Team:Heidelberg/pages/Linker Modeling

From 2014.igem.org

(Difference between revisions)
(sequences to connect the alpha helix to the protein extremity)
(In silico refinement)
Line 73: Line 73:
=In silico refinement=
=In silico refinement=
-
As some of the interesting patterns could not be found often enough to be statistically significant, we decided to make a further refinement in silico by modeling the structure of proteins with circularizing linkers. To perform this for realistic situations, we selected from the RCSB database structures of non-homologous target proteins with extremities that are separated enough to require a linker for circularization. For setting up an environment as close as possible to the application of the patterns were we designed the following workflow.
+
As some of the interesting patterns could not be found often enough to be statistically significant, we decided to make a further refinement ''in silico'' by modeling the structure of proteins with circularizing linkers. To perform this for realistic situations, we selected from the RCSB database structures of non-homologous target proteins with extremities that are separated enough to require a linker for circularization. First, the [https://2014.igem.org/Team:Heidelberg/Software/Linker_Software CRAUT software] generates possible fitting linkers for various proteins. From these possible linkers, the 100 shortest were taken and the angle patterns that should produce the same angles. Like you can easily see from [[table 1 ###]] KTA, LVA and AAIAP nearly produce the same angles. These then should be tested in comparison. On the other hand the software could not handle different patterns for the same angle, so afterwards we exchanged the sequences in the predicted linkers. The linkers connect the ends of the protein without setting tension on the protein, so that the protein can fold in its natural way. For further information on the generation of the linker sequences please follow to the [[Linkergenerator|###link###]]
-
First, the ###linker software### generates possible fitting linkers for various proteins. From these possible linkers, the 100 shortest were taken and the angle patterns that should produce the same angles. Like you can easily see from [[table 1 ###]] KTA, LVA and AAIAP nearly produce the same angles. These then should be tested in comparison. On the other hand the software could not handle different patterns for the same angle, so afterwards we exchanged the sequences in the predicted linkers. The linkers connect the ends of the protein without setting tension on the protein, so that the protein can fold in its natural way. For further information on the generation of the linker sequences please follow to the [[Linkergenerator|###link###]]
+
After this the circularized proteins with the specific linkers are modelled using a software called Modeller.[[#References|[8]]] This software is widely used for comparative structure prediction. It is well established in the scientific community and should be most suitable for prediction of loop regions attached to existing structures. [[#References|[9]]] Modeller is a program that is able to predict the 3d structure of a given sequence based on an alignment with a given structure. In our script modeller needed to be provided with to things, a sequence with the linker attached and the PDB file of the protein of interest. The result of a prediction for lambda-lysozyme can be seen in [[circ_lam_lys_nils.png]] It is freely available for academical usage from the  [http://salilab.org/modeller/ salilab] webpages. As we just want to determine the properties of our linker patterns attached to proteins, it perfectly fitted our purpose. Most important for our purpose is that Modeller does not rely on structural databases like ArchDB database, but does an ab initio modelling of our linkers by minimizing energy functions with different methods like conjugate gradients and molecular dynamics. Even though it recommends only to simulate loops of up to 8 aminoacids, we chose to use it, because the similarity of the sequence with the provided PDB is as a matter of fact at about 90%. Modeller is recommended for usage from about 30% sequence similarity.
After this the circularized proteins with the specific linkers are modelled using a software called Modeller.[[#References|[8]]] This software is widely used for comparative structure prediction. It is well established in the scientific community and should be most suitable for prediction of loop regions attached to existing structures. [[#References|[9]]] Modeller is a program that is able to predict the 3d structure of a given sequence based on an alignment with a given structure. In our script modeller needed to be provided with to things, a sequence with the linker attached and the PDB file of the protein of interest. The result of a prediction for lambda-lysozyme can be seen in [[circ_lam_lys_nils.png]] It is freely available for academical usage from the  [http://salilab.org/modeller/ salilab] webpages. As we just want to determine the properties of our linker patterns attached to proteins, it perfectly fitted our purpose. Most important for our purpose is that Modeller does not rely on structural databases like ArchDB database, but does an ab initio modelling of our linkers by minimizing energy functions with different methods like conjugate gradients and molecular dynamics. Even though it recommends only to simulate loops of up to 8 aminoacids, we chose to use it, because the similarity of the sequence with the provided PDB is as a matter of fact at about 90%. Modeller is recommended for usage from about 30% sequence similarity.
Each modeled structure is provided with energy values, thanks to which different models of the same structure can be compared. From Modeller we received about 8 different models and choose the one with the best energy scores to further proceed. At first modeller makes an alignment between the provided structure and the sequence identifying the regions, that can not be found in the structure . Based on that modeller generates 4 initial models. One of the strengths of modeller is it's capability to further refine only certain parts of the protein. Thus we let modeller refine the loops. A loop for modeller is defined as any part of the protein, that could not be found in the structure file. For these refinement steps, one can choose different levels of optimization. We always decided for accuracy instead of velocity of the program.
Each modeled structure is provided with energy values, thanks to which different models of the same structure can be compared. From Modeller we received about 8 different models and choose the one with the best energy scores to further proceed. At first modeller makes an alignment between the provided structure and the sequence identifying the regions, that can not be found in the structure . Based on that modeller generates 4 initial models. One of the strengths of modeller is it's capability to further refine only certain parts of the protein. Thus we let modeller refine the loops. A loop for modeller is defined as any part of the protein, that could not be found in the structure file. For these refinement steps, one can choose different levels of optimization. We always decided for accuracy instead of velocity of the program.

