Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Microbial Cell Factories https://doi.org/10.1186/s12934-023-02182-3 RESEARCH Open Access A3DyDB: exploring structural aggregation propensities in the yeast proteome Javier Garcia-Pardo1†, Aleksandra E. Badaczewska-Dawid2†, Carlos Pintado-Grima1, Valentín Iglesias1, Aleksander Kuriata3, Sebastian Kmiecik3* and Salvador Ventura1* Abstract Background The budding yeast Saccharomyces cerevisiae (S. cerevisiae) is a well-established model system for studying protein aggregation due to the conservation of essential cellular structures and pathways found across eukaryotes. However, limited structural knowledge of its proteome has prevented a deeper understanding of yeast functionalities, interactions, and aggregation. Results In this study, we introduce the A3D yeast database (A3DyDB), which offers an extensive catalog of aggregation propensity predictions for the S. cerevisiae proteome. We used Aggrescan 3D (A3D) and the newly released protein models from AlphaFold2 (AF2) to compute the structure-based aggregation predictions for 6039 yeast proteins. The A3D algorithm exploits the information from 3D protein structures to calculate their intrinsic aggregation propensities. To facilitate simple and intuitive data analysis, A3DyDB provides a user-friendly interface for querying, browsing, and visualizing information on aggregation predictions from yeast protein structures. The A3DyDB also allows for the evaluation of the influence of natural or engineered mutations on protein stability and solubility. The A3DyDB is freely available at http://biocomp.chem.uw.edu.pl/A3D2/yeast. Conclusion The A3DyDB addresses a gap in yeast resources by facilitating the exploration of correlations between structural aggregation propensity and diverse protein properties at the proteome level. We anticipate that this comprehensive database will become a standard tool in the modeling of protein aggregation and its implications in budding yeast. Keywords Protein aggregation, Aggrescan 3D, AlphaFold, Yeast, Saccharomyces cerevisiae Background † Javier Garcia-Pardo and Aleksandra E. Badaczewska-Dawid are co- The notable conservation of essential cellular structures first authors. and pathways such as cell cycle regulation, DNA repair, *Correspondence: RNA processing, signal transduction pathways, metab- Sebastian Kmiecik olism, protein quality control mechanisms, or stress sekmi@chem.uw.edu.pl Salvador Ventura response, has positioned the budding yeast Saccharomy- Salvador.Ventura@uab.cat ces cerevisiae (S. cerevisiae) as an ideal eukaryotic model 1 Institut de Biotecnologia i de Biomedicina (IBB) and Departament de organism. S. cerevisiae is also a reference expression sys- Bioquímica i Biologia Molecular, Universitat Autònoma de Barcelona, Bellaterra, Barcelona 08193, Spain tem for the heterologous production of recombinant pro- 2 Genome Informatics Facility, Office of Biotechnology, Iowa State teins. One of the major advantages of S. cerevisiae resides University, Ames, IA 50011, USA on its well-characterized and fully annotated genome. 3 Biological and Chemical Research Center, Faculty of Chemistry, University of Warsaw, Pasteura 1, Warsaw 02-093, Poland Indeed, several databases dedicated to yeast have been © The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 2 of 10 published, including the Saccharomyces Genome Data- information. However, the newly developed AlphaFold2 base (SGD) [1], the Yeast Metabolome Database (YMDB) (AF2) database has released the prediction of thousands [2] or YEASTRACT+ [3]. Among them, the SGD data- of structures from different organisms, including S. cere- base stands as the gold-standard repository with a wealth visiae [20]. The overall quality of these computed models of integrated biological information for this microorgan- was shown to be comparable to experimentally deter- ism. It also provides access to analytic tools to explore mined structures [21]. Therefore, the newly reported these data, enabling the discovery of functional relation- structural information from AF2 allows the generation of ships between sequence and gene products in fungi and proteome-wide repositories reporting yeast globular pro- higher organisms. In addition, the availability of compre- teins’ aggregation properties. hensive genomic resources, such as deletion libraries and Herein, we present the A3D yeast database (A3DyDB), collections of temperature-sensitive mutants, facilitates which compiles the structure-based aggregation propen- the identification and characterization of factors involved sity predictions for 6039 S. saccharomyces protein models in different biological processes [4]. Indeed, S. cerevisiae from the AF database [20]. Since many yeast proteins are stands out as one of the most widely employed model not fully structured, the A3DyDB allows