Curli amyloid fibers in Escherichia coli biofilms: the influence of water availability on their structure and functional properties

E. coli biofilms consist of bacteria embedded in a self-produced matrix mainly made of protein fibers and polysaccharides. The curli amyloid fibers found in the biofilm matrix are promising versatile building blocks to design sustainable bio-sourced materials. To exploit this potential, it is crucial to understand i) how environmental cues during biofilm growth influence the molecular structure of these amyloid fibers, and ii) how this translates at higher length scales. To explore these questions, we studied the effect of water availability during biofilm growth on the conformation and functions of curli. We used microscopy and spectroscopy to characterize the amyloid fibers purified from biofilms grown on nutritive substrates with different water contents, and micro-indentation to measure the rigidity of the respective biofilms. The purified curli amyloid fibers present differences in the yield, structure and functional properties upon biofilm growth conditions. Fiber packing and β-sheets content correlate with their hydrophobicity and chemical stability, and with the rigidity of the biofilms. Our study highlights how E. coli biofilm growth conditions impact curli structure and functions contributing to macroscopic materials properties. These fundamental findings infer an alternative strategy to tune curli structure, which will ultimately benefit to engineer hierarchical and functional curli-based materials. Graphical Abstract

their growth conditions. 14,16 For instance, the presence of salts, oxidative stress, surface charges 1 and/or the water content of the substrate influence biofilm morphogenesis via different 2 mechanisms. 3,7,16-20 For example, the flux of water driven by osmotic gradients induces biofilm 3 swelling and nutrient transport, which in turn governs bacteria proliferation and matrix 4 production. 7,16,18 Moreover, the confinement of bacteria growth by the substrate influences 5 biofilm density, gives rise to complex biofilm morphologies and thereby modulates bacterial 6 access to nutrients. These processes have been proposed to determine several properties of the 7 biofilms, 7,19,20 including their mechanical properties. 8, 14,15,17 8 In addition to the numerous studies focusing on the macroscopic scale, much in vitro work has 9 been done to understand the assembly and aggregation kinetics of curli in aqueous solutions. 5,7,21 10 For this, a collection of biophysical techniques, namely circular dichroism, infrared spectroscopy 11 and fluorescence spectroscopy, have proved to be powerful tools to characterize amyloid 12 fibers. 5,7,21,22 Nevertheless, it is not still clear whether and how environmental cues experienced 13 by the bacteria during biofilm growth influence the molecular structure of the matrix fibers. Yet, 14 this piece of fundamental knowledge would be of great interest not only to further leverage the 15 full potential of curli amyloid fibers as tunable building blocks to make bio-sourced materials, 16 but also to develop strategies to prevent biofilm formation where they are detrimental, e.g. in 17 medical and industrial contexts. 11,12,23 18 The present work aims at bridging this gap by studying the effect of water availability during E. 19 coli biofilm growth on the conformation and the functions of curli. For this, we cultured biofilm-20 forming E. coli bacteria of the strain W3110 on nutritive substrates with different water 21 contents. 3 After 5 days of growth, we purified the curli fibers from the biofilms obtained in each 22 condition and performed a comparative study of the final fiber conformation using well-23 established microscopy and spectroscopy techniques. This approach enabled us to demonstrate 1 that the overall mass, the structure, and the properties of the curli fibers assembled in the E. coli 2 biofilms depend on the availability of water during growth. Indeed, curli amyloid fibers purified 3 from biofilms grown on substrates with low water content presented higher hydrophobicity and 4 chemical stability than those obtained from biofilms grown on substrates with high water 5 content. Moreover, the changes in the structure of the matrix fibers reflected the changes in the 6 biofilm mechanical properties. These results highlight the versatility of the curli fibers and their 7 key role in the adaptation of biofilms properties to the physicochemical cues of their 8 environment. As such, this work will greatly benefit to the research aiming at engineering 9 biofilm properties and use their matrix components as key build blocks to make programmable 10 living and/or bio-sourced materials. 11 (i.e. yield) per gram of dry biofilm in each condition. All data presented here come from N=4 independent biofilm cultures for 8 each condition tested, and the statistical analysis was done with One-way ANOVA (p<0.001, *** | p<0.01, ** | p<0.05, * | ns = 9 non-significant).

