Caught in the Act: Anthrax Toxin Translocation Complex Reveals insight into the Lethal Factor Unfolding and Refolding mechanism

ABSTRACT Translocation is essential to the anthrax toxin mechanism. Protective antigen, the translocon component of this AB toxin, forms an oligomeric pore with three key clamp sites that aid in the efficient entry of lethal factor or edema factor, the enzymatic components of the toxin, into the cell. LF and EF translocate through the protective antigen pore with the pH gradient between the endosome and the cytosol acting as the driving force. Structural details of the translocation process have remained elusive despite their biological importance. To overcome the technical challenges of studying translocation intermediates, we developed a novel method to immobilize, transition, and stabilize anthrax toxin. Here, we report a cryoEM snapshot of PApore translocating the N-terminal domain of LF. The resulting 3.1 and 3.2 A structures of the complex trace LFN as it unfolds near the α clamp, translocates through the Φ clamp, and begins to refold in the charge clamp. In addition, helical density inside the β barrel channel suggests LF secondary structural elements begin to refold at the charge clamp site. We conclude the anthrax toxin uses an extended β barrel to efficiently fold the enzymatic payload prior to channel exit. This refolding mechanism has broader implications for pore length of other protein translocating toxins.


INTRODUCTION 16
The anthrax toxin is not only a deadly Bacillus anthracis virulence factor, but also serves 17 as a model system of protein translocation and as a peptide therapeutic delivery platform (Young 18 and Collier 2007, Thoren and Krantz 2011). It's biological importance and biotechnology utility 19 have spurred significant biochemical and biophysical advances in understanding the anthrax 20 intoxication mechanism. In order to gain entry into the cell, this archetypical AB toxin must cross 21 the endosomal membrane. This is accomplished by the B component of anthrax toxin, termed 22 protective antigen (PA). PA forms a translocon pore through which lethal factor (LF) or edema 23 factor (EF), the A component, translocate. Here, we developed an approach to elucidate the 24 structural and mechanistic details of the anthrax toxin during translocation in an effort to 25 understand how LF unfolds in the endosome, translocates through PA, and refolds in the cytosol. 26 An overview of the anthrax toxin mechanism has been reviewed by the Collier lab (Young 27 and Collier 2007) and is briefly summarized here. The first step in intoxication is the 85 kDa 28 monomeric PA binding to host cell receptors. Then the pro-domain of PA is cleaved leaving the 29 63 kDa PA to oligomerize into heptameric or octameric prepore (PAprepore) (Santelli, Bankston et 30 al. 2004, Kintzer, Thoren et al. 2009). Up to three LF and/or EF components can bind to the 31 PAprepore heptamer (Mogridge, Cunningham et al. 2002, Kintzer, Thoren et al. 2009, Antoni, 32 Quentin et al. 2020). The AB toxin complex is endocytosed through clatherin mediated 33 endocytosis (Abrami, Liu et al. 2003). As the endosome acidifies, PAprepore undergoes a 34 conformational change to a pore (PApore) (Miller, Elliott et al. 1999). This pore inserts into the 35 endosomal membrane to form a channel and create a pH gradient. With the pH gradient as the 36 driving force, LF or EF unfold and translocate into the cytosol using a hypothesized Brownian 37 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint 7 slurry and associated with the transmembrane portion of PApore ( Figure 1E). To promote lipid 106 bilayer formation we dialyzed away excess detergent ( Figure 1F). The soluble LFN-PApore-107 nanodisc complexes were then eluted off the thiol sepharose beads using a reducing agent. Eluted 108 complexes at pH 7.5 were added to the cryoEM grid. Then the pH of solution on the grid was 109 dropped to pH 5.5 to capture the native complex at low pH prior to blotting and plunge freezing 110 ( Figure 1G). Acidification of complexes was time sensitive as unfolded, partially translocated 111 LFN would rapidly aggregate in solution at low pH and nanodiscs tended to migrate to the air-112 water interface given enough time (data not shown). Therefore, we waited to acidify complexes 113 until they were on the cryoEM grid and plunge freeze grids within 30 seconds of sample 114 application. 115 Using our TITaNS methodology, we were able to reconstruct a 3.1 Å and a 3.2 Å cryoEM 116 map of LFN translocating through PApore (Figure 2A). We traced density of unfolded LFN peptide 117 in the PApore translocon using local refinement, volume subtraction, and 3D variability analysis. 118 These densities were most notable near the three clamp sites within PApore, suggesting the α, Φ, 119 and charge clamps stabilize translocation intermediates. Using the well-defined cryoEM densities 120 inside the pore, molecular models of LFN translocating through the PApore were built ( Figure 2B). 121 122