for the custom- systems for studying protein aggregation in vivo. Notably, ization of jobs to adapt structural predictions according the yeast proteome is relatively small compared to higher to AF2 confident cutoffs. The database also includes a eukaryotes, which makes it a privileged model for devel- tool for evaluating the influence of user-defined muta- oping systematic studies and high-throughput screenings tions on protein solubility and stability. We believe that targeting protein aggregation [5]. Despite the relevance the A3DyDB will serve as a useful resource for the study of this model organism in the field, aggregation propen- of protein aggregation in yeast. It will also allow the sity predictions are not currently reported in dedicated investigation of correlations between structural aggre- databases. gation propensity and protein function, stability, archi- The identification of intermolecular interactions medi- tecture, location, and protein abundance, among other ated by solvent-exposed aggregation-prone regions factors associated with protein aggregation. Ultimately, (APRs) embedded in the protein sequence has proven we illustrate the performance and utility of the database successful in predicting protein aggregation [6–10]. This with selected case reports. approach is particularly suitable in the context of intrin- sically disordered proteins (IDPs) or for newly synthe- Results sized polypeptide chains, but it often overpredicts when A3DyDB summary and interface description applied to folded proteins. Indeed, in folded proteins the The A3DyDB incorporates A3D predictions for 6039 pro- identified APRs are often located within their hydropho- teins from the S. cerevisiae proteome. To perform aggre- bic cores or at inaccessible regions characterized by the gation predictions, we used a large dataset of structural presence of highly stable secondary structures [11, 12]. models generated with AF2, which were downloaded It is now widely accepted that globular proteins aggre- from the AF database [20]. These structures were ana- gate by the spatial clustering of often non-contiguous lyzed using the latest A3D implementation developed sequence regions of hydrophobic amino acids, form- by our group [17, 22]. The resultant A3D data have been ing structural APRs in the protein surface (STAPs) [13]. stored in the first comprehensive database describing the The aggregation may happen by local or global struc- structure-based aggregation predictions for yeast (http:// tural destabilization [14] or by stochastic fluctuations biocomp.chem.uw.edu.pl/A3D2/yeast). The A3DyDB that lead to the exposure of previously buried APRs [15]. is endowed with a search tab on the front page, which Consequently, considering a protein’s spatial environ- allows users to query for the content by using the gene, ment becomes crucial in understanding the underlying protein name, or Uniprot Accession (See Fig. 1a). Select- forces driving its aggregation. In this context, we devel- ing entries from the query list leads to a page containing oped the Aggrescan 3D (A3D) algorithm [16–18], which the A3D analysis. The analysis is distributed in differ- makes use of the experimentally determined Aggrescan’s ent tabs containing the following information: (I) pro- aggregation propensity scale [7, 19] and projects it into tein information and project details, (II) an interactive a three-dimensional protein structure. The versatility and A3D score profile and annotation of transmembrane accuracy of this algorithm have converted it into one of regions (only applicable to membrane proteins), (III) a the default methods to study the aggregation of proteins detailed table containing the precalculated A3D scores in their natively folded states and how dynamic fluctua- and AF structure prediction confidence scores (pLDDTs), tions and mutations impact this reaction. (IV) the protein structure colored by A3D and pLDDTs Over the years, accurate predictions of structure-based scores, (V) custom jobs with pre-calculated pLDDT cut- aggregation propensities in yeast at the proteome level offs and (VI) collection of images. The A3DyDB Project were hampered by the limited availability of structural details tab contains relevant information regarding the Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 3 of 10 Fig. 1 Overview of the A3DyDB home page and example of a search result for CDC19. (a) A3DyDB home page and query result for CDC19. (b) A3DyDB results page showing the project details tab for the CDC19 entry (Uniprot accession P00549) selected entry (Fig. 1b). A3D plot/score tabs display a Membrane proteins were also included in the database detailed analysis of the per-residue aggregation propen- as A3D accurately predicts hydrophobic transmembrane sity scores (A3D scores) and the profile and additional segments as highly aggregation-prone