Peptide segments ¹⁵⁶NTVTFN¹⁶¹, ¹⁶⁸SIAVNF¹⁷³, ¹⁹⁶IATLYV²⁰¹, ³²⁴GIVIVA³²⁹ and ³⁶⁹TSYVGV³⁷⁴ from Als5 (UniProt accession number Q5A8T7) were synthesized at >98% purity and purchased from GL Biochem (Shanghai) Ltd. The peptides were synthesized with unmodified termini for crystallography or with fully capped termini (acetylated in the N-terminus and amidated in the C structural terminus) for fibrillation assays. Thioflavin T (ThT) was purchased from Sigma-Aldrich. Dimethyl sulfoxide (DMSO) was purchased from Merck. Ultra-pure water was purchased from Biological Industries.

Amyloidogenic propensities of Als5 segments were predicted using combined information from several computational methods, including TANGO (Fernandez-Escamilla et al., 2004), AmylPred 2 (Tsolis et al., 2013) and FiSH Amyloid (Gasior and Kotulska, 2014). TANGO and FiSH Amyloid were used in their default settings, and AmylPred 2 was used with the default consensus option of only 4 of 9 predictors (TANGO and Amyloid Mutants were omitted from its prediction). We used TANGO, AmylPred 2, and FiSH Amyloid because they were the most sensitive and consistent predictors with quantifiable scores. The spine segments were defined as short segments of 6 residues, detected by at least two methods out of three at the epicenter of the predicted amyloidogenic area in the sequence (like in the case of ³⁶⁹TSYVGV³⁷⁴). Only in the case of the segment ³²²SNGIVIVATTRTV³³⁴, which is known to be critical to fibrillation of the Als5 protein (Garcia et al., 2011), we worked with both its full length and a shorter version of 6 residues.

The three-dimensional (3D) model of the entire Als5 sequence from Candida albicans, taken directly from the UniProt accession number Q5A8T7 was generated by AlphaFold v2.1.0 pipeline (Jumper et al., 2021; Varadi et al., 2022). Using Chimera (Goddard et al., 2007), we focused only on residues 20–433 which include the Ig-like/invasin and T domains. The per-residue confidence score (pLDDT) of the Ig-like domain ranged above 90 indicating a reliable prediction (Mariani et al., 2013). For the T domain, the per-residue confidence score ranged between 70 and 90 indicating lower confidence in the prediction.

We used ConSurf (Landau et al., 2005; Ashkenazy et al., 2016) with default settings to examine the evolutionary conservation patterns of Candida albicans Als5 residues 20–433 which include the Ig-like and T domains. The calculation was run using ConSeq (Berezin et al., 2004) (no structure was provided). One hundred and twenty-one unique sequences of homologs with 35%–95% sequence identity were retrieved from the UniRef90 database. Homologs with large gaps (more than 5 residues) in their alignment were filtered using Jalview2 multiple sequence alignment editing program (Waterhouse et al., 2009). The alignment of the remaining 71 homologs was used for the calculations presented here. The alignment is accessible at https://drive.google.com/file/d/1jWLOldYEx5aZz5VnrlnYv7lvYcN6DgkO/view?usp=sharing. The alignment of the seven orthologs used to generate the information presented in Figure 5 is accessible at https://drive.google.com/file/d/1SFTarMIUj762evoOPkWXi-S56jTezG45/view?usp=sharing.

