Solution structural studies of pre-amyloid oligomer states of the biofilm protein Aap

Abstract

Staphylococcus epidermidis is a commensal bacterium on human skin that is also the leading cause of medical device-related infections. The accumulation-associated protein (Aap) from S. epidermidis is a critical factor for infection via its ability to mediate biofilm formation. The B-repeat superdomain of Aap is composed of 5 to 17 Zn²⁺-binding B-repeats, which undergo rapid, reversible assembly to form dimer and tetramer species. The tetramer can then undergo a conformational change and nucleate highly stable functional amyloid fibrils. In this study, multiple techniques including analytical ultracentrifugation (AUC) and small-angle X-ray scattering (SAXS) are used to probe a panel of B-repeat mutant constructs that assemble to distinct oligomeric states to define the structural characteristics of B-repeat dimer and tetramer species. The B-repeat region from Aap forms an extremely elongated conformation that presents several challenges for standard SAXS analyses. Specialized approaches, such as cross-sectional analyses, allowed for in-depth interpretation of data, while explicit-solvent calculations via WAXSiS allowed for accurate evaluation of atomistic models. The resulting models suggest mechanisms by which Aap functional amyloid fibrils form, illuminating an important contributing factor to recurrent staphylococcal infections.

Graphical Abstract

Introduction

Despite the long-known role of Staphylococcus epidermidis as the leading cause of medical device-related infections, many questions remain regarding the mechanism of biofilm formation and infection [1–4]. Although S. epidermidis is considered less pathogenic than S. aureus, its abundance on the surface of the skin allows for contamination of medical devices upon insertion [3]. After initial colonization via surface proteins [4, 5] or cell surface chemistry [6], cell-to-cell accumulation occurs via polysaccharide production or protein-dependent interactions into characteristic, resilient clusters of bacteria, known as biofilms. One such protein, the accumulation-associated protein (Aap), is expressed in 90% of biofilm-positive clinical isolates [7] and has been shown to be required for infection in a rat catheter model [8].

Aap is a large, multidomain, multifunctional protein anchored to the cell wall of S. epidermidis at its C-terminus (Figure 1A). Starting at the N-terminus, Aap contains several short A-repeats, followed by a lectin domain implicated in surface attachment [9–11], a SepA protease cleavage site [12], 5–17 Zn²⁺-binding B-repeats [13], a proline/glycine-rich extended stalk [14], and an LPXTG sortase motif. Whereas the lectin domain mediates adhesion to corneocytes [9], accumulation of staphylococcal cells into a biofilm requires exposure of the B-repeat superdomain via proteolysis of the N-terminal A-repeats and lectin (i.e., the A domain) of Aap [12, 15, 16].

Figure 1. The accumulation-associated protein (Aap) from S. epidermidis RP62A.

(A) Domain arrangement of Aap. Scissors designate a SepA proteolytic cleavage site. At the C-terminus of the protein, an anchor designates the LPXTG Sortase II cell wall-anchoring motif. (B) A model of the Brpt5.5 construct containing five full B-repeats and the C-terminal half repeat (a G5 domain without a spacer). (C) shows the crystal structure of Brpt1.5 (one full B-repeat and the C-terminal half repeat) in a Zn²⁺-dependent dimer state (PDB: 4FUN). The dimer interface (residues within 10 Å of the opposite protomer) is highlighted in magenta. Black arrows point toward the location of Zn²⁺ ions bound in the dimer. The Zn²⁺-coordinating residues involved in dimerization are shown in (D), while the region shown in (E) highlights the H85 residue previously shown to be involved in tetramerization [20]. Note that the H85 residue shown in (E) is pointed outward and is not in the dimer interface.

Much work has been done investigating the mechanism of Zn²⁺-dependent self-assembly of the B-repeats. Each B-repeat contains a G5 subdomain and a spacer subdomain (also called an E domain), while the C-terminal “half-repeat” consists only of a G5 domain (Figure 1A). The B-repeats of Aap are 89–100% identical in sequence but fall into one of two variant sequences with subtle differences in folding stability and dimerization potential [17]. Several crystal structures of minimal Brpt1.5 constructs, containing 1 full B-repeat and the half-repeat, have been solved using various B-repeats throughout Aap [17, 18]. In the presence of Zn²⁺, Brpt1.5 forms an anti-parallel dimer (Figure 1C), and residues responsible for coordination of the Zn²⁺ ions (Figure 1D) have been thoroughly investigated [13, 18, 19]. While the Brpt1.5 dimer can be rapidly dissociated by chelation of the Zn²⁺ and S. epidermidis biofilms can be inhibited by a chelator, a mature biofilm is resistant to chelator [13, 16]. This disconnect was recently explained by studying a longer, more physiologically relevant Brpt5.5 construct (Figure 1B), which forms Zn²⁺-induced functional amyloid-like fibrils that are refractory to Zn²⁺ chelation. Similar fibrils observed in the matrix of mature S. epidermidis biofilms contained Aap as a major constituent. These observations suggest that amyloid fibrils present in the mature biofilm could be at least partially responsible for the strong cohesion and resilience of such device-related biofilms that lead to recurrent, hard-to-treat infections [16]. Interestingly, a detailed thermodynamic analysis of Brpt5.5 assembly revealed the formation of a tetramer intermediate that undergoes a Zn²⁺- and temperature-dependent conformational change that triggers amyloidogenesis [20]. After locating residues which mediated the Zn²⁺-dependent tetramer (Figure 1E), a mutant was designed which was unable to assemble beyond the dimer and unable to aggregate into fibrils [20]. Clearly, understanding the initial assembly events are crucial in our attempt to understand how these fibrils are formed.

Several other bacterial species have been shown to produce functional amyloids in the context of biofilm growth [21–24]. In most known cases, the amyloid is produced by a system of proteins encoded by an operon that typically include amyloid precursor proteins, chaperones, and initiator proteins. In contrast, Aap acts as a single-molecule amyloidogenic system that is expressed as a full-length protein attached to the cell wall and activated to its amyloidogenic state via proteolysis and binding of Zn²⁺. Likewise, the mechanism for amyloidogenesis adopted by Aap is highly unusual. The majority of well characterized amyloid-forming systems assemble via short peptide-length precursors that are generated either through proteolysis, localized unfolding of a larger protein domain, or domain swapping by a secondary structural element such as a β-hairpin [25, 26]. Rarely, small folded proteins such as transthyretin act as the nucleating species and assemble into amyloid fibers [27]. In contrast, a large region of the extremely elongated B-repeat region of Aap (e.g., 720-residue Brpt5.5) reversibly assembles into a tetramer that nucleates the amyloid fiber after a temperature-dependent conformational change [16, 20].

In the present study, we elucidate structural and mechanistic details of the critical Brpt5.5 dimer and tetramer intermediate assembly states that precede amyloidogenesis using analytical ultracentrifugation (AUC) and small-angle X-ray scattering (SAXS). The large size and extraordinarily elongated nature of this protein led to substantial challenges regarding experimental SAXS setup and analysis. The approaches and analyses used here outline best practices that should be taken when analyzing highly elongated protein systems. Further considerations were required given the potential flexibility of the system, as well as the reversible assembly of the dimer and tetramer. Nevertheless, we incorporated complementary data from AUC and other hydrodynamic techniques, as well as in silico modelling to explicitly test models of the Brpt5.5 monomer, dimer, and tetramer. Finally, we propose a mechanism of Zn²⁺-dependent amyloidogenesis of Brpt5.5 based on the tetramer model that is most consistent with the full range of experimental results.

Results

Single H85A point mutations in Brpt5.5 can inhibit Zn²⁺-dependent assembly.

