Iglesias et al. BMC Bioinformatics (2019) 20:24 https://doi.org/10.1186/s12859-019-2601-3 SOFTWARE Open Access AMYCO: evaluation of mutational impact on prion-like proteins aggregation propensity Valentin Iglesias1,2†, Oscar Conchillo-Sole1†, Cristina Batlle1,2 and Salvador Ventura1,2* Abstract Background: Around 1% of human proteins are predicted to contain a disordered and low complexity prion-like domain (PrLD). Mutations in PrLDs have been shown promote a transition towards an aggregation-prone state in several diseases. Results: Recently, we have shown that an algorithm that considers the effects of mutations on PrLDs composition, as well as on localized amyloid propensity can predict the impact of these amino acid changes on protein intracellular aggregation. In this application note, we implement this concept into the AMYCO web server, a refined algorithm that forecasts the influence of amino acid changes in prion-like proteins aggregation propensity better than state-of-the-art predictors. Conclusions: The AMYCO web server allows for a fast and automated evaluation of the effect of mutations on the aggregation properties of prion-like proteins. This might uncover novel disease-linked amino acid changes in the sequences of human prion-like proteins. Additionally, it can find application in the in silico design of synthetic prion-like proteins with tuned aggregation propensities for different purposes. AMYCO does not require previous registration and is freely available to all users at: http://bioinf.uab.cat/amyco/. Keywords: Prion-like domain, Protein aggregation, Amyloid, Protein mutation Background demixing [7, 8]. A number of mutations in human PrLDs Prions are proteins able to adopt multiple structural con- have been shown to convert these liquid compartments formations from which at least one has self-propagating into solid aggregates, abolishing their dynamic nature and properties [1]. Yeast prions are the best understood subset leading to the onset of neurodegenerative disorders [8, 9]. of functional prions. A common feature of most yeast The development of tools able to anticipate the impact of prions is the presence of an intrinsically disordered and such pathogenic amino acid changes is attracting increas- low complexity prion domain (PrD), which is necessary ing interest. and sufficient for prion conversion and propagation. Pro- The self-assembling properties of prion-like pro- teins bearing prion-like domains (PrLDs) sharing these teins have been traditionally thought to rely on the properties seem to exist in all kingdoms of life [2–6]. In biased amino acid composition of their PrLDs [10]. particular, around 1% of the human proteome has been Disease-linked mutations would enhance the predicted to correspond to prion-like proteins [7]. This self-association of these domains, facilitating the human protein subset is enriched in nucleic acid-binding transition to amyloid-like states [11, 12]. We have proteins and involved in the formation of membraneless recently shown that the impact of point and multiple compartments through highly dynamic liquid-liquid mutations or deletions on the aggregation of the model ALS-associated prion-like hnRNPA2 protein is * Correspondence: salvador.ventura@uab.cat best predicted by a function that takes into account † Valentin Iglesias and Oscar Conchillo-Sole contributed equally to this work. both compositional features and amyloidogenic pro- 1 Institut de Biotecnologia i Biomedicina, Universitat Autònoma de Barcelona, pensities [13]. Here we introduce the AMYCO (com- Bellaterra 08193, Spain 2 Departament de Bioquímica i Biologia Molecular, Universitat Autònoma de bined AMYloid and Composition based prediction of Barcelona, Bellaterra 08193, Spain prion-like aggregation propensity) web server, which © The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. 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. Iglesias et al. BMC Bioinformatics (2019) 20:24 Page 2 of 5 implements this approach to perform automated and CentOS 5 server with Apache 2.2.3 using Intel Xeon fast predictions on top of prion-like protein ‘Clovertown’ processors. sequences. AMYCO pipeline Implementation AMYCO evaluates the impact of mutations on the ag- AMYCO is written in Python and uses python2.7 as the gregation propensity of PrLDs in prion-like proteins. interpreter (Anaconda distribution). The web interface They can be single or multiple residues substitutions, as has been build using html/css and inputs and outputs well as deletions and insertions. It exploits the highly are processed by a cgi written in perl. It all runs in a significant correlation between the scores obtained from Fig. 1 Correlation between AMYCO and pRANK predictions