugs targeting some of the world's most serious diseases, including tuberculosis, malaria, and a range of sexually transmitted infections (6,7,29). These enzymes are absent from humans, provide the pathogen's only source of IPP and DMAPP, and have been validated (i.e. proven essential for survival) as therapeutic targets by genetic approaches (29). Such genetic validation is complemented by the chemical validation of DOXP reductoisomerase using the antimicrobial drug fosmidomycin (17,30).
Studies on the DOXP pathway, which represents one of the best examples of modern proteomic research (6 -9, 31), have concentrated on identifying which organisms utilize this biosynthetic route, delineating the intermediates of the pathway and characterizing the structure-function relationships of individual enzymes. Whereas detailed descriptions of the pathway and individual components are now available, we lack knowledge of how the pathway is regulated.
Genes encoding enzymes of the DOXP pathway are dispersed throughout the individual genomes, with the exception of ispD and ispF, which are transcriptionally coupled or, in a few organisms, fused together coding for a bifunctional enzyme (18). The fused gene is ispDF, and recent analysis has shown that the encoded product IspDF indeed has two catalytic functions (33). In the vast majority of cases, bifunctional enzymes associated with a metabolic pathway catalyze consecutive reactions in that pathway. IspDF is therefore highly unusual because it catalyzes nonconsecutive steps in the pathway on either side of the intervening kinase, IspE (Fig. 1). This suggests that there may be a physical coupling between IspDF and IspE, with implications for the organization and regulation of the DOXP pathway. The quaternary structure of the bifunctional enzyme is of interest because the monofunctional IspD is a dimer (20,34), and the monofunctional IspF is a trimer (35)(36)(37).
To investigate these aspects of structure and organization in the non-mevalonate biosynthetic pathway, we have determined the crystal structure of IspDF from the highly prevalent foodborne pathogen Campylobacter jejuni (CjIspDF). In particular, with the use of analytical ultracentrifugation, we studied the possible association of CjIspDF with CjIspE, the possible association of Agrobacterium tumefaciens IspDF (AtIspDF) with AtIspE, and the possible association of monofunctional Escherichia coli enzymes (EcIspD, EcIspE, and EcIspF) with each other.

Sample Preparation-Recombinant CjIspDF, EcIspD, EcIspE, and
EcIspF were prepared by established methods, and purity was checked by SDS-PAGE and matrix-assisted laser desorption ionization time-offlight mass spectrometry (33,38,39). The genes (ispE from C. jejuni and ispDF and ispE from A. tumefaciens) were obtained by PCR of genomic DNA using the oligonucleotide primers listed in Table I. The forward primers contained an NdeI restriction site, and the reverse primers contained a BamHI restriction site. The genes were cloned into PCR-Blunt II TOPO vector (Invitrogen) and then into pET15B expression vector (Novagen). Protein expression and purification again followed published protocols.
Crystallization and Diffraction Data Collection of CjIspDF-Purified enzyme was dialyzed against 100 mM Tris-HCl (pH 7.6) and 50 mM NaCl and then concentrated to ϳ10 mg ml Ϫ1 . Hexagonal rods were obtained at 4°C in 3 l of hanging drops containing a 2:1 mixture of 10 mg ml Ϫ1 protein solution (100 mM Tris, pH 7.6, 50 mM NaCl, 10 mM MgCl 2 , and 2 mM CMP and CTP) with reservoir (35% (v/v) ethylene glycol). The crystals have space group P6 3 22 with one monomer per asymmetric unit and a solvent content of ϳ62% (structure I). Isomorphous crystals, subsequently shown to contain Zn 2ϩ in the IspF active site (structure II), were obtained from a separate protein preparation using 2 l of hanging drops containing a 1:1 mixture of 12 mg ml Ϫ1 protein solution (100 mM Tris-HCl, pH 7.6, 50 mM NaCl, 10 mM MgCl 2 , 2 mM CMP and CTP, and 1 mM Na 2 VO 3 ) with reservoir (25% (v/v) ethylene glycol). Crystals were harvested directly from the drops and then cooled in a nitrogen gas stream at Ϫ170°C for data collection. Diffraction data were measured with Q4 charge-coupled device detectors (Area Detector Systems Corp.) at the European Synchrotron Radiation Facility, processed, and scaled using the HKL (40) and CCP4 program suites (41).