Cell surface expression of 5 and a half B-repeats is required to support staphylococcal biofilm formation [28], which corresponds to our Brpt5.5 construct. This construct is comprised of alternating G5 and spacer (also called E) subdomains with the pattern G5⁰⁸sp⁰⁸G5⁰⁹sp⁰⁹G5¹⁰sp¹⁰G5¹¹sp¹¹G5¹²sp¹²G5¹³, in which G5⁰⁸ and sp⁰⁸ compose the 8^th B-repeat, for example (Figure 1 and Figure 2A). Yarawsky and Herr recently demonstrated that Brpt5.5 assembles in the presence of Zn²⁺ into dimer and tetramer species, and then after a conformational change, the tetramer further assembles to form insoluble aggregate [20]. Previous work indicated this insoluble aggregate is amyloid-like in nature, and is responsible for the Zn²⁺-chelator resistance of mature biofilms [16]. Chemical modification and mutagenesis experiments identified a surface histidine in each B-repeat (called H85 based on its position in a canonical B-repeat), located outside the Brpt1.5 dimer interface, as a critical residue for Zn²⁺-induced tetramer formation. A Brpt5.5 variant with the H85 position in each B-repeat mutated to alanine (Brpt5.5 5xH85A) was unable to assemble into the tetramer and was highly resistant to Zn²⁺-dependent aggregation [20]. These data indicated the importance of the H85 position in coordinating Zn²⁺ in trans across dimers to form the tetramer. Here, a panel of Brpt5.5 constructs with single H85A mutations was used to characterize the mechanism of assembly for the Zn²⁺-induced tetrameric amyloid precursor.

Figure 2. Measuring the assembly capability of H85A mutants.

(A) A schematic of the domain arrangement of Brpt5.5. Each full B-repeat is composed of a G5 domain and spacer (sp) region, while the C-terminal half-repeat contains only the G5 domain. The superscript on the G5 or sp indicates the parent B-repeat. (B) Sedimentation velocity experiments were performed on Brpt5.5 single-H85A mutants in the absence of Zn²⁺, and data were analyzed in DCDT+ [29, 30]. (C) Each construct was also examined in the presence of 3.50 mM ZnCl₂. (D) Sedimentation velocity data is shown for multi-H85A mutants in the absence (dashed lines) and presence of 3.50 mM ZnCl₂ (solid lines). (E) The weight-averaged sedimentation coefficient (s_w) from each analysis in (B)-(D) is shown – solid bars refer to samples without Zn²⁺, while patterned bars refer to samples with ZnCl₂. In panel (F), the weight-averaged molecular weight of samples (fit to a single, ideal species) was determined by sedimentation equilibrium experiments in the absence or presence of 3.50 mM ZnCl₂. Horizontal dashed lines represent the expected molecular weights of monomer, dimer, and tetramer species. (G) Turbidity measurements upon addition of ZnCl₂ to the wild-type and multi-H85A mutants. ZnCl₂ additions to buffer are shown as empty circles.

Analytical ultracentrifugation (AUC) sedimentation velocity experiments with wild-type Brpt5.5 and each of the single H85A mutants revealed that in the absence of Zn²⁺, each protein is monomeric, sedimenting near 2 S (Figure 2B). In the presence of Zn²⁺, H85A mutations in sp⁰⁹ or sp¹¹ allow assembly to a similar degree as wild-type: a mix of monomer (~2 S), dimer (~4 S), and tetramer (~7 S) species, as described [20]. Note that due to rapid kinetics of assembly, discrete peaks are not observed for each species present, but a reaction boundary with contributions from each species is instead observed. Single H85A mutations in sp⁰⁸, sp¹⁰, and sp¹² limit assembly to an apparent dimer, similar to the 5xH85A construct (Figure 2C and Figure 2D) [20]. Interestingly, there was a periodicity in the assembly behavior of single-H85A mutated constructs: the H85 side chain in the first, third, and fifth B-repeat was critical for tetramer assembly, whereas H85 in the second or fourth repeats was dispensable for tetramer formation.

This periodicity of the H85A mutations could provide important insight into the structure of the Brpt5.5 tetramer. Circular dichroism data demonstrated that all single H85A mutants fold properly and show only a minor decrease in melting temperature (T_m) from ~51° C to ~49° C (Supplemental Figure 1). Some loss of thermostability is expected due to the involvement of H85 in a hydrogen-bonding network [17, 18, 20]. Supplemental Figure 2 shows a sequence alignment of each full B-repeat and the C-terminal half-repeat. Shelton, et al. have previously shown that B-repeats in Aap exist as either consensus or variant subtypes; the variant B-repeats contain an alternate sequence cassette with 6 amino acid changes relative to the consensus sequence, leading to an increase in thermostability and a decrease in the extent of Zn²⁺-mediated dimerization [17]. The wild-type Brpt5.5 construct used here is all-consensus except for variant B-repeat 8. With no sequence-based patterns to explain the periodicity, it was hypothesized that this trend may be explained by the mode of tetramer assembly. A proposed structural model for a Brpt5.5 dimer based on crystal structures of Brpt1.5 constructs suggested that the two protomers exhibit significant twisting, like the strands of fiber in a rope [17, 18]. One could imagine that the Brpt5.5 dimer will be twisting along the length of the molecule, and a second dimer might line up parallel to this molecule such that only some of the H85 positions will be in contact (Figure 1E) [20].

Introducing multiple H85A mutations does not have additive effects on inhibition of assembly.

After observing the potent effect of a single H85A mutation on the ability of Brpt5.5 to form tetramers, the effect of multiple H85A mutations in the same construct was examined (Figure 2D). Three constructs were tested. The first, 2xH85A, contained H85A mutations in sp⁰⁹ and sp¹¹ – neither of which inhibited tetramer formation when examined as single mutants. Next, 3xH85A, contained H85A mutations in the three other spacer regions (sp⁰⁸, sp¹⁰, and sp¹²). The three H85A mutations of 3xH85A, individually, were each able to inhibit tetramer formation. Lastly, the 5xH85A mutant contains the H85A mutation in every spacer region of Brpt5.5 and has been characterized in detail in our previous study [20]. It is included here as a tetramer-incompetent mutant, which also does not undergo conformational changes to form the amyloid-like aggregates as wild-type Brpt5.5 does.

Sedimentation coefficient distributions of wild-type and multi-H85A mutants in the absence and presence of Zn²⁺ reveal that the 2xH85A mutant assembles similarly to wild-type, while 3xH85A and 5xH85A have limited assembly (Figure 2D). Comparison of the weight-averaged s* (s_w) values makes it apparent that the multi-H85A mutants behave similarly to their single-H85A mutant counterparts (Figure 2E). This indicates that there is no significant additive effect on assembly when introducing multiple H85A mutations – one H85A in the appropriate position is sufficient to inhibit tetramer formation. Again, circular dichroism spectra verified each mutant is properly folded (Supplemental Figure 1).

Sedimentation equilibrium experiments were performed on a subset of the constructs to determine whether or not the tetramer is populated in situations where decreased assembly was observed relative to wild-type in the sedimentation velocity experiments. The observed trends in average MW (determined by fitting the data to a single, ideal species) align with the sedimentation velocity data (Figure 2F). Datasets were then fit globally to a monomer, monomer-dimer, and/or monomer-dimertetramer self-association model (Supplemental Figure 3). The global fits confirmed that the 5xH85A, 3xH85A, and G5¹²sp H85A mutant were unable to form tetramer at this Zn²⁺ concentration (3.5 mM).

The effect of multiple H85A mutations on Zn²⁺-dependent aggregation was then evaluated using a light-scattering assay to monitor aggregation upon titration of ZnCl₂ into the sample (Figure 2G). Wild-type Brpt5.5 shows a sigmoidal increase in light-scattering with a midpoint near ~25 mM ZnCl₂, which is accompanied by visible aggregation in the sample cuvette. The 2xH85A and 3xH85A mutants both show decreased propensity to aggregate compared to wild-type, while the 5xH85A mutant is most resistant to aggregation, consistent with previous findings [20]. Interestingly, there is a significant difference in the aggregation propensity of 3xH85A and 5xH85A, even though neither can form tetramer. This indicates that Zn²⁺-induced aggregation is not only dependent on formation of the tetramer, but also on the availability of certain H85 residues. This idea is also supported by dynamic light scattering (DLS) data collected for single-H85A mutants examining temperature-dependent aggregation in the presence of Zn²⁺ (Supplemental Figure 4). Furthermore, a temperature-dependent conformational change can be observed by CD upon formation of Zn²⁺-dependent amyloid fibrils [16, 20]. Supplemental Figure 4 shows CD data indicating that Brpt5.5 WT undergoes the conformational change at ~40°C, while 2xH85A (tetramercompetent) requires much higher temperatures to undergo the conformational change (~80°C). The 3xH85A and 5xH85A variants fail to undergo this conformational change at any temperature up to 90°C. These complementary analyses of the single- and multi-H85A mutants suggest that while some H85 positions are important in tetramer formation, the others are important for mediating Zn²⁺-dependent aggregation.