Comparison between the segments ¹⁵⁶NTVTFN¹⁶¹, ¹⁹⁶IATLYV²⁰¹, ³⁶⁹TSYVGV³⁷⁴, and ³²²SNGIVIVATTRTV³³⁴ evolutionary conservation was done by averaging the amino acid conservation scores attained from the ConSurf results for each segment. The segments ¹⁷⁷TVDQSG¹⁸² and ²³⁸VNDWNH²⁴³ were taken as controls due to their lack of amyloidogenic propensity and their different conservation values, namely low and high respectively. The 71 retrieved homologs were ranked by their sequence similarity to C. albicans Als5, with the lowest e-value corresponding to the closer homologs of C. albicans Als5. Out of the 71 homologs we selected one along every five ranked homologs (in descending order of similarity to C. albicans Als5) for a total of fourteen sequences representing diverse sequences in the multiple sequence alignment. We then calculated the amyloidogenic propensity for each of the fourteen homologs using TANGO (Fernandez-Escamilla et al., 2004), AmylPred 2 (Tsolis et al., 2013), and FiSH Amyloid (Gasior and Kotulska, 2014) (Figures 4B, 5). The multiple sequence alignment was used, via Jalview2 (Waterhouse et al., 2009), to identify equivalent regions the segments of C. albicans discussed here. We then calculated the average predicted amyloidogenic propensity values in these corresponding sequences in each of the 14 homologs. TANGO averages were calculated by dividing the sum of each residue’s amyloidogenic propensity by its length. AmylPred 2 averages were calculated by assigning to each residue of the segments either a Boolean value of zero or one (based on consensus recognition of the method) and calculating the average similarly to TANGO averages. Using the default sliding window settings of the method, FiSH Amyloid averages were calculated by dividing the sum of the values assigned to all residues covering the segment by its length.

For the calculations of the amyloidogenic propensity of the entire Ig-like/invasin and T domains, we selected seven orthologs (homologs with a common ancestral parent in different species). Six of those were chosen along the ranked 71 homologs in gaps of 10 (starting from C. dubliniensis orthologs located five places away from the original C. albicans Als5). To these six, we added the emerging pathogen Candida auris, which was not a part of the homologs collected by the ConSurf webserver due to low sequence identity to the C. albicans Als5p (Willaert, 2018; Willaert et al., 2021). The multiple sequence alignment of the 71 homologs along with the C. auris sequence was obtained by reproducing the ConSurf search with similar parameters but lower sequence identity of 25%–95%, and then filtering out the irrelevant homologs using Jalview2.

Thioflavin T (ThT) is a commonly used for identifying and investigating the formation of amyloid fibrils in vivo and in vitro. In the presence of ThT, fibrillation curves commonly show delayed nucleation followed by rapid aggregation. Fibrillation kinetics of the peptides ¹⁵⁶NTVTFN¹⁶¹, ¹⁹⁶IATLYV²⁰¹, ³⁶⁹TSYVGV³⁷⁴, and ³²²SNGIVIVATTRTV³³⁴ were monitored using ThT. All of the above peptides were synthesized with fully capped termini (acetylated in the N-terminus and amidated in the C structural terminus) to mimic its chemical nature in the full-length protein, and were dissolved to 10 mM in DMSO. Each of the freshly dissolved peptides was mixed with 50 mM Tris-HCl buffer with pH 7.3 and with 2 mM filtered ThT stock (made in ultra-pure water) to reach final concentrations of 300 μM of peptide and 20 μM ThT in the final volume of 100 μl reaction. The reaction mixture was carried out in a black 96-well flat-bottom plate (Greiner bio-one) covered with a thermal seal film (EXCEL scientific) and incubated in a plate reader (CLARIOstar, BMG Labtech), at 37°C, with orbital shaking at 300 rpm, for 30 s before each measurement. ThT fluorescence was recorded every 2 minutes for a total time of ∼48 h (we show only 5 h of the measurements) using an excitation of 438 ± 20 nm and an emission of 490 ± 20 nm. The measurements were conducted in triplicates, and the entire experiment was repeated at least three times.

TEM was used to visualize fibrils. Samples for TEM were taken directly from the ThT kinetic assay plate, which was left to incubate at 37°C for 2 days with 300 rpm shaking in the plate reader. The TEM grids were prepared by applying 5 µl samples of each 300 μM sample of peptide on 400 mesh copper grids with support films of Formvar/Carbon (Ted Pella), that were charged by high-voltage, alternating current glow-discharge (PELCO easiGlow, Ted Pella) immediately before use. Samples were allowed to adhere for 1 min followed by negative staining with 1% uranyl acetate for 1 min. Micrographs were recorded using a FEI Tecnai G2 T20 S-Twin transmission electron microscope at an accelerating voltage of 200 KeV at the MIKA electron microscopy center of the Department of Material Science & Engineering at the Technion.

The peptide SNGIVIVATTRTV was dissolved to 10 mg/ml in ultra-pure water. A few microliters were applied between two sealed glass capillaries until completely dried. X-ray diffraction of the samples was collected at the micro-focused beam P14 at the high brilliance 3rd Generation Synchrotron Radiation Source at DESY: PETRA III, Hamburg, Germany.

