D352–D357 Nucleic Acids Research, 2023, Vol. 51, Database issue Published online 16 October 2022 https://doi.org/10.1093/nar/gkac882 AmyloGraph: a comprehensive database of amyloid–amyloid interactions Michał Burdukiewicz 1,2,* , Dominik Rafacz 3 , Agnieszka Barbach 4 , Katarzyna Hubicka4 , 3 ˛ Laura Bakała , Anna Lassota 5 , Jakub Stecko 6 , Natalia Szymańska6 , Jakub W. Wojciechowski 4 , Dominika Kozakiewicz 7 , Natalia Szulc 4 , Jarosław Chilimoniuk 8 , Izabela Je˛ śkowiak 9 , Marlena Gasior-Głogowska ˛ 4 and 4 Małgorzata Kotulska 1 Institute of Biotechnology and Biomedicine, Autonomous University of Barcelona, Campus Universitat Autònoma de Barcelona Plaça Cı́vica Bellaterra, s/n, 08193 Cerdanyola del Vallès, Barcelona, Spain, 2 Clinical Research Centre, Medical University of Białystok, Kilińskiego 1, 15-369 Białystok, Poland, 3 Faculty of Mathematics and Information Science, Warsaw University of Technology, Koszykowa 75, 00-662 Warsaw, Poland, 4 Department of Biomedical Engineering, Faculty of Fundamental Problems of Technology, Wrocław University of Science and Technology, Wybrzeże Wyspiańskiego 27, 50-370 Wrocław, Poland, 5 School of Biosciences, College of Life and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham B15 2TT, United Kingdom, 6 Faculty of Medicine, Wrocław Medical University, Ludwika Pasteura 1, 50-367 Wrocław, Poland, 7 Laboratory of Microbiome Immunobiology, Hirszfeld Institute of Immunology and Experimental Therapy, Polish Academy of Sciences, Weigla 12, 53-114 Wrocław, Poland, 8 Department of Genomics, Faculty of Biotechnology, University of Wrocław, Fryderyka Joliot-Curie 14a, 50-383 Wrocław, Poland and 9 Department of Pharmacology, Wroclaw Medical University, Mikulicza-Radeckiego 2, 50-345 Wrocław, Poland Received August 14, 2022; Revised September 22, 2022; Editorial Decision September 27, 2022; Accepted September 30, 2022 ABSTRACT only publicly available repository for experimentally determined amyloid–amyloid interactions. Information about the impact of interactions be- tween amyloid proteins on their fibrillization propen- sity is scattered among many experimental articles INTRODUCTION and presented in unstructured form. We manually Amyloids are proteins able to self-assembly into insoluble curated information located in almost 200 publica- ␤-sheet supra-molecular fibrils characterized by very regu- tions (selected out of 562 initially considered), ob- lar beta-cross structures. Some of them interact with each taining details of 883 experimentally studied inter- other during fibrillization, which may accelerate or slow actions between 46 amyloid proteins or peptides. down development of fibrils or even lead to the formation We also proposed a novel standardized terminol- of heterogeneous fibrils (1). Interactions between amyloid ogy for the description of amyloid–amyloid inter- proteins raise a growing interest since they may contribute to amyloid-related diseases. The aggregation of amyloid fib- actions, which is included in our database, cover- rils can be associated with pathologies observed in a wide ing all currently known types of such a cross-talk, range of diseases known as amyloidoses. For example, amy- including inhibition of fibrillization, cross-seeding loid fibrils which aggregate in the brain and central nervous and other phenomena. The new approach allows for system, are related to Alzheimer’s and Parkinson’s diseases more specific studies on amyloids and their inter- (2). Another example is a prion conversion, where ingested actions, by providing very well-defined data. Amy- misfolded proteins can seed the aggregation of their homol- loGraph, an online database presenting informa- ogous polypeptide sequence (3). Similar mechanisms can tion on amyloid–amyloid interactions, is available trigger other amyloidoses (4), but experimental data con- at (http://AmyloGraph.com/). Its functionalities are sidering such phenomena are dispersed and often very in- also accessible as the R package (https://github. compatible. com/KotulskaLab/AmyloGraph). AmyloGraph is the The importance of interactions between amyloid proteins makes them a subject of numerous experimental studies. * To whom correspondence should be addressed. Email: michalburdukiewicz@gmail.com  C The Author(s) 2022. