Structural mechanism for modulation of functional amyloid and biofilm formation by Staphylococcal Bap protein switch

Abstract

The Staphylococcal Bap proteins sense environmental signals (such as pH, [Ca²⁺]) to build amyloid scaffold biofilm matrices via unknown mechanisms. We here report the crystal structure of the aggregation‐prone region of Staphylococcus aureus Bap which adopts a dumbbell‐shaped fold. The middle module (MM) connecting the N‐terminal and C‐terminal lobes consists of a tandem of novel double‐Ca²⁺‐binding motifs involved in cooperative interaction networks, which undergoes Ca²⁺‐dependent order–disorder conformational switches. The N‐terminal lobe is sufficient to mediate amyloid aggregation through liquid–liquid phase separation and maturation, and subsequent biofilm formation under acidic conditions. Such processes are promoted by disordered MM at low [Ca²⁺] but inhibited by ordered MM stabilized by Ca²⁺ binding, with inhibition efficiency depending on structural integrity of the interaction networks. These studies illustrate a novel protein switch in pathogenic bacteria and provide insights into the mechanistic understanding of Bap proteins in modulation of functional amyloid and biofilm formation, which could be implemented in the anti‐biofilm drug design.

In response to environmental signals, the biofilm‐forming Staphylococcus aureus Bap protein forms amyloid aggregates in a liquid‐liquid phase separation dependent manner.

Introduction

Amyloids are insoluble, highly structured protein aggregates (Eisenberg & Sawaya, 2017) best known for their association with debilitating human diseases such as amyotrophic lateral sclerosis (ALS) or Alzheimer's disease (AD) (Chiti & Dobson, 2017). Recently, the amyloids have also been found to perform physiological functions from prokaryotes to humans (Pham et␣al, 2014). In bacteria, functional amyloids are fulfilling diverse roles relevant for bacterial growth and survival in the environment, including as part of the extracellular biopolymer matrix that ties the bacteria together into multicellular communities called biofilm (Van Gerven et␣al, 2018). Biofilm formation provides bacteria with many advantages such as high tolerance to harsh environments and increased recalcitrance to antimicrobial agents (Santos et␣al, 2018), thus extremely difficult to eradicate and poses significant challenges to the healthcare system today (Bjarnsholt et␣al, 2013; Santos et␣al, 2018).

Biofilm‐associated proteins (Baps) are a group of bacterial surface proteins mainly involved in intercellular adhesion and mediating biofilm development (Lasa & Penades, 2006; Latasa et␣al, 2006). Its first member is identified in bovine Staphylococcus aureus which is a large multidomain protein of 2,276 amino acids (aa) (Cucarella et␣al, 2001) (Fig 1A), of which region B contains three putative EF‐hand motifs that may regulate Bap functionality upon Ca²⁺ binding (Arrizubieta et␣al, 2004). Although bap gene has never been found in S. aureus human isolates, bap ortholog genes are present in the core genomes of several coagulase‐negative staphylococcal species that are frequent colonizers of human skin (Tormo et␣al, 2005). Moreover, functionally related proteins homologous to Bap exist in many phylogenetically unrelated Gram‐positive and Gram‐negative bacteria (Lasa & Penades, 2006), suggesting its functional importance in a diverse group of bacteria.

Figure 1. Overall architecture of aggregation‐prone region of Staphylococcus aureus Bap.

• A
  Schematic domain organization of S. aureus Bap. S, signal peptide (aa 1–44); A, region A (aa 45–360); B, region B (aa 361–818); SP, spacer region (aa 819–947); C, repeat region (aa 948–2,147); D, serine‐aspartate repeats region (aa 2,148–2,208); W, cell wall anchor region. The aggregation‐prone regions of BSP consists of five modules: 1 (magenta), 2 (orange), 3 (green), 4 (blue), and 5 (cyan) based on its crystal structure.
• B
  A topology diagram of BSP.
• C
  Overall structure of Ca²⁺‐bound BSP monomer. Ca²⁺ ions are shown as gray spheres. Color code is the same in (A–C).

Recent results indicate that the N‐terminal of Bap is proteolytically processed, released, and form amyloid‐like aggregates able to induce bacterial biofilm formation under acidic pHs and low concentrations of Ca²⁺ ([Ca²⁺]). Restoring pH to neutrality reverses the amyloids formed at acidic buffer; addition of Ca²⁺ in the millimolar range inhibits biofilm formation despite acidification of the medium (Taglialegna et␣al, 2016). These findings suggest that staphylococcal Baps can sense the bacterial environment through pH and [Ca²⁺] changes and modulate biofilm formation in an amyloid‐dependent way. However, how Ca²⁺ and pH signals reconcile to regulate Bap‐mediated amyloid and biofilm formation is largely unknown.

Here, we report the crystal structure of the aggregation‐prone BSP region of S. aureus Bap which adopts a dumbbell‐shaped fold and exhibits tandem features. We show that the middle module (MM) connecting the N‐terminal and C‐terminal lobes consists of a tandem of novel double‐Ca²⁺‐binding motifs involved in cooperative interaction networks. While the N‐terminal lobe is sufficient to mediate amyloid aggregation through liquid–liquid phase separation (LLPS) and maturation, and subsequent biofilm formation under acidic condition, such processes are regulated by Ca²⁺‐dependent order–disorder conformational switches of MM. The disordered MM at low [Ca²⁺] promotes LLPS, functional amyloid, and biofilm formation, but the ordered MM stabilized by millimolar [Ca²⁺] dictates inhibitory effects which efficiency depends on the structural integrity of the interaction networks. Given the wide distribution of Bap orthologs and high conservation of the MM among pathogenic bacteria, our work lays the framework for mechanistic understanding of Bap proteins and development of antibacterial and anti‐biofilm therapeutics.

Results

The B and spacer regions cover the amyloidogenic hotspots in␣Bap

A variety of bioinformatics tools have been developed to identify amyloidogenic hotspots within proteins that are prone to aggregation and may induce amyloid formation (Hamodrakas, 2011). Previously, two segments (aa 487–496 and aa 579–584) within region B of Bap from S. aureus and a short peptide within C‐repeats region of Bap from Staphylococcus epidermidis had been predicated as amyloidogenic hotspots and confirmed to spontaneously form amyloid fibrils in␣vitro (Lembre et␣al, 2014; Paharik & Horswill, 2016). To have a thorough analysis on the amyloidogenic hotspots in S. aureus Bap, the propensity of the peptide segments across full‐length Bap to form amyloid fibrils is analyzed using AMYLPRED2 (Frousios et␣al, 2009; Espargaro et␣al, 2015), which determines amyloidogenic regions based on the consensus score of 11 of such tools and thus provides an estimate of the amyloidogenic propensity. Around 14 consensus short peptide segments are predicted which are mainly covered by the B and SP region (hereafter BSP, aa 361–947) (Appendix␣Fig S1), consistent with the previous mass spectrometry analysis of Bap amyloid aggregates that include Bap fragments from the N‐terminal region (Paharik & Horswill, 2016). Interestingly, no consensus amyloidogenic hotspots are predicted in C‐repeats by the AMYLPRED2 analysis, whether C‐repeats of S. aureus Bap form amyloids awaits further study. We therefore target the BSP for biochemical and structural analysis.

Overall architecture of the Ca²⁺‐bound BSP

A plasmid vector encoding BSP with an upstream hexa‐histidine tag is generated for protein expression and purification. The size‐exclusion chromatography (SEC) profile of recombinant BSP after Ni‐NTA column in the running buffer of 20 mM Tris–HCl pH 7.5, 100 mM NaCl at 4°C indicates two different oligomeric states for BSP, presumably as monomer and dimer (Appendix␣Fig S2A). Secondary structure prediction analysis of BSP suggests a putative domain boundary around aa 783; therefore, a SP‐truncated construct (B, aa 361–783) is also generated. The elution profile for B has similar feature as BSP (Appendix␣Fig S2A).

Previous report suggested that Bap B domain (aa 361–819) adopts a transient molten globular‐like structure in the absence of calcium which is stabilized upon calcium binding (Taglialegna et␣al, 2016). To analyze the effects of calcium binding on BSP and B folding, SEC profiles of both monomer and dimer of BSP and B in 10 mM Ca²⁺ are compared with that in 10 mM EDTA, and the elution volumes for the respective monomers and dimers become larger in the presence of Ca²⁺ (Appendix␣Fig S2B). These data suggest that both monomer and dimer of BSP and B become more compact upon Ca²⁺ binding, which encourages us to solve their high‐resolution crystal structures. Both BSP and B were used for crystallization trials. The structure of B is determined at 3.07 Å resolution by single‐wavelength anomalous diffraction (SAD) phasing using crystals grew from selenomethionine‐labeled B protein. The structure of BSP is determined at 1.93 Å resolution by molecular replacement using B structure as a search model. The data collection and refinement statistics are summarized in Appendix␣Table S1.