10
To estimate how the water content of the substrate influences biofilm composition, we 11 considered the whole mass harvested from these E. coli W3110 biofilms to be the sum of the 12 masses of i) the water, ii) the curli fibers and iii) the bacteria, the remaining nutrients and other 13 matrix biomolecules in minor proportions. The sum of the two latter is the dry mass of the 14 biofilm (Table 1). Although all the growth conditions rendered biofilms with similar 15 distributions of wet and dry mass (Figure 1b and Figure S2), there are some differences in the 16 content of their dry masses ( Table 1). Biofilms grown on dryer substrates (2.5 % salt-free LB-17 agar) present slightly lower content of dry mass, while those grown on wet substrates (0.5 % salt-18 free LB-agar) present higher content of dry mass. Both conditions, 0.5 % and 2.5 % salt-free LB-19 agar, yield biofilms with lower curli content in their dry mass composition compared to the other 20 growth conditions. 21 Table 1 Composition of E. coli W3110 biofilms grown on salt-free LB-agar substrates with different water contents. The 22 total mass corresponds to the sum of the water content and the dry mass of the biofilms. a The mass of "bacteria + rest" was 23 estimated by subtracting the curli mass, which is given by the quantification of CsgA monomers after purification from the dry 24 mass. The percentages of curli and bacteria + rest are given with respect to the dry mass. N=4  To make sure that the purification yielded the expected product, the purified materials were 3 examined by transmission electron microscopy (TEM), by fluorescence confocal microscopy 4 after Thioflavin S (ThioS) staining and by circular dichroism (CD) and attenuated total 5 reflectance Fourier transform infrared (ATR-FTIR) spectroscopy ( Figure S3 and Figure 2). The 6 TEM images first confirmed that the purified samples show an amyloid-like morphology with 7 needle-like structures ( Figure S3). ThioS is a fluorescence probe that reports the presence of β -8 sheet pleated structures. 24 As the samples stained positive for ThioS, the amyloid nature of the 9 fibers was confirmed (Figure 2a). 24 The differences in the ThioS intensity of the fibers 10 suggested that biofilm growth conditions render fibers with different conformations (Figure 2a). 11 In all the samples, agglomerated structures were detected and could be interpreted as clumps of 12 tightly associated fibrils and protofibrils. These clumps remained after extended sonication prior 13 to imaging, indicating strong adhesive forces linking the aggregates together. 5 Despite not 14 presenting significant differences, the intensity of the ThioS signal (relative to background) in 15 the samples obtained from biofilms grown on 0.5 % salt-free LB-agar was slightly lower than in samples obtained from the other conditions (Figure 2a). To further assess the differences in 1 fibers structure, we studied conformational characteristics by fluorescence (Figure 2b), CD 2 (Figure 2c), and ATR-FTIR spectroscopy (Figure 2d-e). 3 In order to allow for reliable comparison in the following experiments, the fiber solutions were 4 all normalized to 5µM of monomeric CsgA units. ThioT was added to the different solutions to 5 study fiber conformations. Indeed, it has been demonstrated that ThioT fluorescence intensity 6 relates to the number of exposed β -sheets and/or the spacing between them, 25 therefore 7 qualitatively reporting fibrils packing. 26,27 It was also reported that changes in ThioT emission 8 could also reflect differences in the electrostatic interaction between the dye and the fiber . 28 9 Nonetheless, non-significant differences were observed among the fibers after measuring their ζ -10 potential in the buffer used for the ThioT binding experiment (50 mM glycine buffer pH 8) 11 ( Figure S4). Note that in water, the fibers have negative ζ -potential values, and turned positive 12 in buffer, as expected when increasing pH. This result suggests that differences in the ThioT 13 spectra emission cannot be attributed to electrostatic interactions. Here, all fiber samples showed 14 a 10 to 30 fold increase of the ThioT intensity compared to the reference without fibers ( Figure  15 2b). Fibers assembled in the biofilms grown on 0.5 % salt-free LB-agar present the lowest ThioT 16 intensity, which indicates that the corresponding amyloid-like β -sheets differ in nature, extent 17 and/or packing of their β -strands. 27 18 CD spectroscopy showed the signature of β -sheet signaling for all fiber solutions, with a 19 maximum at ~195 nm and a minimum at ~216 nm. 5 The differences observed between the 20 different spectra indicate changes in the secondary structure content of the purified curli fibers 21 (Figure 2c). Such changes are more pronounced for the biofilms grown in wet conditions (0.5 % 22 salt-free LB-agar) (Figure 2c, inset). ATR-FTIR spectroscopy further confirmed the presence of 23 amyloid fibers and gave additional insights into their secondary structure. Indeed, this technique 1 is widely used to study protein structure and is especially suited for amyloids. 27,29, 30 The analysis 2 was focused on the amide I′ band (from 1700 to 1600 cm −1 ) that mostly represents the amide 3 C=O stretching and is especially sensitive to protein secondary structure. 29-31 All the samples 4 showed spectra similar to Figure 2d. The predominant band (Figure 2d, purple) was centered 5 around 1620 cm -1 and assigned to β -sheet structures. 31 Indeed, a narrow and intense absorption 6 band between 1615 and 1630 cm -1 is a hallmark for amyloid fibers as it indicates a large, planar 7 and extremely well-ordered cross-β spine. 29,30,32 The second band located around 1660 cm -1 is 8 usually assigned to β -turns structures (Figure 2d, green). 5,6,31 Finally, using band fitting to 9 decompose the spectra into its overlapping components systematically revealed a minor 10 contribution around 1650 cm -1 , which is assigned to random and α -helix structures (Figure 2d, 11 pink). 29 In the case of curli fibers purified from E. coli biofilms, the contribution of each of these 12 two components overlapped and could not be distinguished.