Revision as of 12:56, 17 October 2014

Contents

Abstract

Artificially circularized proteins can acquire remarkable properties reviewed in our introduction on circularization. These properties include heat stability thanks to the constrain of the relative positioning of the C- and the N-termini [10]. circularization requires a linker that provide this constrain without changing the natural conformation of the protein. Designing such a linker is particularly non-trivial if the protein extremities are far from each other. Peptidic linkers are generally used in protein modification, not only for connecting protein subdomains but also for ligating their extremities to circularize them. Traditionally, flexible linkers like Glycin-Serine peptides are used for this purpose [1]. However, to keep domains of chimeric proteins in a certain distance, rigid peptides built of helical patterns are also often applied [4] [0]. Here we show a novel approach to build customized rigid linkers that follow a desired shape. This is achieved by connecting peptides forming rigid helices with amino acids that produce a certain angle between them. We herein describe the building blocks with which one can build customized linkers. The building blocks we designed fall into two categories, the alpha helix blocks forming rigid rods, and the angle patterns. The latter were obtained from statistical analysis of supersecondary structures from structural databases, normally used for protein structure prediction. The potential linkers were tested in a large screening in silico, in order to refine their properties. Additionally they were tested in vitro for circularizing lysozyme from bacteriophage lambda as a model enzyme. This experimental work was also used to measure and calibrate the contribution of different linker features to heat stabilization. The outputs of our modeling approach and of the experimental work were combined in a software that designs circularization linkers for any protein with a known structure. Not only the modularity but also the reliability of our linkers are huge advantages compared to normally used linkers made of alanine or of a mix of glycine and serine. This part explains the statistical analysis performed to identify the building blocks and how we refined them by structure modeling. Finally, we show how this approach was used to design linkers used in our experiments.


Background

Primary, secondary, tertiary and quaternary structures are the main levels of protein structure characterization. Primary structure designates the amino acid sequence, while the secondary structure describes the arrangement of consecutive amino acids through their two dihedral angles $\phi$ and $\psi$. The Ramachandran plot, which represents the amino acid position in the space of those two angles, shows two particular arrangement commonly found in proteins: alpha helices and beta sheets. The next level of protein organization is the tertiary structure, which describes how the protein is organized in the three spatial dimensions, whereas the quaternary structure describes how different subunits of proteins cluster. Finally, closely related to these standard structures, the supersecondary structure describes how secondary structure elements are connected to each other. While these connections look undefined at first sight, further analysis revealed that this wide variety of supersecondary structure motifs can be clustered to certain patterns [5].