The BSP monomer has an asymmetric dumbbell‐shaped structure that can be further divided into five modules (1–5): modules 1 (aa 361–541) and 2 (aa 542–728) constitute the N‐terminal lobe, whereas modules 4 (aa 775–859) and 5 (aa 860–947) constitute the C‐terminal lobe. The N‐terminal and C‐terminal lobes are held together by a middle module (MM) 3 (aa 729–774; Fig 1A–C). Interestingly, BSP exhibits tandem features (Fig 1B and C). Modules 1 and 2 share a highly similar β‐sandwich core architecture and pack closely against each other with their edge side of β‐sandwiches at a near perpendicular angle. The middle module 3 folds into a tandem of novel double‐Ca²⁺‐binding motifs (aa 729–751 and 752–774) which will be discussed in detail in the next section. Modules 4 and 5 possess the same canonical all‐beta, immunoglobulin (Ig)‐like fold, and the tandem modules 4 and 5 exhibit a V‐shaped conformation, in which the Ig‐like modules stack on each face of β‐sheet.

The B monomer (aa 361–783) covering modules 1–3 in addition to aa 775–783 (hereafter dubbed as BSP₁₂₃) shares a very similar structure with the corresponding region in the BSP, with an average RMSD of ˜ 1.37 Å over all atoms. However, superimposition of the two crystal structures of B and BSP monomer against modules 1 and 2 reveals 14° twisting of module 3 together with the extruding loop adjacent to module 3 relative to module 2 (Appendix␣Fig S2C), suggesting flexibility in module 3 and the extruding loop.

Middle module 3 folds into a tandem of novel double‐Ca²⁺‐binding motifs

Previously, three potential Ca²⁺‐binding EF‐hand motifs (designated as EF1: aa 598–610, EF2: aa 729–741, EF3: aa 752–764, respectively) had been predicted in the BSP region of Bap (Arrizubieta et␣al, 2004). The crystal structure of BSP monomer reveals five Ca²⁺ ions bound to BSP, with one Ca²⁺ bound to module 1 and four Ca²⁺ bound to module 3, respectively (Fig 1C). The putative EF1 located in module 2 does not form at all. Instead, one Ca²⁺‐binding site is observed near the ends of two paired strands from module 1 (Fig 1C).

Unexpectedly, the putative EF2 and EF3 located in MM form a tandem of novel Loop‐Helix‐Loop (LHL) motifs (LHL1: 729–751 and LHL2: 752–774, respectively) that each binds two Ca²⁺ ions (Figs 1C and, 2A and B) instead of the canonical Helix–Loop–Helix (HLH) EF‐hand motif that usually binds with one Ca²⁺ ion (Fig 2C) (Gifford et␣al, 2007). While each LHL motif consists of 23 residues, six residues shorter than the canonical 29‐residue HLH EF‐hand motif, the first 12 residues of LHL1 and LHL2 share 91 and 89% similarity in sequence to the loop of the consensus EF‐hand motif, respectively (Fig 2D). A comparison of the amino acid composition clearly shows that LHL motif is highly hydrophilic and negatively charged and therefore it lacks the hydrophobic core characteristic of canonical EF‐hand motif (Fig 2E). In LHL1, the first Ca²⁺ binds to the incoming loop in a similar manner as Ca²⁺ binding to the canonical EF‐hand loop (Gifford et␣al, 2007), acidic residues 729, 731, 733, and 740 coordinate the Ca²⁺ through one of the side‐chain oxygen atoms in a monodentate mode, residue 735 binds to the Ca²⁺ with backbone oxygen, and residue 737 immobilizes Ca²⁺‐bound water through side‐chain oxygen (Fig 2A and C). The difference between LHL motif and canonical EF‐hand lies in residue 740. This residue donates only one oxygen atom to the first Ca²⁺ in LHL motif whereas it donates two side‐chain oxygen atoms to Ca²⁺ in a bidentate mode in the canonical EF‐hand loop (Fig 2A and C). Furthermore, Ca²⁺‐bound water also participates in hydrogen bond to backbone oxygen of residue 748 from exiting loop, therefore forming a hydrogen bond network along with residue 737. Given that the O_737th‐O_water‐O_748th angle (110°) is close to the bond‐angle of water, and they are almost coplanar with the Ca²⁺ ions, such hydrogen bonds network is expected to be very stable and thus promotes the coordination of water with Ca²⁺. Along with the backbone oxygen atoms of residues 747 and 750 from the exiting loop, the other side‐chain oxygen atoms of residues 731, 733, 740, and 760 from the incoming loop coordinate with the second Ca²⁺, thus bridging the two Ca²⁺ ions into a coupled Ca²⁺‐pair with a distance as close as ˜ 4 Å. The LHL2 motif shares an almost identical backbone (the RMSD is 0.34 Å) to the LHL1 and harbors a pair of Ca²⁺ as observed in LHL1 (Fig 2B). Interestingly, Ca²⁺–Ca²⁺ coupling exists not only within each motif, but between motifs. As shown in Fig 2A, residue 9 in LHL2 (D760) which stabilizes the Ca²⁺‐bound water donates another side‐chain oxygen atom to the second Ca²⁺ in LHL1. Such interactions are thought to further strengthen the packing between motifs, therefore promote the folding of module 3. Additionally, E849 from module 4 was found to coordinate the second Ca²⁺ in LHL2.

Figure 2. Structural details of the novel double‐Ca²⁺‐binding motifs in tandem.

• A
  Ribbon diagram of module 3 showing a tandem of double‐Ca²⁺‐binding motifs with loop–helix–loop (LHL) topology (LHL1 [limegreen] and LHL2 [limon]). Residues involved in Ca²⁺ coordination are highlighted in stick representation. The Ca²⁺ ions and Ca²⁺‐bound water are shown in magenta and red spheres, respectively.
• B
  Superimposition of the LHL1 and LHL2 motifs together with the Ca²⁺ ions, with a backbone RMSD of ˜ 0.3 Å.
• C
  Ribbon diagram of a canonical EF‐hand (PDB: 2NXQ) and details of its Ca²⁺‐binding sites.
• D
  Sequence alignment of module 3 orthologs using ESPript 3.0. NCBI accession numbers for the respective bacteria are given in parentheses. The positions of secondary structure elements are indicated above the sequences (α represents α‐helix).
• E
  Amino acid composition comparison of the LHL motifs and a canonical EF‐hand (PDB: 1J7O).
• F, G
  The interaction networks between modules 3 and 2 (F) as well as between modules 3 and 4 (G)
• H
  Schematic diagram of LHL1m, LHL2m, and LHL12m mutants, which have D1N, D3A, E12N mutations in LHL1 and/or LHL2.

A search for homologous 3D structures in the Protein Data Bank (PDB) using the Dali Server did not reveal any protein with structures similar to MM, highlighting the novelty of the LHL double‐Ca²⁺ binding motifs in tandem. A subsequent search in the database of non‐redundant protein sequences (nr) showed that homologues to MM are mainly present in the genus Staphylococcus, and a few also exists in Firmicutes and Listeria monocytogenes. Sequence alignment revealed that residues involved in Ca²⁺ binding in each LHL motif as well as their surrounding residues are highly conserved across species, indicating that the integrity of the novel double‐Ca²⁺‐binding motifs in tandem could be crucial to its function (Fig 2D).

Cooperative interaction networks dictate high affinity Ca²⁺ binding

Complex protein interaction networks are observed among modules 2, 3, and 4. Specifically, the backbone oxygen and the side‐chain amide group of C‐terminal residue Q728 of module 2 form two hydrogen bonds with the backbone amide group and one side‐chain oxygen of D737 in the helix of LHL1, respectively, thus immobilizing the orientation of the LHL1 motif relative to module 2 and promoting the coordination of D737 with Ca²⁺ (Fig 2F). The long loop (aa 678–698) extruding from the end of module 2 also participates in long‐range interactions with the LHL1 and LHL2 motifs through formation of one salt bridge network (D731‐R693‐E751) and 7 hydrogen bonds (Fig 2F). Similarly, N847 in module 4 provides its backbone and one side‐chain amide groups to form two hydrogen bonds with the backbone oxygen atoms of Y767 and K768 in the helix of LHL2 (Fig 2G). The backbone oxygen of P793 and amide group of K795 in a long loop (aa 792–801) which adjoins the two β‐strands in the two different β‐sheets within module 4 form two hydrogen bonds with the backbone amide of G774 and the backbone oxygen of L772 from the exiting loop of LHL2, respectively. Besides coordination with the second Ca²⁺ in the LHL2, E849 in module 4 also forms a salt bridge with K758 in the LHL2 (Fig 2G). As module 3 is highly hydrophilic and negatively charged (Fig 2D and E), the interactions of module 3 with its neighboring modules 2 and 4 are predominated by hydrogen bonds or electrostatic interactions (Fig 2F and G).