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Nucleic Acids Research, 2023, Vol. 51, Database issue D353 However, reviews on interactions between amyloid proteins tor. In this case, such interactions were reported bidirection- show that available experimental results are often contra- ally, where protein A acts as an interactor and protein B as dictory (5). Although the information on amyloid proteins the interactee, and vice versa. is collected in several databases (6–10), until now there has Our descriptors do not replace existing terminology, but been no database consolidating the results of numerous ex- rather standardize it. For example, a combination of an- periments studying interactions of amyloids. Importantly, swers to descriptor I ‘The impact on the speed of fibril- existing efforts to present the overview of amyloid–amyloid lization’: ‘Faster aggregation’ and descriptor II ‘Physical interactions do not allow for a more in-depth inspection of binding between interactor and interactee’: ‘Yes, direct evi- data (11). dence’ or ‘Yes, implied by kinetics’ can be related to either Other problems arise from the lack of clear definitions of cross-seeding or co-incubation (12). These two experiments field. Although there have been attempts to standardize the are drastically different design-wise: co-incubation requires vocabulary (12,13) or a list of requirements necessary in re- both interactee and interactor to be in monomeric form, porting amyloid studies (14), the practices are still not being while in cross-seeding experiment interactee is monomeric fully implemented. Therefore, comparing different studies, and interactor in the form of small aggregates. However, especially those regarding amyloid–amyloid interactions, is studies frequently do not mention the exact form of inter- problematic and even fundamental concepts may be under- actee and interactor making general descriptors easier to stood incompatibly. use and more accurate. Therefore, we designed a structured vocabulary to de- scribe amyloid–amyloid interactions more rigorously. It Database scope covers descriptors that fully define the exact nature of the The scope of the current AmyloGraph version is limited influence of an interactor on an interactee. Using the pro- to interactions between two proteins, each of them able to posed methodology, we manually curated a majority of re- form an amyloid-like aggregate by its wild type. Addition- ported interactions between amyloids and presented this in- ally, we allowed for non-aggregating homologs of a well- formation in the form of an interactive graph and a tabular known amyloid protein, such as rat amylin (15). database. While selecting the source publications, we focused on in vitro studies published after 2000. The complete list of eligi- MATERIALS AND METHODS bility criteria is available in the Supplementary Information, Standardized terminology section Manuscript collection. To describe interactions between amyloid proteins, we cre- Data acquisition and curation ated a precisely controlled vocabulary. First, we defined six possible scenarios of amyloid–amyloid interactions (Fig- To ensure the highest possible quality of the collected data, ure 1A). All scenarios assume that there are only two partic- the data acquisition and curation were executed in a three- ipants in each interaction, and an interactor modulates self- stage pipeline, including: pre-screen of manuscripts, man- assembly of an interactee. We are aware that in reality the ual curation and independent final validation. Importantly, distinction between the interactee and interactor may not the first two steps were supported by dedicated forms which always be clear or, depending on other factors, a specific in- played a crucial role in standardizing annotations provided teraction can fall under more than one scenario at the same by curators. time. However, this simplification allowed us to better de- The pre-screen of manuscripts started with our in-house sign the standardized terminology and led to a more struc- collection of 24 publications. Next, we expanded the search tured description of interactions between amyloid proteins. by repeatedly adding manuscripts cited by manuscripts or Next, we developed three descriptors to more rigorously referencing manuscripts