W300–W307 Nucleic Acids Research, 2019, Vol. 47, Web Server issue Published online 3 May 2019 doi: 10.1093/nar/gkz321 Aggrescan3D (A3D) 2.0: prediction and engineering of protein solubility Aleksander Kuriata1,† , Valentin Iglesias2,† , Jordi Pujols2 , Mateusz Kurcinski1 , Sebastian Kmiecik 1,* and Salvador Ventura2,* 1 Biological and Chemical Research Centre, Faculty of Chemistry, University of Warsaw, 02-089 Warsaw, Poland and 2 Institut de Biotecnologia i Biomedicina and Departament de Bioquı́mica I Biologia Molecular Universitat Autònoma de Barcelona, Bellaterra, Spain Received February 12, 2019; Revised March 29, 2019; Editorial Decision April 18, 2019; Accepted April 20, 2019 ABSTRACT The growing concern about protein aggregation has fu- eled the development of over twenty predictive algorithms Protein aggregation is a hallmark of a growing num- (4,5). A majority of methods identify and score protein ag- ber of human disorders and constitutes a major bot- gregation prone regions (APRs) relying only on protein se- tleneck in the manufacturing of therapeutic proteins. quence. Those programs find difficulties predicting APRs Therefore, there is a strong need of in-silico meth- of folded globular proteins, failing to detect APRs when ods that can anticipate the aggregative properties of residues are not contiguous in sequence or mistaking APRs protein variants linked to disease and assist the engi- for the buried hydrophobic core. These problems motivated neering of soluble protein-based drugs. A few years the development of a second generation of algorithms that ago, we developed a method for structure-based use structure-based approaches for their predictions (6). In prediction of aggregation properties that takes into 2015, we developed the Aggrescan3D (A3D) web server for account the dynamic fluctuations of proteins. The prediction of aggregation properties of protein structures (7). The A3D method was shown to outperform sequence- method has been made available as the Aggrescan3D and composition-based algorithms when dealing with pro- (A3D) web server and applied in numerous studies of teins in their native-like states (7,8). protein structure-aggregation relationship. Here, we A3D integrates the 3D information of protein structures present a major update of the A3D web server to and evaluates the contribution of solvent-exposed APRs. version 2.0. The new features include: extension of The method works by projecting experimental aggregation dynamic calculations to significantly larger and mul- propensities onto a protein structure. Aggregation propen- timeric proteins, simultaneous prediction of changes sity is calculated for spherical regions centred on every in protein solubility and stability upon mutation, residue alpha-carbon using the intrinsic amino acid aggre- rapid screening for functional protein variants with gation scale from the Aggrescan method (9,10), the first improved solubility, a REST-ful service to incorpo- sequence-based algorithm to exploit empirical in vivo data. rate A3D calculations in automatic pipelines, and a This provides a structurally corrected aggregation value (A3D score) for each particular amino acid, depending on new, enhanced web server interface. A3D 2.0 is freely its specific conformational context, discarding the negligi- available at: http://biocomp.chem.uw.edu.pl/A3D2/ ble contribution of hydrophobic residues buried in the core of folded proteins and focusing on protein surfaces. The dy- INTRODUCTION namic structural fluctuations of proteins in solution influ- Protein aggregation lies behind more than 40 human dis- ences the degree of exposure of APRs. For this reason, A3D eases, ranging from neurodegenerative disorders to some incorporates the CABS-flex approach (11,12) for fast simu- types of cancers or diabetes type II (1,2). In addition, aggre- lations of protein flexibility in its dynamic mode. Moreover, gation is a major limitation in the production, storage and A3D allows the introduction of user-defined mutations to administration of life-saving protein pharmaceuticals, like rationally design more soluble protein variants or to test antibodies and replacement enzymes, since it both reduces the impact of disease-linked mutations on the aggregation the percentage of therapeutically active molecules and in- propensity. creases immunogenic responses (3). * Towhom correspondence should be addressed. Tel: +34 93 586 8956; Fax: +34 93 581 2011; Email: salvador.ventura@uab.es Correspondence may also be addressed to Sebastian Kmiecik. Tel: +48 22 8220211 (Ext. 310); Fax: +48 22 8220211 (Ext. 320); Email: sekmi@chem.uw.edu.pl † The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors. C The Author(s) 2019. 