To analyze the importance of the interaction networks, BSP subconstructs including BSP₁₂, BSP₃, BSP₄₅, BSP₁₂₃, BSP₃₄₅, and BSP₁₂₃₄, which consist of the first two, the third, the last two, the first three, the last three, and the first four modules of BSP, respectively, were constructed, assuming that only one player (BSP₁₂, BSP₃, BSP₄₅), two players (BSP₁₂₃, BSP₃₄₅), and all three players (BSP₁₂₃₄) of the interaction networks among modules 2, 3, and 4, are taking their places in the respective subconstructs. The interactions of Ca²⁺ with BSP and its subconstructs were studied using isothermal titration calorimetry (ITC) (Fig EV1, Appendix␣Table S2). Consistent with the crystal structure, the isotherm profile of BSP could be best fitted with a 5‐site sequential binding model that gave apparent dissociation constant K _ds of 0.12, 3.6, 14, 17, and 170 μM for the high to low affinity binding sites, respectively, or fitted with five equivalent single‐site model that gave K _d of 29 μM. The isotherm profiles of both BSP₁₂ and BSP₄₅ could be best fitted with a one‐site binding model and 100‐fold decreased binding affinities. Surprisingly, BSP₃ was unable to bind with Ca²⁺ at all. The conjunction of module 3 to BSP₁₂ or BSP₄₅ results in increased Ca²⁺‐binding affinity to BSP₁₂₃ and BSP₃₄₅ relative to BSP₁₂ and BSP₄₅, respectively. The concurrence of both modules 2 and 4 with module 3 in BSP₁₂₃₄ leads to similar mode and high affinity of Ca²⁺ binding as in BSP. These data support cooperative interaction networks among modules 2, 3, and 4 and suggest the importance of its structural integrity in dictating high affinity Ca²⁺ binding. These findings are further supported by decreased Ca²⁺‐binding affinities for BSP mutants of LHL1m, LHL2m, and LHL12m, which Ca²⁺‐binding sites were disrupted with residues corresponding to positions 1, 3, and 12 of LHL1 and LHL2 alone or in combination replaced by Asn, Ala, and Asn, respectively (Fig 2H).

Figure EV1. Ca²⁺ binding to BSP monomer and its subconstructs measured by ITC.

Representative ITC isotherms for Ca²⁺ binding to BSP, its subconstructs, and its mutants with calcium binding sites mutated in module 3. The respective isotherms fitted with N‐equivalent single‐site binding model (black) or N‐site sequential binding model (red) are shown.

The middle module 3 undergoes Ca²⁺‐dependent conformational␣switches

We further characterized BSP monomer in solution in the presence of 10 mM Ca²⁺ or 10 mM EDTA at pH 7.5 using SAXS (Fig EV2A–C). The experimental scattering profiles were fitted to the crystal structure of BSP monomer; however, neither fits well (χ² = 24.63 and χ² = 2.66, respectively; Fig 3A), indicating that BSP monomer in solution adopts conformation significantly different from the crystal structure. To dissect which modules cause the difference, SAXS data were collected for all the aforementioned BSP subconstructs under both conditions (Fig 3A). Interestingly, the experimental SAXS profiles of BSP₁₂, BSP₁₂₃, and BSP₁₂₃₄ in 10 mM Ca²⁺ agree well with the crystal structures, but not for BSP₄₅. Ab initio 3D shape reconstruction by DAMMIN (Svergun, 1999) as well as rigid‐body modeling using Xplor‐NIH (Schwieters et␣al, 2003) suggest an extended conformation for BSP₄₅ in solution, suggesting that the linker connecting module 4 and 5 is flexible in solution (Fig EV2D). The respective SAXS profiles in 10 mM EDTA were also fitted to the corresponding crystal structures and the rigid‐body model of BSP₄₅, respectively. While the SAXS profiles of BSP₁₂ and BSP₄₅ in 10 mM EDTA fit nicely to the crystal structure of BSP₁₂ and the rigid‐body model of BSP₄₅, respectively, the SAXS profiles of BSP₁₂₃, BSP₁₂₃₄, and BSP which all contain module 3 do not fit to the respective Ca²⁺‐bound crystal structures (Fig 3A). These data indicate that module 3 is responsible for the Ca²⁺‐dependent conformational changes in BSP and its subconstructs.

Figure EV2. Structural analysis of BSP, its subconstructs including BSP₄₅, BSP₃, and BSP₃₄, and its mutants including LHL1m, LHL2m, and LHL12m using SAXS and NMR.

• A–C
  The scattering intensities I(q) plotted against momentum transfer q (A), the pair distance distribution function PDDFs transformed from scattering profiles (B) and the dimensionless Kratky plots plotted as (qR _g)² I(q)/I(0) vs qR _g (C) of Bap_BSP monomer in the absence and presence of Ca²⁺. The inset in (A) is the Guinier analysis of the experimental scattering curves with q _max∙R _g < 1.3.
• D
  Two views of the ab␣initio 3D shape envelope of BSP₄₅ in 10 mM Ca²⁺ by DAMMIN (purple) superimposed with the rigid‐body atomic model of BSP₄₅ (purple cartoon) from Xplor‐NIH refinement.
• E, F
  Two views of the ab␣initio 3D shape envelopes of Bap_BSP monomer by DAMMIN superimposed with the best‐fit rigid‐body models derived from Xplor‐NIH refinement in 10 mM Ca²⁺ (E, blue) or 10 mM EDTA (F, green), respectively.
• G, H
  2D ¹H‐¹⁵N HSQC of BSP₃ in the absence (G) or presence (H) of Ca²⁺.
• I, J
  2D ¹H‐¹⁵N HSQC of BSP₃₄ in the absence (I) or presence (J) of Ca²⁺.
• K–M
  The scattering profiles (K), the PDDFs (L), and the dimensionless Kratky plots (M) of BSP monomer and its mutants of LHL1m, LHL2m, and LHL12m in 10 mM Ca²⁺.

Figure 3. The middle module 3 undergoes Ca²⁺‐dependent order–disorder conformational switches.

• A
  Experimental SAXS profiles of monomeric BSP and its subconstructs such as BSP₁₂, BSP₁₂₃, BSP₁₂₃₄, BSP₄₅, and BSP₃₄₅ in 10 mM Ca²⁺ (green) and 10 mM EDTA (red) are fitted to the crystal structures directly (black, 10 mM Ca²⁺; blue, 10 mM EDTA) or to the rigid‐body models refined using Xplor‐NIH against SAXS profiles (magenta, 10 mM Ca²⁺; cyan, 10 mM EDTA). The residual is defined as I _calc/I _exp.
• B
  The dimensionless Kratky plots of SAXS profiles of BSP₃ in 10 mM Ca²⁺ (red) and 10 mM EDTA (black).
• C, D
  Overlays of 2D ¹H‐¹⁵N heteronuclear single quantum coherence (HSQC) NMR spectra of BSP₃ (C) and BSP₃₄ (D) in 10 mM Ca²⁺ (red) and 10 mM EDTA (black).

The dimensionless Kratky plots for BSP₃ under both conditions exhibit steady increase at larger q (Fig 3B), suggesting that BSP₃ alone is intrinsically disordered protein (IDP) (Kikhney & Svergun, 2015). The IDP feature is further supported with heavily overlapped peaks in 2D ¹H‐¹⁵N HSQC NMR spectra of BSP₃ at both conditions (Figs 3C and EV2G and H). Given that module 3 is well folded in the Ca²⁺‐bound BSP structure (Fig 1C), Ca²⁺ depletion (in 10 mM EDTA) has a marginal effect on BSP₁₂ conformation but affects BSP, BSP₁₂₃, BSP₃₄₅, and BSP₁₂₃₄ significantly (Fig 3A; Appendix␣Table S3). Module 3 in such constructs must undergo Ca²⁺‐dependent order–disorder conformational switches and the interaction networks is essential for Ca²⁺‐induced folding of module 3. This prediction is further supported by the HSQC spectra of BSP₃₄ at both conditions (Figs 3D and EV2I and J), which show strong, well‐dispersed peaks (ordered) and weak, overlapped peaks (disordered), respectively.

Atomic models for BSP monomer in 10 mM Ca²⁺ or 10 mM EDTA were then built up to fit the experimental SAXS profiles by rigid‐body modeling using Xplor‐NIH (see Materials and Methods). The best fitting models can be nicely fitted into the respective ab␣initio 3D shape envelopes (Fig EV2E and F). Notably, the resulting atomic models for BSP₃₄₅ under both conditions agree well with the respective SAXS profiles (Fig 3A), further supporting that module 3 undergoes order‐to‐disorder conformational switch upon Ca²⁺ depletion.