18
The hydrophobicity of the curli fibers depends on the water content of the biofilm substrate 19 To further detail the properties of the curli fibers obtained from biofilms grown in different 20 conditions, we used the solvatochromic dye 1-anilinonaphthalene-8-sulfonic acid (1,8-ANS), 21 which probes the local nano-environment around its moiety. 1,8-ANS emits a weak fluorescence 22 signal that both increases and shifts its maximum upon binding to hydrophobic pockets. 33 It is 23 also a suitable dye to distinguish fiber polymorphism. 34,35 In the presence of the fibers tested, 24 1,8-ANS showed a hyperchromic shift compared to the probe in buffer, and with a stronger shift 25 for fibers obtained from biofilms grown in dry conditions (Figure 3a and 3b). Phasor plot 26 analysis helps to visualize the variations in the hydrophobicity of the fibers by means of the shift 27 of the center of mass of each spectrum (Figure 3b and S5). In brief, phasor analysis consists in a 1 Fourier transform of the spectral information (or lifetime, see Experimental section) into a two-2 dimensional space where the axes represent the real and imaginary components. 36,37 In the case 3 of spectral phasors, the angle carries the information of the spectral center of mass and the radial 4 direction is related with the spectral width, providing a fingerprint of the system (See 5 Experimental section and Figure S5 for more details). 36,37 As such, a visual inspection of the 6 phasor plot allows to interpret the changes taking place between the systems, since points 7 presenting different coordinates indicate that the dye is sensing a different nano-environment. In 8

12
Nile red (NR) is another solvatochromic dye that has a high sensitivity for the tertiary structure 13 of different amyloid fibers. 40 NR has the advantage to be fluorescent at a visible wavelength 14 (λ em =585 nm), making it convenient for fluorescence microscopy. Here we use a combined 15 FLIM-phasor approach to assess the changes in NR lifetime within the fibers obtained in 16 different biofilm growth conditions (see experimental section and S6). 41-43 We show the 17 representative confocal microscopy images of fibers stained with NR and the corresponding 18 phasor plot for the lifetime of the fibers obtained from biofilms grown on the different substrates 19 (Figure 3f-g). An increase in the lifetime of NR indicates a less polar, i.e. more hydrophobic 20 environment at the binding site of the probe. 44,45 By using colored cursors (blue and yellow 21 circles), pixels of a given lifetime can be selected and the corresponding pixels are mapped back 22 onto the image using the same color code (Figure 3f). In this case, the blue cursor corresponds to 23 shorter lifetimes (more polar environments) and the yellow cursor to longer lifetimes (more 24 hydrophobic environments). An increase in the agar concentration (i.e. decrease in water 25 availability during biofilm growth) correlates with a higher lifetime of NR in the fibers (i.e. higher hydrophobicity, more yellow in Figure 3g). As such, changes in the NR lifetime, i.e. 1 environment hydrophobicity, between the fibers obtained in different growth conditions can be 2 easily identified without fitting procedures. A statistical quantification of these changes can be 3 found in the supporting information ( Figure S7). 4 The tryptophan population of all fibers is accessible to the solvent 5 Considering the potential of curli as building blocks for bio-sourced materials, we aimed at 6 getting more information on the supramolecular organization of the fibers. For this, we 7 investigated their intrinsic fluorescence, which can be exploited to study changes in solvent 8 accessibility of the side chains of aromatic amino acids. 46 Curli is made of CsgA monomers, 9 each one containing a single tryptophan (Trp), four tyrosine, and three phenylalanine residues. 32 10 When curli is excited at 280 nm, the intensity of the intrinsic fluorescence emission is mainly 11 due to Trp in CsgA. 46,47 When we excite the purified fibers at 280 nm, a single maximum 12 emission can be observed around 350 nm for all the spectra (Figure 4a). This position suggests 13 that the environment of Trp is relatively polar and structureless. The data acquired is consistent 14 with position of the Trp in the CsgA predicted by AlphaFold (Figure 4b). 47,48 15 The steady-state emission spectrum of fibers assembled in biofilms grown in standard growth 16 conditions (1.8 % salt-free LB-agar) have the highest emission intensities, whereas the fibers 17 assembled in biofilms grown on substrates with high water content (0.5 % salt-free LB-agar) 18 present the lowest Trp emission intensities. As explained above, the differences observed in the 19 intensity of the purified fibers reflect changes in solvent accessibility of the side chains of 20 aromatic amino acids. Hence, we explored the exposure of the Trp residues in each fiber using 21 neutral and ionic dynamic fluorescence quenchers (Figure 4c-d), namely acrylamide (polar and 22 uncharged) and iodine (highly hydrated, large and negatively charged). 49 N-acetyl-L-23 tryptophanamide (NATA), a soluble version of tryptophan, was used as reference for the 1 maximum quenching possible (Figure 4c-d insets).