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 Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com Nucleic Acids Research, 2019, Vol. 47, Web Server issue W301 Among other applications, A3D has been exploited to understand the binding of chaperones to their targets (13), to study the binding of antimicrobial proteins to mem- branes (14), to rationalize the yield of engineered nanobod- ies (15), to study the aggregation properties of pathogenic (16,17) and non-pathogenic (18) globular proteins or to assist the design of biotechnologically relevant proteins (19,20). In this work, we present a major update of the original A3D, which significantly extends its capabilities. A3D 2.0 incorporates three major feature upgrades. • protein flexibility simulations using new CABS-flex stan- dalone package (21), which extends the dynamic mode analysis range to proteins up to 4000 residues long and consisting of up to 10 chains. • protein stability calculations using the FoldX force field (22), allowing to account for the impact of amino acid substitutions on the overall structure stability. Figure 1. The pipeline of Aggrescan3D 2.0 server. • an ‘automated mutations’ tool that identifies high scoring residues in structural APRs and suggests protein variants with optimized solubility. decreased and increased stabilities upon mutation, respec- tively. These features were implemented to address the major A3D drawbacks according to users’ feedback. (i) Protein Automated mutation workflow size limitations in the dynamic mode, restricted only to single-chain proteins shorter than 400 amino acids; (ii) the The input structure is optimized using FoldX (22) and user-introduced mutations might impact negatively protein the most aggregation-prone residues identified according to stability, resulting in unfolding and increased aggregation; their A3D score. These residues are individually mutated (iii) the design of improved solubility variants required sig- to solubilizing charged amino acids (arginine, aspartic acid, nificant knowledge about the structural and aggregational glutamic acid and lysine), excluding those positions speci- determinants of proteins and, thus, was not accessible to fied by the user. The changes in aggregation propensity and many potential users. stability are calculated for each potential point mutant and Additionally, A3D 2.0 incorporates an updated REST- short-listed according to these values, up to a maximum of full service that allows the user to incorporate its calcula- 12 suggested changes. Only the two most solubilizing mu- tions in automatic pipelines and a newly designed interface tations for each particular position are shown, in order to that facilitates extended in situ interactive result analysis maximize the number of positions that can be potentially and data interpretation. engineered (up to 6). METHODS NEW FEATURES AND UPDATES A3D prediction protocol Analysis of the impact of protein flexibility in the aggregation properties of large and multimeric proteins The original A3D server was described in detail elsewhere (7). A3D server can be run in Static Mode (default) or Dy- In its dynamic mode, A3D was able to capture the influ- namic Mode. The static mode was validated by predicting ence of structural flexibility on protein aggregation by in- the solubility of a large set of protein mutational variants, corporating the CABS-flex protocol, an efficient alternative whereas the dynamic mode allowed to uncover disease rel- to classical all-atom Molecular Dynamics (11,12,21,23,24). evant