BSP amyloid aggregation is pH‐dependent and Ca²⁺‐inhibited

The N‐terminal region of Bap was previously reported as aggregation‐prone and to self‐assemble into amyloid structure under acidic pHs and low [Ca²⁺] (Taglialegna et␣al, 2016). Similarly, BSP assembles into amyloid fibrils under acidic pH and low [Ca²⁺], as confirmed with Transmission Electron Microscopy (TEM) analysis (Fig 4A). To investigate the effects of pH and [Ca²⁺] on BSP amyloid formation, the kinetics of agitation‐induced aggregation was measured for BSP monomer at 5 μM in the pH range of 4.5–7.5 using the Thioflavin T (ThT) fluorescence assay (Arosio et␣al, 2015). BSP monomer exhibited a prominent pH‐dependent on/off aggregation switch in the presence of 10 mM EDTA (Fig 4B). BSP monomer did not show any obvious amyloid formation over 24 h when pH > 4.5, but aggregated rapidly at pH 4.5. As expected, the aggregation rate at pH 4.5 was faster at higher protein concentration (2–10 μM; Fig 4C). However, amyloid formation at each protein concentration was inhibited in the presence of 10 mM Ca²⁺ (Fig 4D).

Figure 4. Amyloid aggregation in␣vitro and biofilm formation in␣vivo by BSP and its subconstructs.

• A
  TEM image showing fibril morphology of BSP.
• B–D
  BSP amyloid aggregation in␣vitro is pH‐dependent in 10 mM EDTA (B), protein concentration‐dependent in 10 mM EDTA and at pH 4.5 (C) and strongly inhibited by millimolar Ca²⁺ at various protein concentrations (D).
• E–H
  TEM images showing fibril morphology of BSP₁₂ (E), BSP₁₂₃ (F), BSP₃ (G), and BSP₄₅ (H).
• I, J
  Amyloid aggregation by BSP subconstructs. BSP₁₂ exhibits more prominent aggregation propensity than BSP₃ and BSP₄₅ in 10 mM EDTA and at pH 4.5 (I). The conjunction of module 3 with modules 1 and 2 promotes aggregation (J).
• K
  Millimolar Ca²⁺ inhibits aggregation of BSP and its subconstructs. The inhibition efficiency is defined as the ratio of maximal fluorescence fold change at the end stage of measurement achieved in 10 mM EDTA to that in 10 mM Ca²⁺. Each experiment was repeated three times, data shown as mean ± SD.
• L
  Structural organization of BSP chimeric subconstructs. S (red), signal peptide; module 1 (magenta); module 2 (orange); module 3 (green); module 4 (blue); module 5 (cyan); R‐ClfA (yellow), clumping factor R domain; W (purple), cell wall anchor.
• M
  Bacterial clumping and biofilm formation of Staphylococcus aureus V329 and the corresponding S. aureus Δbap strain. For bacterial clumping (bottom), bacteria were cultured overnight with agitation (220 rpm) in LB‐glu (pH < 5). For biofilm formation (top), bacteria were cultured overnight at 37°C in TSB‐glu (pH < 5) using sterile 96‐well polystyrene microtiter TC‐plates under static conditions, assays were repeated for each bacteria.
• N
  Biofilm formation of S. aureus Δbap mutants expressing BSP chimeric subconstructs or carrying empty vector (EV) in the absence (cyan) or presence (red) of Ca²⁺ assayed with crystal violet staining. The EV is a control for S. aureus Δbap bacteria. Each experiment was repeated three times, data shown as mean ± SD.

Roles of individual modules in BSP amyloid aggregation in␣vitro

The predicted consensus amyloidogenic hotspots (Appendix␣Fig S1) were mapped to BSP monomer structure (Appendix␣Fig S3). Interestingly, the hotspots are distributed across all the modules except for module 3. To confirm whether the individual modules form amyloid fibrils, TEM analyses were performed. Isolated fibers can be observed in the aggregates formed by purified BSP₁₂, BSP₃, BSP₄₅, and BSP₁₂₃, although their morphology differs significantly. Specifically, fibrils of BSP₁₂ (Fig 4E) and BSP₁₂₃ (Fig 4F) display wave‐like structures, but aggregates of BSP₃ (Fig 4G) and BSP₄₅ (Fig 4H) show as ribbon‐like and rod‐like fibrils, respectively. These results ratify the amyloidogenic nature of the individual modules of BSP.

To probe the roles of the respective modules in BSP amyloid formation, the kinetics of agitation‐induced aggregation of the aforementioned BSP subconstructs in 10 mM EDTA and at pH 4.5 was measured and compared with BSP. As shown in Fig 4I, BSP₁₂ exhibited prominent aggregation at 5 μM, whereas BSP₃, BSP₄₅ did not show any obvious aggregation at 5 μM. It is very likely the aggregation kinetics of BSP₃ and BSP₄₅ are slow at 5 μM, as amyloid fibrils formed after longer incubation time by TEM analysis (Fig 4G and H), and BSP₃ and BSP₄₅ display obvious aggregation at much higher concentration (Fig EV3A). Therefore, the aggregation propensity of BSP could be mainly attributed to modules 1 and 2. Remarkably, BSP₁₂₃ displayed enhanced aggregation compared with BSP₁₂ and BSP₃ alone (Fig 4J), but aggregation was not enhanced when BSP₁₂ was mixed with BSP₃ in 1:1 ratio (Fig 4I), suggesting that the interactions among modules 1, 2, and 3 promotes BSP₁₂₃ amyloid formation. The effects of protein concentrations on aggregation kinetics for all the subconstructs were analyzed. Overall, increasing protein concentration accelerated aggregation of all the subconstructs (Fig EV3A). Moderate aggregations could be observed for BSP₃ and BSP₃₄₅ at 50 μM, and not for BSP₄₅ until at a concentration of 200 μM (Fig EV3A). Thus, the aggregation propensity of the Bap subconstructs at pH 4.5 and low [Ca²⁺] can be ranked as BSP₁₂₃ > BSP₁₂₃₄ > BSP > BSP₁₂ ≫ BSP₃ > BSP₃₄₅ > BSP₄₅ (Figs 4I and J, and EV3A). It is interesting to note that the aggregation of both BSP and BSP₃₄₅ was retarded compared with BSP₁₂₃ and BSP₃ (Figs 4I and J, and EV3A), respectively, suggesting that modules 4 and 5 play a role in retarding BSP amyloid formation at pH 4.5.

Figure EV3. Amyloid aggregation of BSP and its subconstructs at various solution conditions monitored by Thioflavin T fluorescence assay.

• A
  Time course of aggregation of BSP subconstructs in the absence (left) or presence (right) of Ca²⁺ at pH 4.5 as a function of protein concentrations.
• B
  pH‐dependent on/off aggregation switches are observed for BSP subconstructs in the absence (10 mM EDTA) of Ca²⁺.

Data information: The subconstructs in (A) and (B) include BSP₁₂, BSP₁₂₃, BSP₁₂₃₄, BSP₃, BSP₄₅, and BSP₃₄₅.

Aggregation kinetics of all the subconstructs were also investigated at elevated pH range of 5.5–7.5 in the presence of 10 mM EDTA. Striking pH‐dependent on/off aggregation switches were observed for all the subconstructs containing modules 1 and 2, such as BSP₁₂, BSP₁₂₃, BSP₁₂₃₄, BSP, and subconstruct BSP₃, whereas amyloid fibrils formation by BSP₄₅ and BSP₃₄₅ can be observed at all the pHs tested (Fig EV3B), indicating that module 1, 2, and 3 play an essential role in acidic pH‐induced BSP amyloid formation.

Next, we investigated the effects of Ca²⁺ on the aggregation of the respective subconstructs at pH 4.5. The kinetics of agitation‐induced aggregation for each subconstruct in 10 mM EDTA or 10 mM Ca²⁺ were measured simultaneously using the ThT fluorescence assay. The fluorescence fold change in each sample reaches its maximum at the end stage of the measurement. The inhibitory efficiency of Ca²⁺ to each subconstruct is defined by the ratio of the achieved maximum fluorescence fold change in 10 mM EDTA to that in 10 mM Ca²⁺. As shown in Fig 4K, the presence of Ca²⁺ exhibited prevailing inhibitory effects on all the subconstructs; interestingly, the strongest inhibitory effects were observed for BSP₁₂₃₄ and BSP. Unexpectedly, the inhibitory effects of Ca²⁺ on aggregation of BSP₁₂₃ and BSP₃₄₅ are subtle, despite the fact that module 3 in BSP₁₂₃ and BSP₃₄₅ in 10 mM Ca²⁺ is ordered. It is thus very likely that the cooperative interaction networks among modules 2, 3, and 4 in BSP₁₂₃₄ and BSP are preserved for the strongest inhibition, but the integrity of such interaction networks among modules 2, 3, and 4 in BSP₁₂₃ and BSP₃₄₅ is disrupted, as they lack of the interactions with module 4 or module 2, respectively, resulting in subtle inhibitory effects of Ca²⁺. The significance of the structural integrity of such interaction networks is further supported by structural and aggregation kinetics analyses on BSP mutants of LHL1m, LHL2m, and LHL12m which the Ca²⁺‐binding sites are disrupted. SAXS data shows that all mutants have larger D _max and R _g and enhanced flexibility as compared to BSP wild type in the presence of 10 mM Ca²⁺ (Fig EV2, EV3, EV4, EV5, Appendix␣Table S3), indicating that LHL1 and/or LHL2 mutations result in extended, disordered conformation of module 3, even in the presence of Ca²⁺. As the disordered module 3 loses its interaction networks with neighboring modules 2 and 4, the cooperative interaction networks are disrupted in the BSP mutants of LHL1m, LHL2m, and LHL12m; it is thus not surprising that only subtle inhibitory effects of Ca²⁺ on aggregations of such mutants are observed in 10 mM Ca²⁺. Accordingly, their aggregation kinetics are comparable to BSP in 10 mM EDTA under acidic condition (Fig 4K, Appendix␣Fig S4).