Structure Determination, Model Building, Refinement, and Analysis-Poly-Ala models for single subunits of EcIspD and EcIspF were constructed from Protein Data Bank coordinates 1I52 (20) and 1GX1 (35), respectively, and positioned by molecular replacement (AMORE) (42) with the structure I dataset. A complete model for CjIspDF was then obtained using ARP/wARP (43) with R-work and R-free of 23% and 36%. This model was refined using REFMAC (44) and the graphics program O (45), with inclusion of solvent positions, CMP, GPP, and an ethylene glycol molecule. The R-free and R-work were monitored as guides for refinement, and stereochemistry was assessed with PRO-CHECK (46). Structure II was derived from structure I by rigid body refinement and then completed using similar refinement protocols. Crystallographic statistics are presented in Table II. Surface areas were calculated using GRASP (47); molecular figures were generated using MOLSCRIPT (48) and Raster3D (49).
Size Exclusion Chromatography and Analytical Ultracentrifugation-Size exclusion chromatography was performed with samples applied to a 24-ml (1 ϫ 30-cm) Superdex 200 HR 10/30 size exclusion column equilibrated with buffer (50 mM HEPES, 0.3 M NaCl, 0.01% (w/v) NaN 3 , pH 7.5) and eluted at a flow rate of 0.5 ml min Ϫ1 . The analytical standards were carbonic anhydrase (29 kDa), BSA (66 kDa), alcohol dehydrogenase (150 kDa), ␣-amylase (200 kDa), apoferritin (443 kDa), and thyroglobulin (669 kDa). Blue dextran was used to measure the void volume (V 0 ). The elution volume (V e ) of each standard was measured and corrected by the subtraction of half the sample volume. The retention coefficient (R) for each standard was calculated as R ϭ V e /V o . A plot of log 10 molecular mass of standards against R gave a straight line used to calculate molecular mass values (data not shown).
Sedimentation velocity experiments were performed with a Beckman XL-1 analytical ultracentrifuge, An-50Ti rotor (28,000 rpm at 20°C), using absorption optics at a wavelength of 280 nm. Three different sample concentrations (0.25, 0.5, and 1 mg ml Ϫ1 ) in 100 mM Tris-HCl buffer, pH 7.6, 50 mM NaCl, with the addition of 10 mM dithiothreitol in the presence of EcIspD and IspE, were used in the study, and all data were analyzed with SEDFIT (50). The IspDF samples were analyzed alone and then in combination with the IspE derived from the same organism. All combinations of the two and then three protein mixtures were analyzed for the E. coli enzymes.

Subunit, Domain, and Quaternary Structure of CjIspDF-
Two structures (I and II) have been determined at resolutions of 2.3 and 3.1 Å, respectively. The structure of CjIspDF presents one subunit of 371 amino acids (ϳ41.7 kDa) per asymmetric unit and comprises two distinct domains (Fig. 2a). The N-terminal domain (residues 1-209) corresponds to the cytidyltransferase IspD, and the C-terminal domain (residues 210 -371) corresponds to the synthase IspF. The N-terminal CjIspD segment forms a Rossmann fold-like ␣/␤ domain, into which is inserted an extended "␤-arm" (␤8 and ␤9). The core of the domain consists of a seven-stranded twisted ␤ sheet (␤5, ␤4, ␤1, ␤6, ␤10, ␤7, and ␤11), where all but ␤10 are parallel. This domain shares 25% sequence identity with EcIspD (Fig. 2b), and an overlay of the two structures (Protein Data Bank code 1I52) gives a root mean square deviation of 1.75 Å for 191 C␣ atoms. The C-terminal CjIspF domain of 163 residues displays an ␣/␤ fold constructed with a four-stranded ␤-sheet of ␤12, ␤18, ␤16, and ␤17, with ␤18 antiparallel to the others, on one side and a four-helix bundle of ␣8 to ␣12 on the other. This domain shares a sequence identity of ϳ48% with EcIspF (Fig. 2b), and this pair is more similar than the CjIspD and EcIspD domains, reflected in a root mean square deviation of 0.83 Å for superposition of 149 C␣ atoms of CjIspF and EcIspF (Protein Data Bank 1GX1). The major difference between these two IspF structures occurs in the loop between ␣9 and ␣10 (CjIspD numbering) on one side of the active site, where deviations of nearly 8 Å occur between several equivalent C␣ pairs (data not shown).
