NMR insights into the pre-amyloid ensemble and secretion targeting of the curli subunit CsgA

The biofilms of Enterobacteriaceae are fortified by assembly of curli amyloid fibres on the cell surface. Curli not only provides structural reinforcement, but also facilitates surface adhesion. To prevent toxic intracellular accumulation of amyloid precipitate, secretion of the major curli subunit, CsgA, is tightly regulated. In this work, we have employed solution state NMR spectroscopy to characterise the structural ensemble of the pre-fibrillar state of CsgA within the bacterial periplasm, and upon recruitment to the curli pore, CsgG, and the secretion chaperone, CsgE. We show that the N-terminal targeting sequence (N) of CsgA binds specifically to CsgG and that its subsequent sequestration induces a marked transition in the conformational ensemble, which is coupled to a preference for CsgE binding. These observations lead us to suggest a sequential model for binding and structural rearrangement of CsgA at the periplasmic face of the secretion machinery.

chaperones CsgA through the CsgG pore 22 . CsgF forms a tight complex with the extracellular face of CsgG, where it co-ordinates templating of curli fibres via an interaction with the nucleator CsgB 27-29 , The amyloid fold of CsgA within the curli fibre has been extensively characterized, using data from a wide range of biophysical techniques [30][31][32] , Complementing experimental work on the fibrous state of CsgA, a model of CsgA nucleation and assembly has been proposed using ThT-binding studies 33 , and H/D exchange experiments have identified the most important repeats for driving amyloid formation 34 . Whilst these have provided valuable insights into fibre assembly, structural analysis of the pre-amyloid state of CsgA present within the perisplam has been more difficult. This is largely due to its disordered and transient nature. Indeed, much of our understanding of CsgA as an IDP comes from investigating other csg-encoded proteins and their relationship with CsgA. Previous work revealed a transient electrostatic mechanism for CsgC inhibition of CsgA aggregation 35 . Similarly, studies from the CsgE perspective have also highlighted the importance of transient electrostatic interfaces in controlling CsgA polymerization 36 . Studies have also shown that CsgA can cross seed homologs from other bacterial species, which raises the notion that curli fibers may facilitate multispecies biofilms 37 . The seeding specificity of curli is a more general phenomenon, as CsgA is able to alter the fibrillation kinetics of several human amyloidogenic proteins 38 .
This study provides new structural insights into the pre-amyloid CsgA ensemble and its targeting to the curli secretion system using solution-state nuclear magnetic resonance (NMR) spectroscopy. Structural features adopted by CsgA in solution were characterised by gleaning information from chemical shifts and transverse relaxation. These highlight the importance of prion-like motifs as points of conformational exchange in the repeat regions, and the abundance of polyproline II (PPII) in the disordered ensemble. Removal of the N-terminal targeting region (N 22 ) causes a significant conformational redistribution within the repeat regions of CsgA, particularly near important regions that gate-keep amyloid formation, suggesting a regulatory role. Furthermore, we show that CsgE recognises the truncate in preference to the mature protein that includes N 22 . While the extrapolation of our conclusions to the bacterial cell has limitations, as we ignore the influence of other cellular factors, our work supports a model in which a step-wise progression of structural re-arrangements in CsgA occur during export and secretion. First, intramolecular interactions involving N 22 together with CsgC assist in preventing inappropriate amyloid formation and premature binding of CsgE. Then, N 22 is bound specifically by the periplasmic face of outermembrane CsgG where it is sequestered from pre-fibrillar CsgA. Finally, this alters the conformation distribution of the CsgA ensemble and primes CsgA for interaction with CsgE. Subsequent interaction with of the CsgE ring with the base of the CsgG pore drives encapsulation of CsgA and entropic release through the pore.

Results
Backbone assignment of the disordered, pre-fibril CsgA. Prior characterisation of CsgA by solution-state NMR had been hampered by low signal intensity, spectral overlap, and limited life-time of samples. To address these issues, we optimised several aspects of sample and experimental conditions. Low sensitivity is largely a consequence of low sample concentrations, typically sub 20 μM. We chose to use C-terminally His-tagged CsgA for rapid purification and increased yields. Previous studies have shown that his-tagged and untagged CsgA behave similar in amyloid aggrateion assays 39 .As efforts to increase the sample concentration by centrifugal methods promotes amyloid formation and precipitation, induction times for expression were lengthened in order to produce protein at higher intracellular concentrations. To abate fibril formation and also mimic the periplasmic state of CsgA, the curli inhibitory chaperone CsgC was added to samples at a 1:40 ratio. 1 H-15 N HSQC spectra were recorded in the presence and absence of CsgC, revealing no prominent chemical shift perturbations, consistent with previous work revealing the transient nature of the CsgC-CsgA encounter 35 (Supplementary Figure S1). Further improvements were obtained by the reduction of sample temperature to 10 °C, which also favourably reduced the rate of unwanted amyloidogenesis. Use of higher strength magnetic fields (950 MHz) improved resolution for regions with the limited dispersion that is typical of IDPs. Final samples were 150 μM and stable for several days.