APRs not identified by alternative approaches (7,8). A set of protein models (in an all-atom resolution) reflect- The key principles of A3D 2.0 remain similar to these of the ing the most dominant structural fluctuations in the near- original web server and here we only detail major method- native ensemble are generated with CABS-flex for each ological modifications. The overview of the method pipeline input structure. Then, the highest A3D scoring model is is presented in Figure 1. selected as a proxy of the most aggregation-prone con- former in solution. Although this feature uncovered struc- tural APRs not accessible to other structure-based predic- Calculation of the impact of introduced mutations on protein tors (7), its use was restricted to relatively small, single chain thermodynamic stability proteins, which impeded the analysis of many biomedical Users can introduce individual or multiple mutations before and biotechnologically important proteins. With A3D 2.0 or after running A3D 2.0. The selected mutations are mod- we extended the dynamic mode to larger and multimeric eled on top of the protein structure using FoldX (22) and proteins by dedicating significantly larger computational re- the predicted change in stability, relative to the reference sources to web server jobs and rewriting the CABS-flex code molecule, is calculated. Positive and negative values indicate (21). W302 Nucleic Acids Research, 2019, Vol. 47, Web Server issue We used A3D 2.0 to analyse the influence of protein dy- analysis of the original GFP structure with A3D 2.0, indi- namics on the aggregation properties of multimeric pro- cated the existence of three hydrophobic residues exposed at teins, using a data set of 163 proteins (69 homodimers, 54 the protein surface, whose mutation to either K or D would heterodimers and 60 antibodies) (Supplementary Table S1). be equally solubilizing. However, the energetic analysis indi- In the dynamic mode, A3D 2.0 rendered 12 models for each cated that mutations to K would be neutral, whereas muta- input structure and calculated their individual A3D scores. tions to D would destabilize the protein. Two GFP variants Afterward, these values were compared with the ones ob- in which the three hydrophobic residues were changed either tained for the same proteins ran in static mode. We found to K or D were recombinantly expressed. As predicted, the the input static structures to be the least aggregation-prone triple K GFP mutant (GFP/KKK) was highly soluble, pre- in a large majority of cases, both for the complete set and served the native structure and was fully functional, whereas when the three protein categories were analyzed separately the triple D variant (GFP/DDD) was inactive and could (Figure 2A). We averaged the A3D scores of the 12 models not be purified. Importantly, the behavior of GFP/KKK for each individual protein as a proxy for the aggregation and GFP/DDD designs, could not be anticipated by any propensity of its native-like ensemble. The resulting aver- other alternative sequence- or structure-based algorithm. age value was higher than that of the static structure in 80% A3D 2.0 advises now against the experimental characteri- of the cases (Supplementary Table S1). These results have zation of destabilized re-designs (G > 1 kcal/mol), irre- important implications, since most alternative structure- spective of their A3D scores. G values are provided in based aggregation predictors work directly on PDB struc- the ‘Project details’ tab. tures and, therefore, they might underscore the aggregation of multimeric proteins by ignoring the contribution of tran- Automated design of solubility improved protein variants siently exposed APRs. This effect is illustrated in Figure 2B for bevacizumab, a humanized monoclonal antibody pre- The search for soluble functional variants of therapeutic scribed for the treatment of different types of cancers (25). proteins is a challenging task, usually addressed using com- The Fab domain of bevacizumab is very aggregation-prone, binatorial experimental approaches, such us phage display and, accordingly, the