|Home | About | Journals | Submit | Contact Us | Français|
BinAB is a naturally occurring paracrystalline larvicide distributed worldwide to combat the devastating diseases borne by mosquitoes. These crystals are composed of homologous molecules, BinA and BinB, which play distinct roles in the multi-step intoxication process, transforming from harmless, robust crystals, to soluble protoxin heterodimers, to internalized mature toxin, and finally toxic oligomeric pores. The small size of the crystals, 50 unit cells per edge, on average, has impeded structural characterization by conventional means. Here, we report the structure of BinAB solved de novo by serial-femtosecond crystallography at an X-ray free-electron laser (XFEL). The structure reveals tyrosine and carboxylate-mediated contacts acting as pH switches to release soluble protoxin in the alkaline larval midgut. An enormous heterodimeric interface appears responsible for anchoring BinA to receptor-bound BinB for co-internalization. Remarkably, this interface is largely composed of propeptides, suggesting that proteolytic maturation would trigger dissociation of the heterodimer and progression to pore formation.
Mosquitoes continue to be the insects most injurious to human health, spreading devastating diseases such as malaria, filariasis, Dengue fever, West Nile encephalitis [WHO 2012], and more recently, Zika virus. Synthetic chemical insecticides are a cost effective means for reducing mosquito vector populations, but their intensive application results in resistance to them. Most recently, resistance to pyrethroid insecticides threatens the control of malaria. To deal with the problems of resistance and develop more environmentally sound and sustainable approaches to vector control, various biological and genetic-based control strategies are under development. These include the use of microorganisms such as wolbachial endosymbionts that interfere with pathogen development and transmission1, genetically engineered female mosquitoes incapable of flight or pathogen transmission2, and pathogens such as bacteria or fungi engineered for higher efficacy against larvae or adults3,4.
Currently, one of the environmentally safest and most efficient means of controlling mosquitos is the distribution of naturally occurring protein crystals from Bacillus thuringiensis subsp. israelensis (Bti) and Lysinibacillus sphaericus (formerly Bacillus sphaericus) as practiced in the United States, Germany, China, Thailand and Africa5. These proteins are toxic to their targets, but harmless to humans and other animals. The exquisite specificity and efficiency of the L. sphaericus binary toxin, BinAB, presumably arise from its structural complexity which allows it to navigate through a series of barriers en route and recognize only its intended target. Originating as protoxins inside L. sphaericus, BinA and BinB, in a 1:1 ratio, pack into crystals, themselves stored inside spores. The spores are distributed in aquatic environments where mosquitos breed. Larvae ingest the crystals, whereupon they dissolve in the alkaline midgut juices (pH 8 – 11), releasing heterodimers. These are activated by cleavage of four terminal propeptides. BinB recognizes a glycolipid-anchored maltase receptor located at the microvillar surface of midgut cells6 and assists the internalization of BinA, which carries the toxic function and encodes the host range7. Finally, the homologs re-associate into a toxic pore. Recently, the structure of BinB from L. sphaericus was determined8. Here, we elucidate the structure of BinAB crystals, revealing features which endow BinA and BinB with their respective functions, and we suggest how alkalinity and proteolytic activation trigger a series of structural rearrangements that navigate BinAB past barriers to reach its target while maintaining the resilient 1:1 association throughout the life cycle9.
We collected data from the native BinAB complex and three heavy atom derivatives using serial femtosecond crystallography (SFX) methods at the Linear Coherent Light Source (LCLS) Coherent X-ray Imaging (CXI) instrument10, with the aim of phasing by multiple isomorphous replacement with anomalous scattering (MIRAS). We produced nanocrystals of L. sphaericus BinAB (Fig. 1a, b) by recombinant expression in B. thuringiensis cells and jetted them across the pulsed XFEL beam using either a gas dynamic virtual nozzle (pH 7 crystals)11 or electrospin injector (pH 5 and pH 10 crystals)12 (Extended Data Fig. 1a). Initial structure factor amplitudes were obtained with cctbx.xfel13, and then corrected for partiality with cctbx.prime14 (Extended Data Table 1). This procedure was essential to the success of phasing (Fig. 1c). Heavy-atom sites were successfully located for all three derivatives (Extended Data Fig. 1b-d), and their phasing information combined (Extended Data Fig. 1e and Extended Data Fig. 2a) producing a map of sufficient quality that 60% of the BinAB complex could be traced automatically (Extended Data Fig. 2a-c). Subsequent manual building led to a model with Rwork/Rfree of 0.164/0.200 at 2.25 Å resolution. (Extended Data Fig. 2b, c and Extended Data Table 1).
The maps revealed two rod-shaped molecules, BinA and BinB, each approximately 100 Å long and 25-30 Å in diameter (Fig. 2a). They resemble each other closely (1.7 Å RMSD for 329 pairs of α-carbons) (Extended Data Fig. 2d), reflecting the 28% sequence identity (46% similarity) covering nearly their full lengths (Extended Data Fig. 2e). Each subunit comprises two domains: an N-terminal β-trefoil domain (BinA residues 1-155, BinB residues 1-198) and a C-terminal pore-forming domain (PFD) (BinA residues 156-370, BinB residues 199-448).
The most notable structural differences between BinA and BinB are located on the carbohydrate binding modules of the trefoil domains, suggesting that these may contribute to the distinct roles of BinA and BinB in intoxication. All three modules (α, β, γ) of BinA appear structurally capable of binding carbohydrate; however BinB contains four loop insertions relative to BinA (residues 62-70, 111-117, 139-143, and 179-185), which distort the trefoil's pseudo threefold-symmetry (Fig. 2c, Extended Data Figs. 2e and and3a).3a). One insertion, 62-70, completely obstructs the α-module through a disulphide-linked tether, Cys67-Cys161 (Fig. 2c and Extended Data Fig. 3b). These differences suggest a lesser role for BinB in carbohydrate binding, or perhaps an adaptation to diverse carbohydrates. Carbohydrate binding has been linked with toxicity. Numerous mutations in these modules reduce toxicity (Supplementary Discussion; Supplementary Tables 1 and 2) and sugars such as chitobiose are potent antagonists of the BinA toxin15. The outward facing orientation of the modules in the BinAB dimer (Fig. 3a) indicates accessibility to cell surface glycoproteins and glycolipids, which may aid in concentrating BinAB at the cell surface, before receptor binding (Fig. 3b).