SEC-MALS-SAXS of Brpt5.5 monomer and assembly states.

The primary goal of this study was to gain structural insights into the dimer and tetramer assembly states. After determining several important molecular constraints from AUC data, SAXS was performed to provide complementary structural data appropriate for higher-resolution analysis of the system. Figure 3 and Supplemental Figure 5 show SEC-MALS and SEC-SAXS data for Brpt5.5 wild-type at several protein concentrations (without Zn²⁺ present). At the highest protein loading concentration (~10 mg/ml), an extremely broad and abnormally shaped peak was observed, while lower concentration datasets showed a sharper, Gaussian-shaped peak. Interestingly, the MALS-based molar mass estimate was consistent across the peak (Figure 3A, symbols), as was the radius of hydration (R_h; Supplemental Figure 5). The radius of gyration (R_g) from the SAXS data calculated via Guinier fit (Supplemental Figure 5A, symbols) was also consistent across the width of the peak. The SAXS scattering profile obtained from averaging frames across the 10 mg/ml loading concentration peak was, as expected, much less noisy than that from the 3 mg/ml and 1 mg/ml samples, which led to better confidence in the resulting R_g and D_max estimates (Supplemental Figure 5). Along with these analyses, Supplemental Figure 5 helps demonstrate that the broad, asymmetric peak at high concentration is not due to protein heterogeneity or impurity, but that it is a concentration-dependent effect of viscous fingering [31, 32]. While such aberrant elution behavior is usually not expected until relatively high protein concentrations (~50 mg/ml for BSA [33]), the hydrodynamic nonideality constant (k_s) measured for Brpt5.5 is nearly 10-fold higher than that of BSA. Given the ability for weak interactions (like those reflected by nonideality) to increase solution viscosity, we hypothesize that the high hydrodynamic nonideality measured for Brpt5.5 would translate to higher solution viscosity at much lower protein concentrations than BSA [34–36]. Brpt5.5 5xH85A (Figure 3B) and 2xH85A (Figure 3C) were analyzed by SEC-MALS in the absence and presence of ZnCl₂. Brpt5.5 WT was not examined in the presence of ZnCl₂ due to the high propensity for aggregation. Loading concentrations between 5 and 10 mg/ml were maintained for all samples to ensure high signal-to-noise and to populate the largest assembly states. MALS data indicated a similar monomer molar mass for all three proteins and revealed dimer and tetramer molar masses approximately 2- or 4-fold that of the monomer (Supplemental Figure 5).

Figure 3. SEC-MALS and SAXS Analysis of Brpt5.5 in the absence and presence of ZnCl₂.

(A) SEC-MALS data collected during elution from a Superose 6 Increase (10/300; ~24mL column volume) showing differential refractive index. Molar mass determined by MALS is shown as markers overlaid on the elution peaks. Data were collected at three different concentrations (colored according to the legend). (B) SEC-MALS data collected for Brpt5.5 5xH85A monomer (light red) and dimer (red). (C) SEC-MALS data collected for Brpt5.5 2xH85A monomer (light blue) and tetramer (blue). (D) The SAXS scattering curves for each dataset on a log-linear plot. The I(q) is offset to allow each dataset to be visible. The monomer datasets (-Zn²⁺) are shown in grey, light red, and light blue, while the dimer and tetramer (+Zn²⁺) are shown in red and blue, respectively, in accordance with the color scheme in (A)-(C). (E) The Guinier fit and residuals are shown for each dataset. Data points outlined in black were used in the Guinier Fit, and the fit itself is shown as a solid line. (F) The P(r) distribution (normalized by area) is shown. (G) The dimensionless Kratky plot, with dashed reference lines that intersect where a spherical particle would exhibit the maximum of a Gaussian peak. (H) displays the fit of a modified Guinier that describes the cross-sectional R_g (R_c), and (I) shows the cross-sectional P(r) distribution (P_c(r) normalized by area).

SAXS analysis of Brpt5.5 monomer and the assembly states.

SAXS scattering curves for the monomer datasets are relatively featureless, typical of rod-like particles [37] (Figure 3D). More pronounced features can be observed in the dimer (dark blue symbols) and tetramer (dark red symbols) scattering curves. The fit of the Guinier range to determine the R_g is shown in Figure 3E. The P(r) distributions are plotted in Figure 3F, which are again characteristic of rod-like particles – with a sharp peak at low radii, followed by a gentle and consistently sloping tail [37]. The point at which the P(r) distribution intersects the x-axis represents the maximum particle dimension (D_max). Interestingly, there is only a modest increase in the D_max value of the dimer and tetramer – a clear indication that the dimer and tetramer are formed by monomers stacking side-by-side rather than end-to-end (Figure 3F, Table 1, and Supplemental Table 1).

Table 1.

SAXS analysis of Brpt5.5 assembly states.

	Standard			Cross-Sectional
Sample	Rg (Å) (Guinier)	Rg (Å) (P(r))	Dmax (Å) (P(r))	Rc (Å) (Guinier)	Rc (Å) (Pc(r))	Dc (Å) (Pc(r))
WT monomer	146.8 ± 2.4	158 ± 1.61	558.0	56.3 ± 0.8	57.6 ± 0.3	19.0
5xH85A monomer	141.5 ± 2.2	155 ± 1.28	572.0	54.8 ± 0.4	57.5 ± 0.2	19.0
5xH85A dimer	150.0 ± 0.9	167 ± 0.86	633.0	109.8 ± 0.3	112.5 ± 0.2	37.0
2xH85A monomer	143.6 ± 3.9	157 ± 1.96	566.0	53.8 ± 0.7	56.4 ± 0.3	19.0
2xH85A tetramer	157.2 ± 1.0	177 ± 0.87	700.0	206.8 ± 0.4	209.5 ± 0.6	76.0

The dimensionless Kratky plot provides further detail on the shape of the particles; a compact, spherical protein will exhibit an approximately Gaussian peak centered about the intersection of the dashed lines at its max peak intensity [38]. For the Brpt5.5 monomers, the dimensionless Kratky exhibits a major shift in the maximum towards higher qR_g values, indicating extremely elongated shape (Figure 3G). The dimer and tetramer exhibit an increasingly less pronounced shift, indicating decreased asymmetry with higher oligomeric state. These trends in shape are consistent with our previously reported frictional ratios measured by AUC, which decreased from monomer to tetramer, suggesting less and less asymmetry in the particle shape [20]. Kratky plots are often used to distinguish folded from unfolded (or intrinsically disordered) proteins based on the typical expansion and elongation that occurs upon unfolding of a globular protein. However, the Brpt5.5 monomer is unusual in that it is highly elongated in the folded state – as seen in the P(r) distribution and our previous structural and biophysical work [13, 17–20, 39] – and actually becomes more compact when unfolded [20]. To assist our interpretation of the experimental Kratky plot, a dataset was generated from 10,000 random coil models of Brpt5.5 to simulate an ensemble of unfolded conformations of the protein. The resulting Kratky plot and P(r) distribution for the ensemble are very different from the experimental observation (Supplemental Figure 6). A comparison of this simulated dataset to our experimental observations, as well as analyses by circular dichroism (Supplemental Figure 1), makes it clear that Brpt5.5 is not unfolded but is simply highly elongated in solution.