from our collections. The final col- describe details of the scenarios, based on one of the follow- lection included 562 manuscripts, out of which 364 were pu- ing: the fibrillization speed, presence of physical binding be- tatively suitable for the database. Although our collection tween both interacting proteins and appearance of hetero- system was laborious, we found it to be more effective than geneous fibrils (Figure 1 B). a search of PubMed records based on its annotations (Sup- Each descriptor provides specific dictionary terms of pos- plementary Figure S2A). sible states. For example, descriptor I, ‘The impact on the Next, the database curators manually extracted informa- speed of fibrillization’, enables the choice of one of the tion on interactions from the suitable manuscripts. It should following states: ‘faster fibrillization’, ‘slower fibrillization’, be emphasized that, in the curation procedure, we did not ‘no fibrillization’, ‘no effect’ and ‘no information’. The de- re-interpret data and conclusions provided by their authors. signed states are mutually exclusive and provide in-depth Curators interpreted the data only if the authors did not description to relate them to relevant experimental results provide a description of the results or if the existing descrip- (Supplementary Information, section Descriptors and Sup- tion was too limited. To help curators interpret the results, plementary Figure S1). we enhanced the descriptors with the specifics of experimen- It is essential to stress that most publications provided tal procedures, which helped them to identify a descriptor information only on the interactee’s (a protein whose level and draw the best final conclusion. Moreover, during self-assembly is modulated by interactee) ability to create the project, we developed a FAQ list of almost 100 questions amyloid-like fibrils. Therefore, the descriptors focus only on related to the curation procedure, which helped the curators. the behavior of interactee. However, some manuscripts re- Thanks to all these precautions, the curators could provide ported on the self-assembly of both interactee and interac- data of a higher quality. D354 Nucleic Acids Research, 2023, Vol. 51, Database issue A B Figure 1. (A) Six scenarios of interactions between amyloid proteins. Orange and blue tiles denotes molecules of amyloid-like proteins participating in in- teractions. Roman numerals denote different interactions scenarios. (B) Three descriptors of AmyloGraph. Rectangles represent the descriptors. Rectangles with round edges represent the levels of the descriptors. Green, yellow and purple represent descriptors 1, 2 and 3, respectively. During the initial curation, curators reviewed all col- the Supplementary Information, section Supplementary ref- lected manuscripts. Curators annotated these interactions erences. using our three descriptors and collected information on se- quences of proteins participating in the interaction, focus- Implementation ing on the presence of mutations or other alterations. Cu- rators were also obliged to preserve parts of the publication One of the main limitations of web-based tools is their in- (in the graphical or textual form) supporting their decisions built reliance on the external servers which reduces their regarding final description in the records. persistence (16). Therefore, we made AmyloGraph fully de- After the initial curation, we validated all the collected ployable and usable even if the main server is no longer data to further increase their quality. In this procedure, available. To do so, we implemented our tool as an R new curators reviewed the assigned interaction records. The package (17). The package contains also the front-end semi-random assignment procedure ensured that the cura- of our database, available as a Shiny app (18). The lo- tor who validated a specific record was not involved in its cal deployment of AmyloGraph only requires a very rudi- initial curation. Finally, the correct records were accepted mentary knowledge of R and is described in the Amy- in the database. The manual curation resulted in 172 publi- loGraph main repository at https://kotulskalab.github.io/ cations and 883 interactions (Supplementary Figure S2B). AmyloGraph/. The AmyloGraph codebase is open and