Relatively few structural features distinguish BinA from BinB in the PFD, suggesting pore assembly may be heteromeric16. The PFD topology is characteristic of the aerolysin family of pore-forming toxins, comprising a 60 Å long antiparallel ß-sheet, folded into a sandwich at one end (sandwich subdomain), and adherent to a putative membrane-spanning segment (TM subdomain)17 at the other end (sheet subdomain) (Fig. 2a and Extended Data Fig. 4). This structural homology suggests that BinAB may form pores by a mechanism similar to aerolysin18, in which the TM subdomain refolds into a hairpin, oligomerizes into a β-barrel, and inserts into the membrane. However, no experiment has revealed whether these pores would be homo- or hetero-oligomeric. We find that the degree of structural similarity between the PFDs of BinA and BinB (1.4 Å RMSD over 203 α-carbons) (Extended Data Fig. 2d) is even closer than that observed between LukF and Hlg2 (1.5 Å RMSD over 241 α-carbons) which adopt alternate positions around the ring-shaped heterooctameric pore of γ-hemolysin19. A heteromeric assembly analogous to γ-hemolysin might explain why BinA and BinB form pores more efficiently in combination than either component separately16,20.
The crystal packing reveals a unique and exceptionally large heterodimer interface, specifying a resilient 1:1 association between the two components. In this interface, the 100 Å-long BinA and BinB molecules cross each other at a 30° angle, forming an “X” shape (Fig. 3a, b). This interface is so large (burying 1855 Å2 on BinA and 1800 Å2 on BinB) that it accounts for nearly half of all the intermolecular interface area in the crystal. It exceeds by far the threshold value (856 Å2 per monomer) estimated to discriminate between biological and artificial dimers21. It is three times larger than the next largest interface in the crystal, and therefore it is likely to be the only one of the total seven intermolecular interfaces preserved after dissolution of the crystals (Extended Data Figs. 5 and and6,6, and Supplementary Tables 3-9). Its shape complementarity (0.65) is similar to that observed in antibody-antigen interfaces22 and electrostatic complementarity is evident (Extended Data Fig. 7a). The fit and extent of this heterodimer association implies stability, which likely contributes to attaining the 1:1 molar ratio shown to be optimal for receptor binding and toxicity (Supplementary Tables 1 and 2)
Further examination of the heterodimer interface reveals that its size would be severely reduced by proteolytic activation and this loss would appear to threaten access of BinA into the cell. Remarkably, 42% of the heterodimer interface involves propeptide segments, i.e. 1539 Å2 combined interface surface area (Fig. 3c, d, Extended Data Fig. 5c, Extended Data Fig. 6a, c, d, f, and Supplementary Tables 3 and 10). Therefore, removal of these cleavable segments before internalization would dismantle what may be the essential interface tethering BinA to its chaperone, the cell surface receptor-bound BinB. Indeed, deletion of the largest of the four propeptides (BinB residues 396-448) reduces the ability of this truncated mutant to direct the regional binding of BinA to target cells23. However, we note that due to the large surface area it buries in BinAB (3790 Å2), this 53-residue propeptide would be slow to release. Slow release may delay heterodimer dissociation until after internalization, when at the targeted location, it signals transformation into a pore.
Of all the transformations undergone by BinAB over its life cycle, our data bears most directly on the pH-signalled transformation from crystal to soluble dimers. Using crystallography as a means of structure elucidation, we can see in detail the crystal contacts which are the focus of this transformation and monitor their perturbation with elevation of pH from 7 to 10. The most fascinating attribute of this transformation is the seemingly contradictory combination of both stability and sensitivity that evolved to adapt the crystal to different stages of the life cycle. Crystal stability preserves and stores potency in harsh environments before ingestion, yet alkalinity readily dissolves the crystal in the larval midgut (pH 8-11) after ingestion, releasing BinAB to access cell surface receptors and activating proteases. The mechanism by which alkalinity lowers the high barrier to crystal dissolution is evident at three levels: amino acid composition, local, and global structural changes.
Alkalinity may facilitate crystal dissolution through the concerted deprotonation of BinAB's 49 tyrosine hydroxyl groups, occurring around pH 10 (pKaTyr = 10.1) (Extended Data Fig. 7b). The frequency of tyrosine residues (6%) is almost twice greater than the average for proteins (3.5%)24. We calculate that pH elevation from 7 to 10 would change the net charge on BinAB from -13.9 to -73.5 e, thereby destabilizing the crystals through negative electrostatic repulsion (Extended Data Fig. 7c, d). An analogous mechanism was proposed for the dissolution of the ultra-stable viral spindle crystals, which also contain an unusually high frequency of tyrosines (8.6%), some located strategically at crystal contacts25.
Alkalinity observably perturbs four regions in BinAB crystals, identified by peaks in a difference Fourier map (Fo-Fo) computed between datasets collected at pH 7 and pH 10 (Fig. 4, Extended Data Fig. 8 and Supplementary Tables 11 and 12). These regions involve contacts between molecules or subdomains. Two of these perturbations increase accessibility of propeptide segments to proteases. The N-terminal propeptide of BinA (residues 1-10) unravels from a helix to an extended conformation, following the loss of a hydrogen bond between Gly15(O) and Tyr213(OH) of BinA (Fig. 4c and Extended Data Fig. 9). Similarly, a pair of hydrogen bonds is broken in the BinAB dimer interface between the C-terminal carboxylate of the BinB propeptide (Gln448) and BinA Asp22 side-chain carboxylate (Fig. 4d). The third region is a lattice contact outside the dimer interface, between BinA Tyr134(OH) and BinB Glu59(OE2) which breaks at elevated pH (Fig. 4e and Supplementary Table 7). The fourth region involves loss of a hydrogen-bond between BinA Asp342(OD1) and Glu240(OE1). Located at the junction between the TM and sandwich subdomains, this break might be an early step toward pore formation (Fig. 4f and Extended Data Fig. 8). In each of these four regions, deprotonation is presumably the cause of hydrogen bond disruption, involving either a tyrosine hydroxyl or a carboxylate group paired with an obligate hydrogen bond acceptor (Fig. 4 and Supplementary Discussion).
