Electron cryo-microscopy of Bacteriophage PR772 reveals the composition and structure of the elusive vertex complex and the capsid architecture

Bacteriophage PR772, a member of the Tectiviridae family, has a 70-nm diameter icosahedral protein capsid that encapsulates a lipid membrane, dsDNA, and various internal proteins. An icosahedrally averaged CryoEM reconstruction of the wild-type virion and a localized reconstruction of the vertex region reveals the composition and the structure of the vertex complex along with new protein conformations that play a vital role in maintaining the capsid architecture of the virion. The overall resolution of the virion is 2.75 Å, while the resolution of the protein capsid is 2.3 Å. The conventional penta-symmetron formed by the capsomeres is replaced by a large vertex complex in the pseudo T=25 capsid. All the vertices contain the host-recognition protein, P5; two of these vertices show the presence of the receptor-binding protein, P2. The 3D structure of the vertex complex shows interactions with the viral membrane, indicating a possible mechanism for viral infection.

inner lipid membrane and lack of a tail in the dormant viral particle. During the 50 process of infection, these viruses produce a membranous tube derived from the inner 51 membrane of the viral particle, lined from the inside by proteins P7, P11 and P32 52 (Peralta et al., 2013;Saren et al., 2005). This tube is used to inject the viral dsDNA 53 into the host (Peralta et al., 2013;Saren et al., 2005). 54 55 Genome analysis of PR772 identified 32 open reading frames (ORFs) containing at 56 least 40 codons (Lute et al., 2004). Twenty-eight annotated proteins are known to be 57 expressed from the genome of which 3 do not make it into the final assembly 58 (Butcher, Manole, & Karhu, 2012;Lute et al., 2004). In previous studies on PRD1, a 59 close relative of PR772, it was shown that the capsid is formed by proteins P3 (major 60 capsid protein), P30 (capsid associated protein) and the vertex complex. The vertex 61 complex, includes the penton protein (P31), the host-recognition protein (P5) and 62 not two as previously believed (Abrescia et al., 2004). The new N-terminal P3 88 conformation plays an important role in accommodating the P30 protein during 89 particle assembly. The C-terminal region of P3 not only shows the formation of a b-90 sheet with P30 but also helps in locking the adjacent trisymmetron through a hinge 91 mechanism, thus facilitating the formation of icosahedral particles and regulating their 92 size. Localized asymmetric reconstruction of the vertex region of PR772 revealed a 93 P5-P31 heteropentameric base and the binding of P2 to P5 in the complex. A 94 combination of high-resolution icosahedral symmetrized single-particle 95 reconstruction and localized asymmetric reconstruction has enabled us to answer 96 some of the intriguing questions about the particle architecture, composition of the 97 penton base and arrangement of the vertex complex. showed that most of the capsid was resolved to 2.30 Å ( Figure 1A   The CryoEM 3D reconstruction of PR772 shows that the viral particle follows a 139 pseudo T=25 lattice architecture with a (h,k) of (0,5) resulting in an icosahedral 140 structure with 20 large trisymmetrons and 12 penta-symmetrons. However, the 141 analysis revealed that the penta-symmetrons were hetero-pentamers (Caspar & Klug, 142 1962;Sinkovits & Baker, 2010). Each of these trisymmetrons have 36 copies of P3, 143 the major capsid protein (MCP), arranged as 12 trimers, each of which structurally 144 appears to be hexagonal in shape. At the fivefold vertices, the penta-symmetrons are 145 replaced by a vertex complex to complete the icosahedral shell. With a pseudo T=25 146 architecture, PR772 is one of the larger wild type viruses resolved to a resolution 147 below 3 Å.  between P31 and N-terminal P5 domain we could not rule out that any of these two 234 proteins or a mixture could potentially form the penton base. Therefore, two initial 235 models were generated. One model used the P31 protein sequence and the other used 236 the P5 protein sequence as part of the input. In the initial observation of all the 237 predicted segments of the two models, the P31 protein segment with residues 112-126 238 and the P5 protein segment with residues 107-121 occupied the same region of the 239 sectioned map ( Figure 1E, F). On closer inspection of the side chains from the two 240 models and their fit into the CryoEM map densities in this region, the P5 protein side 241 chains fit clearly at 3.24 RMS and the P31 side chains could only be fitted at RMS 242 values lower than 2.64. As the CryoEM map was generated by icosahedral 243 symmetrized reconstruction, we assumed that the densities of P5 and P31 were 244 averaged. This hinted that the penton base could be a heteropentamer formed by both 245 P5 and P31. 246