The cross-sectional Guinier analysis (Figure 3H), which represents the cross-sectional radius of gyration (R_c) along the shorter axis of the rod, and the cross-sectional P_c(r) distribution (Figure 3I) provide additional insight into the conformation of Brpt5.5 assembly states. Each monomer shows a maximum particle thickness (D_c) value of approximately 20 Å, whereas the dimer and tetramer are approximately 2- and 4-fold thicker. A similar trend is observed in the R_c estimates. Combined with the traditional P(r)-based D_max estimates, the cross-sectional analyses point toward a side-by-side mode of assembly, where the particle becomes thicker rather than more extended. Table 1 and Supplemental Table 1 show the R_g and D_max data, as well as cross-sectional data for each sample.

Characterizing the flexibility of the Brpt5.5 monomer.

One result of highly flexible or dynamic systems is a smoothing or loss of features in the scattering data and Kratky plot [40], which can recapitulate the smooth, featureless scattering curves of rod-like particles [37]. Given that the B-repeat superdomain comprises a series of elongated, unsupported 3-stranded β-sheets and that a loss of crystallographic electron density at each end of monomeric Brpt1.5 constructs has previously been observed [17], the degree of flexibility in the Brpt5.5 monomer was evaluated using the ensemble optimization method (EOM) [41–43]. To perform EOM on this system, a large pool of models was constructed using the Monomer Monte Carlo function of the SASSIE web server [44]. Briefly, a monomer model was built by superimposition of Brpt1.5 protomers [19], and the 10 interface regions between G5 and spacer regions were designated as flexible regions (Figure 4A). A total of 5 pools were simulated, 4 of which were targeted toward higher or lower R_g values within the Monte Carlo simulations to better capture the full degree of flexibility that could be experienced by the monomer. This led to a total number of ~90,000 models input to our EOM calculations.

Figure 4. EOM analysis of Brpt5.5 monomer.

The representative EOM-selected model is shown in (A), with designated flexible regions, the G5 domains, and the spacer regions as indicated. (B) The EOM fit for each monomer is shown along with the experimental scattering curves. Normalized residuals are shown in the bottom plot. Plots (C) and (D) show the R_g and D_max distributions (determined by CRYSOL) of the reference pool, as well as the distributions of the selected ensembles.

Figure 4B indicates the best fit ensembles for the monomer datasets are of good quality. From the R_g and D_max distributions of the most representative ensembles (Figure 4C and Figure 4D), it is evident that Brpt5.5 in solution is skewed heavily toward extended conformations. Indeed, an extended conformation is apparent in the best-fit EOM model (Figure 4A). Perhaps the most critical EOM result is the R_flex value. This value reflects the flexibility of the protein, with 100% indicating the protein is as flexible as the reference pool and a lower R_flex indicating less flexibility than the reference pool. The R_flex for Brpt5.5 monomer is ~45% (Table 2 and Supplemental Table 1). This value confirms that the monomer is relatively rigid compared to the reference pool, which is also obvious from the narrower R_g and D_max distribution peaks compared to the random coil pool (Figure 4C and Figure 4D). However, the absolute degree of flexibility indicated by the data is less obvious. A lower R_flex value indicates less flexibility, but the magnitude of R_flex for a rigid protein will not be 0% but will instead be limited by noise in the data.

Table 2.

EOM analysis of monomer samples.

Sample	Chi2	Rflex	Rpool	Rσ	Rg (Å)	Dmax (Å)
WT	1.087	44.92 ± 0.41	75.80	0.09	168.76 ± 0.06	582.08 ± 2.12
5xH85A	1.517	44.93 ± 0.41	75.80	0.10	168.72 ± 0.06	583.06 ± 2.08
2xH85A	1.051	44.31 ± 0.50	75.80	0.08	168.94 ± 0.04	585.01 ± 2.69
Rigid Model	1.059	42.26 ± 0.89	75.80	0.14	170.89 ± 0.84	596.88 ± 6.54
Rigid Model (0.5x error, 0.5x noise)	1.263	44.01 ± 0.39	75.80	0.14	170.17 ± 0.37	592.08 ± 1.98

To better quantify the degree of flexibility, R_flex was calculated for a rigid model of Brpt5.5 in order to obtain a “baseline” R_flex value based on data quality. A synthetic scattering curve was generated for the EOM-selected model (Figure 4A). Random noise with similar magnitude as the experimental data was added, and R_flex values between 41.03–43.31% (Ave = 42.26% ± 0.89%) across duplicate EOM runs on five different datasets with randomly generated noise were observed. This suggests that if Brpt5.5 were totally rigid, the expected R_flex value would be near 42%. If the error and synthetic noise were reduced by half, a slight increase in the χ² and R_flex value of ~44% is produced (Table 2 and Supplemental Table 1). Compared to the simulated rigid model datasets, there is a slight but significant increase in R_flex for each monomer seen in the experimental EOM results. However, given the magnitude of the R_flex overall and the small difference from the rigid simulated system to the measured dataset, it is concluded that there is minimal flexibility in the monomer. This is also supported by the ability to achieve good fits of the SAXS datasets using single models (Supplemental Figure 7), rather than requiring multiple models as is the case for highly flexible systems. This conclusion agrees with the findings of Gruszka, et al. for monomeric B-repeats from SasG – the Aap ortholog from S. aureus [45].

Evaluating Brpt5.5 dimer models.

Our previous analysis of the 5xH85A mutant afforded some important constraints on how the dimer is formed. For example, the trend in the frictional ratio between monomer, dimer, and tetramer, as well as a rigorous analysis of the linked equilibria between dimerization and Zn²⁺-binding strongly suggested that the two protomers of the dimer overlap substantially, rather than being offset or out of register. Note that for Brpt1.5 and Brpt5.5, there are no significant conformational changes upon reversible Zn²⁺-dependent assembly [13, 20]. To test specific dimer configurations with atomic models, the experimental SAXS dimer data were compared against the simulated data from various Brpt5.5 dimer models built using the best fit monomer model (Figure 5). The dimer models were built to maintain a dimer interface representative of the Brpt1.5 dimer crystal structure, in which residues corresponding to H75, E203, and D21 (or E19) in successive G5 domains ligate the Zn²⁺ ion in trans [18]. Tested dimer models ranged from fully overlapping protomers (Figure 5A, top), comparable to Brpt1.5 dimer crystal structures, to a dimer that is offset to the point where only one G5 domain is overlapping (Figure 5A, bottom). Due to the importance of the hydration layer in defining the SAXS scattering of a particle, as well as the highly elongated shape of Brpt5.5, the WAXSiS server was used to explicitly model the hydration of the dimers [46, 47]. For the totally overlapping dimer, the χ² suggests a reasonable fit, and the R_g and D_max values are on par with the experimental values (Table 1 and Table 3). The deviation in the fit at high q-values (Figure 5B) could be an indication that there are slight differences in the finer details of the dimer, such as the bending and twisting of the molecule, or that there are minor deviations owing to conformational flexibility. The possibility of a mixture of monomer and various dimer configurations was tested, with minimal impact on the resulting χ² value (Supplemental Table 2).

Figure 5. Brpt5.5 dimer models and experimental fits.

(A) Surface representations of Brpt5.5 dimer models. Models are labeled 0 – 5 according to the number of B-repeats by which the two monomers are offset. Gray arrows mark the approximate location of Zn²⁺-binding sites. (B) WAXSiS-generated theoretical scattering curves of each model are shown as solid, colored lines (0=black, 1=orange, 2=red, 3=blue, 4=yellow, 5=purple). The fitted experimental data are shown as gray markers in each panel. The insets focus on the low scattering angle data and fit. The normalized residuals of each fit are shown in the bottom portion of the panels. Axis limits are maintained across the panels.

Table 3.

Evaluation of dimer models.

	WAXSiS			HullRad
Model	Chi2	Rg (Å)	Rg (anhydrous)	S20,w	f/fo	Dmax (Å)	Rg (anhydrous)	#Zn (EQ)&
Exp data	-	167*	-	3.97^	2.99	633*	-	N/A (7–9)
Dimer_0	2.92	162.7	170.5	4.29	2.45	636.1	170.2	6 (9)
Dimer_1	5.23	172.6	183.8	3.90	2.69	738.2	183.7	5 (7.5)
Dimer_2	10.55	200.9	209.1	3.68	2.85	838.2	209.1	4 (6)
Dimer_3	13.46	231.5	242.7	3.45	3.05	940.3	242.8	3 (4.5)
Dimer_4	27.26	264.0	281.5	3.29	3.19	1043.4	281.6	2 (3)
Dimer_5	38.92	318.2	323.6	3.09	3.40	1145.4	323.8	1 (1.5)

^*Exp R_g and D_max values are from GNOM P(r) analysis. Model R_g values from Guinier Fit performed by WAXSiS.