Alkalinity also induces global hinge motions, potentially straining the crystal lattice and thereby contributing to its dissolution. The trefoils of BinA and BinB move about 0.5 Å closer to their respective PFDs (Extended Data Fig. 9 and Supplementary Discussion). These motions might also foreshadow a rearrangement of the dimer interface, to expose the receptor binding motif, which is otherwise buried in the dimer interface (Fig. 3a, b).
To validate our model for alkaline-induced crystal dissolution and address concerns that the structural changes we attribute to pH elevation may have instead originated from unintended radiation damage (Supplementary Discussion), we mutated one of the four pH sensitive switches that we identified, BinA Asp22. Recall that upon pH elevation, withdrawal of donor hydrogen atoms breaks the bifurcated hydrogen bond between carboxylates of Asp22 and the C-terminus of BinB (Fig. 4d). Mutation from Asp to Asn should reduce toxicity by decreasing sensitivity of this interaction to alkaline-induced rupture and thereby delay release of the C-terminus for proteolytic activation. Affirmatively, this mutation did not change the appearance of the crystals; however, it reduced crystal solubility by 30% in the period 30-90 minutes after pH elevation (Extended Data Fig. 7e) and increased LC50 and LC95 by 11.6-fold and 24-fold, respectively (Supplementary Table 13). The striking effect of this single conservative substitution in a 93 kDa complex provides a convincing validation of our model.
Our structure of BinAB illuminates several important molecular events in the life cycle of the toxin. These include discovery of (1) four pH sensitive switches that facilitate crystal dissolution in the larval midgut, (2) a large heterodimer interface explaining how heterodimers persist after dissolution, (3) three competent carbohydrate binding modules in BinA that may assist in directing heterodimers to the cell surface, and (4) the potential to disrupt the heterodimer interface by proteolytic activation, thereby signalling remodelling. Our success in de novo MIRAS phasing of a crystallographic asymmetric unit nearly three times larger than any previously phased de novo by SFX26–28, and from crystals approximately 50 unit cells per edge, suggests this approach could be applied again in other cases where crystal size is limiting.
L. sphaericus BinAB crystals were grown in a Bacillus thuringiensis strain engineered to improve crystal expression, which allowed growth of much larger crystals than occur naturally in L. sphaericus. There is no evidence that growth of the L. sphaericus BinAB crystals in B. thuringiensis in anyway affected the structure of BinAB; the toxicity per unit mass of the Bin crystals produced in B. thuringiensis is the same as that produced in L. sphaericus. Assuming that the cytosolic composition of L. sphaericus and B. thuringiensis cells do not strongly differ, crystal packing interactions are expected to persist and crystals formed by BinAB in the two types of cells should be identical in terms of space group and unit cell constants. Five hundred millilitres of glucose-yeast-salts (GYS) growth medium [0.1% glucose, 0.2% yeast extract, 0.05% K2HPO4, 0.2% (NH4)2SO4, 0.002% MgSO4, 0.005% MnSO4, and 0.008% CaCl2] supplemented with 25 μg/mL erythromycin was sterilized in a 2 L baffled flask and inoculated with spores from a lyophilized 5-day lysate of B. thuringiensis subsp. israelensis strain 4Q7 containing plasmid pPHSP-1 (Bti4Q7/pPHSP-1), which encodes BinA and BinB from L. sphaericus strain 2362 (Park et al., 1998)31. Of all known L. sphaericus strains, this strain produces the most potent toxin. Cultures were harvested after growth for 5 days at 30° C with shaking at 250 rpm, monitored by phase contrast light microscopy until sporulation and cell lysis were observed.
We attempted to obtain crystals of BinAB toxin incorporating the unnatural amino acid 3-iodo-tyrosine (3iTyr) at tyrosine positions in BinA and BinB. By exposing early sporulating cells to excess 3iTyr, phenylalanine, and tryptophan, we aimed to block aromatic amino acid biosynthesis and force the utilization of the iodine-labeled amino acid. Bti4Q7/pPHSP-1 cells were grown to late log phase and exposed to 3iTyr, phenylalanine, and tryptophan at concentrations of 100 mM each. Two further additions of the three amino acids, each increasing the concentration by 65 mM, were made at 24-hours intervals. Cultures were harvested after 4 days. However, SDS-PAGE analysis showed partial or no incorporation of 3iTyr in the BinAB crystals. Furthermore, there is no evidence of incorporation in the final structure of BinAB. Following culture lysis, spores, crystals, cells, and cell debris were pelleted by centrifugation at 6,000 × g for 30 min, then resuspended in 50 mL sterile water. The concentrated lysate was sonicated for 3 min on ice [1 s on, 1 s off (6 min elapsed time); 60% intensity] to lyse remaining cells. The sonicated lysate was pelleted at 6,000 × g for 30 min at 4° C. All subsequent steps were performed at 4° C. The pellet was washed in 50 mL cold water to remove soluble material and some of the spores, then re-pelleted, and resuspended in 15 mL water. Crystals were isolated from the suspension by sucrose gradient centrifugation as described in Sawaya et al., 2014 32. Before injecting the sample into the beam, large particles were removed by filtering the material with a 10 μm stainless steel frit.