12
The Stern-Volmer plots showed no upward curvature, meaning that the quenching effect in all 1 cases was predominately dynamic quenching, i.e. that fluorescence deactivation occurs because 2 of the collision of the quencher with the fluorophore and not because of a binding interaction 3 between both molecules. 49,51 Moreover, the addition of the quencher (acrylamide or iodide), even 4 at high concentrations did not induce a shift in the spectra of the Trp in the purified fibers 5 ( Figure S8). This suggests that there is no alteration of the hydrophobic character of the fibers 6 throughout the experiment. 7 Quenching experiments indicated no buried Trp residue within the fibers, suggesting that the Trp 8 population of all the fibers studied is exposed to the solvent (Figure 4c-d). 52 The plots of the 9 ratio between the fluorescence intensities in the absence and the presence of quenchers (I/I (Table  13 S2). Of note, the K SV values for acrylamide are much lower than those obtained with NATA 14 denoting that the emitters are not exposed to the solvent in the fibers. This implies rapid 15 diffusion of the quencher to the Trp. 49,52 16 Fiber chemical stability and biofilm rigidity follow fiber molecular structure 17 To study the implications of the structural differences described above, we tested the thermal and 18 chemical stability of the purified fibers against exposure to a temperature ramp and different urea 19 concentrations, but also the mechanical properties of the biofilms (Figure 5 and Figure S9). 20 Thermal denaturation was performed by heating the fibers from 25 to 80°C and cooling them 21 down back to 25°C (see Experimental section and Figure S9 for details). At each temperature step, the structure of the fibers was monitored using ATR-FTIR spectroscopy. All the samples of 1 fibers showed a remarkable stability, with no significant change of structure as monitored by 2 band analysis (Figure 5a-d). 3 The chemical stability of the fibers was tested against increasing urea concentrations from 0.1 to 4 8M (Figure 5e). The presence of amyloid fibers was monitored by the fluorescence intensity 5 emitted by ThioT. The stability of the fibers significantly decreased as the water content of the 6 corresponding biofilm the substrate increased (Figure 5e). We observed no linear correlation 7 between the fiber chemical stability and its packing arrangement (Figure 5f). Nevertheless, we 8 observed that the higher the β -sheet content in the fiber, the higher their chemical stability 9 (Figure 5g). 10 Biofilm stiffness was assessed by micro-indentation experiments in the central area of the 11 biofilms grown on different substrates with different water contents ( Figure S10). 17 For each 12 biofilm, loading curves were obtained upon approximatively 10 nm indentation (Figure 5h). 13 Although there is no clear trend, we observed that biofilms grown in wet conditions (0.5 % salt-14 free LB-agar) presented the lowest reduced Young moduli. Their stiffness even revealed to be 15 significantly lower than the one of biofilms grown in other conditions. In contrast, biofilms 16 grown on 1.0 % and 2.5 % salt-free LB-agar were significantly stiffer than those grown on 1.8 % 17 salt-free LB-agar (Figure 5i). These results on the mechanical properties of the biofilm could be 18 understood as macroscopic consequences of the changes of the curli structure in the matrix. We 19 thus plotted biofilm reduced Young moduli as a function of the structure of the curli fibers 20 formed in each corresponding condition (Figure 5j). We found that the higher the fiber packing 21 (ThioT ratio), the stiffer the biofilm. sequences, and lead to structure variations when trapped inside proteins. 55 In a crowded 20 environment, protein hydration shells tend to change and amyloid proteins are more prone to 21 aggregation. 27 As shown by the binding differences to ThioS and ThioT, the packing and the β -22 sheet content of the curli fibers depend on the water content of the biofilm substrate (Figure 2b-23  e). 24 These structural differences in the fibers are further supported by the differences observed 1 in tryptophan (Trp) emission signal, which reflect differences in Trp interactions with 2 surrounding groups such as threonine (Thr), aspartic acid (Asp), glutamine (Gln), and asparagine 3 (Asn) (Figure 4b). 27,47 Indeed, these polar and charged residues have the ability to quench Trp 4 fluorescence. Hence, the lower intensity in the spectra of fibers obtained from biofilms grown in 5 wet conditions (0.5 % agar) can be further explained by changes in the structure of these fibers. 6 The fibers obtained from biofilms grown on substrates with high water content (0.5 % agar) also 7 showed lower hydrophobicity, as sensed by the 1,8-ANS and the NR probes (Figure 3). 