Model D_max values are reported from HullRad.

^{^}S_20,w of Brpt5.5 5xH85A dimer at 0.5 mg/ml (as reported in Yarawsky & Herr, JBC 2020 [20])

^&We consistently observe ~1.5 Zn ions binding per G5 domain in linked equilibrium (EQ) AUC analyses, despite observing clear electron density for only a single Zn²⁺ ion bound per G5 domain in the crystal structure of Brpt1.5. (See previous reports for more details [13, 18, 20].) The first value listed is expected based on electron density in the crystal structure of Brpt1.5, while the number in parenthesis is expected to match the experimental linked equilibrium AUC data reported previously [20].

Additionally, hydrodynamic calculations on the dimer models were performed using HullRad [48], which also pointed to the totally overlapping model being a reasonable option (Table 3). Finally, the number of potential Zn²⁺ sites in the dimer interface was evaluated for each configuration. By crystallography, 1 Zn²⁺ ion per G5 domain of the Brpt1.5 dimer has been observed, while linked equilibria analyses of Brpt1.5, Brpt2.5 and Brpt5.5 show a consistent 1–2 Zn²⁺ ions per G5 domain [13, 20, 49]. The previous analysis of linked equilibria indicated that 7–9 Zn²⁺ ions were bound upon dimerization of Brpt5.5 [20], which is consistent with the totally overlapping dimer model (Table 3). Figure 5A shows grey arrows pointing toward the location where the crystallographic Zn²⁺ binding sites would occur in each dimer configuration. The fully overlapping dimer is consistent with both the number of Zn²⁺ binding sites and number of Zn²⁺ ions bound during assembly.

Elucidating Brpt5.5 tetramer configurations.

Our analysis of the tetramer SAXS scattering curve in Figure 3 indicated only a marginal increase in both R_g and D_max compared to the dimer, and an approximate doubling of the cross-sectional parameters (R_c and D_c) (Table 1 and Supplemental Table 1). These results point toward a tetramer formed by dimers that line up side-by-side, which would result in a tetramer that is twice as thick as the dimer. The second major consideration is the effect on assembly seen with the single- and multi-H85A mutants. In the context of the Brpt5.5 dimer, the ligation of Zn²⁺ in trans by residues in the H75, E203 and D21 (or E19) positions of successive G5 domains places residues in the H85 position on the outer surface of the dimer, where they can ligate additional Zn²⁺ in trans to form the tetramer [20]. Using the best-fit dimer, there is a feasible tetramer orientation which places each of the H85A residues of the 3xH85A mutant (tetramer-incompetent) at the interface between the two dimers. If fewer than all three H85A residues were at the interface of the tetramer, then each of the respective single H85A mutants should not have inhibited tetramer formation (as seen in Figure 2). In Figure 6, a simplified illustration of a possible tetramer configuration that satisfies the H85A restraint is provided. The two assembling dimer species are shown as slightly offset spirals. The lines protruding outward from the spirals represent the approximate position of the H85 residue based on crystallography data of the Brpt1.5 dimer [18] and the best-fit Brpt5.5 dimer model. The dashed lines are the H85A mutations, while the solid lines are the remaining, intact H85 residues which would be capable of binding Zn²⁺. Importantly, the same tetramer configuration is able to explain why 3xH85A and its respective single H85A mutants inhibit tetramer formation (Figure 6A), while also explaining why the 2xH85A mutant and its respective single H85A mutants do not inhibit tetramer formation – the H85A positions would be placed outside of the tetramer interface (Figure 6B). Given that 2xH85A was able to form the tetramer, but not aggregate as readily as wild-type Brpt5.5, a mechanism for Zn-induced polymerization and aggregation can also be hypothesized where the H85A positions of the 2xH85A tetramer are mediating interaction with other tetramers (Figure 6B, bottom right).

Figure 6. Rational tetramer configurations based on experimental observations.

The Brpt5.5 dimer model (based on the best fitting dimer model from Figure 5) is shown in a simplified, spiral representation. Black and grey spirals represent the two protomers of the dimer, while the protruding lines represent approximate positions of the H85 (solid) and H85A (dashed) residues. The top scheme shows 3xH85A, which cannot form the tetramer. The configuration proposed for the tetramer can explain why the 3xH85A mutations inhibit the tetramer, as well as why each single mutant can inhibit tetramer. The configuration in the top right could explain the 3xH85A mutant behavior, but not why the middle H85A mutation (marked by a star) would inhibit tetramer. The red slash designates that an interface is disrupted by the H85A mutations. The bottom set of models shows the 2xH85A mutant. In the same tetramer configuration as proposed for the 3xH85A mutant, 2xH85A still has H85s in the interface – which can mediate tetramer formation. We then propose a mechanism of polymerization where the H85A mutations are disrupting the interface, inhibiting Zn²⁺-dependent aggregation.

After determining logical restraints from the data mentioned above, several atomistic models of the tetramer were built (Supplemental Figure 8A) and tested against the experimental SAXS data using the WAXSiS server to explicitly model the solvent [46, 47]. The fits to the experimental data are not as good as observed for the dimer models; this probably reflects the extreme size and degree of elongation of the tetramer species, which is pushing the boundaries of what can be fitted using available tools. Mixtures of tetramer configurations or monomer/dimer/tetramer combinations were also tested, with no improvement in the χ² value (Supplemental Table 3). A possible reason for the deviations in the fit could be differences between zinc coordination geometries within an ensemble of tetrameric species, as observed in the crystal structure of the Zn²⁺-bound Brpt1.5 dimer [18]. Furthermore, without high resolution structural data to model the interface between dimers, there may be a strong impact of errors in the solvation of the tetramer models. The model with the lowest χ² to the experimental SAXS curve was the one offset by one B-repeat, followed by the totally overlapping model (Supplemental Figure 8B, Table 4). The R_g and D_max also pointed toward these two models as being the most appropriate. However, our AUC data indicated that the H85 residue in sp⁰⁸, sp¹⁰, and sp¹² each are essential for tetramer assembly – supporting the totally overlapping model. Based on our previous linked equilibrium AUC analysis, we determined that two dimers assemble upon addition of 4–7 Zn²⁺. The simplified “spiral” illustration of the tetramer (Figure 6) would place six residues in the H85 position at the interface, suggesting that each H85 position represents a binding site, presumably ligating a Zn²⁺ ion in trans along with yet unidentified residues on the opposing dimer. Considering these complimentary results, the totally overlapping tetramer model seems the most reasonable, though higher resolution approaches will be useful to elucidate more details regarding the important tetramer assembly state.

Table 4.

Evaluation of tetramer models.

		WAXSiS			HullRad
Model	Chi2	Rg (Å)	Rg (anhydrous)	S20,w	f/f 0	Dmax (Å)	Rg (anhydrous)
Exp data	-	177*	-	7.07^	1.44	700*	-
Tetramer_0	72.46	163.5	171.2	7.56	2.20	639.0	170.9
Tetramer_1	42.82	168.4	178.5	7.14	2.33	736.6	178.2
Tetramer_2	84.18	190.7	199.2	6.84	2.44	838.2	198.9
Tetramer_3	135.97	218.9	229.8	6.35	2.62	940.7	229.6
Tetramer_4	259.41	250.4	266.5	6.03	2.76	1042.8	266.3
Tetramer_5	396.42	301.7	307.2	5.69	2.92	1144.3	307.0

^*Exp R_g and D_max values are from GNOM P(r) analysis. Model R_g values from Guinier Fit performed by WAXSiS. Model D_max values are reported from HullRad.

^{^}S_20,w of Brpt5.5 WT tetramer at 0.5 mg/ml (as reported in Yarawsky & Herr, JBC 2020 [20]).