The recombinant plasmid, pPHSP-1, containing BinA and BinB open reading frames under control of the pSTAB promoter 33 was used as template to amplify two DNA fragments each containing the D22N mutation in the BinA ORF, using Q5 polymerase (New England Biolabs, Beverly, MA). The first fragment began at the MluI site near the 3′ end of the BinB ORF and went through the D22 region of the BinA ORF, and contained the BinA D22N mutation; this fragment was amplified using primer pairs . The second fragment began at the D22 region and continued to the end of the Bin operon terminator, and was amplified with primer pair . Both fragments had approximately 15 bp overlapping homology regions with the vector on one end and the second fragment on the other. The two fragments were assembled along with pPHSP-1 linearized with MluI and PstI by recombineering using the Choo-Choo Cloning Kit (MCLAB, S. San Francisco, CA). The entire Bin operon of the resulting clone was sequenced to confirm the presence of the D22N mutation. Full-length fragments of the entire Bin operon of either wild-type pPHSP-1 or pPHSP-1-D22N, including approximately 15 bp vector homology regions on both ends, were amplified using primers , and assembled by recombineering in pBU434 linearized with KpnI and SalI. The resulting plasmids, pBUSP-1 and pBUSP-1 D22N, which also contained the 20-kDa helper protein gene as previously described35,36 were sequenced and used to transform Bacillus thuringiensis strain 4Q7 by electroporation33.
Lyophilized cultures containing spores and parasporal bodies of the L. sphaericus BinAB and BinAD22NB strains were resuspended in ddH2O. Suspensions were diluted to 6 to 7 different concentrations, ranging from 0.5 ng/ml to 1 μg/ml, in 6 oz cups in a final volume of 100 ml. Bioassays were replicated three times using 30 fourth-instars of S-Lab (Bin-sensitive) strains of Cx. quinquefasciatus. After 24 h of exposure at 28°C, dead larvae were counted and the 50% and 95% lethal concentrations, respectively, LC50 and LC95, were calculated by Probit analysis (POLO-PC; LeOra Software, Berkeley, CA).
The strains producing BinAB (4Q7/pPHSP-1) and BinAB(D22N) (4Q7/pBU-D22N) were grown in Nutrient Broth + Glucose (NBG) supplemented with, respectively, erythromycin (25 mg/ml) and tetracycline (3 mg/ml), for 5 days until > 95% of cells had sporulated and lysed. To isolate crystalline inclusions, spore/crystal mixtures collected from 50 ml cultures were resuspended in 15 ml ddH2O and sonicated twice at 50% duty cycle for 15 s using the Ultrasonic Homogenizer 4710 (Cole-Parmer Instrument Co.). Five-milliliter samples were loaded onto a sucrose gradient cushion (30-65% w/v), which was then centrifuged at 20,000 g for 45 min at 20°C in a Beckman L7-55 ultracentrifuge using the SW28 rotor. Bands containing inclusions were collected and washed three times in ddH2O, followed by centrifugation at 6,500 g for 15 min at 4°C after each wash, then lyophilized for storage at -20 °C until use. For solubilization assays, 200 μg or 800 μg of each crystal preparation was resuspended in, respectively, 100 μl of 10 mM Tris-Cl (pH 7; baseline control) or 400 μl of 10 mM Tris-Cl (pH 10; solubilization buffer). Assays were performed in triplicate. For each sample, three-25 μl aliquots were collected and spun at 16,300 × g for 2 min to pellet undissolved crystals, and supernatants were transferred to fresh microfuge tubes containing 75 μl of 10 mM Tris-Cl (pH 10). Absorbance at A280 nm was recorded at various time intervals.
We produced the mercury derivative by soaking the BinAB crystals in 1 mM p-chloromercuribenzene sulfonate (PCMBS) for five months. Specifically, we added 200 μL of 0.1M PCMBS to 15 mL of crystal slurry at room temperature, and then we passed the mixture through a stainless steel frit with 10 μm pore size. We estimated the concentration of crystals in the slurry by pelleting the crystals in a centrifuge, then noting the crystals occupied approximately 25 μL of the 1 mL sample volume. The five months incubation was not intentional. We intended to use the derivative on the day it was prepared, which was during a previous data collection trip; however, this sample immediately clogged the gas dynamic virtual nozzle (GDVN) injector upon its first use. So, virtually none of this large volume of sample passed through the injector. In retrospect, we think the clogging was the result of mercuric iodide precipitation induced by residual iodide ion in the delivery lines; a derivative containing 0.5 M potassium iodide had indeed been run just previously to injecting the PCMBS derivative. To reduce the chance of this same PCMBS sample clogging again, we removed the non-covalently bound PCMBS by washing the crystals twice in deionized water a week before the LCLS experiment. The crystals used for derivatization in this experiment have the same origin as those we used to produce our effectively “native” data set. They were prepared under conditions intended to substitute 3iTyr in place of tyrosine, but the substitution was not efficient and so were effectively native before derivatizing with PCMBS.
We produced the gadolinium derivative by soaking the BinAB crystals in 5 mM GdCl3. More specifically, we added 50 μL of 0.1M GdCl3 to 1 mL of crystal slurry at room temperature one day before the diffraction experiment. We estimated the concentration of crystals in the slurry by pelleting the crystals in a centrifuge, then noting the crystals occupied approximately 25 μL of the 1 mL sample volume. In order to conserve the amount of crystals used in screening for a heavy atom derivative, we recycled the crystal sample that had been used in previous diffraction experiments performed in helium vapour atmosphere (that is, not vacuum). For example, the crystals used in this experiment had been soaking for five months in KI and CsCl since the time they cycled through a GDVN nozzle for data collection in a previous XFEL experiment. The sample could be recycled without fear that the previous experiment damaged the crystals since over 99.9% of the crystals never intercepted the XFEL beam during the previous XFEL experiment. We washed the KI and CsCl from the crystals by pelleting the crystals and replacing the supernatant with water three times before adding GdCl3.