regions (Fig. 2b). annotations for transmembrane and intermembrane pro- Although they do not contribute to typical aggregation teins can be found in the subsequent sections. For each mechanisms, these highly hydrophobic stretches are entry, two different structural models are provided in often inserted in membrane lipidic bilayers or involved the Structure tab (top panel in Fig. 2a). The top structure in protein oligomerization. Based on Uniprot annota- shows a per-residue prediction of the A3D aggregation tions, a total of 1219 transmembrane and intramembrane propensity (A3D score), while at the lower structural rep- proteins from S. cerevisiae were identified. For these resentation AF2 confidence score (pLDDT) is depicted cases, we have included a complementary tab with rel- (lower panel in Fig. 2a). evant information, including predictions of consensus Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 4 of 10 Fig. 2 Examples of protein structural models as reported in the A3D Yeast database. Visualization of A3D predicted aggregation propensity ranges from blue (more solubilizing) to red (more aggregation-prone). A3D Yeast database provides the per-residue local confidence score (pDDLT), a metric that has been shown to inversely correlate with protein disorder [29]. Recent analysis suggests pDDLT scores < 70 usually correspond to unstructured in isolation. Examples of the A3D Yeast database output for (a) a globular protein, (b) a membrane protein and (c) a disordered protein are shown membrane segments from protein sequences by means of single or multiple mutations in a custom A3D analysis. of the TOPCONS server [23]. TOPCONS reports five These new jobs will be immediately listed and accessible different predictive algorithms, including OCTOPUS in the A3DyDB queue. [24], Philius v [25], Polyphobius [26], SCAMPI [27], and SPOCTOPUS algorithms [28]. The inclusion of this Discussion module will facilitate the visual inspection and analysis Protein aggregation is increasingly recognized as a con- of transmembrane regions, with the aim of performing tributing factor to various pathologies in eukaryotes comparative analyses. [32] and constitutes a major limitation to produce func- A significant number of proteins from the A3DyDB tional recombinant proteins in yeast [33]. Numerous contain structurally disordered regions. This is consis- past studies, mostly performed using simple prokary- tent with the observation that eukaryotes have a higher otic and eukaryotic model organisms such as bacteria proportion of IDPs relative to bacteria or archaea [30]. and yeast, have led to a detailed understanding of how AlphaFold2 outputs the pDDLT score, a per-residue esti- highly aggregation-prone proteins form insoluble spe- mate of the model’s confidence on a scale from 0 to 100 cies and how these proteins are toxic for the cells [34]. [20] which reports on the quality of the AF prediction These seminal investigations have allowed the identifi- [21]. Interestingly, regions of low confidence often cor- cation of important principles of protein aggregation, respond to intrinsically disordered regions (IDRs) [29]. which has led during the last decade to the development Not surprisingly, manual data curation revealed that low of a series of predictive algorithms to identify aggrega- pLDDT scores might result in misleading A3D predic- tion-prone sites [35]. Our current understanding of the tions, either because these solvent-exposed regions are structural landscape of the yeast proteome has radically more exposed or compact in the model than they should changed with the development of deep-learning-based be. A significant number of yeast proteins showed the approaches, such as RoseTTA fold [36] or AF2 [21]. This presence of regions with low predicted accuracy (low new wealth of structural data can be exploited to predict pLDDT score in Fig. 2c). After testing different pLDDT the aggregation properties of the whole yeast proteome thresholds, we decided to precompute A3D on top of and undertake the redesign of yeast proteins to improve three different AF models for each protein entry: the full- their solubility and stability for diverse purposes. Herein, length protein model and two additional models in which we have launched the A3DyDB, which contains the pre- residues with pLDDT < 70 or residues with pLDDT < 50 computed aggregation predictions for the S. cerevisiae were removed (see Figure S1). These precalculated mod- proteome, and we have tested the performance of the els can be directly accessed from the Custom Jobs tab. The database in a variety of case reports. Below, we provide A3DyDB implementation also allows users to model the selected case examples demonstrating the suitability of effects of custom mutations on the stability and aggrega- our comprehensive repository in diverse scenarios. tion