and the aggregation propensity of human hnRNPA2 prion-like protein variants. a Graphic representation of the correlation between the mutants' AMYCO (a) and pRANK (b) scores and their ability to form prionic colonies (Ade +) when expressed in yeast, a direct reporter of their aggregation propensity [13] Iglesias et al. BMC Bioinformatics (2019) 20:24 Page 3 of 5 a parameterized linear function, that balances the contri- Sequences mode, or all its possible mutants for bution of both PrLDs composition and amyloid propen- position 290 for Single Mutation mode. sity [13], and the intracellular aggregation of hnRNPA2 The input interface allows two working modes. In variants; the unique prion-like protein for which a large Compare Sequences mode (default mode); the user set of mutations, both natural and artificial have been should introduce a reference sequence and the mutated experimentally validated (Fig. 1a). The contribution of variants (one or several) in the left and right text boxes, PrLDs amino acid composition to prion-like proteins respectively; all in FASTA format. In the Single Mutation aggregation is calculated using the PAPA algorithm [10], mode, the user should introduce a single sequence as a program that exploits a scale of prion propensity well as the position to be scanned. Protein sequences scores for natural amino acids derived from mutagenesis should be at least 60-residues long and only the 20 experiments in the Sup35 yeast prion domain [14]. The standard proteinogenic amino acids are allowed. impact of amyloidogenic sequences within the PrLDs is After submission, the output page will display a job calculated with pWALTZ [12], a program specially identification number along with the names of the input intended to identify short sequences of moderate amyl- sequences and the mutation position if applicable. The oid propensity able to nucleate the aggregation reaction, algorithm will return the AMYCO score for each as those found in yeast prion domains [15]. The outputs sequence, together with a description of the mutations of the two different programs are combined in a linear impact of the overall aggregation propensity. In addition, manner, as described in detail in the Additional file 1. a graphical representation of the mutation/s effect will The AMYCO web server is free and open to all users, be displayed (Fig. 2). We set two arbitrary thresholds of and no previous login or registration is required. low (< 0.45) and high (> 0.78) AMYCO scores to The home page of AMYCO displays three clickable visualize better the overall aggregation propensities of links in its upper margin: (i) a help page containing a the variants. hnRNPA2 mutants scoring < 0.45 were brief description of the method, the output explan- shown to decrease or increase < 5 times the propensity ation and information on examples, (ii) references for of the non-aggregating wild type protein, whereas, the methodology and the web application and (iii) a mutants scoring > 0.78 increased its aggregation by > 50 contact e-mail. Immediately below two links allow to times [17]. Therefore, mutations rendering an AMYCO switch between the Compare Sequences mode in score < 0.45 are considered of low aggregation propen- which multiple sequences can be compared to a refer- sity and labeled in blue. Mutations that increase the ag- ence one, and the Single Mutation mode in which all gregation propensity of the protein, but whose AMYCO possible mutations for a given protein residue are score is below 0.78 are labeled in red, whereas mutations evaluated. The adjacent “Example” button fills the in- above this threshold are considered to be of high aggre- put text areas with the full-length sequences of wild gation propensity are labeled in red and bold. Sequences type (wt) human hnRNPA2 protein and its might display AMYCO scores > 1.0, indicating that they aggregation-prone D290V mutant [16] for Compare are predicted to be more aggregation-prone than the Fig. 2 Graphical representation of the AMYCO score. AMYCO output representation of a low aggregation-prone (D290V Y283K), the wild type, the natural pathogenic mutant D290V and a high aggregation-prone (D290V D276V) hnRNPA2 prion-like protein variants Iglesias et al. BMC Bioinformatics (2019) 20:24 Page 4 of 5 highest scoring hnRNPA2 variant used in the paramet- Table 2 AMYCO correctly predicts prion converting mutations rization of the prediction function. The output files can on yeast proteins be downloaded for further analyses, as a ZIP file con- Protein variant AMYCO score taining the resulting text explanation, a machine read- PUF4 wt 0 able JSON file, the visualizations as a PNG file and PUF4mut 0.69 FASTA files with the introduced sequences and the 