antibody must be formulated at low (31). A goal of any aggregation prediction algorithm is to concentrations (26,27). The comparative static and dynamic provide a routine that can substitute for these experiments, analysis of bevacizumab Fab fragment (two chains, four do- saving time and costs. Ideally, this routine should be simple mains) suggests that structural fluctuations result in an in- enough to be accessible to non-expert users. With these two creased aggregation-prone area, with newly exposed APRs objectives in mind, we implemented the ‘automated muta- ready to establish intermolecular interactions. The same ef- tions’ tool in A3D 2.0, accessible at the server front page fect was observed for other therapeutic antibodies, replace- through the ‘Enhance protein solubility’ option. ment enzymes like ␣-galactosidase or pharmaceutically im- The ‘automated mutations’ tool identifies the most portant targets such as insulin and androgen receptors. aggregation-prone patches at the protein surface and vir- tually mutates their residues by charged amino acids, under the assumption that they would act as ‘gatekeepers’, coun- Simultaneous analysis of the impact of user-selected muta- teracting protein self-association. Then it provides a ranked tions in protein solubility and stability list of point mutations, where both the solubilizing and en- The A3D server allowed users to mutate one or more se- ergetic effects are taken into account, in such a way that the lected residues in the structure, pre- or post-analysis, in or- user can discard potentially solubilizing, but destabilizing der to evaluate the impact in protein aggregation. How- mutations. ever, these mutations might also affect the protein thermo- The optimization of the solubility of antibodies is espe- dynamic stability, an effect that was not taken into account cially difficult, because, in these molecules, the tight binding at that time. Indeed, we have shown that there exists a strict to their targets depends on the presence of exposed APRs at correlation between the destabilizing impact of a given mu- their complementarity-determining regions (CDRs). This is tation and the increase it promotes in protein aggregation the reason why computer- (32) or experiment- (33) based de- (28,29). Thus, the solubilizing impact of a residue substitu- signs usually target residues within or close to CDRs; how- tion can be completely cancelled if it impacts the protein ever, these changes might compromise significantly the an- stability. tibody affinity. A3D 2.0 addresses this problem by allow- Mutations at the protein surface are generally better tol- ing users to exclude from the virtual screening functionally erated that residue changes in the protein interior (30). relevant residues, i.e. CDRs in antibodies or active sites in However, when we used A3D to identify the top solubi- enzymes. lizing point mutations for a set of 75 globular proteins, it The ‘automated mutation’ tool has been used for the re- turned out that 10% of these superficial changes (32/324) design of an aggregation-prone Variable Heavy (VH) seg- destabilized the protein >1 kcal/mol according to FoldX ment of the human antibody germline (34). Soluble vari- (Supplementary Table S2). This motivated us to introduce ants of this antibody were previously evolved by phage dis- a simultaneous prediction of protein solubility and stabil- play, but all the introduced mutations clustered at one of ity changes upon mutation in A3D 2.0, to identify muta- the CDRs (35). A3D 2.0 was ran pre-excluding residues tions that decrease globular proteins aggregation propen- at the CDRs. Mutations at three different residues outside sities without compromising their stability and function. these domains were automatically suggested (Figure 4A). This approach was exploited to design of a fast-folding, A designed VH variant containing the 3 top ranked muta- aggregation-resistant GFP variant (19) (Figure 3). The tions was recombinantly expressed and characterized (Fig- Nucleic Acids Research, 2019, Vol. 47, Web Server issue W303 Figure 2. Aggregation propensity for different multimeric