propensity of a given particular protein entry using the FoldX force field [31]. Using the mutation editor at the Custom Jobs tab it is possible to evaluate the effects Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 5 of 10 Case examples proteins to form assemblies is closely linked to their pro- Exploiting the A3DyDB to study cellular organization and pensity for aggregation. Linear aggregation predictors metabolism in yeast such as TANGO [39] have been used to study differences Yeast organisms live in a wide range of different environ- in protein aggregation in these dynamic assemblies [38], ments which require local adaptations to transient con- but most of the proteins involved in foci formation con- ditions [37]. For this reason, proteins have evolved to tained globular regions which require dedicated struc- self-organize in the cellular milieu in response to specific ture-based aggregation predictive tools. stimuli such as nutrient starvation. Under these circum- Herein, we used the data from the A3DyDB to inves- stances, metabolic enzymes from yeast proteins undergo tigate possible differences in structural aggregation a widespread reorganization into reversible punctate between the 180 foci and 27 non-foci forming proteins cytoplasmic foci that are disassembled when the stress described by Narayanaswamy et al. [38] (Supplemen- is released [38]. Interestingly, the evidence strongly sug- tary Table 1). Our results showed that foci-forming pro- gests that the formation of reversible protein assemblies, teins had a significantly higher A3D average score than specific to metabolites, is potentially widespread in the non-foci-forming proteins (Fig. 3a). Indeed, the visual realm of cell biology [38]. In this context, the ability of inspection of proteins from the two independent datasets Fig. 3 Structural aggregation propensity differences between foci and non-foci forming proteins from yeast. (a) The A3D average score is higher for proteins observed to form punctate foci (in red) than those that did not form (in blue). A statistically significant difference between the two groups was observed (Mann-Whitney-Wilcoxon two-sided test-t; p = 0.0047). The A3D predictions from AF ARO2 (AF-P28777-F1) (b) and TIF2 (AF-P10081-F1) (c) structures, two representative cases of foci and non-foci proteins respectively. The foci-forming protein ARO2 presents a larger number of STAPs than the non-foci protein TIF2. Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 6 of 10 revealed that foci proteins, such as ARO2, contain a Predicting STAPs to study functional protein assemblies larger number of STAPs than non-foci proteins like TIF2 We have investigated a case example in which STAPs (Fig. 3b and c, respectively). Overall, it seems that protein are important for mediating functional protein-protein organization in yeast is a very dynamic process that could interactions. The actin fold is found in cytoskeletal poly- be, at least, partially understood by protein aggregation. mers, chaperones, and various yeast enzymes involved in In this framework, the release of thousands of struc- metabolic pathways. Most actin-fold proteins, such as the tural predictions in the yeast A3D database can help in carbohydrate kinases are monomeric proteins and do not identifying STAPs in several proteins involved in cellular polymerize. However, it has been recently reported that reorganizations. the S. cerevisiae glucokinase GLK1 can form polymers in response to its substrates and products (Fig. 4a) [40]. Fig. 4 A3D predictions for yeast hexokinase GLK1 self-assembly. (a) Cryo-EM structure of the yeast hexokinase GLK1 filament (PDB: 6PDT). (b) Structure of a GLK1 monomer showing the location of the N-terminal Phenylalanine residue involved in filament formation (Phe3, indicated with a circle). (c) A3D prediction of the wild-type active monomeric GLK1 obtained from the A3DyDB. Predicted aggregation propensities are colored in the structure ranging from blue (more solubilizing) to red (more aggregation-prone). The position of the two predicted hydrophobic STAPs is indicated with arrows. (d) Detailed representation of the insertion site of Phe3 into the C-terminus hydrophobic cavity. (e) A3D prediction of the Phe3Ser (P3S) solubilizing mutant. This solubilizing mutation eliminates filament polymerization Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 7 of 10 The polymerization of this actin-fold protein inhibits its and coworkers mutated the N-terminal Phe3 involved in kinase catalytic activity, a mechanism directly coupled to GLK1 assembly contacts to Ser, to change the non-polar cell viability and the adaptation of the yeast to stochastic character of this [21] protein position [40]. Based on our changes in the environment. Recently, cryo-EM studies A3D predictions, this mutation significantly decreases have revealed