Peptides synthesized with free (unmodified) termini were used for crystallization experiments to facilitate crystal contacts. All peptides were dissolved to 10 mM in ultrapure water if possible, and in DMSO in case they were water-insoluble. ¹⁹⁶IATLYV²⁰¹ was dissolved in 95% DMSO, ¹⁵⁶NTVTFN^{161 168}SIAVNF¹⁷³ and ³²⁴GIVIVA³²⁹ were dissolved in 100% DMSO. ³⁶⁹TSYVGV³⁷⁴ was dissolved in 100% ultrapure water. Peptide solution drops (100 nl) were dispensed onto crystallization screening plates, using the Mosquito automated liquid dispensing robot (TTP Labtech, United Kingdom) located at the Technion Center for Structural Biology (TCSB). Crystallization using the hanging drop method, was performed in 96-well plates, with 100 µl solution in each well. The drop volumes were 150–300 nl. All plates were incubated in a Rock imager 1,000 robot (Formulatrix), at 293K. Micro-crystals grew after few days and were mounted on glass needles glued to brass pins. No cryogenic protection was used. Crystals were kept at room temperature prior to data collection. Structures were obtained from drops that were a mixture of the following peptide and reservoir solutions: ¹⁹⁶IATLYV²⁰¹: 10 mM IATLYV, 0.1 M tri-Sodium citrate pH 5.6 and 1.0 M Ammonium phosphate; ¹⁵⁶NTVTFN¹⁶¹: 10mM NTVTFN, 0.1 M HEPES pH 7.5, 0.8 M NaH₂PO₄, 0.8 M KH₂PO₄; ³⁶⁹TSYVGV³⁷⁴: 10 mM TSYVGV, 0.2 M Sodium chloride, 0.1 M Sodium acetate pH 4.6, 30% (v/v) 2-Methyl-2,4-pentanediol (MPD).

X-ray diffraction data were collected at 100K, using 5° oscillation. The X-ray diffraction data were collected at the micro-focus beamline ID23-EH2 of the European Synchrotron Radiation Facility (ESRF) in Grenoble, France; wavelength of data collection was 0.8729 Å. Data indexation, integration and scaling were performed using XDS/XSCALE (Kabsch, 2010). Molecular replacement solutions for all segments were obtained using the program Phaser within the CCP4 suite (Murshudov et al., 1997; McCoy et al., 2007; Winn et al., 2011). The search models consisted of geometrically idealized β-strands. Crystallographic refinements were performed with the program Refmac5 (Murshudov et al., 1997). Model building was performed with Coot (Emsley et al., 2010) and illustrated with Chimera (Goddard et al., 2007). There were no residues that fell in the disallowed region of the Ramachandran plot. Crystallographic statistics are listed in Table 1.

The Lawrence and Colman’s shape complementarity index (Lawrence and Colman, 1993) was used to calculate the shape complementarity between pairs of sheets forming the dry interface (Supplementary Table S1). The buried surface area was calculated with Chimera (UCSF) (Goddard et al., 2007), with a default probe radius and vertex density of 1.4 Å and 2.0/Ų, respectively. The number of solvent accessible buried surface areas was calculated as the average area buried of one strand within two β-sheets (total area buried from both sides is therefore double the reported number in Supplementary Table S1.

A prediction of the multimeric assembly of Als5 Ig-like and T-domains was done with AlphaFoldv2 Advanced pipeline (modified AlphaFold v2.1.0 Multimer pipeline) (Jumper et al., 2021) (Richard Evans et al. bioRxiv 2021). The pipeline was executed via the Google Colab service available online to each AlphaFold program (AlphaFold v2.1.0—https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb, AlphaFold v2.1.0 advanced—https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/beta/AlphaFold2_advanced.ipynb#scrollTo=pc5-mbsX9PZC). The prediction was based on the sequence of Als5 20–433 residues (UniProt accession number Q5A8T7). A total of five homodimers of Als5 20–433 were retrieved and were ranked based on their per-residue confidence score. Models of homodimers were illustrated, colored, and analyzed using UCSF Chimera (Goddard et al., 2007).