doc- Next, we contacted the authors of all 172 publications umented in the roxygen2 standard. included in AmyloGraph to obtain their validation of our records. We always tried to reach the corresponding au- DATABASE OVERVIEW thors or, in case of their unavailability, the publication’s first author. The authors were provided with customized links Manually curated data, obtained in the previously described to Google Sheets only containing data from manuscripts procedure, are available in the AmyloGraph database. Cur- they authored. In total, we contacted 122 authors, and 11 rently, the database includes 883 interactions between 46 authors (9.04%) confirmed 81 interactions (9.17%) in 21 proteins reported in 172 manuscripts. Furthermore, one of manuscripts (12.14%). Despite our efforts, we could not find the main objectives of the database is to present the inter- correct and up-to-date contact information to authors of actions between amyloid proteins in a standardized man- three manuscripts. It is important to notice that no inter- ner and user-friendly presentations, such as a graph format actions were removed or added after the contact with the (Figure 2A). Here, nodes represent individual amyloid pro- authors, which implies a high data quality (Supplementary teins and edges stand for interactions between them. No- Figure S2C). tably, a single edge represents all interactions between two The in-depth description of the curation procedure is amyloid proteins. Tooltips of the edges represent digital ob- available in the Supplementary Information, section Data ject identifiers (DOIs) of manuscripts reporting the interac- acquisition. The full list of 172 manuscripts is available in tions. Nucleic Acids Research, 2023, Vol. 51, Database issue D355 A B C Figure 2. Overview of the AmyloGraph database. (A) Graph view of interactions between amyloid proteins. The interactions (edges of the graph) are colored according to the levels of descriptor 2, ‘physical binding’. The panel on the right-hand side represents an overview of the amyloid-␤ interactions. (B) Tabular view of interactions. The top section of this card contains download options allowing to obtain data in a selected flat-file format. (C) View of a single interaction with the sequential information. D356 Nucleic Acids Research, 2023, Vol. 51, Database issue After clicking on a node, a panel on the right-hand side AmyloGraph offers a submission form for authors involved opens. It presents brief information on a protein, its name in relevant research to report their results directly to the and links to its UniProt record. If a single amyloid pro- database. Thus, we are going to implement a highly struc- tein in AmyloGraph is associated with several records in the tured system for finding publications by annotating records UniProt, we provide links to all of them. This panel also acquired in searches of the PubMed database. contains two tables presenting all interactees and interac- One of other challenges regarding AmyloGraph is rep- tors of the protein. resentation of protein data. Right now, AmyloGraph is Aside from the graph, AmyloGraph enables tabular rep- very protein-centric and treats whole families of homologs resentation of the interaction data (Figure 2B). The table or variants of a single protein as a single entity. The ac- is interactive and searchable. A user can also download se- tual situation is much more complicated as we often deal lected rows in a flat-table format (.csv or .xlsx). As a result, with fragments of recombinant proteins or even protein the downloaded table contains all available information, in- grafts (21). In the future, we want to extend AmyloGraph to cluding the sequences of amyloid proteins participating in contain more information about the protein sequence and the interactions. allow proteins with non-standard amino acids or even non- Both, graph and tabular representations of the data can amino-acid modifications. This change is also necessary to be filtered out using filters available on the left-hand side add to AmyloGraph information on the impact of small of the user interface. The filters cover all three descriptors. molecules on amyloid fibrillization. Moreover, a user can color the edges on the graph, accord- Another limitation of