247
With an assumption that the penton base could be a heteropentamer, the P5 N-248 terminal domain (residues 1-124) and P31 (residues 1-126) were modelled using the 249 icosahedrally averaged map at ~3.04 and ~2.2 RMS respectively in Coot (Emsley,   From the icosahedrally symmetrised reconstruction, at lower contour levels (0.065 in 296 chimera) the map showed smeared densities above the 5-fold vertex of the viral 297 particle ( Figure 4A-B). This could be due to a symmetry mismatch of the proteins 298 present in the region. Accordingly, the smeared region over the five-fold vertex was 299 isolated and resolved by capsid signal subtraction followed by localized asymmetric 300 (C1) reconstruction (see Methods). All the classes generated by 3D classification 301 showed a single protruding density except one of the classes, which revealed two 302 significant densities; one poorly resolved knob-like density and another more well 303 resolved density closely interacting with one of the monomers of the penton ( Figure  304 5-figure supplement 1). On closer inspection of every 3D class generated during the 305 process of localized asymmetric reconstruction, we noticed that three of the subunits 306 of the penton base had a stem-like protrusion close to the 5-fold axis, extending 307 outwards and interacting with each other forming a thick stalk ( Figure 4C

344
With the localized asymmetric reconstruction, the structure and composition of the 345 penton base was revealed. It is now known that the stalk-like density observed, 346 emanating from the penton base, is built with three copies of the P5 protein, forming a 347 collagen-like motif. As described previously, one of the 3D classes, with ~41,000 348 sub-particles from a total of 276,000 sub-particles, also revealed an extra rigid density 349 closely interacting with the penton base and the protruding stalk ( Thr384 -Leu386 from subunit c of P3 trimer 1 and residues Thr384 -Asn388 from 388 subunit c of P3 trimer 2 respectively (Figure 2A

474
Our studies show that unlike what was predicted for PRD1, the penton base of PR772 475 is an asymmetric heteropentamer consisting of three copies of P5 and two copies of 476 P31 ( Figure 4F). P31 has high sequence similarity with the N-terminal domain of P5. 477 Previously, it was shown that P31 can replace P5 to form the penton and produce an 478 intact viral particle . Even though the viral particles were 479 intact, they were non-infectious due to the lack of the viral receptor-binding protein, 480 P2, that is bound to the host recognition protein, P5 ( sus525 mutant that lacks P31, it was also shown that these particles lacked the vertex 483 complex (Rydman et al., 1999). With the current model, P5 will be unable to replace 484 P31 to form an intact viral particle due to obstructions that occur during the formation 485 of the vertex complex. A penton with only P5 subunits will inhibit (i) the formation of 486 a stable collagen-like motif (Figure 7)  The localized asymmetric reconstruction shows that P2 is bound to P5 ( Figure 5). In 501 the current CryoEM data, the orientation of P2 with respect to the penton base is 502 reversed when compared to what was described for PRD1 (Huiskonen et al., 2007;Xu 503 et al., 2003). Here, the beta-propeller motif of P2 with domain I and II, interacts with 504 the N-terminal base of P5 and also the stalk. This was also speculated on in a previous 505 study, where it was noted that members of the Tectiviridae family with large sequence 506 variation in the beta-propeller region of P2 also showed variations in the P5 subunit as 507 a compensatory effect (Saren et al., 2005). P2 interacts with both the N-terminal base 508 and the stalk region of P5 but does not interact directly with any other structural 509 protein in PR772. The interaction of P2 with the P5 stalk stabilises the C-terminal 510 region of P5 ( Figure 5). In the case of PRD1, P2 seems to occupy all vertices expect 511 for the unique packaging vertex (Butcher et al., 2012;Javier Caldentey et al., 2000;512 Saren et al., 2005). In this scenario, the special conformation in which the N-terminal 513 residues of P5 are wedged in between the neighbouring P3 subunits was found to be 514 prominent (Abrescia et al., 2004). In the current model of PR772, P2 seems to occupy 515 only ≈2 vertices of the icosahedral shell. This was also indicated in a previous study 516 comparing the abundance of different viral proteins from various members of the 517 Tectiviridae family using western blots (Saren et al., 2005). 518