Sequence- and structure-based amyloid predictors use a variety of criteria including solubility and physicochemical properties, geometry, homology, and secondary structure propensity to predict occurrence of amyloidogenic core regions that can form cross-β aggregates and steric zippers. Accordingly, we scanned the primary structure of Als5 Ig/invasin and T domains (residues 20–433) with three amyloid prediction programs. TANGO has been useful for predictions in Als adhesins, is a thermodynamics state-based predictor (Fernandez-Escamilla et al., 2004; Lipke et al., 2017). AmylPred 2 is a consensus method based on eleven predictors (Tsolis et al., 2013). Lastly, FiSH Amyloid detects co-occurrence patterns in sequence data based on machine-learning techniques (Gasior and Kotulska, 2014). Nine segments scored above threshold in at least two of the three predictors, marked on the Als5 sequence in Figure 1A. Six segments are in the Ig-like/invasin region: ⁶⁵DTFILN⁷⁰, ¹³²IAFNVG¹³⁷, ¹⁵⁶NTVTFN¹⁶¹, ¹⁶⁸SIAVNF¹⁷³, ¹⁹⁶IATLYV²⁰¹, and ²⁵⁹FGISIT²⁶⁴, and three sequences are in the T domain: ³²⁴GIVIVA³²⁹, a section from the longer segment ³²²SNGIVIVATTRTV³³⁴ at the Ig/T domain interface, ³⁶⁹TSYVGV³⁷⁴, and ³⁸⁸TATVIV³⁹³. These segments were mapped onto an AlphaFold Monomer v2.0 model of Als5p (Jumper et al., 2021; Varadi et al., 2022) (Figure 1). AlphaFoldv2 predicted a folded structure for the Ig/T domains and the first four tandem repeats regions. As expected, there was no structural prediction for the highly glycosylated low complexity C-terminal stalk region, which is unstructured and in extended conformation in vivo (Lipke et al., 2017).

We determined the atomic structures of ¹⁵⁶NTVTFN¹⁶¹, ¹⁹⁶IATLYV²⁰¹ and ³⁶⁹TSYVGV³⁷⁴ segments (PDB IDs 6RHA, 6RHB, and 6RHD, respectively; Figure 2, Table 1, and Supplementary Table S1). The locations of the amyloid segments on the AlphaFoldv2 predicted structure of Als5 are shown in Figure 1B. The crystal structure of the ¹⁵⁶NTVTFN¹⁶¹ spine segment from the Ig-like domain formed a canonical steric-zipper structure of tightly mated parallel β-sheets with a dry interface. The ¹⁹⁶IATLYV²⁰¹ spine segment, also from the Ig-like subdomain 2, revealed a partial mating between antiparallel β-sheets. The ³⁶⁹TSYVGV³⁷⁴ spine segment from the T domain of Als5 exhibited an atypical amyloid structure, with antiparallel β-sheets and kinked β-strands (Figure 2 and Supplementary Figure S1).

These peptides induced development of thioflavin-T (ThT) fluorescence typical of amyloid aggregation (Figure 3A) and formed amyloid-like fibrils visualized by transmission electron microscopy (Figure 3B). The peptides ¹⁵⁶NTVTFN¹⁶¹ and ¹⁹⁶IATLYV²⁰¹ each formed abundant and long fibrils, although without the typical lag associated with amyloid formation, possibly due to presence of seeds in the original sample. ³⁶⁹TSYVGV³⁷⁴ formed abundant short plump bundled fibrils. The longer peptide ³²²SNGIVIVATTRTV³³⁴ segment, known to be critical for biological activity of the adhesin and crucial for fibrillation of soluble forms of Als5 protein, formed long and relatively straight fibrils that also induced ThT fluorescence (Lipke et al., 2017). This sequence contains a strongly predicted amyloid core sequence ³²⁴GIVIVA³²⁹. Although we did not successfully crystallize either the 6-residue or the 13-residue segments, the X-ray fiber diffraction of ³²²SNGIVIVATTRTV³³⁴ showed a canonical amyloid cross-β pattern with orthogonal reflection arches at 4.6 and ∼9 Å (Figure 2).