56 8 Interestingly, the phasor plot obtained from 1,8-ANS spectra revealed that the fibers purified 9 from biofilms grown on 1.8 % salt-free LB-agar present more hydrophobic binding sites ( Figure  10   3a-b). Moreover, the similar trends of the ThioT and the 1,8-ANS intensities as a function of 11 growth condition, suggest that the β -sheet content can be correlated to the hydrophobicity in the 12 fibers (Figure 3c-e). 5 NR fluorescence studies confirmed that fibers obtained in wet conditions 13 are less hydrophobic than fibers obtained in dry conditions (Figure 3f-g). The difference 14 between the results obtained from the 1,8-ANS and the NR experiments on the 1.8 % salt-free 15 LB-agar are more likely due to the different binding sites and affinities of the probes. 39,57,58 16 Nonetheless, both results indicate clear differences between fibers fibrillated in dry and wet 17 conditions. 18 Amyloids are highly ordered and arranged into intermolecular β-sheets and cross β-structures, 19 which contain many strong hydrogen bonds that enhance the overall stability of the fibers. 5,59 20 Models of curli fibers also suggest that the hydrogen bonds between Gln and Asn residues from 21 different monomers form a network that contributes to the stability of these fibers. 5,32,60 22 Considering the importance of the structure-function relationship in proteins and the materials 23 they form, the implications of the structural differences observed in the purified curli fibers were 1 assessed with stability assays. While all fibers showed high stability against thermal 2 denaturation, 9 their stability upon urea exposure revealed significant differences (Figure 5h). 3 The thermal stability can be attributed to the high content of β-sheets in fibers obtained from all 4 conditions (Figure 2e). However, the relatively light packing and lower β-sheets content 5 measured in curli fibers obtained at high water concentrations could explain their lower chemical 6 stability compared to those obtained on dryer substrates (Figure 5h-i). Indeed, less densely 7 packed fiber structures not only result in weaker interactions between the distant Gln and Asn 8 residues, but they also provide better access for urea to denature the fibers by interactions with 9 their hydrophobic and amide groups (Figure 6). 61 Note that the fibers extracted from biofilms 10 grown on 0.5% salt-free LB-agar not only contain less than 50% of β -sheet but almost twice as 11 much β -turns compared to the other conditions (Figure 2e), which indicates that β -turn structure 12 is less chemically stable than β -sheet ( Figure S11). 13 Previous studies involving different bacteria strains showed that biofilms grown on substrates 14 with high water content (e.g. 0.5 % agar) have a less dense matrix, which is heterogeneously 15 distributed across the biofilm thickness. 15,17 Our results would attribute these observations to a 16 lower production of curli fibers in these conditions, as well as to different final fiber 17 conformation (Figure 1e, Table 2). By affecting the quantity and the molecular structure of curli 18 fibers in the biofilm matrix (Figure 2c and Figure 6), the water content of the agar substrate is 19 therefore expected to affect biofilm macroscopic features like their growth kinetics 7 and their 20 mechanical properties. 3,15,19 For example, the high content of β-turns in the curli fibers purified 21 from biofilm grown on substrates with high water content (0.5 % agar) results in a lower packing 22 compared to the fibers obtained from dryer conditions (1.0 -2.5 % agar) (Figure 2e). Such difference in structure could be explained by larger average spacing between strands (Figure  1   6a), and the resulting flexibility of the curli fibers in the matrix contribute to the lower rigidity of 2 biofilms grown on wet substrates as measured by micro-indentation experiments (Figure 5b). 17 3 It is important to note that previous studies decoupled the role of the water content and of the 4 stiffness of the substrate using semipermeable membranes, and revealed that differences in 5 biofilm growth behavior in such conditions are rather due to the water availability than substrates 6 mechanics. 62 Moreover, wet substrates were proposed to promote bacteria motility by enabling 7 swimming, which would explain the larger biofilms obtained in such conditions, as well as their 8 less dense matrix and softer mechanical properties. 17 Such differences in the microenvironment 9 of the curli fibers could in turn influence their packing and conformation. 10 In bacterial biofilms, curli amyloid fibers influence cell properties such as their ability to 11 withstand drying, as well as the thickness, the hydrophobicity and the rigidity of the biofilm. 