that glucokinase GLK1 from S. cerevisiae the potency of the N-terminal STAP (Fig. 4e), which is able to form two-stranded filaments with a molecular coincides with the experimental observation that it elimi- architecture different from that of cytoskeletal polymers nated polymerization both in vitro and in vivo. [40]. These filaments are built up by GLK1 monomer sta- bilized by hydrophobic interactions between GLK1 sub- Using the A3DyDB to study membrane proteins units along a strand. In this structure, a solvent exposed The endoplasmic reticulum (ER) network is built up by Phe3 of one GLK1 subunit is inserted into the hydropho- tubules with high membrane curvature in cross-sec- bic pocket at the C-terminus of the next GLK1 moiety, tion, which are generated and stabilized by reticulons effectively mediating the stabilization of the filament and receptor expression-enhancing proteins (REEPs). (Fig. 4b). Reticulons and REEPs are integral membrane proteins A detailed analysis of the structural aggregation pro- resident at the ER that are evolutionary conserved across pensity of the monomeric GLK1 was obtained from the all eukaryotes [42, 43]. These proteins share a common AF2-derived model collected in A3DyDB (Fig. 4c). Our architecture that has been also identified in other human analysis showed the presence of two strong STAPs in this proteins that function as ER-phagy intramembrane protein that overlap with the observed GLK1 oligomer- receptors (i.e. ATG40 in S. cerevisiae and FAM134B ization interfaces, where the N-terminus Phe3 is inserted in mammals) [44, 45]. Here, we used the A3DyDB to in the hydrophobic C-terminus cavity (Fig. 4d). This is investigate the structure and aggregation propensities of consistent with the view that functional and aberrant YOP1, as a model membrane protein (Fig. 5). polymerization surfaces share very similar physicochemi- YOP1 is a yeast reticulon that is highly enriched in the cal properties and do frequently overlap [13, 41]. Garner tubular portions of the ER and virtually excluded from Fig. 5 A3DyDB allows the study of membrane proteins and their functional implications. (a) A3DyDB results page or Transmembrane regions tab for YOP1 entry (Uniprot Accession Q12402). (b) Scheme of YOP1 predicted topology deduced from A2DyDB TOPCONS predictions and previous experi- mental data [44, 46]. (c) YOP1 AF2 model (AF-Q12402-F1), showing the location of predicted TMs and APH regions. (d) A3D analysis (upper panel) and AlphaFold2 (AF2) model confidence score (lower panel). Note that YOP1 exposes STAPs to the hydrophobic interior of the membrane bilayer and at the APH region Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 8 of 10 other regions, which induces high membrane curvature could become more stable and soluble. The presented of tubules by an unknown mechanism. Indeed, dele- here in silico approach can serve to make faster and more tion of the reticulons and YOP1 in yeast has been linked cost-efficient yeast mutants for different applications to the loss of tubular ER. Previous studies have sug- such as reconstructing metabolic networks, improving gested that YOP1 generates high membrane curvature the solubility of endogenous proteins, recombinant pro- by hydrophobic insertion and scaffolding mechanisms, tein production, and anticipating improved protein vari- which is mediated by the insertion of the TM regions on ants in synthetic biology approaches. the highly hydrophobic lipid bilayers [47–49]. The sug- gested mechanism relies on functional data and the pro- Methods tein’s structure, but its experimental verification remains Data collection and A3D analysis incomplete at this stage. S. cerevisiae (UP000002311) protein structures We have analyzed the presence of membrane segments (n = 6039) were downloaded from the AF database at the transmembrane regions tab from the A3DyDB out- (October 20, 2022; structural model version v4, avail- put. We found that YOP1 has two pairs of trans-mem- able at https://ftp.ebi.ac.uk/pub/databases/alphafold/v4/ brane (TM) segments, which is one of the characteristic UP000002311_559292_YEAST_v4.tar) and run with A3D features of reticulons (Fig. 5a). These TM regions are in static mode with a distance of aggregation analysis of well predicted by the different algorithms implemented 10Å and FoldX-based energy minimization for stability in the database and by the consensus TOPCONS predic- calculations. Custom jobs were created for all predicted tion, which is accompanied by an overall high prediction structures with two defined AF cutoffs 50 and 70, with reliability (Fig. 5a). Next, we have compared the struc- residues of pLDDT < = 50 or < = 70 removed for the A3D ture-based aggregation predictions from A3D with the aggregation prediction. proposed membrane topology. Most of the TM regions were predicted as large highly aggregation-prone