Because steric-zipper fibrils are unusual in that pairs of β-sheets mate more closely than the adjoining surfaces in other protein complexes, quantitative measures of amyloid stability are based on solvent-accessible surface area buried at the interface between the mating sheets, and shape complementarity indicating on the closeness of fit of two protein surfaces (Tayeb-Fligelman and Landau, 2017). The shape complementarity, inter-strand distance, and solvent-exposed surface area buried were calculated for Als5 spine structures and compared to the NNQQNY segment from yeast prion Sup35 steric zipper (PDB ID 1YJO) (Lawrence and Colman, 1993). The NNQQNY structure was chosen for this comparison as it shows one of the highest values of shape complementarity and surface area buried among steric zipper structures (Eisenberg and Sawaya, 2017). The shape complementarity was similar for all Als5 spines segment, all lower than prion NNQQYY (Supplementary Table S1). However, the T domain ³⁶⁹TSYVGV³⁷⁴ spine segment, displaying kinked β-strands, showed a significantly lower solvent-exposed surface area buried compared to the other structures (Supplementary Table S1), indicating a less stable structure that might also suggest reversible fibril formation.

The ¹⁵⁶NTVTFN¹⁶¹ and ¹⁹⁶IATLYV²⁰¹ segments in the Ig-like/invasin region are each close to a disulfide bonded Cys residue (Figure 1). Therefore, it is likely that these sequences would not have the conformational freedom to lead to solvent exposure or interaction necessary for amyloid formation. In contrast, the amyloid-forming segments in the T region, which is conformationally variable in vivo, are not constrained by disulfides or other structured regions. Thus, the amyloid propensity in this region is likely to be expressed in vivo, as demonstrated before for ³²²SNGIVIVATTRTV³³⁴ (Lipke et al., 2017).

Evolution can shift and alter traits in all biological systems. Selection pressure will work to conserve traits that possess an important beneficial function. It is also true for functional amyloids, where purifying selection acts to conserve the amyloidogenic potential, amino acid sequence, and accessible conformation; however, the amyloidogenic potential and amino acid sequence are not always conserved equally. Occasionally, amino acid sequences can change considerably without greatly affecting their amyloidogenic potential (Maji et al., 2009). In light of this difference, we have examined the evolutionary conservation patterns of the amyloidogenic segments correlated with the trajectory of the change in amyloidogenic potential from close to more distant homologs of C. albicans Als5.

To evaluate the conservation of the amyloidogenic segments in protein homologous to Als5 we used the ConSurf webserver (Berezin et al., 2004; Ashkenazy et al., 2016). First, we searched for homologs for Als5 residues 20–433 which include the Ig-like/invasin and T domains. Seventy-one unique sequences of homologs with 35%–95% sequence identity were retrieved from the UniProt (UniRef90) database. We excluded homologs with large gaps in the multiple sequence alignment. The segments ¹⁵⁶NTVTFN¹⁶¹, ¹⁹⁶IATLYV²⁰¹, ³²²SNGIVIVATTRTV³³⁴, ³²⁴GIVIVA³²⁹ and ³⁶⁹TSYVGV³⁷⁴ showed diverse pattern of sequence conservation across the Als5 homologs. In the Ig-like/invasin region, the sequence ¹⁵⁶NTVTFN¹⁶¹ was remarkably conserved across the homologs (Figure 4A), with five strongly conserved positions (Supplementary Figure S2), implying a conserved functional or structural role. In contrast, ¹⁹⁶IATLYV²⁰¹ was not well conserved and showed low average conservation value with four highly variable positions in its sequence. In the T domain, the ³²²SNGIVIVATTRTV³³⁴ region showed intermediate to above-average conservation. This observation is in agreement with previous reports showing that the functional amyloid-forming sequence ³²²SNGIVIVA³²⁹ is strongly conserved within the C. albicans ALS gene family (Otoo et al., 2008; Ho et al., 2019). The LARKS-like sequence ³⁶⁹TSYVGV³⁷⁴ also showed intermediate average conservation value with three highly conserved positions, with highest value at Gly³⁷³.