Assignment experiments were carried out using triple resonance methodology, with the hNcocaNNH pulse sequence particularly useful for overcoming ambiguity in overlapped regions. Despite narrow 1 H chemical shift dispersion (7.6-8.4 ppm), 1 H-15 N correlations were assigned for 118/134 (88.1%) of residues (Fig. 2). The C-terminal His-tag is excluded from this count. A majority of the unassigned residues are located in, or proximal to, the glycine-rich repeats within the N 22 . Prolines, P24 and P41, are also unassigned, as are residues located after the N-terminal methionine (G21-V23). (red) are flanked by linker regions (grey). Residues annotated in red are located in the most amyloidogenic repeats (R1 and R5), and have been reported to be critical for curli formation 40 . Residues annotated in blue act as gatekeeper residues that act to reduce the rate of CsgA aggregation prior to templating at the cell surface 41 . Numbering includes the signal sequence that is cleaved after perisplamic translocation. www.nature.com/scientificreports www.nature.com/scientificreports/ the n 22 region influences the conformational ensemble of CsgA. Next, using our backbone assignment, we sought to characterize the curli pathway from the perspective of CsgA. The binding affinities of the export complex were assessed using surface plasmon resonance (SPR). In this experiment, an NTA sensor chip was activated using Ni + and the CsgG-His complex was immobilized via the His-tag, with a peptide comprising N 22 from CsgA being injected. Experiments with full length CsgA were not successful, due to the aggregation of CsgA on the SPR chip. Our N 22 peptide data showed a specific interaction with a dissociation constant in the micromolar range (0.53 μM; Fig. 3). This is broadly consistent with a value of 28.3 μM, recently measured for the www.nature.com/scientificreports www.nature.com/scientificreports/ shorter N 6 peptide using isothermal calorimetry. The CsgG-N 22 interaction likely reflects the specific interaction with full-length CsgA, as it has been shown the N 22 peptide is not part of the amyloid fold and it is sufficient to direct secretion of heterologous proteins to the cell surface 21 . The interaction between the N 22 and CsgG would therefore sequester the N-terminal region from the dynamic solution state ensemble. Therefore, to probe differences in the conformational preference of CsgA between sequestered and un-sequestered N 22 , we produced a truncation of CsgA lacking this region (CsgA ΔN22 ).
The 1 H-15 N HSQC spectra of CsgA ΔN22 reveals significant chemical shift differences (Fig. 4A), which indicates changes in the conformational distribution of the disordered ensemble. In order to map these changes to the CsgA sequence, independent assignment experiments were carried out for CsgA ΔN22 and chemical shift perturbations (CSPs) plotted against the CsgA sequence (Fig. 4B). Outside of shifts proximal to the N 22 region, which would be a direct result of the truncation, two main clusters are observed where the largest changes occur. Residues L59, T61, and D62 are located within repeat R1, which is essential for amyloid formation 40 . Interestingly, G108, N110, S111, and M113 are located in the linker between the amyloidogenic R3 and the non-amyloidogenic R4 40,19 , and are flanked by gatekeeper residues D91, D104, G123 and D127, which prevent the repeat regions from participating in nucleation events 41 .
To further investigate differences between the molecular ensembles of CsgA and CsgA ΔN22 , transverse relaxation rates (R 2 ) were measured for the amide resonances. This parameter is sensitive to conformational dynamics on the fast time scale (ps-ns), and also influenced by slower exchange processes (μs-ms) 42 . It is often used to identify areas of backbone flexibility and conformational exchange processes. The R 2 rates of CsgA are generally low across the sequence, in the range of 10 s −1 , typical of an IDP due to the inherent and high flexibility (Fig. 4C). In contrast, CsgA ΔN22 has significantly elevated R 2 values across the whole sequence, approximately twice that for CsgA, suggesting that the truncate exhibits conformational exchange on a μs-ms time scale. In order to probe for μs-ms dynamics further, CPMG experiments were conducted on both samples, however no dispersion was detected. The dynamics indicated by the increased R 2 values suggest they may arise from intramolecular conformational exchange or self association on the faster end of the μs timescale. A few short sequence regions of CsgA, proximal to the glycine-rich motifs of repeats, exhibit elevated transverse relaxation (as denoted by the blue bars in Fig. 4C). A previous study identified these hexapeptide motifs (Q-X-G-G/F-G-N) as similar to to the highly amyloidogenic peptide regions of animal and yeast prion proteins, and demonstrated they are vital for curli formation 43 . The R 2 values observed here appear to support their findings, and highlight that important structural transitions appear to occur in these motifs. Although it has been shown in vitro that both CsgA ΔN22 and CsgA readily formed amyloid fibrils on broadly similar timescales, the aggregation kinetics display different protein concentration dependencies 44 .