We produced an iodinated tyrosine derivative by adapting the “vaporizing iodine labelling” (VIL) methods described by Miyatake et al. 37. The method uses gaseous iodine vapour to derivatize the ortho positions of accessible tyrosine residues. In our case, rather than derivatizing a few crystals in a standard 1-5 μL drop, we derivatized 1 mL of crystal slurry. To improve accessibility of the iodine vapour to this relatively larger volume of crystals, we distributed the 1 mL of BinAB crystal slurry evenly over nine wells of a glass depression dish (i.e. 111 μL per well). Four 10 μL aliquots of a KI/I2 mixture (0.67 M / 0.47 M concentration) were placed on the dish, adjacent, but not contacting the crystal slurry in the wells. The plate was sealed inside a sandwich box with 5 mL of water to act as a reservoir. The three components (drops of crystal slurry, the iodine, and the water reservoir) were separated in space but in vapour contact. The sandwich plate was sealed with tape at room temperature. The drops of crystal slurry turned yellow after 3.5 h. The crystals were harvested after 21 h. We estimated the concentration of crystals in the slurry by pelleting the crystals in a centrifuge, then noting the crystals occupied approximately 25 μL of the 1 mL sample volume. The crystals used in this experiment had the same history of origin as those used for the gadolinium derivative. They had been soaking for five months in KI and CsCl for a previous XFEL experiment, then jetted across the beam in helium vapour atmosphere, and washed three times to remove the non-covalently bound heavy atoms before derivatizing with iodine.
BinAB nanocrystals of the native and derivatives (average volume is 0.75*0.5*0.25 μm3) were injected using a gas-focused dynamic virtual nozzle (GDVN) liquid microinjection system38 at the Coherent X-ray Imaging (CXI) instrument of LCLS10. Data were collected on two occasions, one for native data and one for heavy atom data, for a combined time of 245.9 minutes, with the detector framing at 120 Hz.
Crystals of native BinAB at pH 7 were pelleted by centrifugation at 10,000 × g for five minutes, and then resuspended in pH 5 buffer (0.1 M N-cyclohexyl-3-aminopropanesulfonate (CAPS) pH 10.0, 30% glycerol (v/v), 10% PEG 2000 MME, and 0.1 M NaCl). This preparation was originally intended to be at pH 10; however, it was later discovered that one of the components, polyethylene glycol monomethyl ether 2000 (PEG 2000 MME) was acidic (probably due to age), overpowering the pH 10 CAPS buffer, and bringing the final pH to 5. The pH of the crystal slurry was indicated by colour change on pH paper. The crystals were soaked at pH 5 for 10 minutes before the diffraction experiment.
Crystals of native BinAB at pH 7 were pelleted by centrifugation at 10,000 × g for five minutes, and then resuspended in pH 10 buffer (0.11 M CAPS pH 10.0, 33% glycerol (v/v), 11% methylpentanediol, and 0.11 M NaCl). The pH of the crystal suspension was verified by colour change on pH paper. The crystals were soaked at pH 10.0 for 5 hours before the diffraction experiment.
Crystals of native BinAB in pH 5 and pH 10 buffers were delivered in the microfluidic electrokinetic sample holder (MESH) method, described more fully in Sierra et al. 201212. The MESH injection system was modified (Extended Data Fig. 1a) to interface more readily with the CXI nozzle rods that mount interchangeably into the liquid sample delivery system. The crystal slurries were prepared as described above. A continuous 1.5 m long fused silica, polyamide coated capillary of 100 μm inner diameter and 360 μm outer diameter was used to deliver the sample into the CXI vacuum chamber. Approximately 800 μl of sample slurry with glycerol additive was placed in a microcentrifuge tube and placed in a small pressurized sample holder. The capillary and platinum wire were fed through the pressure cell and immersed in the slurry. A small backing pressure of 5 psig nitrogen gas was applied to aid the injection. The voltage was applied by a Stanford Research Systems PS350 (Sunnyvale, CA) high voltage source and was held between 4300-4500 V (currents < 1 μA) while the counter electrode was grounded. The flow rate was not directly measured, but we estimate that the sample consumption was approximately 2 μl/min, as judged from crude measurements of left-over sample volume. The pH 10 structure presented here was collected in less than one hour of continuous beamtime and consumed less than a millilitre of sample. The sample injection compared favourably to prior attempts at data collection by giving comparable resolution with only hundreds of microliters of sample consumed.
The sample chamber was at room temperature, under vacuum. Native and derivatives pH 7 datasets were collected with a XFEL beam focused to a 1 μm FWHM spot, and characterized by a wavelength of 1.21 Å (native crystals; 1.9 ×1011 photons per pulse) or 1.41 Å (heavy atom soaked crystals; 7.2 × 1011 photons per pulse). We chose a single wavelength for all three derivatives as a compromise of beam stability, time efficiency, and significant f″ for the elements used (Extended Data Fig. 1b). The pH 5 and pH 10 dataset were collected with a XFEL beam focused to a 100 nm FWHM spot (6.4×1011 photons per pulse), and characterized by a wavelength of 1.46 Å. Thus, the pH 5 and pH 10 datasets were collected with a ~500 fold higher dose than the pH 7 dataset (Extended Data Table 1).