Supersecondary structure

When the properties of supersecondary structures were first described, only very few patterns were identified, mainly due to the lack of highly resolved protein structures. At that time the structures were mainly classified by the Ramachandran plot regions ($\alpha, \beta, \gamma$ etc. ) where the amino acids could be found [6]. With growing amount of known crystal structures, the analysis of supersecondary structure improved and lead to databases with about 150 000 classified loop structures and elaborate clustering [7]. Nowadays supersecondary structures are defined as the structures built when two secondary structure elements are combined by a small peptide that is not clustered into one of the secondary structures. These loop peptides range from 1 to 9 amino acids. Our aim was to build reliable stable linkers out of alpha helices connected by supersecondary structure motifs that produce certain angles. To achieve that, we searched for the most reliable alpha helix patterns that would form rigid rods and angle patterns covering the whole range of angles from 0 to 180 degrees.

linker building block design

Helix patterns

Various different patterns have been used to build helical linkers to connect protein ends [3]. Moreover, in known protein structures, linkers between subdomains can be identified and their properties have already been analyzed [2]. Two main criteria were used to build the alpha helix patterns: they should robustly for alpha helices, and they should be soluble in aqueous solution. Therefore, we could not just use linkers built of Alanine. So we decided to add some charged aminoacids to the pattern, and to position them physically close to each other so that they could stabilize themselves by Coulomb interaction. These amino acids needed to be separated by 3 amino acids as a helical turn takes about 3.6 aminoacids. The pattern we chose as most suitable for our purpose was also described to be one of the most stable [4]. 8 alpha helix building blocks were eventually chosen: AEAAAK, AEAAAKA, AEAAAKAA, AEAAAKEAAAK, AEAAAKEAAAKA, AEAAAKEAAAKEAAAKA, AEAAAKEAAAKEAAAKEAAAKA, AEAAAKEAAAKEAAAKEAAAKEAAAKA, with a respective estimated length of 9, 10.5, 12, 16.5, 18, 25.5 and 33 Å.

Angle patterns

The angle patterns for our model were obtained from the ArchDB database [5], which classifies loops from known proteins structures. About 17 000 non-homologous proteins from PDB database were analyzed and from them, 150 000 loop structures, i.e. regions connecting two secondary structure elements, were identified. The classification took into account not only the length of the loop, its conformation, meaning φ and ψ backbone dihedral angles of the residues in the loop, but also the distance between the attachments of the loop to the surrounding secondary structures. Furthermore the secondary structures surrounding the loop and the geometry defined by the super-secondary structure motifs can be found in the database. To extract from ArchDB the relevant supersecondary structure motifs for our linker design, the complete database was downloaded and helix-loop-helix motifs were extracted using a self-written script in Python programming language. From them we only took into account loops composed of 1 and 2 amino acids, because the longer the loops, the less frequent and therefore the less reliable they are, and the further the ends are from each other. The interesting information for us was (1) the angle produced between the vectors defining the surrounding alpha helices, (2) the distance between the ends of the loop, and (3) the type of amino acids surrounding the loop. Furthermore we analyzed the statistical significance of the conformation. For each amino acid combination in the loop region, the angle distribution between the embracing alpha helices, the loop length distribution and a 2d heatmap of the surrounding amino acids were automatically plotted. These distributions were then visually analyzed to identify loops of interest for linker design. We focused on loops that showed a narrow angle distribution and that appeared frequently in the database. We extended the distribution analysis to the surrounding amino acids. For example, when we have identified, that K in a turn produces an interesting distribution, we would elongate it by K in front and elongate it by A in the end and see how the distribution behaves. By restraining the possibilites, the occurrences go down tremendously, but the properties become more interesting ###plot_of__T_.png### plot_of_K_T_.png plot_of__T_A.png plot_of_K_T_A.png. This step had two main goals: narrowing down the angle distribution and finding loops with no preferences for the amino acid surrounding them. This last point was important for the modularity of our approach: the angle blocks should not be affected by the surrounding alpha helices. Using this approach, 10 different angle motifs could be identified producing different angles. Importantly, all these motifs were chosen so that the length of the two segments starting from the turning point was the same for all of them (put figure turning point).


table 1: The span of parameters.
Angle Patterns
Pattern NVL KTA LVA AAIAP AADGTL VNLTA AAAHPEA ASLPAA ATGDLA
Mean 29.7 38.7 35 36.5 60 74.5 117 140 160
Variation 8.5 30 29 27 12. 27. 12. 15 . 5.

Sequences to connect the alpha helix to the protein extremity

Helical patterns often affect the folding of the attached sequences. To prevent them from affected the structure of the protein of interest, we analyzed the effects of various aminoacids in silico using online tools like [http://bioserv.rpbs.univ-paris-diderot.fr/services/PEP-FOLD/ pep-fold] on helix formation. We identified glycines and prolines as reliable amino acids to interrupt helix formation. We then decided to use glycine pairs to connect the protein of interest to the linker, because they give more flexibility to the initial orientation of the initial helix. This only concerns the protein end that was connected to the linker through the coding sequence. The other end is ligated through exteins or sortase scar, both treated as unstructured flexible regions.