15 In 12 this work, a more hydrophobic behavior of the curli fibers and a stiffer behavior of the biofilm 13 follow a denser packing (Figure 3e, 5c) 15   biofilm materials properties under the given conditions. For example, the biofilms grown in the 23 extreme conditions (0.5% and 2.5% salt-free LB-agar) contained less curli and had less water 1 uptake capacity (Figure 1f-g). These results suggest that, in addition to providing biofilms with 2 adhesion and rigidity, 3,17 curli fibers also promote biofilm water uptake from the surroundings, 3 thereby contributing to their hydration capacity. This ability constitutes a significant advantage 4 for biofilms growing at solid-air interfaces, i.e. in environments where an influx of water carries 5 the nutrients from the substrate to the biofilm. While osmotic gradients have been proposed to be 6 involved in this transport function of the biofilm matrix, their origin is yet not clear. 16 Since i) 7 bacteria are able to regulate the number and location of the curli fibers formed, 7 and ii) less curli 8 fibers were found in the biofilms grown in wet conditions than those grown in dryer conditions 9 (Figure 1e), then curli production could be proposed as a way for bacteria to create these 10 osmotic gradients in conditions where water is more difficult to reach. The discrepancy of this 11 hypothesis with the lower quantity of curli fibers measured in biofilms grown on the driest 12 substrates (2.5 % agar) could be explained by the extreme difficulties to reach the nutrients, 13 which could impair the overall biofilm growth and/or matrix production. 14 Biofilms can be seen as natural hydrogels, in which bacteria are embedded in a cross-linked 15 polymer matrix. 64 This perspective contributes to the emerging trend of considering E.coli-grown 16 amyloid fibers (i.e. curli) are promising building blocks for making sustainable and bio-sourced 17 materials with interesting functional properties. As such, these biopolymers have been used to 18 produce hydrogels, 12 functional templates for protein immobilization, 11 conductive composite, 13 19 among others. For instance, adding curli fibers to alginate hydrogels was shown to increase their 20 stiffness. 12 Our work suggests that the stiffness of such hydrogels could also be adjusted by 21 tuning the structure of the added curli fibers rather than by adjusting their amount (Figure 6a). 22 This alternative can become interesting in situations where the mechanical properties of the 23 materials need to be decoupled from their composition, e.g., in tissue engineering research. 65 1 While genetic engineering remains the preferred approach to design curli fibers with specific 2 characteristics for further use in functional materials, 11 our work shows the potential of using 3 biofilm growth conditions as an interesting alternative to tune the physico-chemical properties of 4 curli amyloid fibers. As such, a given strain of biofilm-forming E. coli could yield curli fibers of 5 different secondary structures on demands. 6 Overall, this work helps to understand better the effect of water availability during E. coli 7 biofilm growth on the secondary structure of the curli amyloid fibers extracted from the biofilm 8 matrix. The biophysics-based methods used to characterize curli at the molecular scale appear to 9 be of great value to inform materials engineers about the physico-chemical properties of this 10 promising building block. Moreover, our study reports on how these differences further impact 11 other functional properties of the curli fibers on their own (chemical and thermal stability) or of a 12 material where they greatly contribute (e.g. biofilm mechanical properties). Finally, the findings 13 reported provide valuable knowledge regarding the structure -function relationship that spans 14 biofilm scales from the molecular to the tissue levels, and give insights into the adaptation 15 response of biofilms in specific environments. As such, this study will also contribute to uncover 16 the full potential of biofilm matrix as building blocks to engineer living and/or bio-sourced 17 materials, which can eventually become construction materials, sustainable plastic-like materials 18 or scaffolds for tissue engineering (Figure 6b). 11,12,23 19 20 1 Figure 6 A graphical summary of the results. (a) By varying the water content in the substrates used for biofilm growth, we 2 showed a dependence of the production of curli fibers. The packing of the fibers also varies as a function of the substrate water 3 content, which in turn influences their chemical stability and the mechanical properties of the materials they constitute (e.g. a 4 biofilm). (b) The potential of curli as a building block for bio-sourced materials 10-13,23 .