Images were reduced using cctbx.xfel13,39. No thresholding was applied to filter out blank shots; we attempted to index every frame. Further, for the heavy atom soaked data, if a frame indexed successfully, we removed the primary lattice from the list of candidate reflections and attempted to index again, searching for a secondary lattice. 8% of the final number of indexed images from the heavy atom data was from secondary lattices. In total, 24.5% of all recorded frames from the heavy atom data contained indexable patterns (13.2% for the native data), for a final total of 398,971 indexed patterns across the three derivatives and the native data set collected at pH 7. The dataset collected at pH 5 and pH 10 respectively consisted of 17,099 and 27,792 indexed patterns. First we merged and post-refined our data without negative intensities included using cctbx.prime14, which determines relative scale factors for each image, along with partiality estimates for each structure factor measurement. Post-refinement in this manner was critical for multiple isomorphous replacement with anomalous scattering (MIRAS) phasing as described below. Figure 2 illustrates, for the three derivatives and their combination thereof, the extent to which cctbx.prime improved the quality of the intensities, leading to improved phases, and thereby rendering the path of the peptidic chain visible in electron density map, where only broken density was visible before. Resolution cut-offs were determined based on completeness (> 90%), redundancy (> 4) and CC1/2 (> 0.14) ; in all datasets, <I/sigI> in the highest resolution shells are greater than 6. For all data sets, postrefinement corrected for the inherent partiality of SFX data and thus improved the σ-weighted (i.e. variance weighted) average estimate of structure factor amplitudes. Handling of negative intensities was then implemented in cctbx.xfel, allowing the production of meliorated post-refined datasets that are not plagued apparent twinning. We used the same criteria for the resolution cut-off, based on completeness on (> 90%), redundancy (> 4) and CC1/2 (> 0.14); in all datasets, the resulting <I/sigI> in the highest resolution shells are greater than 0.5. While inclusion of negative intensities dramatically affects <I/sigI>, it results in a stabilization of the refinement for the three native structures (pH 5, pH 7 and pH 10), requiring to impose a lower weight on geometry in phenix.refine to obtain a structure with a plateauing Rfree. Inclusion of negative intensities in the datasets also resulted in a decrease of noise levels in q-weighted structure factor amplitude Fourier difference maps (see below). At the atomic level, the only noticeable change between structures refined with and without negative intensities included is an increase in the B-factors, reflecting the increase of the Wilson-B – and possibly giving a more realistic description of the BinAB structures at room-temperature.
Heavy-atom sites were successfully located for all three derivatives using ShelxD40 (Extended Data Fig. 1c, d). The individual derivatives demonstrated limited phasing power, and the maps they produced were not interpretable (Extended Data Figs. 1d and and2a).2a). Fortunately, the three derivatives were isomorphous (Extended Data Fig. 1e), and so could be combined to obtain a map of sufficient quality to automatically trace a partial BinAB model using the Phenix suite41 (Fig. 1c and Extended Data Fig. 2a, b), and extend using Arp/Warp42. The remaining residues were built manually, leading to a model with Rwork/Rfree of 0.164/0.200 at a resolution of 2.25 Å (Extended Data Table 1). After each cycle of model rebuilding, reciprocal space refinement (including refinement of coordinates, atomic displacement parameters, TLS parameters and occupancy) was carried out using phenix.refine. The final MIRAS model features residues 1-357 of BinA and 28-448 of BinB. Figure 2 illustrates the benefit of post-refinement to the accuracy of experimental phasing. For this figure, all SIRAS and MIRAS maps were calculated using MLPhaRe, followed by ten cycles of density modification with DM43 enforcing 58% solvent content, and displayed at 2.8 Å resolution. The CC quantifies the improvement gained from post-refinement. It reports on the whole asymmetric unit, not just the region illustrated in the figure.
We attempted to determine heavy-atom substructures for each of the three derivatives by supplying both isomorphous and anomalous difference signals (SIRAS), or only anomalous difference signal (SAD) to the program ShelxD 40 in super-sharpening mode (‘PSMF=-4′). Success was evidenced by the clear separation between populations of false and true solutions ranked by goodness-of-fit. Super-sharpening enhanced this separation. We found that SAD substructure determination was successful for the PCBMS (mercury) and Gd (gadolinium) derivatives, but not for the VIL (iodide) derivative (Extended Data Fig. 1d). SIRAS substructure determination were however successful for all three derivatives (Extended Data Fig. 1d). SIRAS phases of the mercury and iodide derivatives are of better quality than their SAD counterparts, as evidenced by a sharper separation between populations of false and true solutions ranked by goodness-of-fit (Extended Data Fig. 1d). In contrast, the SAD phases of the gadolinium derivative are better quality than its SIRAS phases.
Yet, for each individual derivative, the correct choice of hand remained ambiguous after initial density modification of SIRAS phased maps and chain tracing with ShelxE, as judged by a lack of separation in map quality statistics between the two hands, such as contrast, connectivity, and correlation coefficient for auto-traced chains (Extended Data Fig. 2a). We sought to maximize the distinction between hands by systematically computing a series of maps at different estimated solvent contents, from 55 to 73% and comparing the map quality statistics from opposing hands (Extended Data Fig. 2a). The true solvent content of BinAB crystals is 59.7%. We found that for each value of solvent content tested between 55% and 65%, a consistent choice of hand was indicated by the correlation coefficient of the automatically traced chain. However, at higher solvent content values tested, the opposite hand was indicated. Correlation coefficient of traces (CCs) were below 10% in all cases, and the differences between the CCs of the two hands in the 0.5-2.5 % range. So, the distinction between hands appeared to remain ambiguous at this point. For each derivative, we chose the hand having the most consistently (repeatedly) better correlation coefficient over the range of solvent content values tested. The refined heavy atom sites were input to Phenix.autosol from the Phenix 41 software suite for phase refinement, density modification and initial model building.
Attempts to solve the structure based on any of the single derivative (including single isomorphous replacement, single anomalous dispersion and single isomorphous replacement with anomalous scattering) were all unsuccessful. The combination of phase information from the three derivatives nonetheless produced usable maps, allowing us to obtain an experimental BinAB model. Our previous inability to distinguish the correct hand for the heavy atom substructure for the individual derivatives was overcome in Phenix through use of cross difference Fourier techniques, an option that becomes available with multiple derivatives (Extended Data Fig. 2a). Briefly, phenix.autosol (using Solve for phase combination) 44 was used to test various phase combinations schemes and to refine phases. While quality maps were obtained by combining anomalous and isomorphous signals from all three derivatives (MIRAS), the best map was obtained combining the anomalous signal of the gadolinium derivative with the combined anomalous and isomorphous differences of the mercury and iodide derivatives. Initial figures of merits (FOMs) were 0.12, 0.14 and 0.19 for the PCMBS, Gd and VIL substructures, respectively. The FOM of the best phase-combined map was 0.19 before density modification in Resolve 44, and 0.26 after density modification. The phase-and-build routine of Phenix was able to trace 296 residues from this map, resulting in a model displaying Rfree and Rfactor of 0.41 and 0.39, respectively (Extended Data Fig. 2b). Of importance, we had to optimize the parameters of phenix.autosol to obtain this model; in particular, we specified that the data were weak, that we needed to perform a thorough search, and that we needed to use “extreme-dm.”. The FOM of the density-modified map obtained after running this script was 0.72. From this map, phenix.autobuild was in 30 cycles able to reconstruct 501 residues, of which 183 were placed, corresponding to Rfree and Rfactor of 0.35 and 0.32, respectively.