Figure EV4. Liquid–liquid phase separation and maturation of BSP and its subconstructs regulated by pH and Ca²⁺ .

• A
  Representative fluorescence microscopy images of BSP liquid droplets at various protein concentrations under pH 5.0. Scale bar, 25 μm.
• B
  Time‐dependent changes of liquid droplets of BSP analyzed by TEM imaging. After 10 days, amyloid‐like fibrils projects from droplets.
• C
  Aging of BSP droplets observed by confocal microscopy 83 days after the initiation of mixing, showing that some mesh‐like structures appeared with clustered amyloid‐like fibers. Condition: protein concentration, 20 μM; pH, 4.5; Scale bar, 5 μm.
• D
  Representative fluorescence microscopy images of BPS₁₂ (top) and BSP₁₂₃ (down) proteins at various protein concentrations under pH 5.5. Scale bar, 25 μm.
• E
  Time‐dependent changes of liquid droplets of BSP₁₂ (top) and BSP₁₂₃ (down) analyzed by TEM imaging. At day 2, all of protein shows droplet aggregates. At later time points, amyloid‐like fibrils emerges from droplets. At the end of the incubation period, the protein shows mesh‐like structures.
• F
  Phase separation behaviors of BSP, BSP₁₂, and BSP₁₂₃ are affected by Ca²⁺. Condition: protein concentration, 20 μM; pH, 5.5; Scale bar, 25 μm.

Data information: In (A–D), PEG8K concentration is 8%.

Figure EV5. SAXS analysis of BSP and its subconstructs at varying pHs in the absence or presence of Ca²⁺ .

• A
  Scattering profiles (left) and PDDFs (right) of BSP and its subconstructs at various pHs (4.5, magenta; 5.5, blue; 6.5, green; 7.5 black) in the presence of 10 mM EDTA.
• B
  Scattering profiles (left) and PDDFs (right) of BSP and its subconstructs at various pHs (4.5, magenta; 5.5, blue; 6.5, green; 7.5 black) in the presence of 10 mM Ca²⁺.
• C
  The radius of gyration (R _g) (solid lines and solid circles) and D _max (dashed lines and open circles ) derived from PDDFs of Bap_BSP and its subconstructs in the presence of 10 mM EDTA (black) or 10 mM Ca²⁺ (red) are plotted against varying pHs.

Roles of individual modules in Bap‐mediated biofilm formation␣in␣vivo

To further dissect the roles of the individual modules in Bap‐mediated biofilm formation in␣vivo, BSP and its variants including BSP₁₂, BSP₁₂₃, BSP₄₅, LHL1m, LHL2m, and LHL12m were cloned and introduced into bap‐deficient S. aureus V329 (denoted as S. aureus Δbap) (see Materials and Methods, Fig 4L). The bacteria were grown in media containing glucose, because of the accumulation of acidic by‐products from glucose fermentation, the pH drops below 5 when the bacteria reach stationary. After an overnight incubation S. aureus V329 strain clearly showed a biofilm adhered to the microtiter plate and bacterial clumps at the bottom and wall of the tube, while the corresponding Δbap strain did not (Fig 4M). While the empty vector (EV) showed no effects on mediating biofilm formation to S. aureus Δbap, all the BSP variants except for BSP₄₅ induce biofilm formation to S. aureus Δbap similar to that of the wild type S. aureus V329 (Fig 4N). As all the BSP variants except for BSP₄₅ contain modules 1 and 2, these results indicated that the N‐terminal lobe of BSP was sufficient to mediate biofilm development in␣vivo under acidic culture conditions. Notably, the presence of 10 mM Ca²⁺ inhibited the biofilm formation mediated by BSP, where the interaction networks among modules 2, 3, and 4 are preserved. In contrast, the inhibitory effects of Ca²⁺ were subtle on biofilm formation capacity of other BSP variants. These data were consistent with the inhibitory effect of Ca²⁺ on amyloid aggregation by BSP variants in␣vitro. Together, our data suggest that Bap amyloid aggregation in␣vitro and Bap‐mediated biofilm development in␣vivo under acidic conditions are mainly contributed by its N‐terminal lobe, and the inhibitory effects of Ca²⁺ highly depend on the structural integrity of the interaction networks among modules 2, 3, and 4.

BSP undergoes acidic pH‐induced but Ca²⁺‐inhibited LLPS and␣maturation

To understand the mechanism of acidic pH‐induced BSP amyloid aggregation, we studied the LLPS behavior of BSP. The tandem feature of modules 1 and 2 and the IDP property of module 3 is reminiscent of a variety of proteins that undergo LLPS and underlie the pathological amyloid assembly in a variety of neurodegenerative diseases (Aguzzi & Altmeyer, 2016; Babinchak & Surewicz, 2020), but the roles of LLPS in bacteria functional amyloid formation remain largely unexplored.

To examine the phase separation behavior of BSP, BSP was expressed and purified fused to the enhanced green fluorescent protein (eGFP) (BSP_eGFP) through a C‐terminal flexible linker (Fig 5A). To mimic the crowded environment of extracellular matrix, polyethylene glycol 8000 (PEG 8K) was included into the solutions. Different concentrations of BSP_eGFP protein (2–30 µM) were subjected to a wide range of pH values (7.5–4.5) and PEG 8K concentrations (2–8%), and analyzed by confocal microscopy (Figs 5B and C, and EV4A). BSP_eGFP exhibited pH‐dependent LLPS behavior and large droplets with fused structures can be distinctly observed at lower pH 4.5 (Fig 5B). We next assessed the mobility of BSP_eGFP between the droplet and the bulk phase by fluorescence recovery after photobleaching (FRAP) measurement (Fig 5D). FRAP experiment performed 15 min after mixing showed that the majority of the fluorescence signal (˜ 80%) recovered efficiently, suggesting that BSP_eGFP was highly dynamic, with rapid exchange of molecules between the droplets and the surrounding solution. However, the droplets undergo maturation, evidenced by slower rates and smaller extents of recovery when photobleaching was initiated 2 and 6 h after mixing, respectively (Fig 5D). The transition of BSP_eGFP droplets to amyloid‐like fibrils was further examined at various time points using TEM. While only droplets appeared at day 2, thin fibril‐like structures projected from the droplets at day 10, and larger fibrils were present adjacent to the droplets at day 18 (Fig EV4B). After 83 days, the material of BSP_eGFP droplets appeared with clustered amyloid‐like fibers detected by confocal microscopy (Fig EV4C), consistent with a liquid‐to‐solid transitions.

Figure 5. Liquid–liquid phase separation and maturation of BSP is regulated by pH and Ca²⁺ .

• A
  Schematic of BSP constructs with eGFP fusions.
• B
  Representative fluorescence microscopy images of BSP and its subconstructs (20 μM) at varying pHs. PEG 8K concentration is 2%. Scale bar, 20 μm.
• C
  Phase diagram of 20 μM proteins as a function of pH and PEG 8K concentration, which was scored by the presence or absence of droplets in the samples.
• D
  Time lapse FRAP of liquid droplets formed by fluorescently labeled BSP, BSP₁₂, and BSP₁₂₃. Bleaching was performed at the indicated time points after mixing, and the recovery occurring was recorded. Data are reported as mean ± SD. SD is depicted as shadows in a light color. n = 3 liquid droplets. n.s., not significant (P > 0.05); ***P < 0.001; (one‐way analysis of variance [ANOVA]).
• E
  Quantification of droplet area is shown for each protein under various Ca²⁺ concentration conditions. Each sample is quantified by total area (left) and mean area (right). Each experiment was repeated three times, data shown as mean ± SD.