Unfolding intermediates of LFN during translocation 123
Prior to translocation, LFN is bound to the cap of PA at the interface of two PA protomers 124 with helix α1 bound to the α clamp (Feld, Thoren et al. 2010, Hardenbrook, Liu et al. 2020). In 125 our translocation complex, we observe density for LFN at the interface of the two PA protomers 126 and at the α clamp. In order to explore the heterogeneity of LFN bound to the PA cap, we used 127 symmetry expansion in combination with 3D variability analysis. The resulting vectors revealed a 128 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. To trace LFN from the α clamp to the Φ clamp, we subtracted PApore model density from 148 our unsharpened map to reveal electron density in the pore lumen not accountable for the PApore. 149 Density started in the α clamp and traveled down the pore lumen, interacting with β sheet 39 of 150 PApore. Density near the top of the PApore cap was in proximity to several hydrophobic residues of 151 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint PApore ( Figure 4A). Some of these residues, including Phe202, Phe236 and Phe464 were in the α 152 clamp and have previously been proposed to aid in unfolding LF and stabilizing unfolded 153 intermediates(Feld, Thoren et al. 2010). Additional hydrophobic residues further down the pore 154 lumen, such as Trp226, Tyr456, and Tyr375 were also near LFN density and may help to stabilize 155 the unfolded peptide. The unfolded peptide also interacts with acidic residues Glu465, Asp426, 156 and Glu393 closer to the Φ clamp. Previously, molecular modeling using milestoning noted these 157 residues are important for early translocation events (Ma, Cardenas et al. 2017). Our results are 158 consistent with these residues facilitating translocation of LFN into the pore in an unfolded state. 159 Added asymmetric density in and around the Φ clamp ( Figure 4C) was seen in the final 160 refined map without further processing (e.g., local refinement, map subtraction, sharpening). 161 Compared to the previously published apo PApore cryoEM structure (Jiang, Pentelute et al. 2015), 162 the Φ clamp region in our translocating structure has added density. Specifically, density for each 163 of the Phe427 was smeared in plane with the benzyl ring suggesting rotameric states moving up 164 and down ( Figure 4D). There is also density in the center of the Φ clamp ( Figure 4C). We attribute 165 this density to the unfolded LFN interacting with PApore Φ clamp loop as LFN is translocating 166 through the pore. 167 Density was also observed in the β barrel of the PApore. Focused refinement of the β barrel 168 interior revealed an α helical density with a portion of unfolded peptide above and below the helix 169 ( Figure 5). Notably, the helical density starts in the PApore charge clamp suggesting the 170 deprotonated state favors helix formation (Jas, Childs et al. 2019). Canonical charge clamp 171 residues Asp276, Glu434, and Glu335 are shown in Figure 5 with each of the seven PApore 172 subunits contributing one residue and LFN density translocating through the center of the channel. 173 This helical density in the charge clamp provides strong evidence for initial refolding of LF 174 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. favorable path from the α to Φ clamp involves a series of hydrophobic residues that are amenable 197 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. did not evolve to fold in the PApore channel. Interestingly, these non-native substrates require 220 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint obtain more descrete complexes, the method is being expanded to include immoblization on a 244 column, allowing for low pH pulse chase. E126C was selected as the immobilized residue for its 245 location with respect to the PA binding site. It also stalls the complex early in translocation. 246 Moving the immobilization residue further from the N-terminus would capture mid and late 247 translocation complexes. In the future, TITaNS could be adapted to time-lapse cryoEM which has 248 previously been used to characterize ribosome processivity (Frank 2017). was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint length of these toxins is noticeably shorter (Figure 6B). We predict translocon pores have evolved 266 extended pores to faciliate substrate refolding inside the translocon for effective intoxication. averaging all frames of each movie were used for defocus determination and particle picking. 319 Micrographs obtained by averaging frames 2-23 (corresponding to ~30 electrons per square 320 ångström) were used for two-and three-dimensional image classifications. The best 3,223 321 micrographs were selected for the following in-depth data processing. 322

323
Single particle analysis and structure determination 324 Single particle analysis was performed using cryoSPARC v2.14.2 (Punjani, Rubinstein et 325 al. 2017) (Supplemental Figure 2). A random subset of micrographs was selected for blob 326 particle picking. These particles were subjected to 2D classification in order to obtain a set of five 327 particle templates. Using these templates, 1,772,616 particles were selected from 3,223 328 micrographs. After multiple rounds of 2D classification, the remaining 209,513 'good' particles 329 were used to create a C7 symmetric 3D model. C7 symmetry expansion was performed on these 330 particles followed by 3D variability analysis. Three orthogonal principle modes (i.e. eigenvectors 331 of the 3D covariance) were solved with a filter resolution of 6Å. 124,919 particle from four 332 resulting clusters with potential LFN density were selected for further processing. 3D classification, 333 consisting of two rounds of heterogenous refinements were performed. 3D classes without LFN 334 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint density were discarded leaving two classes. A non-uniform refinement on the per particle motion 335 and CTF corrected particles was performed resulting in a 3.06 Å and 3.24 Å model. Resolution 336 was determined using gold standard Fourier shell correlation with a cut off of 0.143. Resolution 337 and particles distribution are shown in Supplemental Figure 3. 338 Local refinement was performed on the cap of the PA pore to further characterize bound 339 LFN using a mask of LFN (PDB 3KWV) low pass filtered to 30Å. To clearly visualize density in 340 the PApore lumen not attributable to the PApore, a 3.0Å volume of PApore was created using Chimera 341 molmap (Pettersen, Goddard et al. 2004). This volume was subtracted from the cryoEM map. The An initial model using PDB 6PSN for the PApore and PDB 3KWV for LFN was docked 348 into the cryoEM map and corresponding LFN local refinement map, respectively, using Chimera 349 map to model (Pettersen, Goddard et al. 2004). The PApore coarse model was then refined 350 using PHENIX real space refine (Adams, Afonine et al. 2010). Individual atomic model side 351 chains were manually adjusted to fit the density map using Coot (Emsley and Cowtan 2004). 352 This process was repeated iteratively until an optimal model was obtained. Ramachandran plots 353 and MolProbity (Chen, Arendall et al. 2010) were used to assess model quality. Supplementary 354 Table 1 is a summary of cryoEM data collection and processing as well as model building and 355 validation. 356 357 was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020. ; https://doi.org/10.1101/2020.04.20.049601 doi: bioRxiv preprint was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which this version posted April 20, 2020.

414
Fisher, M. T. and S. Naik (2019). Systems and methods for identifying protein stabilizers, Google Patents.