This model and the corresponding phases were then input to a first round of Arp/Warp 42, fitting 62 additional residues into the BinA and BinB models (518 residues, 245 placed) (Extended Data Fig. 2b). This experimental yet incomplete BinAB model displayed Rfree and Rfactor of 0.32 and 0.29. The MIRAS model then underwent another cycle of model building in phenix.autobuild and then a second round of Arp/Warp, resulting in a model with 630 residues, of which 450 were putatively in sequence. This automatically fitted model was however unsatisfactory in terms Rfree, Rfactor and goodness of fit of the placed residues in the experimental map. Manual rebuilding was necessary, in order to correct wrongly assigned residues and to fit the complete BinAB sequence into the electron density maps. In one round of manual fitting, an additional 92 residues were built, leading to a total of 722 residues (299 and 423 residues in BinA and BinB, respectively) (Fig. 1c and Extended Data Fig. 2c). A cycle of reciprocal space refinement and yet another cycle of manual rebuilding produced a map which allowed fitting the 59 missing residues of BinA. The final model displays Rfree and Rfactor of 0.200 and 0.164, respectively (Extended Data Table 1), with 96.5% of residues in favoured region of the Ramachandran plot (0.5% Ramachandran outliers, 1.2% rotamer outliers). Of note, we could also phase and build the structure using MIRAS phases of all derivatives, i.e. without excluding the isomorphous differences from Gd, but the initial model turned out to be less complete, further highlighting the higher quality of the SAD phases of the gadolinium derivative, as compared to SIRAS (Extended Data Fig. 2b).
Molecular replacement for BinAB (strain 2362) was performed with Phaser 45 using as a starting model the activated BinB structure 8 (PDB ID: 3WA1) from L. sphaericus strain 2297. BinB from the two strains differ in only five amino acids (A104S, R267K, L314Y, F317L, L389M). This search model became available only after we had begun our search for heavy atom derivatives. BinB shares only 28% identity with BinA. Two copies of BinB were found (corresponding to a solvent content of 68 %), one of which indeed corresponded to BinB, but the other to BinA. Corresponding residues were mutated and fit manually into the electron density maps using Coot 46. After each cycle of model rebuilding, reciprocal space refinement (including refinement of coordinates and atomic displacement parameters) was carried out using phenix.refine 41. The last cycle of refinement was performed using Buster 47 and included TLS refinement. Rwork and Rfree of the final model are 15.8 and 20.3%, respectively, at a resolution of 2.25 Å. Note that residues 1-11 of BinA and residues 1-34 of BinB were disordered in the electron density map and therefore not included in the model. This model is nearly identical to the independently built and refined MIRAS model (0.102 Å RMSD over 773 aligned α-carbons; see Supplementary Table 14). The BinAB model obtained by molecular replacement was never used in building the BinAB model to the experimental map. In fact, the two models were built independently by different authors. The similarity between the independently obtained models provided additional assurance of model accuracy.
We obtained experimental insights into pH induced conformational changes by calculating structure factor amplitude Fourier difference maps (Fo-Fo) between the pH 5, pH 7 and pH 10 datasets. To improve the estimate of structure factor amplitude differences, Fo-Fo maps were q-weighted as described48 and produced using a CNS49 custom-written script50. Application of the Q-weighting scheme to the diffraction data sets was essential to eliminate noise and amplify the difference signal. Figure 4 shows the FopH10-FopH7 map (Riso=0.28) phased with the pH 7 model (pH7) and highlights 4 four regions highly sensitive to pH elevation (Fig. 4c-f). Extended Data Figure 8 shows, for each of these regions (Extended Data Fig. 8a-d) and for the two BinAB disulfides (Extended Data Fig. 8e-f), the six possible FopHj-FopHi, pHi maps calculated from the pH 5, pH 7 and pH 10 datasets and structures. Supplementary Tables 11 and 12 respectively list all BinA and BinB residues that feature peaks higher than ± 3.5 σ and larger than 2 voxels in the FopH10-FopH7, pH7 map. Sequence-wise integrations of the FopH10-FopH7, pH7 and FopH5-FopH7, pH7 map around BinA and BinB are given in Extended Data Figs. 8g and 8h, respectively.
We phased the pH 5 and pH 10 dataset by molecular replacement with Phaser45 using as a starting model the refined MIRAS (pH 7) structure. Conformational changes with respect to the latter were modelled manually. Reciprocal space refinement was performed using phenix.refine and included positional, B-factor, TLS and occupancy refinements. Rwork and Rfree of the final pH 5 model are 0.211 and 0.262, respectively, at a resolution of 2.5 Å, with 97.6% of residues in favoured region of the Ramachandran plot (0.4% Ramachandran outliers, 1.5% rotamer outliers). Rwork and Rfree of the final pH10 model are 0.165 and 0.211, respectively, at a resolution of 2.4 Å, with 98.9% of residues in favoured region of the Ramachandran plot (0.0% Ramachandran outliers, 0.0% rotamer outliers). Difference distance matrices (DDMs) are presented in Extended Data Fig. 9. Fo-Fo maps integration and DDM calculations were performed using custom-written scripts.