Experimental Section 1
Bacterial strain and growth 2 The biofilm-forming bacterial strain E. coli K-12 W3110 was used throughout this study. Salt-3 free LB-agar plates (15mm diameter) were prepared with 0.5%, 1.0%, 1.8%, or 2.5% w/v of 4 bacteriological grade agar−agar (Roth, 2266), supplemented with 1% w/v tryptone (Roth, 8952) 5 and 0.5% w/v yeast extract (Roth, 2363). After agar pouring, the plates were left to dry for 10 6 minutes with the lid open and 10 minutes with the lid partially open to avoid future 7 condensation. Each agar plate was left to rest for 48 hours before bacteria seeding. A suspension 8 of bacteria was prepared from a single colony and grown overnight in Luria−Bertani (LB) 9 medium at 37°C with shaking at 250 rpm. Each plate was inoculated with arrays of 9 drops of 10 5μL of bacterial suspension (OD600 ∼ 0.5 after 10x dilution). After inoculation, the excess of 11 water evaporated from the drops and left bacteria-rich dry traces of comparable sizes from 4 to 12 8mm diameter, depending on the growth condition. Biofilms were grown for 5 days in total 13 (∼120h) inside an incubator at 28°C. Monitoring the relative humidity in the incubator showed 14 that it remains around 30%RH. 15

Biofilm imaging 16
Three biofilms per condition were imaged with a stereomicroscope (AxioZoomV.16, Zeiss, 17 Germany) using the tiling function of the acquisition software (Zen 2.6 Blue edition, Zeiss, 18 Germany). To estimate the biofilm size, 3 independent biofilms were measured at their 19 equatorial line using the Fiji software. 66 An average was then calculated for each growth 20 condition. 21 The water content and water uptake of the biofilms were determined by scraping 7 biofilms per 1 condition from the respective agar substrates after 5 days of growth (~120 h). Biofilms were 2 placed in plastic weighing boats, and dried at 60°C for 3 h in an oven. Wet and dry masses (m wet , 3 m dry ) were determined before and after drying. 17 To determine the water uptake (W up ), we added 4 Millipure water in excess (5 ml) to the biofilms harvested from each condition, covered them 5 with aluminum foils to avoid evaporation and left overnight. The water excess was removed and 6 the biofilm samples were weighed again (m rewet ). The biofilms water content in each growth 7 condition was estimated with Eq. (1) 8 Eq. (1) W = (m wet −m dry )/m wet × 100% w/w 9 The percentage of water uptake of biofilms after rehydration (%W up,w ) was determined with 10 respect to biofilm initial wet mass as described in Eq. All procedures were carried out in four independent experiments. 15

Curli fiber purification and quantification 16
Fiber purification involved a similar process as reported in previous works. 21 Briefly, a total of 17 27 biofilms (~ 1g of biofilm material) were scraped from the surface of the substrates. Biofilms 18 were blended five times on ice with an XENOX MHX 68500 homogenizer for 1min at 2-min 19 intervals. The bacteria were pelleted by centrifuging two times at low speed (5000g at 4°C for 20 10min). A final concentration of NaCl 150 mM was added to the supernatant and the curli pelleted by centrifuging at 12.000g at 4°C for 10 minutes. The pellet was resuspended in 1mL of 1 solution containing 10mM tris (pH 7.4) and 150mM NaCl, and incubated on ice for 30min 2 before being centrifuged at 16.000g at 4°C for 10 minutes. This washing procedure was repeated 3 thrice. The pellet was then resuspended in 1mL of 10mM tris solution (pH 7.4) and pelleted as 4 described above (16.000g at 4°C for 10 minutes). The pellet was again suspended in 1mL of 5 10mM tris (pH 7.4) and centrifuged at 17.000g at 4°C for 10 minutes. This washing step was 6 repeated twice. The pellet was then resuspended in 1mL of SDS 1% v/v solution and incubated 7 for 30min. The fibers were pelleted by centrifuging at 19.000g at 4°C for 15min. The pellet was 8 resuspended in 1mL of Milli-Q water. This washing procedure was repeated thrice. The last 9 resuspension was done in 0.1mL of Milli-Q water supplemented with 0.02% sodium azide. The 10 fiber suspension was stored at 4°C for later use. The protein concentration in monomeric units of 11 the suspensions was determined by the absorbance from an aliquot incubated in 8M urea at 25°C 12 for 2h, a treatment leading to complete dissociation of the fibrils as verified by Thioflavin T 13 measurements. Attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) 4 IR spectra were acquired on a spectrophotometer (Vertex 70v, Bruker Optik GmbH, Germany) 5 equipped with a single reflection diamond reflectance accessory continuously purged with dry air 6 to reduce water vapor distortions in the spectra. Fibers in Milli-Q water samples (∼10μL) were 7 spread on a diamond crystal surface, dried under N 2 flow to obtain the protein spectra. A total of 8 64 accumulations were recorded at 25°C using a nominal resolution of 4cm −1 . 9 Spectra were processed using Kinetic software developed by Dr. Erik Goormaghtigh at the 10 Structure and Function of Membrane Biology Laboratory, Université Libre de Bruxelles, 11 Brussels, Belgium. After subtraction of water vapor and side chain contributions, the spectra 12 were baseline corrected and area normalized between 1700 and 1600cm −1 ( Figure S2). For a 13 better visualization of the overlapping components arising from the distinct structural elements, 14 the spectra were deconvoluted using Lorentzian deconvolution factor with a full width at the half 15 maximum (FWHM) of 20 cm −1 and a Gaussian apodization factor with a FWHM of 30 cm −1 to 16 achieve a line narrowing factor K = 1.5. 