To gain insights into the roles of individual modules, the phase separation behaviors of BSP subconstructs (BSP₁₂, BSP₃, BSP₄₅, and BSP₁₂₃) (Fig 5A) were characterized (Figs 5B and C, and EV4D). Similar pH‐dependent liquid droplets formation was observed for both BSP₁₂ and BSP₁₂₃. Unexpectedly, no obvious phase‐separated liquid droplets was observed for BSP₃ and BSP₄₅ under all the conditions. At pH 5.0 and the same protein concentration, BSP₁₂ formed larger droplets than BSP, and BSP₁₂₃ formed even larger ones when the pH was down to pH 4.5. All of BSP, BSP₁₂, and BSP₁₂₃ formed some fused droplets, but the droplet density for BSP₁₂ and BSP₁₂₃ dropped markedly as compared to BSP (Fig 5B). These data suggest that LLPS of BSP is mainly related to modules 1 and 2, which can be promoted by the IDP module 3 in the context of BSP₁₂₃, although module 3 alone does not undergo phase separation. FRAP experiments for both BSP₁₂ and BSP₁₂₃ also support time‐dependent droplets maturation (Fig 5D), the recovery efficiencies decrease against the time delays to initiate photobleaching after mixing. Notably, the decreasing rates of recovery efficiency for BSP and BSP₁₂₃ (15 min) are much faster than that for BSP₁₂ (6 h) (Fig 5D), indicating a faster rate of liquid‐to‐solid transition and a more loss of dynamicity in BSP and BSP₁₂₃. Such loss of dynamicity has been thought of as an aggregation phenomenon and often coincides with a loss of reversibility (Babinchak & Surewicz, 2020). This explanation is consistent with the above observations that BSP₁₂ has the lowest propensity to aggregation among these three proteins. The liquid droplets‐to‐solid fibrils transitions of BSP₁₂ and BSP₁₂₃ materials were further confirmed as similar to BSP using TEM analysis (Fig EV4E).

The effects of Ca²⁺ on LLPS of BSP₁₂, BSP₁₂₃, and BSP were also assayed (Fig EV4F). Generally, the presence of Ca²⁺ strongly inhibited the droplets size and area. However, at [Ca²⁺] larger than 10 mM, no obvious droplets formation can be observed for BSP (Fig 5E), smaller droplets can still be observed for BSP₁₂₃ and BSP₁₂, indicating that the inhibitory effect of Ca²⁺ on the LLPS of BSP is more efficient than its subconstructs of BSP₁₂₃ and BSP₁₂. Overall, Ca²⁺ binding causes similar inhibitory effects on the phase‐separated liquid droplets formation as that on amyloid aggregation of BSP subconstruct, and the inhibition efficiency also depends on the structural integrity of the interaction networks among modules 2, 3, and 4.

Ca²⁺‐binding counteracts pH‐induced conformational changes

To provide structural insights into the counteracting effects of Ca²⁺ binding on acidic pH‐induced LLPS and amyloid aggregation of BSP, conformational changes of BSP and its subconstructs at various pHs were probed by SAXS. The scattering profiles, the pair distance distribution function PDDF transformed from the scattering profiles, and the structural parameters of R _g and D _max derived from PDDFs for BSP and its subconstructs at various pHs are plotted in Fig EV5. In 10 mM EDTA, that the scattering profiles, the PDDFs, the R _g and D _max of BSP₁₂ and BSP₄₅ were similar across the varying pHs, by contrast, the scattering profiles, the PDDFs, the R _g and D _max of all the constructs containing module 3 such as BSP₃, BSP₁₂₃, BSP₁₂₃₄, BSP₃₄₅, and BSP were prominently affected by varying pH (Fig EV5B and C). These results suggest that module 3 plays an essential role in dictating the pH‐dependent conformational changes at low [Ca²⁺]. In 10 mM Ca²⁺, the conformations of BSP and all its subconstructs except for BSP₃ were very similar across the varying pHs (Fig EV5B and C), suggesting that millimolar [Ca²⁺] were able to counteract the effects of varying pH on BSP and its subconstructs. Notably, the presence of 10 mM Ca²⁺ causes further conformational compaction to all the constructs except for BSP₁₂, BSP₃, and BSP₄₅ under acidic conditions, consistent with disorder‐to‐order transitions of module 3 upon Ca²⁺ binding in the context of BSP₁₂₃, BSP₃₄₅, BSP₁₂₃₄, and BSP. These data further highlight the significance of cooperative interaction networks in mediating the counteracting effects of Ca²⁺ binding on pH‐induced conformational changes.

Discussion

A detailed mechanistic understanding of bacteria amyloid formation will be of great value in development of new anti‐biofilm therapeutics. In this work, we determined the crystal structure of the aggregation‐prone region of S. aureus Bap protein and studied its conformational dynamics in response to [Ca²⁺] and pH changes, and dissected the roles of the individual modules and the cooperative interaction networks in conferring and regulating BSP amyloid aggregation and biofilm formation. Our results identify modules 1 and 2 of BSP as the scaffold of LLPS and amyloid aggregation, and module 3 as the sensor of pH and [Ca²⁺] changes, consistent with the essential role of the N‐terminal in S. aureus biofilm formation (Taglialegna et␣al, 2016), allowing us to propose a mechanistic model for staphylococcal Bap in modulation of amyloid aggregation and biofilm formation in response to environmental signals (Fig 6).

Figure 6. Mechanistic model for staphylococcal Bap proteins to build amyloid scaffold biofilm matrices in response to environmental changes.

After secretion, Bap is proteolytically processed, and the N‐terminal containing regions A and B is released. Module 3 acts as a pH and Ca²⁺ sensor by significant conformational switches. When the pHs become acidic and [Ca²⁺] is low, modules 1 and 2 of BSP act as scaffold and aggregate into amyloid fibrils over time through LLPS and maturation, contributing to subsequent staphylococcal biofilm formation, which process can be promoted by intrinsically disordered module 3. When [Ca²⁺] reaches millimolar in the ECM, Ca²⁺ binding to module 3 stabilizes its ordered folding and the interaction networks among modules 2, 3, and 4, resulting in reduced conformational dynamics across BSP, hindering its self‐assembly and aggregation.

Liquid–liquid phase separation has been implicated in a variety of biological processes and pathological protein amyloid aggregation in neurodegenerative diseases (Aguzzi & Altmeyer, 2016; Babinchak & Surewicz, 2020). However, the role of LLPS in the formation of bacteria functional amyloid remains the least explored. So far, only a few examples of phase separation in bacteria have been reported (Al‐Husini et␣al, 2018; Monterroso et␣al, 2019; Wang et␣al, 2019; Yoo et␣al, 2019). To our knowledge, S. aureus Bap is one of the first bacterial proteins secreted into the extracellular matrices that undergoes LLPS and progressively aggregates into amyloid fibrils by liquid‐to‐solid maturation (Yoo et␣al, 2019). Our results thus support an important role of LLPS and maturation in bacteria functional amyloid formation. Recently, a variety of bacteria proteins including the CsgA, FapC, TasA, Aap, and Sbp have been identified to form functional amyloid (Van Gerven et␣al, 2018; Wang et␣al, 2018; Yarawsky et␣al, 2020). It would be important to decipher whether LLPS plays similar roles in the aggregation process of such proteins.

Our results provide mechanistic insights into amyloidogenesis of Bap proteins. Our structural data reveal that modules 1 and 2 are duplicated folded domains and module 3 is intrinsically disordered region at low [Ca²⁺], reminiscent of those multivalent proteins producing phase‐separated liquid droplets (Banani et␣al, 2017). Thus, LLPS of BSP is likely to be driven by multivalent weak interactions between modules 1 and 2. Our data reveal coherent pH dependence of BSP‐related conformational changes, LLPS, amyloid aggregation, and biofilm formation at low [Ca²⁺], establishing a general paradigm on the conversion between soluble protein and its high‐order amyloid as well as functional relevance (Wu & Fuxreiter, 2016). The pH‐dependent behaviors of BSP can be explained by the changes in surface charges upon varying pHs. BSP contains a total of 89 negatively charged residues (31 glutamic acids and 58 aspartic acids) against 48 positively charged residues (10 arginines and 38 lysines) and four histidines, most of these charged residues are exposed, so the surface of BSP at neutral pH is predicted to be negatively charged, likely resulting in Coulombic repulsion. As pH decreases from neutral to acidic, some aspartic acids and glutamic acids are expected to change from negatively charged to neutral, and some histidines are expected to change from neutral to positively charged. This would result in a reduction of surface electrostatic repulsion that would favor conformational compaction, concomitantly, evolving into liquid droplets. However, acidic pH‐induced phase‐separated liquid droplets are only metastable and prone to conversion into more stable, solid‐like fibrous states over time. Therefore, amyloid aggregation of BSP may proceed from acidic pH‐induced conformational compaction to phase‐separated liquid droplets, and to fibrous aggregates (Wu & Fuxreiter, 2016). Increased amyloid fibrils formed by Bap at low pHs could help biofilm formation under harsh conditions, which is of central importance to bacteria pathogenesis.