proteins, calculated in static or dynamic modes. (A) The aggregation propensity of the static input structure relative to that of the 12 dynamic models is represented for homodimers, heterodimers, antibodies or the complete set. In the color scale, dark blue indicates the static structure being the most soluble (ranking 1) and dark red the static structure being the most aggregation-prone (ranking 13). (B) Monoclonal antibody bevacizumab Fab fragment (PDB: 1BJ1) ran on static (left) or dynamic (right) modes. W304 Nucleic Acids Research, 2019, Vol. 47, Web Server issue Figure 3. A3D 2.0 as a tool for the in silico redesign of more stable and soluble proteins. A3D structures of original GFP (left) (PDB: 2B3Q:A) and engineered GFP/KKK mutant (right) (PDB: 6FWW) coloured according to the A3D score. Mutations lowering aggregation propensity, while maintaining protein stability are encircled. The mutated variant was experimentally shown to be 2-fold more resistant against aggregation (19). ure 4B), turning to be significantly more resistant against structure, which are built using FoldX. Selecting this op- aggregation than the original germline antibody (19). tion will prompt a new window, which allows introducing the desired mutation(s). Note: this option cannot be used with the ‘Enhance protein solubility’ option. DESCRIPTION OF THE WEB SERVER • Distance of aggregation analysis––in the A3D method, Input interface and requirements the intrinsic aggregation propensity of each particular amino acid in the structure is modulated by its specific The only required input is a protein structure in PDB for- structural context. Aggregation propensity is calculated mat (given as a PDB code or uploaded by user in the ‘In- for spherical regions centered on every residue C␣ car- put structure’ panel). Optionally, a user can provide desired bon. This option allows changing the size of said region chain(s) identifier(s) (only provided chains will be used in allowing for more and less granular approaches. the A3D analysis). In the ‘Options’ panel, there are several • Enhance protein solubility––Please see Methods for de- additional options to be chosen: tails on the automated mutations workflow behind this option. Selecting it will prompt a new window to open • Project name––the name under which the project will be upon submitting, where the user can prevent chosen displayed (and which can be used to find it via the project residues from being mutated. Note: This option cannot name search on the top of the page) be used with ‘Mutate residues’ or ‘Dynamic mode’ op- • Email address––provided an email address the server will tions. notify the user when the job has started and ended • Do not show my job in the results page - if the box is • Stability calculations––if selected (default: Yes), the sub- ticked the job will not be visible to other A3D 2.0 users. mitted structure will be energetically minimized before the A3D analysis using FoldX and stability calculated in For more details, see the online documentation available case mutations are defined. from the ‘Tutorial’ tab. • Dynamic mode––in this mode, the input structure’s flex- ibility will be simulated using the CABS-flex software. A Output interface set of predicted models reflecting the flexibility of the in- put structure will be analysed and scored for aggregation For each submitted job, the output interface is organized propensity. Note: this option cannot be used with the ‘En- under the following tabs: ‘Project details’, ‘Aggrescan3D hance protein solubility’ option. plot’, ‘Aggrescan3D score’, ‘Structure’, ‘Automated muta- • Mutate residues––if selected, the option allows a user to tions’ (available only if the job was submitted with the op- select residues and perform point mutations on the input tion ‘Enhance protein solubility’), ‘Dynamic mode details’ Nucleic Acids Research, 2019, Vol. 47, Web Server issue W305 Figure 4. Automated mutations for variable heavy (VH) segment of a human germline antibody. (A) A3D 2.0 automated mutations output (the residues at the three antibody CDRs were excluded from the screening). (B) The blue highlighted