seg- Database construction ments, as expected for such hydrophobic regions. The The user interface of the A3DyDB online database was main four STAPs in YOP1 were followed by an additional developed utilizing HTML and integrated with custom strong short aggregation-prone stretch (residues Ile143- JavaScript functions to enhance interactivity. The visual Ile152), a region that matches with an amphipathic helix design of the website is a combination of standard Boot- (APH) that is required for maintaining the character- strap components along with a touch of custom CSS istic reticulon’s ER-tubule localization (Fig. 5c and d). styles. An Apache2 web server is employed to host the Both the TMs and APH have been shown as essential website, leveraging MySQL integration to handle data elements to generate high membrane curvature and for storage, retrieval, and querying of the pre-calculated maintaining relevant protein-protein interactions [46]. It A3DyDB entries. Interactive plots are dynamically gener- has been previously demonstrated that YOP1 undergoes ated using the D3.js library, while molecular visualization homotypic and heterotypic oligomerization [46, 47, 50, tasks are handled by the Open Source PyMOL tool [52]. 51]. This behavior was mostly due to homotypic interac- Additionally, the database is seamlessly integrated with tions mostly between the TMs regions, as suggested by the A3D Server, enabling direct submission of custom Cystein-based crosslinking experiments [50, 51]. Taken mutation analysis. The transmembrane analysis for mem- together, our results indicate that STAPs are present brane proteins was performed with the consensus algo- in YOP1, and it is likely that these regions are involved rithm TOPCONS [23], which includes predictions from in membrane binding, as well as in maintaining YOP1 OCTOPUS [24], Philius [25], Polyphobius [26], SCAMPI oligomerization interaction interfaces. [27], and SPOCTOPUS [28]. Conclusions Foci and mutation structural analyses The A3DyDB provides a unique repository of structural Foci (n = 180) and non-foci (n = 27) forming proteins were aggregation predictions for thousands of yeast structures obtained from [38]. Structural aggregation propensities collected in the AF database, a resource that for a long for both protein datasets were obtained from the yeast time remained elusive due to the limited amount of avail- A3DyDB. Statistical significance between variables and/ able structural data. Given the importance of S. cerevisiae or datasets was assessed with Mann-Whitney-Wilcoxon as a model organism in aggregation, we see A3DyDB as two-sided test with Bonferroni correction. p-value was a valuable resource to inspect STAPs in yeast proteins marked with asterisks to better convey statistical sig- and find associations between aggregation propensity nificance (p > 0.05 (ns), p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 and other functional aspects of yeast biology. Besides, (***), p ≤ 0.0001 (****)). A3DyDB can be used to analyze the effect of mutations GLK1 mutation analysis was performed with A3DyDB on the 3D surface of proteins and engineer variants that mutation mode under the Custom jobs tab. Structural Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 9 of 10 8. Maurer-Stroh S, Debulpaep M, Kuemmerer N, Lopez de la Paz M, Martins IC, representations of A3D mutants were obtained from the Reumers J, et al. Exploring the sequence determinants of amyloid structure Structure tab visualization tool. using position-specific scoring matrices. Nat Methods. 2010;7(3):237–42. 9. Walsh I, Seno F, Tosatto SC, Trovato A. PASTA 2.0: an improved server for Acknowledgements protein aggregation prediction. Nucleic Acids Res. 2014;42(Web Server Not applicable. issue):W301–7. 10. Santos J, Pujols J, Pallares I, Iglesias V, Ventura S. Computational prediction Authors’ contributions of protein aggregation: advances in proteomics, conformation-specific JG-P, CP-G, and VI performed the analyses and wrote the initial draft of the algorithms and biotechnological applications. Comput Struct Biotechnol J. manuscript. AEB-D, AK and SK contributed with the implementation of the 2020;18:1403–13. database. JG-P, SK and SV acquired funding and reviewed the final version of 11. Castillo V, Chiti F, Ventura S. The N-terminal helix controls the transition the manuscript with inputs from all authors. JG-P, SK and SV conceptualized between the soluble and amyloid states of an FF domain. PLoS ONE. the project. All authors have read and agreed to the final version of the 2013;8(3):e58297. manuscript. 12. Santos J, Iglesias V, Ventura S. Computational prediction and redesign of aber- rant protein oligomerization. Prog Mol Biol Transl Sci. 2020;169:43–83. Funding 13. Castillo V, Ventura S. Amyloidogenic regions and interaction surfaces overlap This work was funded by the Spanish Ministry of Science and Innovation in globular proteins related to conformational diseases. PLoS Comput Biol. (MICINN) PID2019-105017RB-I00 to S.V, by ICREA, ICREA-Academia 2020 and 2009;5(8):e1000476. by EU (PhasAge /H2020-WIDESPREAD-2020-5) to SV. J.G.