AmyloGraph, which we plan to ing to the levels of a chosen descriptor. The user can also use alleviate in the next version, is the lack of experimental amino acid sequences to filter the information presented in information. Amyloid assembly process and their interac- the graph or tabular form. Here, we implemented a simpli- tions are extremely liable to experimental conditions, such fied set of regular expression inspired by the POSIX system as pH (22) or concentration of proteins (23). We are con- to facilitate more advanced searches. vinced that each record in AmyloGraph should be anno- The last card of the graphical interface, ‘Interaction’, tated with more parameters defining the environment of the opens when the user selects a specific interaction (Fig- interaction. ure 2C). This view presents information on a single interac- Even considering all limitations described above, we still tion, including levels of all descriptors and exact sequences believe that the current release of AmyloGraph is a valuable of proteins. In case of multi-chain proteins, such as insulin, tool that provides access to a unique dataset. To our knowl- AmyloGraph presents the sequences of all chains. edge, AmyloGraph is the first effort to collect and present To streamline the use of AmyloGraph, we enhanced it information on interactions between amyloid proteins in a with helpers explaining basic functionalities of the database. unified format. AmyloGraph’s high accessibility and data Moreover, a video tutorial is available, presenting examples quality further enhance its usefulness. of AmyloGraph queries. AmyloGraph is a FAIR-compliant database (19). All in- DATA AVAILABILITY teractions are identifiable by an individual index. They are also linked to original publications using their DOIs pro- The forms supporting collection of manuscripts, initial cu- vided by the Crossref. Proteins participating in interactions ration and validation of manuscripts are available upon are linked to the UniProt database (20). As recommended request to the corresponding author. The procedure of by the FAIR guidelines, we extended the existing vocabu- data acquisition, the in-depth definitions of AmyloGraph lary to describe our data by fully providing our standard- descriptors and references to all curated manuscripts, ized methodology. are available in the Supplementary Information and on- line at https://kotulskalab.github.io/AmyloGraph/articles/ definitions.html. AmyloGraph is available as an online CONCLUSIONS AND FUTURE DIRECTIONS database (http://AmyloGraph.com/). AmyloGraph is the first endeavor to present an overview of experimentally verified interactions between amyloid pro- CODE AVAILABILITY teins. It has also been the first attempt to standardize re- porting of the amyloid–amyloid interactions and present All AmyloGraph functionalities are also accessible as the R them in the interactive database. We believe that, thanks package (https://github.com/KotulskaLab/AmyloGraph). to our rigorous data curation procedure, we have managed to collect and thoroughly systematize the majority of avail- SUPPLEMENTARY DATA able information. Even though AmyloGraph is currently the most comprehensive compendium on the interactions Supplementary Data are available at NAR Online. between amyloid proteins, we see three areas that require an improvement: constant updates, representation of pro- ACKNOWLEDGEMENTS tein data and extending information of experimental condi- tions. Access to Wroclaw Centre for Networking and Supercom- The greatest challenge regarding AmyloGraph, which we puting at Wroclaw University of Science and Technology is envisage, will be to keep it updated. We encountered a sud- greatly acknowledged. We also thank Daniel Otzen (Aarhus den influx of new publications reporting new interactions University, Denmark) and Vytautas Smirnovas (University during our work on the database. To alleviate this issue, of Vilnus, Lithuania) for fruitful discussions. Nucleic Acids Research, 2023, Vol. 51, Database issue D357 FUNDING interactome: exploring protein aggregation. PLOS ONE, 12, e0173163. National Science Centre, Poland [2019/35/B/NZ2/03997]; 12. Bondarev,S.A., Antonets,K.S., Kajava,A.V., Nizhnikov,A.A. and M.B. was supported by the Maria Zambrano grant funded Zhouravleva,G.A. (2018) Protein co-aggregation related to amyloids: by the European Union-NextGenerationEU. methods of investigation, diversity, and classification. Int. J. Mol. Sci., 19, 2292. Conflict of interest statement. None declared. 