mutations in panel A were combined to render triple mutant engineered antibody. Structures of wild type (PDB: 5I19) and the mutant, as predicted by A3D 2.0. Solubilizing mutations are encircled. The engineered antibody variant was experimentally shown to be 3-fold more resistant against aggregation (19). (available only if the job was submitted in the ‘Dynamic • ‘Structure’ tab––allows viewing an analysed structure in mode’) and ‘Gallery’. The content of these tabs is pre- an interactive way. The residues are colored in shades sented and described in the online documentation. Here, we from dark blue (high soluble residues), through white (no present only short descriptions: predicted influence on aggregation properties), to dark red (aggregation prone residues). Various visualization options are possible. • ‘Dynamic mode details’ tab––presents A3D analysis re- • ‘Project details’ tab––contains information about the sults for a set of models reflecting flexibility of the input specified options used to run the job and links to down- structure. The results are organized in the table and inter- load the job data. It also provides stability calculations, active plots presenting scores for particular models. when it applies. • ‘Automated mutations’ tab––presents A3D analysis re- • ‘Aggrescan3D plot’ tab––presents A3D analysis results in sults for a set of mutant models generated using option the form of an interactive online plot for a selected pro- ‘Enhance protein solubility’. The results are organized in tein chain. the table and interactive plots presenting scores for par- • ‘Aggrescan3D score’ tab––presents A3D analysis results ticular protein variants. in the form of an interactive table together with ‘mutate’ • ‘Gallery’ tab - contains all screenshots taken by users (us- buttons in the right side of the table, which allows to re- ing ‘Take snapshot’ buttons) submit the job with chosen mutations. W306 Nucleic Acids Research, 2019, Vol. 47, Web Server issue Online documentation 2015 to S.V.]; National Science Center (NCN, Poland) Grant [MAESTRO2014/14/A/ST6/00088 to M.K and The documentation can be found on the server itself un- S.K.]. der the ‘Tutorial’ tab (available from the main menu). Ad- Conflict of interest statement. None declared. ditionally, the web interface provides short help notes that are available close to the presented content. The online doc- umentation is updated on a regular basis according to users’ REFERENCES needs or the server improvement. 1. Chiti,F. and Dobson,C.M. (2017) Protein misfolding, amyloid formation, and human disease: a summary of progress over the last Command-line availability decade. Annu. Rev. Biochem., 86, 27–68. 2. Invernizzi,G., Papaleo,E., Sabate,R. and Ventura,S. (2012) Protein The A3D 2.0 server can be also operated from the command aggregation: mechanisms and functional consequences. Int. J. line using RESTful web services. The instructions for using Biochem. Cell Biol., 44, 1541–1554. 3. Hamrang,Z., Rattray,N.J. and Pluen,A. (2013) Proteins behaving the RESTful service are available from the online tutorial badly: emerging technologies in profiling biopharmaceutical (accessible from the main menu). aggregation. Trends Biotechnol., 31, 448–458. 4. Pallares,I. and Ventura,S. (2017) Advances in the prediction of protein aggregation propensity. Curr. Med. Chem., Server architecture doi:10.2174/0929867324666170705121754. The Aggrescan3D 2.0 server is a HTML based service dy- 5. Meric,G., Robinson,A.S. and Roberts,C.J. (2017) Driving forces for nonnative protein aggregation and approaches to predict namically generated using the Flask framework and the Aggregation-Prone regions. Annu. Rev. Chem. Biomol. Eng., 8, jinja2 templating engine. The user data is stored using a 139–159. MySQL database upon submission and unique id and a sta- 6. Ricardo Graña-Montes,J.P.-P., Carlota,Gómez-Picanyol and tus are assigned for each job. The server notifies the user Ventura,A.S. (2017) In: Rigden,DJ (ed). From Protein Structure to of its progress by a job status, which is ‘pending’ when Function with Bioinformatics. Springer, pp. 205–263. 