-P. was supported 14. Castillo V, Espargaro A, Gordo V, Vendrell J, Ventura S. Deciphering the role of by the Spanish Ministry of Science and Innovation with a Juan de la Cierva the thermodynamic and kinetic stabilities of SH3 domains on their aggrega- Incorporacion IJC2019-041039-I. V.I. was supported by the Spanish Ministry tion inside bacteria. Proteomics. 2010;10(23):4172–85. of Science and Innovation and the European Union-NextGenerationEU 15. Grana-Montes R, de Groot NS, Castillo V, Sancho J, Velazquez-Campoy A, Ven- (Universitat Autònoma de Barcelona 02/07/2021). C.P.-G. was supported by tura S. Contribution of disulfide bonds to stability, folding, and amyloid fibril the Secretariat of Universities and Research of the Catalan Government and formation: the PI3-SH3 domain case. Antioxid Redox Signal. 2012;16(1):1–15. the European Social Fund (2023 FI_3 00018). This article is partially based upon 16. Zambrano R, Jamroz M, Szczasiuk A, Pujols J, Kmiecik S, Ventura S. AGGR- work from COST Action ML4NGP, CA21160, supported by COST (European ESCAN3D (A3D): server for prediction of aggregation properties of protein Cooperation in Science and Technology). SK acknowledges financial support structures. Nucleic Acids Res. 2015;43(W1):W306–13. by National Science Centre, Sheng grant number 2021/40/Q/NZ2/00078. 17. Kuriata A, Iglesias V, Pujols J, Kurcinski M, Kmiecik S, Ventura S. Aggrescan3D (A3D) 2.0: prediction and engineering of protein solubility. Nucleic Acids Res. Data Availability 2019;47(W1):W300–W7. All data generated or analyzed during this study are included in this published 18. Pujols J, Iglesias V, Santos J, Kuriata A, Kmiecik S, Ventura S. A3D 2.0 update article and the A3DyDB is freely available at http://biocomp.chem.uw.edu.pl/ for the prediction and optimization of protein solubility. Methods Mol Biol. A3D2/yeast. 2022;2406:65–84. 19. Badaczewska-Dawid AE, Garcia-Pardo J, Kuriata A, Pujols J, Ventura S, Kmiecik S. A3D database: structure-based predictions of protein aggregation for the Declarations human proteome. Bioinformatics. 2022;38(11):3121–3. 20. Varadi M, Anyango S, Deshpande M, Nair S, Natassia C, Yordanova G, et al. Ethics approval and consent to particiate AlphaFold protein structure database: massively expanding the structural Not applicable. coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res. 2022;50(D1):D439–D44. Consent for publication 21. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Not applicable. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583–9. Competing interests 22. Kuriata A, Iglesias V, Kurcinski M, Ventura S, Kmiecik S. Aggrescan3D stand- The authors declare no competing interests. alone package for structure-based prediction of protein aggregation proper- ties. Bioinformatics. 2019;35(19):3834–5. Received: 30 June 2023 / Accepted: 18 August 2023 23. Tsirigos KD, Peters C, Shu N, Kall L, Elofsson A. The TOPCONS web server for consensus prediction of membrane protein topology and signal peptides. Nucleic Acids Res. 2015;43(W1):W401–7. 24. Viklund H, Elofsson A. OCTOPUS: improving topology prediction by two-track ANN-based preference scores and an extended topological grammar. Bioin- formatics. 2008;24(15):1662–8. References 25. Reynolds SM, Kall L, Riffle ME, Bilmes JA, Noble WS. Transmembrane topol- 1. Cherry JM, Hong EL, Amundsen C, Balakrishnan R, Binkley G, Chan ET, et al. ogy and signal peptide prediction using dynamic bayesian networks. PLoS Saccharomyces Genome Database: the genomics resource of budding yeast. Comput Biol. 2008;4(11):e1000213. Nucleic Acids Res. 2012;40(Database issue):D700–5. 26. Kall L, Krogh A, Sonnhammer EL. An HMM posterior decoder for sequence 2. Ramirez-Gaona M, Marcu A, Pon A, Guo AC, Sajed T, Wishart NA, et al. YMDB feature prediction that includes homology information. Bioinformatics. 2.0: a significantly expanded version of the yeast metabolome database. 2005;21(Suppl 1):i251–7. Nucleic Acids Res. 2017;45(D1):D440–D5. 27. Bernsel A, Viklund H, Falk J, Lindahl E, von Heijne G, Elofsson A. Prediction of 3. Monteiro PT, Oliveira J, Pais P, Antunes M, Palma M, Cavalheiro M, et al. YEAST- membrane-protein topology from first principles. Proc Natl Acad Sci U S A. RACT+: a portal for cross-species comparative genomics of transcription 2008;105(20):7177–81. regulation in yeasts. Nucleic Acids Res. 2020;48(D1):D642–D9. 28. Viklund H, Bernsel A, Skwark M, Elofsson A. SPOCTOPUS: a combined predic- 4. Jin K, Li J, Vizeacoumar FS, Li Z, Min R, Zamparo L, et al. PhenoM: a database tor of signal peptides and membrane protein topology. Bioinformatics. of morphological phenotypes caused by mutation of essential genes in Sac- 2008;24(24):2928–9. charomyces cerevisiae. Nucleic Acids Res. 2012;40(Database issue):D687–94. 