Eight segments of polypeptide form the CjIspD-IspF domaindomain interface over an area of ϳ850 Å 2 . These segments include the loop leading into 1, the ␤-strands ␤2 and ␤3, and the helix ␣7 on the N-terminal domain, which interact with the link section between ␣7 and ␤12; the C-terminal sections of ␣8 and ␣10; and the C terminus itself. The interface is mainly hydrophobic in character and involves the residues Val 21 , Phe 25  The CjIspF domain is placed near a crystallographic 3-fold axis parallel to c, and the CjIspD domain is beside a crystallographic 2-fold, and these symmetry elements generate a CjIs-pDF hexamer displaying point group D 3 (Fig. 3). This quaternary assembly is consistent with the monofunctional enzymes from other species (i.e. IspD homodimer and IspF homotrimer).  The hexamer, for which subunits are assigned labels A to F, has a total mass of about 250.2 kDa and longest dimensions of ϳ100 ϫ 130 Å in the equatorial and axial directions, respectively. Two trimers (subunits A, B, and C and then subunits D, E, and F) of IspF domains form the apices, and three dimers of IspD domains form the middle of the assembly (Fig. 3). The IspD dimers involve subunit pairs A with D, B with E, and C with F. A distance of about 35 Å separates the IspD active sites within a dimer, the closest distance between two IspD active sites on separate dimers is 56 Å, and the closest distance between the IspD active site and IspF active site on the same monomer is ϳ38 Å. The closest distance between two IspF active sites is ϳ30 Å. The major contribution to the CjIspD dimer interface is provided by the association of the ␤-arm of each subunit, with a lesser contribution from side chain interactions between residues on helix ␣7. The active site is formed at the dimer interface by six segments of polypeptide from one subunit and one segment from the partner subunit. The dimer interface covers an area of ϳ1670 Å 2 , 9% of the subunit surface area, and involves 26 intersubunit hydrogen bonds. The surface area is reduced by about 200 Å 2 in comparison with EcIspD mainly due to a smaller loop, reduced by 8 residues, between ␤8 and ␤9 in CjIspD (Fig. 2b).
IspF forms a compact homotrimer shaped like an extended trigonal prism. The trimer is formed using a single type of subunit-subunit interface with 18 intersubunit hydrogen bonds, mainly involving side chain atoms. The surface area of a CjIspF monomer buried in the homotrimer is 2630 Å 2 , about 15% of the total surface area of CjIspDF and significantly larger than the value of 2120 Å 2 observed in EcIspF (35). At the core of the trimer, on the 3-fold axis, is a narrow cavity into which is modeled GPP. This assignment was based on the electron density and previous characterization of ligands that bind EcIspF at the central region of the trimer. 2 These ligands also include phosphate, IPP and/or DMAPP, and farnesyl diphosphate. In CjIspDF, the diphosphate moiety of GPP is placed at the entrance to the cavity and participates in 10 hydrogen bonds with the side chains of Arg 351 and amides of Phe 348 from the three subunits (data not shown). The geranyl tail is directed down into the core of the trimer participating in van der Waals interactions with the side chains of Phe 216 , Met 308 , and Phe 348 , which together with Phe 208 , Ile 210 , Val 218 , Leu 346 , and Leu 358 form a hydrophobic lining to the cavity. The presence of GPP likely makes a significant contribution to the stability of the IspF trimer.