We also evaluated the conservation of amyloidogenic potential within the Als5p sequence homologs (Figure 4B). The spine-forming segments ¹⁵⁶NTVTFN¹⁶¹, ¹⁹⁶IATLYV²⁰¹, ³²²SNGIVIVATTRTV³³⁴, and ³⁶⁹TSYVGV³⁷⁴ presented a wide range of differences in amyloidogenic potential calculated by three predictors: TANGO, AmylPred 2, and FiSH Amyloid. For further analyses, we selected as controls two segments with a predicted low amyloidogenic propensity, ¹⁷⁷TVDQSG¹⁸² and ²³⁸VNDWNH²⁴³, showing low and intermediate evolutionary conservation, respectively. The ³²²SNGIVIVATTRTV³³⁴ segment already implicated in Als5 function (Lipke et al., 2017) showed a high amyloidogenic propensity in close homologs, which declined in more distant homologs. The LARKS-like sequence ³⁶⁹TSYVGV³⁷⁴, which is relatively conserved (Figure 4A), showed the highest stability in maintaining amyloidogenic potential across homologs according to Amylpred2 and FiSH amyloid, with more fluctuations in propensity towards more distant homologs (Figure 4B). In contrast, the Ig-like/invasin region segments ¹⁵⁶NTVTFN¹⁶¹ and ¹⁹⁶IATLYV²⁰¹ show a less clear pattern of amyloidogenic propensity across homologs, similar to the control sequences ¹⁷⁷TVDQSG¹⁸² and ²³⁸VNDWNH²⁴³. Control sequence ¹⁷⁷TVDQSG¹⁸² showed high fluctuations in amyloidogenic propensity, despite its low value in the C. albicans proteins. This finding corresponds to its low sequence conservation (Figure 4). Overall, the ⁶⁹TSYVGV³⁷⁴ and ³²²SNGIVIVATTRTV³³⁴ segments in the Als5 T domain appear to have the most conserved amyloidogenic propensity among examined segments, at least across close homologs of C. albicans.

We compared the amyloidogenic propensity of the entire Als5 Ig-like/invasin and T domains in Als5 and homologs from seven species: Candida dubliniensis, C. tropicalis, C. maltosa, C. viswanathii, Spathaspora passalidarum, Spathaspora sp. JA1, and C. auris (Muñoz et al., 2021). The amyloidogenic propensities predicted by AmylPred2 are summarized in Figure 5, with amyloidogenic sequences marked in purple. Each of the adhesins was predicted to have amyloid-forming sequence at the equivalent region to the ³²²SNGIVIVATTRTV³³⁴ segment of C. albicans Als5 (marked green), except for one homolog from Spathaspora sp. JA1. The equivalent region to the LARKS-like segment ³⁶⁹TSYVGV³⁷⁴ also showed a strong amyloidogenic propensity (marked dark yellow). In the C. auris homolog, there was no equivalent sequence to ³⁶⁹TSYVGV³⁷⁴, but other segments in this region (residues 343–370) were predicted to be amyloidogenic (marked purple). In the Ig-like/invasin region, the equivalent sequence to ¹⁹⁶IATLYV²⁰¹ (marked blue in Figure 5) showed no amyloidogenic propensity in C. maltosa and S. passalidarum or Spathaspora sp. JA1, but this segment did show some amyloidogenic propensity in C. auris. The predicted amyloidogenic propensity of this region was the least conserved among the distant homologs examined in Figure 4B. The segment equivalent to ¹⁵⁶NTVTFN¹⁶¹ showed amyloidogenic propensity predictions (marked red in Figure 5) across the homologs except for S. passalidarum and C. auris. Overall, each ortholog shown in Figure 5 contains various regions with strong prediction for amyloidogenic propensity. Among the identified segments of Als5, the T-domain segments displayed more conserved amyloidogenic traits compared to segment in the Ig-like/invasin region.

Als5 polymerizes in solution to form initial oligomers which convert to SDS-resistant oligomers, and then to amyloid fibers (Otoo et al., 2008). To gain insight into these interactions, we used AlphaFoldv2 Advanced (modification of AlphaFold v2.1.0 Multimer pipeline) to model the formation of Als5 dimers in the Ig-like/invasin and TR regions (Jumper et al., 2021) to model interactions between two copies of the Ig-like/invasin and T domains Als5. The model suggested dimerization of the folded monomers with an interface roughly parallel to long axis of the protein (Figure 6). In the three highest-ranked models, the major interface was mediated by the T domains, with minor contacts between the Ig-like/invasin domains. The highest-ranked model had the most extensive interface. The predicted amyloidogenic segments are not a part of the interface, but those in the T domain are close to their equivalent segment in the facing monomer. The models are based on the folded structure of the Als5, and do not show partial misfolding into an amyloid state. It is possible that future development of AlphaFold and other methods will allow modeling the cross-β fibrils.