31 Second derivative was performed on the Fourier self-17 deconvoluted spectra for band assignment. The bands identified by both procedures were used as 18 initial parameters for a least square iterative curve fitting of the original IR band (K = 1) in the 19 amide I' region, using mixed Gaussian/Lorentzian bands. Corrected steady-state emission spectra were acquired with a FluoroMax®-4 spectrofluorometer 2 (HORIBA). Spectra were recorded at 25°C using a 3-mm path cuvette (Hellma® Analytics). transform of the fluorescence spectra. This transformation offers a powerful, model free, graphic 10 method to characterize spectral 36 and lifetime information. 67 A detailed analysis can be found 11 elsewhere. 41,68,69 In this work, ANS spectra can be transformed using the following for x and y 12 coordinates: 13 Eq. (5) are the fluorescence intensity values, λ 0 is the initial wavelength of the spectrum (λ 0 1 =400 nm), L is the length of the spectrum and n is the harmonic value. The values L=300 (from 2 400 to 700 nm) and n=1 were used for phasor calculations. The angular position of the phasor in 3 the plot is related to the center of mass, while the radial position depends on the full width at half 4 maximum of the spectrum. Spectral phasor plots were constructed using Originlab data analysis 5 software. 6 where I 0 and I are the fluorescence intensities in the absence and in the presence of quencher, 10 respectively, at the concentration [Q] and K SV is the dynamic quenching constant. 11 Thermal stability assay 12 5μM fiber samples were incubated at 25, 30, 40, 50, 60, 70 and 80°C for 3 min before their 13 measurement with ATR-FTIR spectroscopy. An extra measurement was done by taking the 14 fibers from 80°C to a bath water at 25°C and incubate the samples there for 3 min. Data were 15 acquired as described above. 16

Chemical stability assay 17
To test the chemical stability of the protein fibers, 5 μ M samples were prepared by incubation of 18 increasing urea concentration (0 -8 M) and left for 2 h at room temperature to ensure 19 equilibrium. Thioflavin T (ThioT) was then added in a final concentration of 1 mM and 20 fluorescence emission spectra of the samples were acquired under excitation at λ exc = 446 nm and 21 spectral bandwidth of 5 nm. Emission of the ThioT in urea was done and no significant signal 1 was detected. 2 The IC50 value was estimated by fitting the data to a linear regression (Y = a * X + b) and then 3 calculating IC50 = (0.5 -b)/a. 4

Micro-indentation on biofilms 5
The micro-indentation experiments were carried out as described in Ziege et al. 17 Briefly, E. coli 6 W3110 biofilms were grown for 5 days on their respective agar substrate and 3 to 4 of them were 7 tested per condition. Ten measurements were performed in the central region of each biofilm. 8 The distance between two measurement points was at least 250 μ m in x and y directions and the 9 depth of the indentation was between 10 and 30 µm, i.e. much less than biofilm the thickness 10 (~100µm). 70 A TI 950 Triboindenter (Hysitron Inc.) was used to determine the load-11 displacement curves after calibration of the instrument in air. Loading rates ranged from 20 to 30 12 μ m/s, which translates to loading and unloading times of 10 s. The loading portion of all curves 13 were fitted with a Hertzian contact model over and indentation range of 0 to 10µm to obtain the 14 reduced Young's modulus E r . 15

Statistical analysis 16
For each experiment, 3 to 4 fiber solutions were used, where each solution came from different 17 fiber purification batches. For each purification, 27 biofilms were cultured in each of the 4 18 growth conditions tested (i.e. on salt-free LB-agar plate containing 0.5%, 1.0%, 1.8% and 2.5% 19 agar respectively). For each batch of biofilm culture, the different samples of fibers obtained for 20 the 4 conditions were treated simultaneously (or in consecutive days) to avoid variability due to 21 unavoidable slight variations in the implementation of the protocols (e.g. temperature and 1 humidity in the laboratory during agar preparation and/or biofilm seeding). 2 For statistical analysis, a Shapiro Wilk test was used to check for data normality. For data with 3 no normal distribution, Kruskal-Wallis non-parametric test was performed. For data with normal 4 distribution, a One-way ANOVA test was carried out. Mechanical properties data was analyzed 5 using a Mann-Whitney U test. Unless otherwise stated in the caption, Dunn's post-test for 6 multiple comparisons were done with respect to the 1.8 % salt-free LB-agar condition, 7 considered as the standard seeding condition. Details of each test are described in the legend of 8 the figures. 9 10 11 12 Supporting Information 1 The following files are available free of charge: 2 • Table S1: Composition of salt-free LB agar subsrates 3 • Figure S1: Characterization of salt-free LB agar substrates 4 • Figure S2: Biofilm dry mass 5 • Figure S3: Purity of curli fibers 6 • Figure S4: ζ -potential of purified curli fibers 7 • Figure S5: Amide I' spectra of curli fibers purified from E. coli biofilms 8 • Figure S6: Spectral phasor plot analysis 9 • Figure S7: FLIM phasor plot 10 • Figure S8: Nile red FLIM analysis 11 • Figure S9: Raw spectra of the quenching experiments 12 • Table S2: Stern Volmer constants of quenching experiments 13 • Figure S10: Complete fiber thermal denaturation ramp (ATR-FTIR analysis) 14 • Figure  The data that support the findings of this study are available from the corresponding authors 19 upon reasonable request.