There is only one small area of interaction, about 210 Å 2 , between the IspD domain of one subunit with the IspF domain of another subunit. This is a single type of interface involving residues on the ␣2 and ␤2 of subunit A interacting with the 7 segment of subunit C, for example. Two hydrogen bonds are involved here; Lys 62 NZ and O donate and accept hydrogen bonds, respectively, with Gln 319 OE1 and Phe 334 N (data not shown).
The Active Sites of IspD and IspF Are Highly Conserved-Previous work has produced an excellent understanding of substrate recognition and the mechanism of catalysis by EcIspD and EcIspF (20, 34 -37, 51). We therefore only comment briefly on the two active sites in CjIspDF.
The active site of CjIspD is formed at the dimer interface by polypeptide segments 9 -16, 73-76, 79, 94 -99, 191-193, 144 -148, and 167-168 from one subunit and segment 127-129 from the other. The recognition and interactions of EcIspD, by direct hydrogen bonds, with substrate and/or product involve 19 residues (Fig. 2b). Ten of these residues only use main chain functional groups, and five are strictly conserved in CjIspDF (Ala 10 , Ala 11 , Gly 12 , Gly 73 , and Asp 74 ). A noteworthy difference with respect to the active site is the Gly 18 to Ser 14 alteration b, the amino acid sequence of CjIspDF with the assignment of secondary structure. Strands are labeled ␤1 to ␤19, helices are labeled ␣1 to ␣12, and 3 10 -helical segments are labeled 1 to 7. The amino acid sequences of EcIspD and EcIspF are aligned with IspDF, taking into consideration a structural overlay. Residues boxed in red are strictly conserved; those in black are similar. OE indicates residues that participate in hydrogen bonding interactions with substrate and or product, whereas ‚ marks residues involved in van der Waals interactions with active site ligands. E identifies residues that bind metal ions, and ૾ indicates the isoleucine-lysine difference in the IspF active sites. and an Arg 19 to Thr 15 change (Fig. 2b). In EcIspD, Arg 19 interacts with a CTP phosphate using main chain amide and side chain groups. Alteration to a threonine does not influence the main chain amide and leaves the hydroxyl in combination with the adjacent Ser 14 hydroxyl to likely interact with substrate.
Eight of the remaining nine residues that use side chains (Arg 16 , Lys 23 , Ser 79 , Asp 96 , Arg 99 , Thr 129 , Arg 139 , and Lys 191 in CjIspDF) are strictly conserved. Arg 16 and Thr 129 use both main chain and side chain functional groups. Four of these conserved residues, Arg 16 , Lys 23 , Lys 191 , and Arg 139 , bind and polarize the substrate for nucleophilic attack and then serve to stabilize the negatively charged transition state (19). Two residues, Arg 76 and Thr 147 in CjIspDF, are important for van der Waals interactions with ligands and are conserved in EcIspD (Fig. 2b).
The active site of CjIspF is also formed with contributions from two subunits, in this case, by polypeptide segments 217-219, 241-243, 251-252, and 265-278 from one subunit and segments 309 -315 and 340 -344 from the partner. The catalytic function of IspF depends on two divalent cations, which orient and polarize the substrate (35)(36)(37), and the residues that bind these cations are strictly conserved. Structure II contains a Zn 2ϩ ion in the CjIspF active site, confirmed by an x-ray Absorption Near Edge Scan and anomalous dispersion measurements (data not shown). The cation is coordinated by Asp 217 , His 219 , His 251 , and a water molecule. In contrast, structure I does not contain Zn 2ϩ , and we presume that the cation leached out of the protein during the longer dialysis step (overnight as opposed to 2 h) employed for the batch of enzyme from which that structure was derived. Structure I carries Mg 2ϩ in the CjIspF active site coordinated to the phosphate of CMP, side chains of Asp 217 and Thr 341 , and water molecules. Although nearby, this is not the same divalent cation-binding site used by EcIspF when it coordinates the diphosphate containing CDP or substrate 4-diphosphocytidyl-2C-methyl-D-erythritol (34, FIG. 2-continued  35). It is likely that Mg 2ϩ is brought into the active site with the ligand, be it substrate or, in this case, CMP. For catalysis by the CjIspF domain, we anticipate that similar coordination would occur as in EcIspF, i.e. the ion interacts with two phosphates, the conserved Glu 344 and water molecules.