Another important finding in this study is the discovery of a tandem of novel double‐Ca²⁺‐binding motifs involved in cooperative interaction networks. Our crystal structure confirms a novel LHL motif in tandem structurally distinct from a variety of Ca²⁺‐binding motifs found in both the eukaryotic and prokaryotic Ca²⁺ binding proteins in recent years, including the canonical EF‐hand motif, EF‐hand like motifs, Ca²⁺‐binding β‐roll motif, Ca²⁺‐binding Greek key motif and the bacterial immunoglobulin‐like domains (Dominguez et␣al, 2015). Module 3 acts as a Ca²⁺ and pH sensor through disorder‐to‐order conformational switches, conferring the ability on bacteria to sense environmental signals and live under hash environmental conditions. The minimal amount of Ca²⁺ able to inhibit biofilm development is around 10–12 mM (Cucarella et␣al, 2004). The presence of Bap is particularly enriched in isolates from mammary glands in ruminant suffering from mastitis. In milk, the Ca²⁺ concentration is around 32 mM, whereas the concentration of free Ca²⁺ in mammalian blood is stringently maintained between 1.1 and 1.3 mM. Thus, fluctuations in the calcium concentration are probably great in the mammary gland environment (for example, between lactation and dry periods). In addition, sites where there are low local Ca²⁺ concentrations likely occur in an inflamed mammary gland as a result of increased permeability of blood vessels and concomitant altered ionic chemical balances. Mammary gland inflammation takes place not only in advanced mastitis infections but also in stress situations or during physiological changes (start of the dry period and around parturition). The observed calcium‐dependent bacterial behavior might have important implications for the propagation of bacteria. Initial rapid planktonic growth and horizontal spread may take place during lactation, and subsequent formation of a biofilm consisting of resistant multicellular communities in the mammary gland may result in chronic infections which persist in the animal, making it a bacterial reservoir. However, the Ca²⁺‐binding affinity, the inhibitory efficiency of Ca²⁺ on aggregation, the in␣vivo biofilm formation, LLPS, and the counteracting pH effects highly depend on the structural integrity of the cooperative interaction networks, suggesting an allosteric pathway to propagate environmental signals (Xiao et␣al, 2017). The detailed regulatory mechanisms of the interaction networks among modules 2, 3, and 4 require further studies.

Recently, the N‐terminal of Esp, a Bap‐orthologous protein produced by Enterococcus faecalis, displays a similar amyloidogenic behavior as Bap under acidic pH (Taglialegna et␣al, 2020). As Bap is present in coagulase‐negative staphylococci, and many other bacteria use Bap‐like proteins to build the biofilm matrix, the structural mechanism described here for S. aureus Bap may be widespread among pathogenic bacteria and of great value in search for antibacterial and anti‐biofilms compounds.

Materials and Methods

Amyloidogenic hotspots prediction

The propensity of the peptide segments across full‐length Bap to form amyloid fibrils is analyzed using AMYLPRED2 (http://aias.biol.uoa.gr/AMYLPRED2/) server, which employs a consensus of different methods that have been specifically developed to predict features related to the formation of amyloid fibrils. The consensus of short peptide segments in Bap is predicted as in Appendix␣Fig S1.

Construct cloning

The DNA fragments encoding Bap_B (amino acids 361–784) and Bap_BSP (amino acids 361–947) of Bap were PCR amplified from the pBT2 plasmid encoding full‐length Bap of S. aureus (GenBank: AAK38834.2), then subcloned into pET‐28a plasmid with an N‐terminal hexa‐histidine tag and verified by DNA sequencing. All the other mutants were generated using the TIANGEN's Fast Mutagenesis System.

Protein expression and purification

Plasmids containing the respective constructs were transformed into Escherichia coli BL21(DE3) competent cells and cultured in Luria‐Bertani media. Protein expression was induced with 0.5 mM IPTG (isopropyl β‐d‐1‐thiogalactopyranoside) when OD600 reached 0.8. Bacteria were further cultured at 37°C for 4 h and then centrifuged at 5,842 g, 4°C for 30 min. Cell pellet was resuspended with lysis buffer (20 mM Tris–HCl pH 7.4, 500 mM NaCl) and then homogenized by a high‐pressure cell disruptor system. 1 mM PMSF (phenylmethyl sulfonylfluoride) was added to lysate during this process. After centrifugation at 18,500 g for 60 min, the supernatant containing the His‐tagged protein was collected and loaded onto a Ni‐NTA affinity resin which was pre‐equilibrated with the binding buffer (20 mM Tris–HCl pH 7.4, 500 mM NaCl). The resin was extensively washed using wash buffer1 (20 mM Tris–HCl pH 7.4, 1 M NaCl, 20 mM imidazole) and wash buffer2 (20 mM Tris–HCl pH 7.4, 200 mM NaCl, 20 mM imidazole) successively. Target protein was eluted using buffer containing 20 mM Tris–HCl pH 7.4, 100 mM NaCl, 250 mM imidazole. His‐tag was cleavaged by TEV protease which then was removed by anion‐exchange chromatography by using HiTrap Q column (GE Healthcare). Preparative size‐exclusion chromatography (SEC) was performed for further purification using HiLoad 16/600 Superdex 75 or 200 column (GE Healthcare) depending on the molecular weights of desired proteins. Fractions containing target protein were visualized with SDS–PAGE gel and pooled together. Selenomethionine‐labeled Bap_B was overexpressed in E. coli BL21(DE3) B834 strain cultured with selenomethionine‐supplemented M9 medium as described by the manufacturer (Molecular Dimensions), and purified using the same protocol as above.

Crystallization, data collection, and structure determination

Proteins prior to crystallization trial were concentrated to about 20 mg/ml in buffer consisting of 20 mM Tris–HCl pH 7.4, 100 mM NaCl, 10 mM CaCl₂. Both initial screening and optimization were performed using the sitting‐drop or hanging‐drop vapor diffusion method. The crystallization optimization of Bap_BSP protein was performed by mixing 1.3 µl protein with 1 µl crystallization buffer containing 0.1 M Sodium Citrate pH 5.0, 10% 2‐propanol, 26% PEG 400 at 16°C. The crystallization optimization of Bap_B protein was performed by mixing 3 μl protein with 1 μl crystallization buffer consisting of 0.1 M Sodium Cacodylate pH 6.5, 40% MPD, 5% PEG 8000 at 16°C. The crystallization optimization of Selenomethionine‐labeled Bap_B protein was performed by mixing 2.5 μl protein with 1 μl crystallization buffer containing 0.1 M Sodium Cacodylate pH 6.5, 38% MPD, 5% PEG 8000. For cryoprotection, the single crystals were fished with nylon loops gently and dipped into crystallization mother liquor for several seconds before transferred into liquid nitrogen.

X‐ray diffraction data from all the crystals were collected at the BL17U and BL19U beamlines of Shanghai Synchrotron Radiation Facility (SSRF). The data were indexed, integrated, merged, and scaled with HKL2000 (http://www.hkl‐xray.com) or X‐ray Detector Software (XDS). The phasing of the Bap_B crystal structure was performed by the selenium single‐wavelength anomalous dispersion method using Phaser from the program PHENIX or SHELX. For the Bap_BSP crystal structure, the Bap_B structure was used as a search model to determine the phase. Initial models were built using the AutoBuild wizard in PHENIX, and these models were further refined by iterative cycles of manual model building using COOT and refinement against initial X‐ray diffraction data with PHENIX. Validation of the refined models was carried out using MolProbity on the PHENIX platform. Ramachandran analysis of the backbone dihedral angles indicated that all residues are located in the most favorable and generally allowed regions. The data collection and structural refinement statistics are summarized in Appendix␣Table S1.

Isothermal titration calorimetry

All the ITC measurements were performed with a MicroCal ITC 200 calorimeter (GE Healthcare) at 25°C, and the data were processed with the Origin7 software package from MicroCal. Purified Bap_BSP and its subconstructs were extensively exchanged into buffer of 50 mM Tris–HCl pH 7.5, 100 mM NaCl using a Superdex 75 or 200 10/300 GL column (GE Healthcare), then concentrated to 20–800 µM and placed in the cell. The syringe was filled with 2–10 mM CaCl₂ solution. Measurements were repeated twice, and similar results were obtained. The background data obtained from the buffer blank titration were subtracted before the data analysis.