To better gauge how differences in the chemical shifts and R 2 relaxation rates related to the conformational distribution in the CsgA/CsgA ΔN22 ensembles, secondary structure propensities (SSPs) were calculated. We used the δ 2D method to calculate propensities from chemical shift data, as it is optimally parameterized for disordered proteins. As would be expected for IDPs, the major SSP in CsgA is random coil, however there are noticeable sub-populations of other motifs (Fig. 5A). In full-length CsgA, the major non-coil propensities are PPII helices. This type of secondary structure has high conformational flexibility, providing a low energy barrier for conversion to other structural motifs, and has been observed as an important intermediate in transitions from disordered states into the amyloid fold 45,46 , Truncation of the N 22 causes a transition away from PPII, particularly in the linker  www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
The rapid polymerization of CsgA into the functional curli amyloid is critical for construction of the E. coli biofilm. However, whilst the time-scale for this process favours the organism, it presents a challenge for structural studies. In this work, we addressed this challenge by optimising solution conditions and present an NMR characterisation of the CsgA pre-amyloid state, both prior to and after capture by the secretion machinery. The utilisation of the NMR assignment has allowed us to observe structural motifs within the intrinsically disordered monomer. Non-coil secondary structure propensities in free CsgA favour the PPII conformation that has been found important in amyloid transitions 45,46 , As expected, there are minimal β-sheet propensities that are concentrated in the core-forming repeat regions. Transverse relaxation rates are elevated in the prion-like motifs of the repeat regions, suggesting these are important sites of conformational exchange en route to amyloid precipitation. Our work has shown that the N-terminal N 22 region binds specifically to CsgG, and the structural details of this has been recently illuminated with the cryo-electron microscopy structure of the CsgG-CsgF-CsgAN 22 complex 47 . A six residue stretch (V22-G27) is bound in a surface channel on the periplasmic face of CsgG, with conserved residues V23, Q25, and Y26 making the most important contacts. We sought to characterize the conformation upon recruitment to the secretion pore, and therefore produced a truncate -CsgA ΔN22 . Widespread chemical shift differences in CsgA ΔN22 suggested a switch in the disordered ensemble, concentrated in clusters of residues previously identified as gatekeepers for prevention of amyloidosis. Further characterisation of the truncate revealed transverse relaxation was approximately twice the rate compared with the full-length protein, and also that PPII propensity collapses, coupled with the emergence of a prominent α-helical motif between R1 and R2. These conformational perturbations suggest that the N 22 has a dual-purpose, first directly targeting CsgA to CsgG and once bound by CsgG inducing a conformation transition in CsgA that facilitates secretion through the CsgG pore.
It is important to note that the N 22 region is not cleaved from CsgA, therefore a mechanism must exist for its release from the periplasmic face of CsgG prior to secretion. Indeed, the structural transitions observed in our truncate provide new insight into how this may occur. Interactions with CsgE, which regulates periplasmic substrate export, were observed from CsgA ΔN22 , but was not detectable with full length CsgA. Therefore, combining results from this study with the structure-function analysis of CsgE 36 , we present a model for CsgA recognition and the interplay with the CsgE cap (Fig. 6C). In short, we propose that CsgA is kept in it monomeric state with in the perisplasm by freely diffusing CsgC, until it is recruited to CsgG via the specific interaction with its N-terminal N 22 region. Subsequent sequestration of N 22 induces a conformational re-distribution of the disordered CsgA ensemble, which promotes interactions with CsgE and could also assist in stabilising the CsgE nonamer. This notion is also consistent with recent pull-down assays showing an interaction between CsgE and CsgA ΔN22 , but not CsgA 47 . The CsgE oligomer then plugs the periplasmic aperture of the CsgG pore and releases the N 22 region. Transient electrostatic interactions from CsgE stabilise the unfolded state of CsgA in its captured state preventing premature amyloid accumulation. Furthermore, oligomerised CsgE encloses the CsgA substrate within the CsgG-CsgE vestibule, reducing its conformational flexibility and the entropy. The release of this high energy intermediate to the extracellular milieu contributes to the driving force for secretion.