6PP,1N *PUF4 0.93 virtually generated mutants in Compare Sequences and PUF44PP 0.60 Single Mutation modes, respectively. YLR177W wt 0 Results *YLR177Wmut 0.85 Performance *YLR177W4PP,1N 1.23 pRANK is a novel multiple-instance machine learning *YLR177W4PP 1.03 method aimed to predict prion propensity based on KC11 wt 0 amino acid composition alone [18]. We compared the *KC11mut 0.97 performance of AMYCO and pRANK web servers in predicting the impact of mutations on human hnRNPA2 PDC2 wt 0 aggregation propensity (Fig. 1). AMYCO clearly outper- PDC2mut 0.78 forms pRANK (Table 1), an observation which is con- AMYCO correctly predicts mutations that induce prionic phenotypes. sistent with the important influence that sequential Mutations predicted to increase and highly increase aggregation propensity are shown in italics and bold, respectively. Variants that do not need features exert on protein aggregation [19]. overexpression to generate a prionic phenotype in yeast are indicated with an AMYCO was further assayed on known mutations asterisk [20] promoting the apparition of a de novo prion-like behav- ior (Table 2). It was able to predict a large increase in mutations occurring in different human prion-like pro- aggregation propensity for mutations that convert the teins. In particular, mutations in hnRNPA1 associated to non-prionic PrLDs of PUF4, YLR177W, KC11 and ALS [16], mutations in hnRNP D0/AUF1 identified in PDC2 yeast proteins into prionic when expressed in familiar cases of Crohn Disease [21] and mutations in yeast [20] (Table 2). Importantly, according to AMYCO, hnRNP DL causing limb-girdle muscular dystrophy 1G five out of the eight variants were predicted to have ac- [22] (Table 3). In contrast, natural variants of these pro- quired a very high aggregation propensity. These vari- teins bearing mutations in the PrLDs not associated to ants are exactly the ones experimentally shown to clinical phenotypes [23] do not have any significant induce a prionic phenotype with basal protein levels, without a need for overexpression [20] (Table 2). Finally, AMYCO is able to predict an increase in ag- Table 3 AMYCO predicts disease-causing mutations on human gregation propensity for a series of disease-linked prion-like proteins Protein variant AMYCO score Table 1 Performance of pRANK and AMYCO approaches in the hnRNPA1 wt 0.34 prediction of mutation impact upon the aggregation of the hnRNPA1 Q277K 0.34 human prion-like protein hnRNPA2 pRANK AMYCO hnRNPA1 G283R 0.34 Sensitivity 0 1 hnRNPA1 P340S 0.36 Specificity 1 1 hnRNPA1 D314V 0.59 Precision – 1 hnRNPA1 D314N 0.53 Accuracy 0.45 1 hnRNP DL wt 1.18 MCC – 1 hnRNP DL D378H 1.26 Mean % error −7.08 −1.25 hnRNP DL D378N 1.30 Standard Deviation (%) 37.71 12.19 hnRNP D0 wt 1.13 SEM (%) 8.04 2.44 hnRNP D0 F225 L 1.13 Coefficient of Determination 0.152 0.882 hnRNP D0 D319V 1.33 P-value (two tailed test) 0.468 < 1.00E-08 hnRNP D0 isoform-2 D300V 1.33 AMYCO identifies multisystem proteinopathy and ALS causing mutations on Rho (ρ) 0.334 0.929 hnRNPA1 [16], Crohn Disease causing mutations on both isoforms of hnRNP The best performance according to each particular parameter is shown in D0/AUF1 [21] and limb-girdle muscular dystrophy 1G (LGMD1G) on hnRNP DL bold. Details on the calculation of the different statistic parameters are [22] are shown in bold. Natural variants not associated to a clinical phenotype provided in the Additional file 1 are shown in italics Iglesias et al. BMC Bioinformatics (2019) 20:24 Page 5 of 5 impact in the polypeptides predicted aggregation Received: 10 July 2018 Accepted: 3 January 2019 propensity. References Conclusion 1. Aguzzi A, Calella AM. Prions: protein aggregation and infectious diseases. Physiol Rev. 2009;89:1105–52. AMYCO has been developed as a web application to as- 2. Malinovska L, Palm S, Gibson K, Verbavatz JM, Alberti S. Dictyostelium sess the impact of mutations on the aggregation propen- discoideum has a highly Q/N-rich proteome and shows an unusual sity of prion-like proteins, allowing a fast and accurate resilience to protein aggregation. Proc Natl Acad Sci U S A. 2015; 112(20):E2620–9. evaluation of the effect of disease-associated mutations 3. Iglesias V, de Groot NS, Ventura S. Computational analysis of candidate in these polypeptides; as well as engineering novel vari- prion-like proteins in bacteria and their role. Front Microbiol. 2015;6:1–13. ants with designed aggregation propensities for different 4. Pallarès I, Iglesias V, Ventura S. The rho termination factor of Clostridium botulinum contains a prion-like domain with a highly Amyloidogenic Core. applications. Front Microbiol. 2016;6:1–12. 5. Yuan AH, Hochschild A. A bacterial global regulator forms a prion. Science. 