Implications for amyloidogenesis.

This overlapping tetramer model that is most consistent with our combined data suggests a plausible mechanism for amyloidogenesis by Brpt5.5. Using the AmylPred2 server [50], we identified five potential amyloidogenic peptide sequences within the consensus B-repeat primary sequence (Figure 7A); four of these occur within the G5 domain, and the 5^th occurs in the spacer domain. The B-repeat fold places all five of these potential amyloidogenic peptide sequences within fairly close proximity (Figure 7B). In particular, the four amyloidogenic sequences within the G5 domain each pair up with another to form adjacent antiparallel β-strands in the G5 fold; the GTEKVV and EIVHYGG sequences form adjacent strands in the 3-stranded β-sheet that coordinates the Zn²⁺ ion, and the GTKTIT and TEKITKQ sequences form adjacent strands in the preceding β-sheet within the G5 domain (Figure 7B; see blue/orange and green/yellow spheres). The fifth sequence, TGEVVT (red spheres in Figure 7B), is found within the spacer domain, but the tertiary fold of the B-repeat places this sequence adjacent to the GETKVV and EIVHYGG strands. Furthermore, all five of these sequences are found near a 2-fold symmetry axis in the B-repeat dimer, such that a total of ten potential amyloidogenic sequences are found in close proximity. Our most plausible tetramer model (Tetramer 0, Supplemental Figure 8A) places two parallel Brpt5.5 dimers in register. Thus, the tetramer provides a framework that places each potential amyloidogenic sequence 30 Å from an identical sequence positioned in parallel in the adjacent dimer. Furthermore, for amyloidogenic sequences near the 2-fold symmetry axis (e.g., the blue/orange GTEKVV/EIVHYGG sequences), each is also within ~30 Å of an identical sequence in the antiparallel orientation. We propose that the conformational change in the tetramer at modestly elevated temperature (~40° C) leading to amyloid formation [16, 20] corresponds to localized unfolding that allows one or more amyloidogenic sequences from the upper dimer to stack onto the parallel (or antiparallel) identical sequence(s) from the lower dimer to form a nucleating species with interdigitating side chains between the planes of the associating strands. These nuclei would then propagate into a fibril through lateral H-bonding between the edges of the strands in each nucleus. This mechanism would allow an arrangement of strands in the fibril resembling well-known classes of steric zippers observed in amyloid crystal structures [51]. For example, a parallel, face-to-back arrangement of single strands across the tetramer interface could form a Class 2 steric zipper, whereas an antiparallel, face-to-face arrangement of single strands could form a Class 1 steric zipper, using the nomenclature of Eisenberg and colleagues [51]. The cartoon in Figure 7C illustrates a similar example of how two antiparallel strands from one Brpt5.5 dimer (blue and orange) could stack on the identical strands from the lower dimer (underlying light blue and light orange) to form an amyloidogenic nucleus and then propagate into a fibril through lateral H-bonding interactions. In general, the Brpt5.5 tetramer framework positions the molecules in a way to greatly diminish the entropic penalty in bringing amyloidogenic peptides from Aap together in a parallel (or antiparallel) orientation to nucleate an amyloid fibril.

Figure 7. Proposed mechanism for amyloid nucleation by Brpt5.5.

(A) Primary sequence of a consensus B-repeat in Aap, showing the five amyloidogenic peptide sequences predicted by AmylPred2. (B) Structural representation of the overlapping tetramer (Tetramer 0 from Supplemental Figure 8), with inset figures illustrating the locations of the amyloidogenic sequences (colored spheres). The side view demonstrates that identical sequences are found in parallel in the top and bottom dimers, separated by 30 Å. (C) Cartoon illustration of one possible mechanism for fibril nucleation in which two antiparallel strands with amyloidogenic potential (orange and blue) in the top dimer could assemble with the identical antiparallel strands from the lower dimer (light orange and light blue) to form a nucleus that can then propagate into a fibril through lateral assembly of nuclei via H-bonding along the lateral edge of the strands.

Discussion

Our present study utilized AUC and SAXS combined with a panel of assembly-restricted mutant proteins to provide novel insights into the structure of tandem B-repeats and their reversible assembly. Furthermore, these data lead to a new hypothesis regarding the mechanism by which the tetramer nucleates amyloidogenesis. The simplest assumption is that the Brpt5.5 monomer would essentially be a simple extension of the Brpt1.5 construct, of which several variants have been crystallized [17, 18]. Although generally accurate, the Brpt5.5 monomer in solution appears to be even more extended than suggested by our Brpt1.5 crystal structures, based on the best-fit models consistent with the SAXS data. Dimer models that adhered to approximate constraints observed in the Brpt1.5 dimer, including maintaining the Zn²⁺-binding sites, dimer interface, and overall twisting of the dimer were evaluated. Along with the experimental SAXS data, several important parameters determined by our previous hydrodynamic and thermodynamic analyses of the Brpt5.5 dimer were used [20]. These complementary data increase confidence in the overall configuration of the best-fit dimer model, although atomic-level structural details remain to be solved. The tetramer presented a much more significant challenge, as no high-resolution structural data exist to provide molecular constraints for modeling the interface. However, basic size and shape constraints provided by sedimentation velocity AUC, along with R_g and D_max values from SAXS analyses, pointed toward a side-by-side tetramer instead of two dimers lining up end-to-end. This configuration is also most reasonable in the biological context, where the B-repeat superdomain is tethered to the bacterial cell surface. Additionally, the orientation of the H85 residues was considered, since the H85A mutation in certain locations precludes tetramer formation. Because any one of the three H85A mutations in the 3xH85A mutant (i.e., H85A in the sp⁰⁸, sp¹⁰, or sp¹² subdomains) could prevent tetramer formation, each of these H85 residues should lie in the tetramer interface, as is the case for the proposed totally overlapping model. Interestingly, when arranging the tetramer in a side-by-side fashion with the three H85 residues in the interface, the two remaining H85 residues (in the sp⁰⁸ and sp¹¹ subdomains), which when mutated to alanine were able to inhibit Zn²⁺-induced aggregation in our assays compared to the wild-type protein, were solvent-exposed such that they could be available for Zn²⁺-mediated polymerization. Accordingly, analysis of the multi-H85A mutants demonstrated independent contributions to Zn²⁺-induced aggregation from tetramer formation and the presence of certain H85 residues (Figure 2).

During SAXS analysis of the Brpt5.5 monomer and assemblies, unusual dimensionless Kratky plots were observed that could easily be misinterpreted if not for complementary data like hydrodynamic data from AUC, folding studies using CD, and comparisons with simulated data for unfolded ensembles. Our results demonstrate what can be expected for a folded protein that is significantly extended in solution.

SAXS is useful for the analysis of flexible proteins and proteins which undergo conformational changes [38]. The EOM approach is the most common method for analyzing this type of SAXS data [41–43]. In order to more accurately quantify the flexibility of the Brpt5.5 monomer, data were simulated for a single (rigid) model with noise similar to the experimental data, and then the resulting R_flex value was compared to the experimentally observed R_flex. This allowed for a better interpretation of the R_flex value in the context of the specific data quality by providing a baseline value of R_flex, i.e., the value expected for a totally rigid model when a dataset has a certain amount of noise. We recommend that this becomes a common practice for those analyzing potentially flexible macromolecules using SAXS.

An interesting observation was made regarding the hydration of Brpt5.5 monomer and oligomers. While for globular proteins, the addition of solvent around the protein is expected to increase the R_g value, this is not the result observed for the highly elongated protein studied here. In this case, the majority of solvent molecules are not added at the end of the protein but are added across the length of the protein. This essentially leads to a thicker molecule, which would yield a smaller R_g measurement (Table 3 and Table 4). It would be expected that other highly elongated molecules would share a similar trend in R_g upon hydration.