7. Zambrano,R., Jamroz,M., Szczasiuk,A., Pujols,J., Kmiecik,S. and the server is waiting for a computational cluster response, Ventura,S. (2015) AGGRESCAN3D (A3D): server for prediction of ‘queue’ when there are no resources available yet, ‘running’ aggregation properties of protein structures. Nucleic Acids Res., 43, and then finally either ‘done’ or ‘error’. The simulation W306–W313. is carried out using the Aggrescan3D standalone software 8. Pujols,J., Pena-Diaz,S. and Ventura,S. (2018) AGGRESCAN3D: Toward the prediction of the aggregation propensities of protein (that is available at http://bitbucket.org/lcbio/aggrescan3d) structures. Methods Mol. Biol., 1762, 427–443. and other previously described programs (with the RSA cal- 9. Conchillo-Sole,O., de Groot,N.S., Aviles,F.X., Vendrell,J., Daura,X. culations done by FreeSASA software (36)). The structures and Ventura,S. (2007) AGGRESCAN: a server for the prediction and are presented in an interactive way using the 3Dmol library evaluation of “hot spots” of aggregation in polypeptides. BMC (HTML5/Javascript). The A3D score plot is done using the Bioinformatics, 8, 65. 10. de Groot,N.S., Castillo,V., Grana-Montes,R. and Ventura,S. (2012) D3.js library (HTML5/Javascript) and the model and mu- AGGRESCAN: method, application, and perspectives for drug tant comparison plots are generated using the Bokeh library design. Methods Mol. Biol., 819, 199–220. (Python/Javascript). The PDB structures are obtained us- 11. Jamroz,M., Kolinski,A. and Kmiecik,S. (2013) CABS-flex: server for ing RESTful services. The A3D 2.0 website handles user’s fast simulation of protein structure fluctuations. Nucleic Acids Res., 41, W427–W431. requests using an Apache2 server. The A3D 2.0 server is 12. Kuriata,A., Gierut,A.M., Oleniecki,T., Ciemny,M.P., Kolinski,A., free, open to all users and there is no login requirement. Kurcinski,M. and Kmiecik,S. (2018) CABS-flex 2.0: a web server for fast simulations of flexibility of protein structures. Nucleic Acids Res., 46, W338–W343. SUMMARY 13. Pulido,P., Llamas,E., Llorente,B., Ventura,S., Wright,L.P. and In this work, we developed an easy-to-use A3D 2.0 web Rodriguez-Concepcion,M. (2016) Specific Hsp100 chaperones determine the fate of the first enzyme of the plastidial isoprenoid server interface for the prediction and engineering of pro- pathway for either refolding or degradation by the stromal Clp tein solubility. The implemented changes boost the server protease in arabidopsis. PLos Genet., 12, e1005824. functionality with an unprecedented combination of fea- 14. Pulido,D., Arranz-Trullen,J., Prats-Ejarque,G., Velazquez,D., tures for APRs identification and design taking into ac- Torrent,M., Moussaoui,M. and Boix,E. (2016) Insights into the count dynamic and thermodynamic aspects in the predic- antimicrobial mechanism of action of human RNase6: structural determinants for bacterial cell agglutination and membrane tions. A3D 2.0 can find application in the identification and permeation. Int. J. Mol. Sci., 17, 552. analysis of pathogenic familial mutations in conformational 15. Soler,M.A., de Marco,A. and Fortuna,S. (2016) Molecular dynamics disorders, but, it is specially intended to assist the design of simulations and docking enable to explore the biophysical factors soluble proteins for biotechnological and biomedical appli- controlling the yields of engineered nanobodies. Sci. Rep., 6, 34869. 16. Bhandare,V.V. and Ramaswamy,A. (2018) The proteinopathy of cations, where it should contribute to reduce development D169G and K263E mutants at the RNA Recognition Motif (RRM) costs and times. domain of tar DNA-binding protein (tdp43) causing neurological disorders: a computational study. J. Biomol. Struct. Dyn., 36, SUPPLEMENTARY DATA 1075–1093. 17. Zerovnik,E. (2017) Putative alternative functions of human stefin B Supplementary Data are available at NAR Online. (cystatin B): binding to amyloid-beta, membranes, and copper. J. Mol. Recognit., 30, e2562. 