2017;355(6321):198–201. Availability and requirements 6. Chakrabortee S, Kayatekin C, Newby GA, Mendillo ML, Lancaster A, Lindquist Project name: AMYCO. S. Luminidependens (LD) is an Arabidopsis protein with prion behavior. Project home page: http://bioinf.uab.cat/amyco/. Proc Natl Acad Sci. 2016;113(21):6065–70 201604478. 7. King OD, Gitler AD, Shorter J. The tip of the iceberg: RNA-binding Operating system(s): Platform independent. proteins with prion-like domains in neurodegenerative disease. Brain Programming language: A computing core coded in Res. 2012;1462:61–80. Python and a front end written in a combination of html 8. Patel A, Lee HO, Jawerth L, Maharana S, Jahnel M, Hein MY, Stoynov S, Mahamid J, Saha S, Franzmann TM, et al. A liquid-to-solid phase and perl cgi. transition of the ALS protein FUS accelerated by disease mutation. Cell. Other requirements: A web browser with a working 2015;162(5):1066–77. internet connection. 9. Polymenidou M, Cleveland DW. Prion-like spread of protein aggregates in neurodegeneration. J Exp Med. 2012;209(5):889–93. License: None. 10. Toombs JA, Petri M, Paul KR, Kan GY, Ben-Hur A, Ross ED. De novo design Any restrictions to use by non-academics: None. of synthetic prion domains. Proc Natl Acad Sci U S A. 2012;109(17):6519–24. 11. Ryan VH, Dignon GL, Zerze GH, Chabata CV, Silva R, Conicella AE, Amaya J, Burke KA, Mittal J, Fawzi NL. Mechanistic view of hnRNPA2 low-complexity Additional file domain structure, interactions, and phase separation altered by mutation and arginine methylation. Mol Cell. 2018;69(3):465–79 e467. 12. Sabate R, Rousseau F, Schymkowitz J, Ventura S. What makes a protein Additional file 1: Dataset obtention and performance analysis. (PDF 219 kb) sequence a prion? PLoS Comput Biol. 2015;11(1):e1004013. 13. Batlle C, Fernandez MR, Iglesias V, Ventura S. Perfecting prediction of mutational impact on the aggregation propensity of the ALS-associated Abbreviations hnRNPA2 prion-like protein. FEBS Lett. 2017;591(13):1966–71. MCC: Matthews correlation coefficient; PrD: Prion domain; PrLD: Prion like 14. Toombs JA, McCarty BR, Ross ED. Compositional determinants of prion domain; SEM: Standard error of the mean formation in yeast. Mol Cell Biol. 2010;30(1):319–32. 15. Sant’Anna R, Fernandez MR, Batlle C, Navarro S, de Groot NS, Serpell L, Funding Ventura S. Characterization of amyloid cores in prion domains. Sci Rep. This work was funded by the Spanish Ministry of Economy and 2016;6:34274. Competitiveness (BIO2016–783-78310-R to SV). SV has been granted an 16. Kim HJ, Kim NC, Wang YD, Scarborough EA, Moore J, Diaz Z, MacLea KS, ICREA ACADEMIA award. Freibaum B, Li S, Molliex A, et al. Mutations in prion-like domains in hnRNPA2B1 and hnRNPA1 cause multisystem proteinopathy and ALS. Availability of data and materials Nature. 2013;495(7442):467–73. All data analyzed during this study are included in articles [16, 17, 20–23]. 17. Paul KR, Molliex A, Cascarina S, Boncella AE, Taylor JP, Ross ED. Effects of mutations on the aggregation propensity of the human prion-like protein hnRNPA2B1. Mol Cell Biol. 2017;37(8):e00652. Authors’ contributions 18. Afsar Minhas FU, Ross ED, Ben-Hur A. Amino acid composition predicts VI and OC analyzed the data and implemented the web server. CB analyzed prion activity. PLoS Comput Biol. 2017;13(4):e1005465. the data. VI drafted the manuscript. SV designed the research, analyzed the 19. Sabate R, Espargaro A, de Groot NS, Valle-Delgado JJ, Fernandez-Busquets X, data and wrote the final version of the manuscript. All authors read and Ventura S. The role of protein sequence and amino acid composition in approved the final manuscript. amyloid formation: scrambling and backward reading of IAPP amyloid fibrils. J Mol Biol. 2010;404(2):337–52. Ethics approval and consent to participate 20. Paul KR, Hendrich CG, Waechter A, Harman MR, Ross ED. Generating new Not applicable. prions by targeted mutation or segment duplication. Proc Natl Acad Sci U S A. 2015;112(28):8584–9. Consent for publication 21. Prakash T, Veerappa A. N BR: complex interaction between HNRNPD Not applicable. mutations and risk polymorphisms is associated with discordant Crohn’s disease in monozygotic twins. Autoimmunity. 2017;50(5):275–6. 22. Vieira NM, Naslavsky MS, Licinio L, Kok F, Schlesinger D, Vainzof M, Sanchez Competing interests N, Kitajima JP, Gal L, Cavacana N, et al. A defect in the RNA-processing The authors declare that they have no competing interests. protein HNRPDL causes limb-girdle muscular dystrophy 1G (LGMD1G). Hum Mol Genet. 2014;23(15):4103–10. 23. UniProt C. UniProt: a hub for protein information. Nucleic Acids Res. 2015; Publisher’s Note 43(Database issue):D204–12. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.