Sequence- and structure-based amyloid predictors generally identify amyloidogenic or β-aggregation potential in many protein sequences. Indeed, systematic studies show that the majority of small peptides with high potential amyloidogenic sequences do form amyloids in vitro (Porat et al., 2003; Eisenberg and Sawaya, 2017; Hughes et al., 2018; Bera et al., 2019). However, only few of the proteins containing these sequences form amyloids in vivo, probably because peptides are constrained by being part of a stable protein fold in situ. Formation of cross-β structures depends on conformation, solvent accessibility, and compatible geometry in flanking regions (Shewmaker et al., 2011; Eisenberg and Sawaya, 2017; Bera et al., 2019). Therefore, amyloids can often form only after proteins denature to expose the amyloidogenic sequences and allow flexibility in the peptide chains. In Als5, we identified 9 segments with high potential to form amyloid-like cross-β structures. We argue here two of these sequences are involved in functional amyloid formation, while 6 others probably do not form cross-β structures in situ in the protein. The arguments are based on structural and evolutionary arguments, and on positive evidence for functionality of one segment.

Of the six potential amyloidogenic β-aggregating sequences in the Ig-like/invasin domain of Als5, three are within disulfide-stapled regions of the protein and so these segments are unlikely to be solvent-exposed and flexible enough to form cross-β structures (Figure 1). Of the three potential amyloidogenic sequences between that are not disulfide-constrained, ¹⁵⁶NTVFN¹⁶⁰ and ¹⁹⁶IATLYV²⁰¹ formed cross-β spines in vitro. However, these segments are unlikely to do so in vivo. Although the sequence ¹⁵⁶NTVTFN¹⁶¹ is relatively highly evolutionary conserved (Figure 4A and Supplementary Figure S2), its TANGO amyloidogenic propensity is low and not conserved across homologs. It has greater potential in the other two predictors (Figures 4B, 5). In Ig/invasin subdomain II, ¹⁵⁶NTVTFN¹⁶¹ seals the edge of a β-sheet (Figure 1), and so it contributes to the hydrophobic core of the fold. Therefore, this sequence is expected to be key for the high stability of this region, and this region of the protein unfolds only at high extension forces > 350 pN (Alsteens et al., 2009; Alsteens et al., 2013). In addition, ¹⁵⁶NTVTFN¹⁶¹ is near disulfide-bonded Cys¹⁵⁰ as well as the domain I/II interface, which both constrain its flexibility and geometry in the native protein. Therefore, this sequence is unlikely to form a cross-β structure in vivo in the intact protein.

The sequence ¹⁹⁶IATLYV²⁰¹ is in a β-strand (Figure 1B) and is poorly conserved in sequence (Figure 4A). Its equivalent region is predicted to have amyloidogenic propensity in some of the homologs, but without the specific trend of amyloidogenic propensity conservation along relatively close and distant homologs (Figures 4B, 5). Therefore, it seems unlikely that this segment participates in the formation of functional β-aggregates in Als adhesins. The segment is extremely close to disulfide bonded Cys²⁰⁴, and so it is likely to be too geometrically constrained in vivo to form cross-β structures.

The four other potential amyloid sequences in the Ig-like/invasin region failed to form cross-β spines. Therefore, there is little support for formation of cross-β structures in situ within the Ig/invasin region of the protein.

Two of the potential amyloid sequences in the T domain show conservation of their sequence and amyloidogenic propensity. Figures 4B, 5 demonstrate the strong cross-β potential of ³²²SNGIVIVATTRTV³³⁴ and the conservation of this propensity in this region across homologs. Similarly, the LARKS-like segment ³⁶⁹TSYVGV³⁷⁴ also shows high conservation of amyloidogenic propensity (Figures 4B, 5) and conservation in sequence as well (Figure 4A and Supplementary Figure S2). Both of these segments showed strong cross-β properties (Figures 2, 3). Also contrasting with the Ig/Invasin region, this region is unconstrained by disulfide bonds and becomes unstructured under low shear stress. Thus, these sequences are likely to be exposed and relatively unconstrained in vivo (Alsteens et al., 2009; Alsteens et al., 2010; Lipke et al., 2017). These features are in accord with the known T domain functional amyloid formation.