13. Ren,B., Zhang,Y., Zhang,M., Liu,Y., Zhang,D., Gong,X., Feng,Z., Tang,J., Chang,Y. and Zheng,J. (2019) Fundamentals of cross-seeding of amyloid proteins: An introduction. J. Mat. Chem. B, 7, 7267–7282. REFERENCES 14. Martins,P.M., Navarro,S., Silva,A., Pinto,M.F., Sárkány,Z., 1. Konstantoulea,K., Louros,N., Rousseau,F. and Schymkowitz,J. Figueiredo,F., Pereira,P.J.B., Pinheiro,F., Bednarikova,Z., (2022) Heterotypic interactions in amyloid function and disease. Burdukiewicz,M. et al. (2020) MIRRAGGE – Minimum information FEBS J., 289, 2025–2046. required for reproducible aggregation experiments. Front. Mol. 2. Ivanova,M.I., Lin,Y., Lee,Y.-H., Zheng,J. and Ramamoorthy,A. Neurosci., 13, 582488. (2021) Biophysical processes underlying cross-seeding in amyloid 15. Green,J., Goldsbury,C., Mini,T., Sunderji,S., Frey,P., Kistler,J., aggregation and implications in amyloid pathology. Biophys. Chem., Cooper,G. and Aebi,U. (2003) Full-length rat amylin forms fibrils 269, 106507. following substitution of single residues from human Amylin. J. Mol. 3. Gil-Garcia,M., Iglesias,V., Pallarès,I. and Ventura,S. (2021) Biol., 326, 1147–1156. Prion-like proteins: from computational approaches to 16. Veretnik,S., Fink,J.L. and Bourne,P.E. (2008) Computational biology proteome-wide Analysis. FEBS Open Biol., 11, 2400–2417. resources lack persistence and usability. PLoS Comput. Biol., 4, 4. Friedland,R.P. and Chapman,M.R. (2017) The role of microbial e1000136. amyloid in neurodegeneration. PLoS Path., 13, e1006654. 17. R Core Team (2022) In: R: A Language and Environment for 5. Tran,J., Chang,D., Hsu,F., Wang,H. and Guo,Z. (2017) Cross-seeding Statistical Computing. R Foundation for Statistical Computing, between A␤40 and A␤42 in Alzheimer’s disease. FEBS Lett., 591, Vienna, Austria. 177–185. 18. Chang,W., Cheng,J., Allaire,J., Sievert,C., Schloerke,B., Xie,Y., 6. Wozniak,P.P. and Kotulska,M. (2015) AmyLoad: Website dedicated Allen,J., McPherson,J., Dipert,A. and Borges,B. (2022) shiny: web to amyloidogenic protein fragments. Bioinformatics (Oxford, application framework for R. R package version 1.7.2. England), 31, 3395–3397. 19. Wilkinson,M.D., Dumontier,M., Aalbersberg,I.J., Appleton,G., 7. Louros,N., Konstantoulea,K., De Vleeschouwer,M., Ramakers,M., Axton,M., Baak,A., Blomberg,N., Boiten,J.-W., da Silva Santos,L.B., Schymkowitz,J. and Rousseau,F. (2020) WALTZ-DB 2.0: An updated Bourne,P.E. et al. (2016) The FAIR guiding principles for scientific database containing structural information of experimentally data management and stewardship. Scientific Data, 3, 160018. determined amyloid-forming peptides. Nucleic Acids Res., 48, 20. The UniProt Consortium (2021) UniProt: the universal protein D389–D393. knowledgebase in 2021. Nucleic Acids Res., 49, D480–D489. 8. Varadi,M., De Baets,G., Vranken,W.F., Tompa,P. and Pancsa,R. 21. Azoitei,M.L., Correia,B.E., Ban,Y.-E.A., Carrico,C., Kalyuzhniy,O., (2018) AmyPro: A database of proteins with validated amyloidogenic Chen,L., Schroeter,A., Huang,P.-S., McLellan,J.S., Kwong,P.D. et al. regions. Nucleic Acids Res., 46, D387–D392. (2011) Computation-guided backbone grafting of a discontinuous 9. Rawat,P., Prabakaran,R., Sakthivel,R., Thangakani,M.A., Kumar,S. motif onto a protein scaffold. Science, 334, 373–376. and Gromiha,M. (2020) CPAD 2.0: A repository of curated 22. Pfefferkorn,C.M., McGlinchey,R.P. and Lee,J.C. (2010) Effects of pH experimental data on aggregating proteins and peptides. Amyloid, 27, on aggregation kinetics of the repeat domain of a functional amyloid, 128–133. Pmel17. Proc. Natl. Acad. Sci., 107, 21447–21452. 10. Pawlicki,S., Le Béchec,A. and Delamarche,C. (2008) AMYPdb: A 23. Hu,X., Crick,S.L., Bu,G., Frieden,C., Pappu,R.V. and Lee,J.-M. database dedicated to amyloid precursor proteins. BMC Bioinform., (2009) Amyloid seeds formed by cellular uptake, concentration, and 9, 273. aggregation of the amyloid-beta peptide. Proc. Natl. Acad. Sci. 11. Biza,K.V., Nastou,K.C., Tsiolaki,P.L., Mastrokalou,C.V., U.S.A., 106, 20324–20329. Hamodrakas,S.J. and Iconomidou,V.A. (2017) The Amyloid