Conclusion

The last three parts show how we could design alpha helix and angle pattern blocks and connect them to each other and to the protein of interest. They provide the material that allowed us to transform a linker defined as a geometrical path into a real amino acid sequence in our software.

Thanks to this we could design linkers to circularize DNMT1 and lysozyme.

In silico refinement

As some of the interesting patterns could not be found often enough to be statistically significant, we decided to make a further refinement in silico by modeling the structure of proteins with circularizing linkers. To perform this for realistic situations, we selected from the RCSB database structures of non-homologous target proteins with extremities that are separated enough to require a linker for circularization. First, the CRAUT software generates possible fitting linkers for various proteins. From these possible linkers, the 100 shortest were taken and the angle patterns that should produce the same angles. Like you can easily see from table 1 ### KTA, LVA and AAIAP nearly produce the same angles. These then should be tested in comparison. On the other hand the software could not handle different patterns for the same angle, so afterwards we exchanged the sequences in the predicted linkers. The linkers connect the ends of the protein without setting tension on the protein, so that the protein can fold in its natural way. For further information on the generation of the linker sequences please follow to the ###link### After this the circularized proteins with the specific linkers are modelled using a software called Modeller.[8] This software is widely used for comparative structure prediction. It is well established in the scientific community and should be most suitable for prediction of loop regions attached to existing structures. [9] Modeller is a program that is able to predict the 3d structure of a given sequence based on an alignment with a given structure. In our script modeller needed to be provided with to things, a sequence with the linker attached and the PDB file of the protein of interest. The result of a prediction for lambda-lysozyme can be seen in circ_lam_lys_nils.png It is freely available for academical usage from the [http://salilab.org/modeller/ salilab] webpages. As we just want to determine the properties of our linker patterns attached to proteins, it perfectly fitted our purpose. Most important for our purpose is that Modeller does not rely on structural databases like ArchDB database, but does an ab initio modelling of our linkers by minimizing energy functions with different methods like conjugate gradients and molecular dynamics. Even though it recommends only to simulate loops of up to 8 aminoacids, we chose to use it, because the similarity of the sequence with the provided PDB is as a matter of fact at about 90%. Modeller is recommended for usage from about 30% sequence similarity. Each modeled structure is provided with energy values, thanks to which different models of the same structure can be compared. From Modeller we received about 8 different models and choose the one with the best energy scores to further proceed. At first modeller makes an alignment between the provided structure and the sequence identifying the regions, that can not be found in the structure . Based on that modeller generates 4 initial models. One of the strengths of modeller is it's capability to further refine only certain parts of the protein. Thus we let modeller refine the loops. A loop for modeller is defined as any part of the protein, that could not be found in the structure file. For these refinement steps, one can choose different levels of optimization. We always decided for accuracy instead of velocity of the program. Modeller was run via the ###link to i@h### system, calculating distributedly the structures of various proteins at the same time. The modelling of one linker took about 10 hours of calculation time on average via the iGEM@home system. Actually this value is highly depending on the size of the protein. Then the best model is evaluated by another self-written program to analyze the behaviour of the linker patterns in their natural surroundings.

In the third step all the models for the different structures and the different linkers are analyzed for their properties like the length of the helical patterns, the shape of the attachment structures of the linker and the angles produced by the angle patterns. Figure helix_winkel_messung.png First the modeled structure and the natural structure are fitted, to see how big the differences between those are. If the protein has been disturbed too much, the model is discarded. For the analysis of of the different patterns, the connection between the $C_{\alpha}$ of the first amino acid and the $C_{\alpha}$ of the last aminoacid is defining a vector. For the attachment sequences and the helices the length of this vector is calculated, for the Otherwise length of attachment sequences are calculated just by calculating the distance of the atoms. For the helical patterns a vector is fitted to the C$\alpha$s. For these vectors always distance between the ends, the length and the angles are calculated. Furthermore a possible crossing point is estimated. Afterwards for each helical pattern and for each angle-pattern we obtain a distribution for the different properties, so that we can refine our assumptions on the behaviour of the patterns. With the coordinates of the estimation of crossing points, on can furthermore see, whether the linker really follows a software predicted path and thus verifying the results of the linker-software.