Hydrogen bonding interactions with the substrate/products in EcIspF involve 10 amino acids, of which 7 (Ile 57 , Gly 58 , Phe 61 , Ala 100 , Pro 103 , Met 105 , and Leu 106 ) use only main chain groups. These residues correspond to Ile 266 , Gly 267 , Tyr 270 , Ala 309 , Pro 312 , Leu 314 , and Lys 315 in CjIspDF. Despite the use of only main chain groups to interact with the ligands, these residues remain well conserved (Fig. 2b). Five other amino acids provide side chains to hydrogen bond with active site ligands, and these residues, His 243 , Ser 244 , Asp 265 , Lys 313 , and Thr 342 of CjIspDF, are strictly conserved in EcIspF (Fig. 2b). There is a significant difference in the active site of the two enzymes. In CjIspDF, Lys 318 NZ forms a hydrogen bond with the CMP O 2 Ј ribose hydroxyl. In EcIspF, the corresponding position is Ile 109 placed so the side chain does not contribute to the active site (data not shown).
Size Exclusion Chromatography-Experiments were carried out on all enzymes individually and then carried out on the various mixtures. EcIspD showed the expected monomer and dimer species but also showed a tetramer, which disappeared on the addition of dithiothreitol. Previously, we showed that this enzyme forms a dimer-dimer association due to a disulfide linkage (34). Analysis of the CjIspDF trace showed trimer, hexamer, and dodecamer assemblies. A mixture of CjIspDF and CjIspE only showed peaks corresponding to the individual enzymes. In similar fashion, EcIspD and EcIspE did not appear to associate, nor did EcIspD and EcIspF. However, analysis of EcIspE mixed with EcIspF revealed a peak of ϳ500 kDa not observed in the experiments on individual enzymes (data not shown). This may represent a complex of EcIspF and EcIspE. Encouraged by this observation, we investigated further using analytical ultracentrifugation.
Sedimentation Characteristics-Initial experiments to investigate the association of IspDF with IspE used recombinant C. jejuni enzymes. Aggregation problems with CjIspE were noted, and we therefore generated recombinant AtIspDF and AtIspE, which proved more amenable for analysis and, importantly, provided another system for comparison. The amino acid sequence identity is 32% between the bifunctional enzymes and 25% between the IspE of both species.
The sedimentation velocity runs for CjIspDF and AtIspDF were consistent, and each showed only two peaks, of ϳ125 and ϳ250 kDa (Fig. 4a). These peaks correspond to a trimer and a hexamer, and the absence of any other species suggests that IspDF can best be described as a dimer of trimers. The IspD dimers appear less stable than the IspF trimers, which may be explained by the much larger interface area associated with the latter domain, as described earlier. The bifunctional IspDF samples were mixed with the cognate IspE samples, and the results indicated the presence of the trimeric and hexameric IspDF together with species of higher mass (ϳ380 and ϳ580 kDa) (Fig. 4b). A complex of CjIspDF hexamer plus three IspE dimers would have a mass of about 408 kDa. In the case of AtIspDF ϩ AtIspE (Fig. 4c), an additional high mass species of nearly 1 MDa is observed.