Small angle X‐ray scattering

The data collection and processing procedures are similar as described before (Zhang et␣al, 2019). Briefly, SAXS measurements were carried out at room temperature at the beamline 12 ID‐B of the Advanced Photon Source, Argonne National Laboratory or the beamline BL19U2 of the Shanghai Synchrotron Radiation Facility (SSRF). The scattered X‐ray photons were recorded with a PILATUS 2 M detector (Dectris) at 12 ID‐B and a PILATUS 100k detector (Dectris) at BL19U2. The setups were adjusted to achieve scattering q values of 0.005 < q < 0.89 Å⁻¹ (12ID‐B) or 0.009 < q < 0.415 Å⁻¹ (BL19U2), where q = (4π/λ)sinθ, and 2θ is the scattering angle. Thirty two‐dimensional images were recorded for each buffer or sample solution using a flow cell. The 2D images were reduced to one‐dimensional scattering profiles using MATLAB (12ID‐B) or BioXTAS Raw (BL19U2). Scattering profiles of the proteins were calculated by subtracting the background buffer contribution from the sample‐buffer profile using the program PRIMUS (Konarev et␣al, 2003) following standard procedures. Concentration series measurements (fourfold and twofold dilution and stock solution) for the same sample were carried out to remove the scattering contribution due to inter‐particle interactions and to extrapolate the data to infinite dilution. The forward scattering intensity I(0) and the radius of␣gyration (R _g) were calculated from the data of infinite dilution at␣low q values in the range of qR _g < 1.3, using the Guinier approximation: lnI(q) ≈ ln(I(0)) − R _g ² q ²/3. These parameters were also estimated from the scattering profile with a broader q range of 0.006–0.30 Å⁻¹ using the indirect Fourier transform method implemented in the program GNOM (Svergun, 1992), along with the pair distance distribution function (PDDF), p(r), and the maximum dimension of the protein, D _max. The volume‐of‐correlation (V _c) was calculated using the program Scatter, and the molecular weights of solutes were calculated on a relative scale using the R _g/V _c power law developed by Rambo and Tainer (2013), independently of protein concentration and with minimal user bias. The theoretical scattering intensity of the atomic structure model was calculated and fitted to the experimental scattering intensity using CRYSOL (Svergun et␣al, 1995).

Low resolution bead models were built up with the program DAMMIN, which generate models represented by an ensemble of densely packed beads (Svergun, 1999), using scattering data within the q range of 0.006–0.30 Å⁻¹. 32 independent runs were performed, and the resulting models were subjected to averaging by DAMAVER (Volkov & Svergun, 2003) and were superimposed by SUPCOMB (Kozin & Svergun, 2001) based on the normalized spatial discrepancy (NSD) criteria and were filtered using DAMFILT to generate the final model. NSD is a measure of quantitative similarity between sets of three‐dimensional models, if two models systematically differ from each other, their NSD exceeds 1, for identical objects, it is 0.

With the available coordinates for the individual modules, rigid‐body modeling was carried out to construct atomic models for Bap_BSP in 10 mM Ca²⁺ or 10 mM EDTA using Xplor‐NIH package (Schwieters et␣al, 2003), during which the respective individual modules were kept as rigid bodies (modules 1–4 and module 5 as two rigid bodies for BSP in 10 mM Ca²⁺, module 1–2, module 4, and module 5 as three rigid bodies for BSP in 10 mM EDTA), the linkers between were allowed to translate or rotate freely, and a simulated annealing algorithm was performed to optimize the best position and orientation of the individual modules against SAXS data.

NMR spectroscopy

To make uniformly ¹⁵N‐labeled BSP₃ or BSP₃₄ samples for NMR analysis, cells were grown in 2L LB medium at 37°C until A₆₀₀ was ˜ 0.8. Cells were then spun down, washed with 1 × M9 salt, and transferred into 1 l of M9 minimal medium containing ¹⁵NH₄Cl as the sole nitrogen sources to produce ¹⁵N‐labeled proteins. The samples used for NMR experiment are as follows: 0.5 mM ¹⁵N‐labeled BSP₃ or BSP₃₄ in buffers containing 10% D₂O (v/v), 20 mM K₂HPO₄/NaH₂PO₄, 100 mM NaCl, 50 μM NaN₃, pH 7.20, with or without additional 10 mM Ca²⁺. 2D ¹H‐¹⁵N HSQC spectra were acquired at 25°C on a Bruker 600 MHz spectrometer equipped with a cryogenic probe (Bruker BioSpin GmbH, Germany). All data were processed using the software package NMRPipe (Delaglio et␣al, 1995) and analyzed with NMRView (Johnson, 2004).

Structural illustration

All of the illustrations of atomic models were generated using the PyMOL Molecular Graphics System. The bead models generated by DAMMIN were transformed into volumetric maps using Situs (Wriggers, 2012) and displayed using Chimera (Pettersen et␣al, 2004).

Aggregation kinetics

Protein samples were diluted by different pH buffer with 10 mM EDTA or 10 mM Ca²⁺ to the final concentration. 100 μM thioflavin T (ThT) was added into samples before using PerkinElmer Enspire plate reader monitoring fluorescence during amyloidogenesis. The protein samples were shaken at 800 rpm under the temperature of 25°C. Every 15 min, the fluorescence intensity was measured through the bottom of the plate with excitation filter of 440 nm and emission filter of 480 nm, to record the fluorescence intensity results at this time. At the same time, the buffer without protein was measured as the background ThT fluorescence, so the fluorescence intensity of the protein sample was divided by the fluorescence intensity of the ThT‐only sample after subtracting initial fluorescence intensity value, to calculate fluorescence fold change. In addition, the inhibitory efficiency of Ca²⁺ on aggregation is defined as the ratio of fluorescence fold change achieved at the end stage of the measurements in 10 mM EDTA to that in 10 mM Ca²⁺.

Construction of shuttle vectors

The signal peptide (S) and the R region of clumping factor A gene containing an LPXTG motif were cloned from the pCN51 plasmid; the different domains of the Bap protein were cloned from the BSP‐pET‐28a plasmid. Plasmid DNA was isolated from E. coli or S. aureus RN4220 strains whose cells were lysed by lysostaphin (12.5 μg/ml, Sigma‐Aldrich) at 37°C for 2 h before plasmid purification, then transformed into S. aureus by electroporation, using a previously described procedure (Cucarella et␣al, 2001).

Biofilm formation assay in␣vivo

Briefly, S. aureus strains were grown overnight in TSB at 37°C. The culture was diluted 1:40 in TSB containing 0.15% glucose, 10 μg/ml erythromycin, 1 μM CdCl₂, 10 mM, or 20 mM CaCl₂. 200 μl of this diluted culture was used to inoculate sterile 96‐well polystyrene microtiter TC‐plates (Corning). After 24 h, the wells were gently washed twice by submerging in sterile water, dried overnight, and stained with 0.1% crystal violet for 10 min. Wells were gently rinsed twice with sterile water, dried, and eluted with 33% acetic acid, and then, the absorbance was determined at 595 nm (Perkin EnVision instruments). Each assay was repeated three times.

In vitro amyloid fibril formation

BSP and its subconstructs proteins at a concentration of 5 µM in 20 mM sodium acetate, 80 mM NaCl, pH 4.5 were incubated for 2–10 days at 37°C with rotary shaking at the same speed as that in aggregation kinetics assay. Before negative staining, the protein solution was gently mixed, then subjected to TEM characterization at certain time points.

Transmission electron microscopy

10 μl of protein fibrils was directly deposited onto a glow‐discharged 200 mesh carbon‐coated copper grid (BZ11022a, Zhongjingkeyi, China) and incubated for 5 min, then stained with 2% (w/v) uranyl acetate for 1 min and air‐dried before imaging. The stained samples were imaged using a HT7800 TEM (Hitachi) operated at 80 kV or 120 kV.

In vitro phase separation assay

In vitro phase separation assay was performed in different buffer mixed with PEG 8000 (Sigma). Droplets were assembled in 384 low‐binding multi‐well microscopy glass plates (384‐well microscopy plates), sealed with optically clear adhesive film to prevent evaporation and observed under a Nikon A1 microscope equipped with 100× oil immersion objectives.

Fluorescence recovery after photobleaching (FRAP)

For the in␣vitro experiments, FRAP was carried out with samples in 384‐well microscopy plates using a Nikon A1 microscope equipped with 100× oil immersion objectives. Droplets were bleached with a 488‐nm laser pulse (three repeats, 100% intensity, dwell time 0.5 s). FRAP analysis involved assessment of mean fluorescence intensity from each bleached region of interest as well as non‐bleached regions and background to correct for microscope drift. The recovery from photobleaching was recorded for the indicated time.

Author contributions

Project conception and supervision: XF Majority of the experiments and data analysis: JM, XC. Crystal structure analysis: JM, ZX. SAXS and NMR data analysis: JM, YZ. Diffraction data collection: JM, SF. SAXS data collection: XZ. Bap plasmid, S. aureus strains, and manuscript editing: JV, IL. Manuscript writing: JM, XF.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Review Process File

Appendix

Expanded View Figures PDF

The EMBO Journal (2021) 40: e107500.

Data availability

The datasets produced in this study are available in the following databases:

• Protein structure data: PDB 7C7R (https://www.rcsb.org/structure/7C7R), PDB 7C7U (https://www.rcsb.org/structure/7C7U).

• All the other data supporting the findings of this study are available within the paper and its appendix files and also available from the corresponding author on reasonable request.