Results

Out of this, we decided to set up a modular system for our linkers. All linkers start with two amino acids, that guarantee some flexibility to the ends of the protein and that prohibit the attached helix to continue into the protein and thus making non-helical regions helical. The next building block is one of the alpha helix forming patterns AEAAAK, AEAAAKA, AEAAAKAA, AEAAAKEAAAK, AEAAAKEAAAKA, AEAAAKEAAAKEAAAKA, AEAAAKEAAAKEAAAKEAAAKA, AEAAAKEAAAKEAAAKEAAAKEAAAKA with a well-defined length and shape. Then an angle pattern is attached. All the angle patterns chosen by us, have the same distance from the actual turning point.figure ###turning point###. Thus one can easily exchange different angle patterns and easily calculate the distances between the following turning points, like it is used in our ###software###. To this angle pattern, another helix pattern can easily be attached again. ###figure needed###. All our linkers end with the two exteins because of circularization or the sortase scar, treated both as rather unstructured flexible regions. On the other side we have introduced two amino acids, that prevent the helix from disturbing the protein ends by helix formation. Therefore we identified GG as a suitable pattern.

Application

-DNMT1 -Lysozyme

Verification of patterns

The whole process for the verification of the different linker patterns was set up on the distributed computing system. #### iGEM@home But unfortunately due to lack of time only few results could be analyzed, resulting in distributions for the different helices, see for example ### fig 5. and 6.### plot_of_AEAAAKA.png plot of AEAAAKEAAAKA

Conclusion

The patterns introduced provide an easy and fast tool to build customly shaped peptides. The main achievement is the identification of the angle patterns. These are created in a building block like manner for enhanced applicability. The shapes were identified from a database of non-homologous proteins. The patterns were refined until the distribution of attached amino acids looked randomly distributed. Thus we can exclude to a certain amount, that the angle distributions we have observed is not due to the attached sequences, but due to the identified patterns. We have not observed any evidence for certain helix patterns being preferred in the database. From figures ###5. and 6### we have learned, that the lengths we have assumed for the helices needed to be adjusted to the new values. For example we had assumed the AEAAAKA motif to span a distance of 10.5 ###Å ### but have observed it to be only 10 Å long. This has found direct influence to the Linker_Software ###CRAUT, the software for generating linkers, we have introduced.

References

[0] Wang, C.K.L., Kaas, Q., Chiche, L. & Craik, D.J. CyBase: A database of cyclic protein sequences and structures, with applications in protein discovery and engineering. Nucleic Acids Research 36, (2008).

[10] Vieille, C. & Zeikus, G.J. Hyperthermophilic enzymes: sources, uses, and molecular mechanisms for thermostability. Microbiology and molecular biology reviews : MMBR 65, 1-43 (2001).

[1] Yu, Y. & Lutz, S. Circular permutation: a different way to engineer enzyme structure and function. Trends Biotechnol. 29, 18-–25 (2011).

[2] Chen, X., Zaro, J. L. & Shen, W.-C. Fusion protein linkers: property, design and functionality. Adv. Drug Deliv. Rev. 65, 1357-13–69 (2013).

[3] George, R. a & Heringa, J. An analysis of protein domain linkers: their classification and role in protein folding. Protein Eng. 15, 871-87–9 (2002).

[4] Arai, R., Ueda, H., Kitayama, a, Kamiya, N. & Nagamune, T. Design of the linkers which effectively separate domains of a bifunctional fusion protein. Protein Eng. 14, 529–-532 (2001).

[5] Efimov, a V. Standard structures in proteins. Prog. Biophys. Mol. Biol. 60, 201-2–39 (1993).

[6] Donate, L. E., Rufino, S. D., Canard, L. H. & Blundell, T. L. Conformational analysis and clustering of short and medium size loops connecting regular secondary structures: a database for modeling and prediction. Protein Sci. 5, 2600-26–16 (1996).

[7] Bonet, J. et al. ArchDB 2014: structural classification of loops in proteins. Nucleic Acids Res. 42, D315-D31–9 (2014).

[8] Fiser, a et al. Modeling of loops in protein structures. Protein science : a publication of the Protein Society 9, 1753-73 (2000).

[9] Fiser, A. & Sali, A. ModLoop: Automated modeling of loops in protein structures. Bioinformatics 19, 2500-2501 (2003).