18. Katina,N.S., Balobanov,V.A., Ilyina,N.B., Vasiliev,V.D., FUNDING Marchenkov,V.V., Glukhov,A.S., Nikulin,A.D. and Bychkova,V.E. (2017) sw ApoMb amyloid aggregation under nondenaturing Spanish Ministry of Economy and Competitiveness conditions: the role of native structure stability. Biophys. J., 113, [BIO2016-78310-R to S.V., ICREA, ICREA-Academia 991–1001. Nucleic Acids Research, 2019, Vol. 47, Web Server issue W307 19. Gil-Garcia,M., Bano-Polo,M., Varejao,N., Jamroz,M., Kuriata,A., protein engineering and incorporation of glycosylation motifs applied Diaz Caballero,M., Lascorz,J., Morel,B., Navarro,S., Reverter,D. to bevacizumab. mAbs, 8, 99–112. et al. (2018) Combining structural aggregation propensity and 28. Espargaro,A., Castillo,V., de Groot,N.S. and Ventura,S. (2008) The in stability predictions to re-design protein solubility. Mol. Pharm., 15, vivo and in vitro aggregation properties of globular proteins correlate 3846–3859. with their conformational stability: the SH3 case. J. Mol. Biol., 378, 20. Xia,X., Kumru,O.S., Blaber,S.I., Middaugh,C.R., Li,L., Ornitz,D.M., 1116–1131. Sutherland,M.A., Tenorio,C.A. and Blaber,M. (2016) Engineering a 29. Castillo,V., Espargaro,A., Gordo,V., Vendrell,J. and Ventura,S. (2010) cysteine-free form of human fibroblast growth Factor-1 for “Second Deciphering the role of the thermodynamic and kinetic stabilities of Generation” therapeutic application. J. Pharm. Sci., 105, 1444–1453. SH3 domains on their aggregation inside bacteria. Proteomics, 10, 21. Kurcinski,M., Oleniecki,T., Ciemny,M.P., Kuriata,A., Kolinski,A. 4172–4185. and Kmiecik,S. (2019) CABS-flex standalone: a simulation 30. Franzosa,E.A. and Xia,Y. (2009) Structural determinants of protein environment for fast modeling of protein flexibility. Bioinformatics, evolution are context-sensitive at the residue level. Mol. Biol. Evol., 35, 694–695. 26, 2387–2395. 22. Schymkowitz,J., Borg,J., Stricher,F., Nys,R., Rousseau,F. and 31. Sidhu,S.S. (2000) Phage display in pharmaceutical biotechnology. Serrano,L. (2005) The FoldX web server: an online force field. Nucleic Curr. Opin. Biotechnol., 11, 610–616. Acids Res., 33, W382–W388. 32. Sormanni,P., Aprile,F.A. and Vendruscolo,M. (2015) The CamSol 23. Jamroz,M., Kolinski,A. and Kmiecik,S. (2014) CABS-flex predictions method of rational design of protein mutants with enhanced of protein flexibility compared with NMR ensembles. Bioinformatics, solubility. J. Mol. Biol., 427, 478–490. 30, 2150–2154. 33. Perchiacca,J.M., Lee,C.C. and Tessier,P.M. (2014) Optimal charged 24. Jamroz,M., Orozco,M., Kolinski,A. and Kmiecik,S. (2013) mutations in the complementarity-determining regions that prevent Consistent view of protein fluctuations from all-atom molecular domain antibody aggregation are dependent on the antibody scaffold. dynamics and coarse-grained dynamics with knowledge-based Protein Eng. Des. Sel., 27, 29–39. force-field. J. Chem. Theory Comput., 9, 119–125. 34. Teplyakov,A., Obmolova,G., Malia,T.J., Luo,J., Muzammil,S., 25. Gridelli,C., de Castro Carpeno,J., Dingemans,A.C., Griesinger,F., Sweet,R., Almagro,J.C. and Gilliland,G.L. (2016) Structural diversity Grossi,F., Langer,C., Ohe,Y., Syrigos,K., Thatcher,N., Das-Gupta,A. in a human antibody germline library. mAbs, 8, 1045–1063. et al. (2018) Safety and efficacy of bevacizumab plus 35. Dudgeon,K., Rouet,R., Kokmeijer,I., Schofield,P., Stolp,J., Standard-of-Care treatment beyond disease progression in patients Langley,D., Stock,D. and Christ,D. (2012) General strategy for the with advanced non-small cell lung cancer: the AvaALL randomized generation of human antibody variable domains with increased clinical trial. JAMA Oncol., 4, e183486. aggregation resistance. Proc. Natl. Acad. Sci. U.S.A., 109, 26. Oliva,A., Llabres,M. and Farina,J.B. (2014) Capability measurement 10879–10884. of size-exclusion chromatography with a light-scattering detection 36. Mitternacht,S. (2016) FreeSASA: an open source C library for method in a stability study of bevacizumab using the process solvent accessible surface area calculations [version 1; peer review: 2 capability indices. J. Chromatogr. A., 1353, 89–98. approved]. F1000Research, 5, 189. 27. Courtois,F., Agrawal,N.J., Lauer,T.M. and Trout,B.L. (2016) Rational design of therapeutic mAbs against aggregation through