The T domain and the sequence ³²²SNGIVIVATTRTV³³⁴ are essential for many activities of Als5 and its close paralog Als1 (Lipke et al., 2017; Ho et al., 2019). This sequence is necessary for strong cell-to-cell binding, as well as for clustering the adhesins on the cell surface. A single site mutation V326N reduces the TANGO β-aggregation potential by about 20-fold and also reduces aggregation activity in Als5 without affecting ligand binding to the Ig/invasin domain (Garcia et al., 2011). Furthermore, adhesion activity is enhanced in the presence of a homologous peptide, and activity is reduced in the presence of a homologous non-amyloid peptide with the Val to Asn substitution. These are properties expected from a sequence interacting through β-aggregation. The T region unfolds easily under shear, including under physiological shear stresses (Chan and Lipke, 2014; Lipke et al., 2017). Thus, extension forces unfold the T domain, rendering it unstructured giving the conformational flexibility needed for formation of cross-β structures. Accordingly, we have published extensively on the importance of the ³²²SNGIVIVATTRTV³³⁴ sequence (Garcia et al., 2011; Lipke et al., 2017; Dehullu et al., 2019a). We have also found that the T region is essential for secretion and processing in yeast (Rauceo et al., 2006). This result implies that the deletion affects the entire fold of the domain.

The discovery of the highly-conserved T-region ³⁶⁹TSYVGV³⁷⁴ segment that can form a cross-β structure explains several observations. Specifically, Als5 and Als1 show residual aggregation activity and thioflavin binding after mutations that abolish the amyloid potential of the major T domain sequence ³²²SNGIVIVATTRTV³³⁴ (Alsteens et al., 2010; Garcia et al., 2011; Chan and Lipke, 2014; Dehullu et al., 2019a; Dehullu et al., 2019b; Ho et al., 2019). Similarly, there is 25%–40% residual activity after treatment with a specific peptide that inhibits cross-β formation of the ³²²SNGIVIVATTRTV³³⁴ segment (Garcia et al., 2011; Alsteens et al., 2013; Chan and Lipke, 2014; Dehullu et al., 2019a; Dehullu et al., 2019b). The residual activity has characteristics of cross-β structures because the cell surface remains somewhat thioflavin T fluorescent. Also consistent with cross-β functional bonds, the activity can be inhibited by concentrations of Congo red or thioflavins 1000-fold higher than those used in fluorescence. At these concentrations, these dyes are sequence-independent inhibitors of amyloid formation (Garcia et al., 2011; Landau et al., 2011; Harris, 2012; Chan and Lipke, 2014). These observations support the idea that another cross-β forming sequence mediates the residual activity. In general, LARKS-like sequences generate transient, low energy cross-β aggregates (Guenther et al., 2018; Hughes et al., 2021). These sequences often mediate reversible liquid-liquid phase separations (Hughes et al., 2018). On the surface of stimulated cells, Als proteins form mobile surface nanodomains visible in atomic force microscopy (AFM) and by in vivo staining with thioflavins (Rauceo et al., 2004; Alsteens et al., 2010; Garcia et al., 2011). These nanodomains would be a two-dimensional equivalent of such reversible liquid-like associations.

Therefore, evolutionary conservation, the presence of liquid-like surface nanodomains, the residual amyloid-like behavior in fluorescence and AFM are consistent with the LARKS-like sequence ³⁶⁹TSYVGV³⁷⁴ providing a supportive role in adherence. Thus, the use of our screens for cross-β sequences has resulted in discovery of a new cross-β segment that appears to contribute to the activity of amyloid-like sequences in the T domain of Candida Als adhesins.

In vivo, the T domain is unfolded by flow-induced shear stress over mucosal surfaces as low as 0.1–10 (dyne/cm²) (Chan and Lipke, 2014). These values are well-within the range of shear stresses in mucosal flow and in the blood stream. Shear-triggered unfolding could thus lead to peptide extension and exposure of segments with cross-β potential, and then to the formation of cross-β aggregates in vivo. Models for the transition from the well-folded state to cross-β of Als have been recently published (Lipke et al., 2017; Lipke et al., 2021).