A mixture of EcIspD with EcIspF showed the expected dimer and trimer species for the individual enzymes in a single broad peak, but no higher order complexes were seen (data not shown). In contrast, a mixture of EcIspD with EcIspE displayed a dimer for each enzyme of ϳ51 and ϳ62 kDa, respectively, and also displayed a peak of ϳ130 kDa. The analysis of EcIspE mixed with EcIspF revealed a broad peak corresponding to their dimeric and trimeric states, respectively, as well as a peak at ϳ250 kDa, which fits with a model of two IspF trimers and three IspE dimers. When the individual enzymes EcIspD, EcIspE, and EcIspF were mixed, a complex of ϳ430 kDa was observed (Fig. 4d). This likely corresponds to an assembly of three IspD dimers, three IspE dimers, and two IspF trimers, similar to that observed for the complex assembly of IspDF and IspE derived from both C. jejuni and A. tumefaciens. A summary of these results is given in Table III. Note that in all cases, the individual enzymes were first characterized as controls, and we only observed the higher mass species when enzyme mixtures were analyzed. enzyme catalyzing nonconsecutive steps in the pathway. Bifunctional enzymes are distinctive and highly conserved products of relatively infrequent gene fusion events that generally link proteins with related yet distinct functions. Some interaction with the intervening kinase IspE therefore seemed plausible. Interestingly, associations were also observed for the monofunctional E. coli enzymes. We can now consider likely biological implications of such associations.

Implications of
There are distinct mechanisms by which metabolic pathways are controlled: firstly, covalent modification, as exemplified by phosphorylation of mammalian HMG-CoA reductase (5), which regulates the mevalonate pathway. Secondly, the compartmentation of enzymes in organs or organelles can control pathway flux. For example, the mevalonate pathway occurs in the cytosol and mitochondria of plants, whereas the DOXP pathway is compartmentalized into chloroplasts. Apicomplexan parasites likewise carry the DOXP pathway enzymes in their apicoplast (52). Thirdly, the binding of effector molecules can provide feedback control or link pathways together. There is also control by repression or activation of gene expression. In Gram-positive bacteria that utilize the mevalonate pathway, the relevant genes are organized into operons (7) and are likely regulated at the level of transcription.
The first two types of control mechanisms are not applicable to the DOXP pathway in eubacteria, and there is no evidence of a transcriptional regulator to coordinate expression of the relevant genes. The recent identification of isoprenoids binding to E. coli IspF 2 raises the possibility of feedback regulation in the DOXP pathway, and now our observation of protein-protein interactions between IspDEF entities suggests a role for a multienzyme complex.
Functional links between proteins can be inferred from the genomic context of their genes (53), and recent analyses have exploited this and the presence of gene fusions to predict such connections (32, 54 -56). The STRING data base (32) uses such information to predict functional links, which may or may not involve some physical interaction between the gene products. For the enzymes of the DOXP pathway, the data base predicts that IspD has links to 4-hydroxy-3-methylbut-2-en-l-yl diphosphate synthase (IspG), IspF, DOXP reductoisomerase (IspC), and 4-hydroxy-3-methylbut-2-enyl diphosphate reductase (IspH). There is no mention of IspE because the biochemical context of the enzyme has not been factored into the data base.
A fusion of IspD and IspF to create the bifunctional enzyme would reduce the entropy of dissociation, thereby reducing the association free energy. We did not observe association between the monofunctional EcIspD and EcIspF until EcIspE was present, and we suggest that there may have been a thermodynamic benefit to the fusion of IspD with IspF in some organisms. A complex comprising three IspD and IspE dimers together with two IspF trimers would provide an organized assembly that localizes 18 catalytic centers. This could support efficient catalysis of three consecutive reactions at the core of isoprenoid precursor biosynthesis because there is a shorter distance over which diffusion of product from one active site to become substrate in another takes place. Irrespective of whether such a process exploits the advantages of substrate channeling, complex formation may contribute control of the metabolic flux of this key biosynthetic pathway. Monomer (D), 1 dimer (D) ϩ 1 dimer (e) ϩ 1 trimer (F), 2 trimers (F) ϩ 1 dimer (D) ϩ 2 dimers (E), 2 trimers (F) ϩ 3 dimers (D) ϩ 3 dimers (E), higher order assembly