29. Piovesan D, Monzon AM, Tosatto SCE. Intrinsic protein disorder and condi- 5. Di Gregorio SE, Duennwald ML. Yeast as a model to study protein misfolding tional folding in AlphaFoldDB. Protein Sci. 2022;31(11):e4466. in aged cells. FEMS Yeast Res. 2018;18(6). 30. Pancsa R, Tompa P. Structural disorder in eukaryotes. PLoS ONE. 6. Belli M, Ramazzotti M, Chiti F. Prediction of amyloid aggregation in vivo. 2012;7(4):e34687. EMBO Rep. 2011;12(7):657–63. 31. Schymkowitz J, Borg J, Stricher F, Nys R, Rousseau F, Serrano L. The FoldX 7. Conchillo-Sole O, de Groot NS, Aviles FX, Vendrell J, Daura X, Ventura S. AGGR- web server: an online force field. Nucleic Acids Res. 2005;33(Web Server ESCAN: a server for the prediction and evaluation of hot spots of aggregation issue):W382–8. in polypeptides. BMC Bioinformatics. 2007;8:65. Garcia-Pardo et al. Microbial Cell Factories (2023) 22:186 Page 10 of 10 32. Chiti F, Dobson CM. Protein misfolding, amyloid formation, and Human 43. Yang YS, Strittmatter SM. The reticulons: a family of proteins with diverse Disease: a Summary of Progress over the last decade. Annu Rev Biochem. functions. Genome Biol. 2007;8(12):234. 2017;86:27–68. 44. Khaminets A, Heinrich T, Mari M, Grumati P, Huebner AK, Akutsu M, et al. 33. Romero-Suarez D, Wulff T, Rong Y, Jakociu Nas T, Yuzawa S, Keasling JD, et al. Regulation of endoplasmic reticulum turnover by selective autophagy. A reporter system for cytosolic protein aggregates in yeast. ACS Synth Biol. Nature. 2015;522(7556):354–8. 2021;10(3):466–77. 45. Bhaskara RM, Grumati P, Garcia-Pardo J, Kalayil S, Covarrubias-Pinto A, Chen W, 34. Ibstedt S, Sideri TC, Grant CM, Tamas MJ. Global analysis of protein aggrega- et al. Curvature induction and membrane remodeling by FAM134B reticulon tion in yeast during physiological conditions and arsenite stress. Biol Open. homology domain assist selective ER-phagy. Nat Commun. 2019;10(1):2370. 2014;3(10):913–23. 46. Brady JP, Claridge JK, Smith PG, Schnell JR. A conserved amphipathic helix is 35. Pintado-Grima C, Barcenas O, Bartolomé-Nafría A, Fornt-Suñe M, Iglesias V, required for membrane tubule formation by Yop1p. Proc Natl Acad Sci U S A. Garcia-Pardo J, et al. A review of Fifteen Years developing computational 2015;112(7):E639–48. tools to study protein aggregation. Biophysica. 2023;3(1):1–20. 47. Voeltz GK, Prinz WA, Shibata Y, Rist JM, Rapoport TA. A class of membrane pro- 36. Baek M, DiMaio F, Anishchenko I, Dauparas J, Ovchinnikov S, Lee GR, et al. teins shaping the tubular endoplasmic reticulum. Cell. 2006;124(3):573–86. Accurate prediction of protein structures and interactions using a three-track 48. Campelo F, McMahon HT, Kozlov MM. The hydrophobic insertion neural network. Science. 2021;373(6557):871–6. mechanism of membrane curvature generation by proteins. Biophys J. 37. Dhar R, Sagesser R, Weikert C, Wagner A. Yeast adapts to a changing stressful 2008;95(5):2325–39. environment by evolving cross-protection and anticipatory gene regulation. 49. Wang N, Clark LD, Gao Y, Kozlov MM, Shemesh T, Rapoport TA. Mechanism of Mol Biol Evol. 2013;30(3):573–88. membrane-curvature generation by ER-tubule shaping proteins. Nat Com- 38. Narayanaswamy R, Levy M, Tsechansky M, Stovall GM, O’Connell JD, mun. 2021;12(1):568. Mirrielees J, et al. Widespread reorganization of metabolic enzymes into 50. Shibata Y, Voss C, Rist JM, Hu J, Rapoport TA, Prinz WA, et al. The reticulon and reversible assemblies upon nutrient starvation. Proc Natl Acad Sci U S A. DP1/Yop1p proteins form immobile oligomers in the tubular endoplasmic 2009;106(25):10147–52. reticulum. J Biol Chem. 2008;283(27):18892–904. 39. Fernandez-Escamilla AM, Rousseau F, Schymkowitz J, Serrano L. Prediction of 51. Xiang Y, Lyu R, Hu J. Oligomeric scaffolding for curvature generation by ER sequence-dependent and mutational effects on the aggregation of peptides tubule-forming proteins. Nat Commun. 2023;14(1):2617. and proteins. Nat Biotechnol. 2004;22(10):1302–6. 52. The PyMOL Molecular. Graphics System, Version 2.0 Schrödinger, LLC. 40. Stoddard PR, Lynch EM, Farrell DP, Dosey AM, DiMaio F, Williams TA, et al. Polymerization in the actin ATPase clan regulates hexokinase activity in yeast. Science. 2020;367(6481):1039–42. Publisher’s Note 41. Pechmann S, Levy ED, Tartaglia GG, Vendruscolo M. Physicochemical prin- Springer Nature remains neutral with regard to jurisdictional claims in ciples that regulate the competition between functional and dysfunctional published maps and institutional affiliations. association of proteins. Proc Natl Acad Sci U S A. 2009;106(25):10159–64. 42. Chen S, Novick P, Ferro-Novick S. ER structure and function. Curr Opin Cell Biol. 2013;25(4):428–33.