519
In the icosahedrally averaged model of PR772, we see that the special conformation 520 of P5 where the N-terminal residues are wedged between the P3 subunits of the 521 adjacent trimer 1 is not prominent, but these P5 densities are visible at lower RMS 522 values. The presence of P2 in the vertex complex may coincide with the N-terminal 523 region of P5 adopting the special conformation. It was reported that, compared to the 524 CryoEM map of a wild type PRD1 virion, the CryoEM map of sus690, a mutant 525 PRD1 virion that lacks P2 and P5, showed absence of density in the region where we 526 see the N-terminal of P5 wedged between two neighbouring P3 subunits. The lack of 527 this density was attributed to the conformational changes in P31 due to the binding of 528 P5 or the presence of P5 in between P31 and the P3 trimer (Huiskonen et al., 2007). 529 In light of the current findings in PR772, P5 is known to be part of the 530 heteropentameric base forming the penton and the absence of the above-mentioned 531 density confirms that P5 adopts the special N-terminal conformation and not P31. propose a mechanism for the initiation of infection. In analogy to the studies on 559 PRD1, It is known that P5 is needed for host recognition, but the binding of P5 to the 560 host is transient (A M Grahn et al., 1999). The host binding is stabilised by the high-561 affinity interaction of P2 and its receptor, locking the viral particle to the host (A M 562 Grahn et al., 1999). The relative changes between P5 and P2, triggers the disruption 563 of the vertex complex by pulling the N-terminal region of the P5 trimers that are 564 wedged between the P3 subunits. This will in turn disrupt the interaction between the 565 P5 proteins and P30 as seen in Figure 6. This disruption cascades further and disturbs 566 the interaction between P30 and P16, rendering the whole vertex complex unstable. 567

591
The condition for CryoEM grid preparation was optimised for collecting a large 592 number of particle images. For vitrification of the viral sample by plunge freezing 593 into liquid ethane, we used a Vitrobot Mark IV (ThermoFisher). The best grid 594 condition with uniform sample distribution was obtained by applying 3 µL of 7 595 mg/mL concentrated viral sample solution on a glow-discharged C-Flat grid CF-2/2-596 2C under 100 % humidity at room temperature. 597

598
The data were collected on a Titan KRIOS (ThermoFisher) equipped with a K2 599 Summit (Gatan) direct electron detector and a GIF Quantum LS (Gatan) energy filter. 600 All the data were collected at a magnification of 130 k in EFTEM mode with a pixel 601 size of 1.06 Å. The slit width of the energy filter was 20 eV. The dose rate was 4.4 e -602 per Å 2 per second with a total exposure of 9 seconds resulting in a total dose of ~40 e -603 /Å 2 . The total dose was distributed over 40 frames in each movie. 3220 movies were 604 collected. 605 606 Image Processing

607
The movie frames were corrected for beam induced sample motion and aligned using 608 MotionCor2 (Zheng et al., 2017). The first 3 frames of the movies were skipped and 609 the rest were aligned. These aligned frames were averaged with and without dose 610 weighting. The non-dose weighted image stacks were used to estimate defocus and 611 correct CTF using CTFFIND4 (Rohou & Grigorieff, 2015). All estimated fits of 612 defocus and CTF were visually inspected. All images with significant astigmatism or 613 a prominent ring due to crystalline ice around 3-4 Å were discarded. manually picked particles were used as templates for auto-picking. The auto-picked 620 particles were binned 2´ during the extraction (box size of 429´429 and 2.12 Å/pix). 621 Extensive reference free 2D classification was performed to remove any particle 622 images with ethane contaminants or broken/empty viral particles. The classes with 623 good 2D averages were selected and the particles from these classes were extracted. 624 This resulted in 51893 particles that were used for 3D classification. RELION: 3D 625 initial model tool, which is based on stochastic gradient descent, was used to generate 626 an ab-initio reference map for 3D classification.The 2´ binned particle images were 627 used to generate this low-resolution icosahedrally averaged map. This low-resolution 628 map was used as a reference for 3D classification. The 3D classification was 629 performed with icosahedral symmetry (I4) applied. The most dominant class, with 630 46348 particles, was selected and the particles were extracted for final refinement. 631 Icosahedral symmetry (I4) was also applied during the final refinement. The 632 refinement with the 2´ binned particles reached Nyquist sampling (~4.24 Å here). 633 The particles from the final iteration step of the refinement were re-extracted without 634 binning and further refined. The reference map was also scaled to match the new box 635 and pixel size (858x858 and 1.06 Å/pix respectively) using e2proc3d.py from 636