|Native, pH 7 5F0Ya||PCMBSb||Gdc||VILd||Native, pH 10 5F0Ze||Native, pH 5 5G37f|
|a, b, c (Å)||86.9, 97.4, 128.3||87.1, 97.8, 128.4||87.0, 97.5, 128.1||86.7, 97.3, 127.7||86.7, 97.3, 127.7||86.8, 97.0, 127.0|
|α, β, γ (°)||90, 90, 90||90, 90, 90||90, 90, 90||90, 90, 90||90, 90, 90||90, 90, 90|
|X-ray beam focus (μm)||1.3||1.3||1.3||1.3||0.25||0.25|
|Photons/pulse (× 1011)||1.9||7.2||7.2||7.2||6.4||6.4|
|Pulse duration (fs)||31.9||45.1||45.1||45.1||40.8||40.8|
|Absorbed dose (GGy)g||0.03||0.15||0.15||0.15||14.4||14.4|
|Number of collected frames||312,659||761,052||431,986||264,770||371,386||493,979|
|Number of indexed patterns||41,206||184,091||131,137||42,537||27,792||17099|
|Number of indexed images accepted by cctbx.prime||40,794||170,616||123,484||39,789||26,022||16224|
|Resolution (Å)h||43.5-2.25 (2.29-2.25)||43.5-2.40 (2.44-2.40)||43.5-2.35 (2.39-2.35)||43.5-2.60 (2.64-2.60)||43.5-2.37 (2.44-2.37)||43.5-2.50 (2.54-2.50)|
|Number of observationsi||12,854,588||53,361,546||39,314,393||11,249,082||12,069,895||5,495,157|
|I/σI||2.4 (0.5)||2.8 (0.7)||3.1 (0.6)||3.7 (0.9)||2.7 (0.9)||6.34 (2.46)|
|CCl/2||96.8 (26.2)||98.6 (16.2)||98.4 (3.5)||96.3 (17.5)||97.3 (34.1)||86.8 (28.0)|
|Completeness (%)||99.7 (99.7)||99.8 (96.9)||99.8 (97.0)||99.4 (96.2)||99.9 (100.0)||99.9 (89.4)|
|Multiplicity||65.9 (4.8)||372 (4.3)||260.1 (4.4)||100.4 (4.7)||88.8 (19.2)||76.62 (8.61)|
|Refinement target function||Phased maximum-likelihood||Maximum-likelihood||Maximum-likelihood|
|Resolution (Å)||2.25 (2.30 – 2.25)||2.40 (2.46 – 2.40)||2.50 (2.56 – 2.50)|
|Number of reflections||52379 (3523)||42817 (2923)||37785(2557)|
|Rwork / Rfreeg||0.164(0.288)/ 0.200 (0.327)||0.165 (0.264)/ 0.211(0.311)||0.211(0.357)/ 0.262 (0.427)|
|Number of atoms|
|Protein (BinA/BinB)||47.5 / 40.7||52.8 / 47.3||39.2/36.8|
|Bond lengths (Å)||0.006||0.007||0.002|
|Bond angles (°)||1.359||0.8||0.5|
We acknowledge the help of the following people during data collection: Sangho Lee, Jake Koralek, Robert Shoeman, Sabine Botha, Bruce Doak, Oliver Zeldin. We thank Anne Volveda for advice regarding sequence-wise Fourier difference map integration, Jonathan Brooks-Bartlett and Elspeth Garman for help with dose calculations, and Martin Weik for stimulating discussions and continuing support. We thank the HCIA program of HHMI, the W.M. Keck Foundation (Grant 2843398), NIH (Grant AG-029430), National Science Foundation (Grant MCB 0958111) and DOE (DE-FC02-02ER63421) (to D.S.E.), the France Alzheimer Foundation (FA-AAP-2013-65-101349) and the Agence Nationale de la Recherche (ANR-12-BS07-0008-03) (to J.-P.C.), NIH Grants GM095887 and GM102520 for data-processing methods (to N.K.S.), and NIH Grant AI45817 (to B.A.F.). Support by the CNRS (PEPS-SASLELX-2013, PEPS-SASLELX-2014) funded travel to LCLS. Use of the Linac Coherent Light Source (LCLS), SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy, Office of Science, and Office of Basic Energy Sciences under Contract no. DE-AC02-76SF00515.The CXI instrument was funded by the Linac Coherent Light Source Ultrafast Science Instruments project funded by the DOE Office of Basic Energy Sciences. Parts of the sample injector used at LCLS for this research was funded by the National Institutes of Health, P41GM103393, formerly P41RR001209.
Author Contributions: J.-P.C., M.R.S., M.G., J.A.R., D.C., B.A.F. and D.S.E designed and coordinated the project. M.G. carried out the in vivo production of BinAB nano-crystals. H.W.P., D.K.B., and B.A.F., engineered bacterial strain to produce large BinAB crystals. R.L.R and B.A.F. performed toxicity assays. R.H.H. designed the mutagenesis protocol. M.R.S. performed heavy atom derivatizations. J.-P.C., M.R.S., J.A.R., D.C., A.S.B., T.M.-C., S.B., G.J.W., M.M., D.P.D., R.G.S., H.L., J.E.K., M.S.H., N.C., M.U. and N.K.S. acquired and processed data. J.-P.C. and M.R.S, carried out the MIRAS and MR phasing, and built and refined the atomic models. R.G.S. and H.L. developed the MESH-on-a-stick injector. J.-P.C., M.R.S., and D.S.E prepared the manuscript with input from M.G., J.A.R., D.C., A.S.B., T.M.-C., R.G.S., M.S.H., A.T.B., B.A.F., and N.K.S.
Atomic coordinates and structure factors have been deposited in the Protein Data Bank under accession codes 5FOY (pH 7 structure), 5FOZ (pH 10 structure) and 5G37 (pH 5 structure).
The authors declare no competing financial interests.
Readers are welcome to comment on the online version of the paper.