In addition to providing insights into the use of AUC and SAXS for the structural analysis of a highly elongated and assembling system, potential weaknesses were revealed in the Aap assembly process that could potentially be exploited therapeutically. Early efforts to understand Aap-dependent biofilm formation indicated that the use of DTPA, a Zn²⁺-chelator, could effectively prevent biofilm formation [13]. However, at just two hours into bacterial growth on a polystyrene surface, S. epidermidis develops a resistance to the effect of moderate concentrations of DTPA. This time frame correlated with development of Aap amyloid fibrils [16]. The presence of amyloid fibrils in biofilms forming on indwelling medical devices could help explain the major difficulty in treating nosocomial infections. Amyloid fibrils are known to be extremely rigid, owing to extensive intermolecular hydrogen bonding [52]. Once formed, amyloid fibrils are highly stable and usually considered irreversible under typical conditions [53, 54], although cellular mechanisms by which amyloid fibrils can be degraded or disassembled are being explored given their interest as future therapeutic tools [55]. Our current work has indicated the H85 position as a potential therapeutic target, which when occluded, would likely prevent downstream aggregation into amyloid fibers, based on our H85A mutation studies presented here and in previous work [20].

Finally, the tetramer model that is most consistent with all our experimental data suggests plausible models for amyloid nucleation by Brpt5.5. The structural framework provided by the tetramer places four copies each of five distinct amyloidogenic peptide sequences within close proximity of one another. This positioning would greatly decrease the entropic penalty for bringing the amyloidogenic peptides into contact with one another. Depending on the potential parallel or antiparallel nature of the interactions between amyloidogenic peptides, the resulting fibrils would resemble different well-characterized steric zipper classes from known amyloid crystal structures; further structural and mutagenesis studies will be needed to better define which modes of assembly occur for the Brpt5.5 fibrils. These solution studies of the Brpt5.5 dimer and tetramer assembly states will better equip us to develop approaches to combat the highly problematic and persistent device-related infections caused by Staphylococci.

Materials and Methods

Protein expression and purification.

Brpt5.5 constructs were generated from the wild-type plasmid previously used, containing an N-terminal, TEVp-cleavable His-MBP tag [16, 20], and the H85A mutants were generated from primers listed previously [20]. A C-terminal, TEVp-cleavable Strep-II Tag for SAXS experiments was added with a glycine linker (-G-) following the C-terminal half-repeat of Brpt5.5 (…EYGPT-G-ENLYFQ-GWSHPQFEK). Non-Strep-tagged proteins were expressed and purified as previously described [16]. For purification of the Strep-tagged proteins, a Strep-affinity purification step was added after the initial Ni-affinity using a Cytiva StrepTrap HP (5mL) column. After TEVp-cleavage, a subtractive Ni-affinity step was used to remove uncleaved fusion protein and His-MBP. Next, a subtractive StrepTrap step removed uncleaved Strep-tagged fusion protein and Strep tag. Lastly, a Superdex 200 pg (Cytiva) or Superose 6 Increase (Cytiva) was used as a final purification step. For all experiments (except where denoted explicitly), protein was dialyzed into 50 mM MOPS pH 7.2, 50 mM NaCl.

Analytical ultracentrifugation.

A Beckman Coulter XL-I analytical ultracentrifuge was used for collection of AUC data. All data were collected using the interference optical system with no time delay in between scans. Data were collected in equilibrium mode, rather than velocity mode, such that greater than 999 scans could be collected. Experiments were run at 48,000 rpm, and scans were collected until no further sedimentation was observed. Sedimentation velocity data were analyzed by DCDT+ [29, 30]. Sedimentation equilibrium data were collected using absorbance optics at 240 nm for samples at a loading concentration of 0.5 mg/ml. Data were collected at 10,000, 13,000, 17,000, and 24,000 rpm. Equilibrium data were analyzed globally (at least 3 speeds for each sample) using SEDANAL [56]. The weight-averaged molecular weight shown in Figure 2 was determined by global fits to a single species. A more detailed analysis was performed by fitting data to a monomer-dimer or monomer-dimer-tetramer model (Supplemental Figure 3). For determining the S⁰_20,w value of the monomers, we performed velocity experiments on samples across a concentration range of 0.25 mg/ml to 1 mg/ml in 20 mM KPO₄ pH 7.4, 50 mM NaCl. The buffer density, buffer viscosity, and protein extinction coefficients were calculated using Sednterp 3 [57]. Velocity data were analyzed in SedFit v16.36 using the c(s) analysis model [58] and plotted in GUSSI [59]. In GUSSI, the area and weight-averaged sedimentation coefficient of the monomer peak were determined using the integrate function. The area was used to determine the concentration, based on the Sednterp estimate of the fringes per mg/ml of each protein based on sequence. The concentrations and corrected sedimentation coefficient (s_20,w) values were input into the “evaluate ks and extrapolate to s⁰_20,w” form of Sednterp. This form calculates the hydrodynamic non-ideality term (ks) and s⁰_20,w by linear regression of data plotted as concentration (c; mg/ml) vs 1/s, according to Equation 1. The s(c) term represents the weight-average sedimentation coefficient at a given loading concentration, and s₀ refers to the sedimentation coefficient at infinite dilution.

$$\frac{1}{s(c)}=\frac{1}{s_0}\left(1+k_{s}c\right)$$ Equation 1:

Circular dichroism.

An Aviv 215 spectrometer equipped with a Peltier junction temperature control system was used to collect far-UV data. A 0.5-mm quartz cuvette was used. Wavelength scans and thermal denaturation experiments (Supplemental Figure 1) were collected on 20 μM protein in 50 mM MOPS pH 7.2, 50 mM NaCl. Data were fitted to a two-state transition (folded monomer to unfolded monomer) with pre- and post-transition linear corrections and no change in heat capacity [60]. Wavelength scans to monitor for conformational changes relating to amyloidogenesis were taken at 10 °C increments from 20 °C to 90 °C, then back to 20 °C. Samples were 0.5 mg/ml (~6.5 μM) in 50 mM MOPS pH 7.2, 50 mM NaCl with 5 mM ZnCl₂ added to the protein and buffer solution, directly. Scan parameters for each experiment were consistent with those listed previously [20].

Dynamic light scattering.

Temperature-induced aggregation experiments were performed using a Malvern Zen 3600 Zetasizer Nano. Temperature steps of 2 °C were used with 120 seconds of equilibration time. Three measurements with automatic measurement duration were taken at each temperature. A low-volume quartz cuvette was loaded with 200 μl of 0.5 mg/ml protein in 50 mM MOPS pH 7.2, 50 mM NaCl. Samples were filtered immediately before the experiment began with a 0.22 μm syringe filter to remove any dust or aggregate.

SEC-MALS-SAXS.

Experiments were performed at BioCAT (beamline 18ID at the Advanced Photon Source, Chicago, Illinois, US). Samples were dialyzed into 50 mM MOPS pH 7.2 and 50 mM NaCl or 50 mM MOPS pH 7.2, 50 mM NaCl, and 5 mM ZnCl₂. Samples passed through a Superose 6 (10/300 GL) column (Cytiva) operated by a 1260 Infinity II HPLC (Agilent Technologies) at 0.6 ml/min, followed by the Agilent UV detector, a MALS detector and a DLS detectors (DAWN Helios II, Wyatt Technologies), and an RI detector (Optilab T-rEX, Wyatt). The flow then went through a 1.0 mm ID quartz capillary SAXS flow cell with ~20 μm walls. A coflowing buffer sheath was used to separate sample from the capillary walls to help prevent radiation damage [61]. Scattering intensity was recording with a Pilatus3 X 1M (dectris) detector placed 3.6 m from the sample, yielding a q-range of 0.0045 Å⁻¹ to 0.35 Å⁻¹. Exposures of 0.5 s were acquired every 1 s during elution and data reduction was performed in BioXTAS RAW (version 1.6.3) [62]. Buffer blanks were created by averaging regions flanking the elution peak, or when necessary, by linear baseline correction to regions on either side of the elution peak. Buffer blanks were subtracted from exposure selected from the elution peak to create I(q) vs q scattering curves for subsequent analysis. Molecular weights and hydrodynamic radii were calculated using the MALS and DLS data respectively in the ASTRA 7 software (Wyatt). It has been shown that full information about a scattering object is contained by sampling at the Shannon channels in the data, q=n*π/D_max. Thus, any measurement must measure to a q_min<π/D_max [63]. Given our q_min of 0.0045 1/A, this criterion is met for all samples.