A persistent feature of pre-amyloid in curli and other functional amyloid systems is the role of weak transient interactions in preventing amyloid formation and guiding the protein for export and subsequent templating 35,36,[48][49][50] , The transience of interactions between CsgC/CsgE and CsgA, as observed by fast exchange behaviour in NMR titration experiments 35,36,49 , is likely important for the progressive handling and delicate manipulation of the dynamic pre-amyloid ensemble. Such fine tuning of this early stage of amyloid formation enables the bacteria to deliver subunits in the appropriate state for efficient and homogeneous fibre format at the surface. It also explains why knock-out mutants of AgfE and AgfC in Salmonella typhi do not abrogate secretion but assemble mixed fibre morphology 51 .

Methods
Protein expression and purification. The gene encoding the mature CsgA polypeptide (Uniprot P28307, residues 22-151) was obtained by PCR amplification of E. coli BL21 (DE3) genomic DNA. An expression construct with a C-terminal His-tag was obtained by ligation into a pET-28a vector using NcoI/XhoI sites. The CsgA ΔN22 truncation was cloned using the Q5 Site-Directed Mutagenesis kit (NEB). CsgA variants were transformed into BL21 (DE3) E. coli expressions strains, inoculated into 800 mL of labelled M9 minimal media ( 15 NH 4 Cl/ 13 C 6 D-Glucose), and grown to 1.0 OD 600 . Temperature was decreased to 22 °C, IPTG (isopropyl-BD-thiogalactoside) added to a final concentration of 1 mM, and expression induced for 16 hours. Harvested cell pellets were stored at −80 °C. The pTrc99A expression vector for oligomerization-inhibited W48A/F79A CsgE mutant was kindly provided by Professor Matthew Chapman. W48A/F79A CsgE was purified as outlined previously 52 . CsgC-6xHis and CsgG-6xHis were expressed and purified as described previously 53 .
Denaturing purification of recombinant pre-amyloid CsgA. Cell pellets were resuspended in 15 mL g −1 of denaturation buffer (50 mM Tris-HCl, 6 M guanidine hydrochloride, pH 8). Lysis was achieved using sonication, and cell debris removed by centrifugation at 30,000 RCF for 30 minutes. Supernatant was loaded onto a 5 mL FF HisTrap column (GE Life Sciences) that was pre-equilibrated in denaturation buffer. Immobilized metal affinity chromatography (IMAC) was conducted in a step-wise manner with increasing imidazole concentrations (30/60/120/300 mM imidazole, 50 mM Tris-HCl, 6 M guanidine hydrochloride, pH 8). Fractions containing pure amyloid monomer were determined using SDS-PAGE, then pooled, concentrated to 2.5 mL using 10 kDa cut-off Vivaspin 20 device (Sartorius Biotech), and buffer-exchanged by passing through a PD-10 desalting column (GE Life Sciences) equilibrated in low salt buffer (20 mM Na 2 PO 4 , pH 7.2). Samples were either immediately used, or flash frozen in liquid N 2 and lyophilised. www.nature.com/scientificreports www.nature.com/scientificreports/ NMR data collection, assignments, and analysis. Samples of 150 μM CsgA, [U- 15 N] or [U-13 C, 15 N], were prepared in low salt buffer (20 mM Na 2 PO 4 , 10% D 2 O, pH 7.2. To prevent CsgA from transitioning to amyloid in the time course of assignment experiments (several days), curli-inhibitory chaperone was added at a concentration of 3.75 μM.
All spectra were recorded at 283.2 K. Data were collected on either a Bruker Avance III HD 950 MHz spectrometer, or a Bruker Avance III HD 800 MHz spectrometer. Both were equipped with a 5 mm TCI cryoprobe. Assignment of C α, C β, H N , and N chemical shifts were achieved using the following triple resonance spectra: HNCACB 54 , HNcoCACB 54 , hNcocaNNH 55 . Non-uniform sampling (NUS) methodologies were implemented in data collection. All spectra were processed with NMRPipe 56 , using SMILE to process NUS data 57 . Spectra were visualised and analysed using SPARKY 58 . Semi-automated, iterative backbone assignment was achieved using the algorithm MARS 59 . Secondary structure populations were determined from assigned chemical shifts, using the δ 2D methodology that is optimized for analysing disordered proteins 60 . 15 N transverse relaxation parameters were acquired using Carr-Purcell-Meiboom-Gill (CPMG) experiments 61 . These experiments were recorded with 14 different CPMG frequencies, ranging from 50 to 1600 Hz with three points repeated in duplicate, with a 20 ms mixing time per CPMG block. Values of R 2 were derived from line shape integration of peaks and fitting using the software package NMRPINT 62 . For interaction experiments, CsgA constructs (100 μM) were incubated with W48A/F79A CsgE (400 μM) for 24 hours at 4 °C prior to recording spectra. No CsgC was included in these samples. Receiver gains were set to the same value. CsgE was buffer exchanged into low salt buffer immediately before incubation. Chemical shift perturbations were calculated as a euclidean distance (d), and weighted as previously published 63 .