SAXS analysis.

SAXS analyses were performed in BioXTAS RAW (version 2.1.0) [62, 64], with the exception of the modified Guinier analysis (cross-sectional R_g), which was performed using Primus from the ATSAS package (version 3.0.3) [65, 66]. SASSIE’s monomer Monte Carlo module [44] was used to generate Brpt5.5 models. The flexible regions were specified as discussed in the text, each with a maximum angle sampling of 30° during each iteration. Separate pools of 50,000 trial attempts (resulting in ~13,000– 22,000 models per pool) were generated, with one pool not being directed toward a specific R_g. Other pools were directed toward high or low R_g values to generate a wide range of conformations. A custom python script was written to perform CRYSOL [67] in non-fitting mode with the desired non-default settings (maximum order of harmonics: 99; order of Fibonacci grid: 18; number of points: 101) for all models. Additionally, the script generated the appropriate EOM input files. EOM data were averaged over 10 GAJOE calculations in the case of experimental data. For EOM evaluation of simulated data for a single model (no flexibility), CRYSOL was used to generate the scattering curve and a python script was written to add random noise to the CRYSOL curve that could be scaled in intensity with respect to experimental data. For EOM calculations, we averaged data from 3 GAJOE calculations on 3 randomized datasets. Hydrodynamic calculations were performed using the HullRad [48] code available at https://hullrad.wordpress.com/. The dimer and tetramer models were generated in PyMOL (The PyMOL Molecular Graphics System, Version 2.4 Schrödinger, LLC.).

Due to the elongated nature of this protein, there was concern about the accuracy of methods commonly used for generating SAXS scattering profiles from atomic models. The hydration shell around the protein molecule and the excluded volume both impact the SAXS scattering [46, 47, 68]. Popular methods, such as CRYSOL and FoXS, fit parameters related to the hydration of the protein using the experimental data provided [67, 69]. It is possible these methods will be less accurate when analyzing very highly elongated models that lie outside of the range of molecules which were originally tested by the respective authors. Supplemental Table 4 shows a comparison of χ² values for a subset of monomer models determined using several common programs. WAXSiS was expected to be the most accurate approach to SAXS analysis of this system, due to the explicit solvent modeling [46, 47]. While WAXSiS provided the lowest χ² values, its use for large sets of models is limited by the high computational cost (~3–6 hours per model using the web server). CRYSOL provided the second-best χ² values when increasing the order of harmonics to the maximum value of 99. FoXS is expected to be faster than CRYSOL and yield similar results [69], but FoXS’s potential beyond this limited test was not evaluated. When performing EOM, however, the order of harmonics used by CRYSOL was limited by the program to a maximum value of 50. Therefore, a custom python script was written to perform CRYSOL using the desired value of 99 (which is the maximum when running CRYSOL independently from EOM), then to create the appropriate output files to be used by the GAJOE step of EOM. While SASSIE has its own method for calculating the SAXS scattering of each model [44, 70], its handling of the effect of the hydration shell appears to be inadequate at the present time, resulting in poor χ² values for the models tested, which also did not follow a trend similar to the other methods. Molecular dynamics-based SAXS calculations using the same explicit solvent simulations implemented in WAXSiS are a powerful approach to analyzing data from extended particles which might also be flexible [71]. In this study, however, we did not carry out flexible, free simulations of Brpt5.5 because of the large computational cost.

Explicit-solvent SAXS calculations (WAXSiS).

Because the simulation systems for Brpt5.5 were large, explicit-solvent SAXS calculations [46] were carried out on a local cluster and, if not stated otherwise, not on the WAXSiS web server [47]. Simulations were carried out with the Gromacs software [72] version 2021.3. Interactions of the protein were described with the Amber99SB-ildn [73, 74] force field, and the TIP3P [75] water model was used. The proteins were placed in simulation boxes of a cuboid, where the distance between the protein to the box faces was at least 2 nm. The boxes were subsequently filled with water molecules and the systems were neutralized by adding sodium ions. In total, the systems contained between 757,158 and 1,717,939 atoms. The energies of the systems were minimized with the steepest decent algorithm. Subsequently, the systems were equilibrated for 100 ps and simulated for 1 ns for production with harmonic position restrains applied to the backbone atoms (force constant 1000 kJ mol⁻¹ nm⁻²). The temperature was kept at 289.15 K using velocity rescaling (τ = 0.1 ps) [76] The pressure was controlled at 1 bar with the Berendsen barostat (τ = 2 ps) [77]. The geometry of water molecules was constrained with the SETTLE algorithm [78], and LINCS [79] was used to constrain other bond lengths involving hydrogen atoms. An integration time step of 2 fs was used. The Lennard-Jones potentials with a cut-off at 1.2 nm were used to describe dispersive interactions and short-range repulsion. Electrostatic interactions were computed with the smooth particle-mesh Ewald method [80].

Explicit-solvent SAXS calculations [46, 81] were performed with the rerun functionality of an in-house modification of Gromacs 2018.8, as also implemented on the webserver WAXSiS [47]. Details of explicit-solvent SAXS calculations are presented in Reference [82]. The source code and documentation are available on GitLab at https://gitlab.com/cbjh/gromacs-swaxs and https://cbjh.gitlab.io/gromacs-swaxs-docs, respectively. A spatial envelope was built around all solute frames of the protein at a distance of 0.7 nm from all solute atoms. Solvent atoms inside the envelope contributed to the SAXS calculations, thereby accounting for the modified density of the hydration layer. The buffer subtraction was carried out using 100 simulation frames of a pure-water simulation box, which was simulated for 10 ns and which was large enough to enclose the envelope. The orientational average was carried out using 16,000 q-vectors for each absolute value of q, and the solvent electron density was corrected to the experimental value of 334 e/nm³.

Multi-state SAXS modeling.

OLIGOMER [65] was used to fit the experimental dimer or tetramer data against multiple models to determine whether the presence of multiple species would improve the fits. The input form-factor file was built from the explicit-solvent SAXS calculations (WAXSiS). For fitting the experimental dimer data, the input file contained all 6 dimer models (Figure 5) and the best fit monomer from the EOM analysis (Figure 4). For the tetramer data analysis, the input file contained all 6 tetramer models (Supplemental Figure 8), all 6 dimer models, and the best fit monomer model. The OLIGOMER χ² value was inconsistent with the WAXSiS-calculated χ² value. Because of this, a theoretical scattering curve was built by calculating the weight-average scattering intensities from the individual components. This theoretical scattering data were then fit against the experimental data using the WAXSiS fitting algorithm, such that the χ² values would be directly comparable to the single-state model fitting. This framework also allowed for testing desired mixtures outside of what OLIGOMER suggested.

Aap undergoes Zn²⁺-dependent assembly into amyloid fibrils that stabilize biofilms.

Point mutants were analyzed for their ability to assemble and aggregate into fibrils.

B-repeat monomers from Aap are highly extended, mostly rigid rods.

Data-driven dimer and tetramer models suggest a mechanism for amyloidogenesis.

Challenges inherent in SAXS analysis of highly elongated structures are discussed.

Glossary

Aap

accumulation-associated protein

AUC

analytical ultracentrifugation

CD

circular dichroism

EOM

ensemble optimization method

SAXS

small-angle X-ray scattering

MALS

multi-angle light scattering

TEVp

tobacco etch virus protease

MBP

maltose-binding protein

DTPA

diethylenetriaminepentaacetic acid

SasG

Staphylococcus aureus surface protein G

SepA

Staphyloccocus epidermidis secreted metalloprotease

SEC

size-exclusion chromatography

Script and Data availability

SAXS datasets and select models are available via the SASBDB (SASDP43, SASDP53, SASDP63, SASDP73, SASDP83). A custom script for running CRYSOL independently and generating EOM files is available at Mendeley Data (doi.org/10.17632/zbbmr8mtbb.1) [83]. A modified GROMACS software for explicit-solvent SAXS calculation (GROMACS-SWAXS) is available at https://gitlab.com/cbjh/gromacs-swaxs.