SymProFold - Structural prediction of symmetrical biological assemblies

doi:10.21203/rs.3.rs-3830312/v1

Download PDF

Research Article

SymProFold - Structural prediction of symmetrical biological assemblies

https://doi.org/10.21203/rs.3.rs-3830312/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Symmetry in nature often arises from self-assembly processes and serves a function. Our automated pipeline SymProFold leverages the high accuracy of the AlphaFold-Multimer predictions to derive symmetrical assemblies like two-dimensional S-layer arrays or spherical viral capsids from their protein sequence, verified with available experimental data on the cellular level. S-layers, found on many bacterial and archaeal cells, are vital for facilitating cell adhesion, evading the immune system, and providing protection against stress. However, their experimental structural characterization remains challenging because of their self-assembly property and high sequence variability. SymProFold now offers new avenues for exploring functionalities and designing targeted applications in diverse fields like nanotechnology, biotechnology, medicine, and material and environmental sciences.

S-layers

surface layer proteins

SymProFold

structure prediction

biological assemblies

symmetry

Symmetry patterns are a fundamental and widespread feature, spanning from the microscopic to the macroscopic scale. At the molecular level, the driving force of this symmetry is self-assembly, where individual components autonomously organize into larger, ordered structures driven by their interactions. Many proteins have symmetrical structures, which are essential for their function ¹. Examples of this symmetrical organization include the hexagonal arrangement of transmembrane chemotaxis receptors, the icosahedral capsid proteins of viruses, and most prominent, the repetitive regular array of S-layer proteins ^2–6.

Surface layers (S-layers) are porous 2-dimensional crystalline protein arrays covering the cell envelopes of many eubacterial and archaeal strains ^5,6. The S-layer is built from one or more (glyco)-protein subunits, S-layer proteins (SLPs), that self-assemble into a highly flexible and dynamic lattice through an entropy-driven process, allowing for structural adaptation in response to changing environmental conditions ^4,6–10. So far, S-layers have been shown to serve various functions, including cell stability, adhesion, molecular sieve, and aiding in osmotic stress adaptation ^6,11–13. Still, each S-layer's exact functionality and assembly properties remain unclear. Additionaly, S-layers represent a unique structural basis for generating complex supramolecular assemblies with considerable application potential in (nano)biotechnology, biomimetics, biomedicine, and synthetic biology ^6,14. The ability to self-assemble also plays a crucial role for viral capsid proteins, which form envelopes encapsulating the viral nucleic acid ¹⁵. Successful assembly and disassembly of the virus proteinaceous coat is crucial in virus replication and infection.

Classical structure determination methods are often insufficient for getting atomic resolution insights into fully assembled S-layers. The fast-advancing development and accuracy in structure prediction programs like RoseTTAFold ¹⁶, AlphaFold2 ¹⁷, and AlphaFold-Multimer ¹⁸ are bridging the gap of missing experimental structures and protein interactions. AlphaFold has been extensively used to predict individual virus proteins ^19,20 and a method to combine AlphaFold monomer predictions with symmetric all-atom docking simulations to predict cubic complexes is available ²¹. Here, we present Symmetry Protein Fold (SymProFold), an automated pipeline for predicting the fully assembled proteins with certain symmetry and unit cell parameters, using sequential information as input. The assembly of 19 new S-layers from bacteria and archaea as well as a viral capsid are predicted. The generated SymProFold models were verified using available experimental data at cellular level. The predictions provide comprehensive insight into scarcely understood assembly of S-layers at atomic level and reveal exciting features for S-layers which, considering the central function of S-layers in microorganisms, will be crucial for clarifying S-layer functionality and enabling the design of S-layers for their usage in a broad range of applications.

The underlying idea behind SymProFold is to combine oligomer predictions with the general symmetry patterns found in nature, as observed for S-layers, for generating a model of a fully assembled layers. An overview of the general workflow is shown in Fig. 1 and is described in detail in the Methods section.

SymProFold uses the sequential information of the protein of interest, therefore, no prior knowledge of the symmetry or size of the unit cell of the assembly is needed. The protein analysis process begins with the definition of protein domains, which are defined either manually or via an automated tool 'Domain_Separator'. Subsequently, full-length and truncated variants of the protein, called subchains, are created according to specified domains. These subchains are then used for oligomer predictions, forming different symmetric complexes (dimers to hexamers), which are evaluated for rotational symmetry. The symmetry complexes are then filtered based on a series of quality criteria, including the number of clashes. An example output of the filtering and clustering of symmetry complexes from A. salmonicida S-layer exhibiting a p4 symmetry is shown in Fig. 2. Prediction scores suggest the presence of two rotational symmetry axes (A, B), each exhibiting a 4-fold symmetry. A complete overview of our proposed models' top-scored symmetry complexes and clustering is recapped in Figure S2 and all individual plots for each prediction are shown in Figures S3-S20.

Eventually, the symmetry complexes are clustered by their binding interfaces, and potential repetitive assemblies are tested by superposition and scored with the aim of identifying the most probable representation of a fully assembled S-layer. For SLPs listed in Table 1, high-scored SymProFold S-layer models were obtained (Fig. 3), corresponding well with the experimentally determined S-layer parameters. The primitive unit cell of each S-layer is shown in Figure S21. For p1 S-layers of EA1 from Bacillus anthracis and the main SLP from Bacillus licheniformis, different steps of SymProFold were used to manually build the models of these S-layers (Figures S22, S23).

Validation of Models with Experimental Data

We selected a broad range of S-layer proteins from Gram-positive, Gram-negative bacteria and archaea and validated calculated assembly models with experimentally published data (Table 1).

Table 1

Comparison of lattice constants determined by prediction and experimentally.
Organism	Accession No	Amino acids	Sym.	Unit cell prediction [nm]	Unit cell experimental [nm]	Experimental Data
Bacillus anthracis	P94217	862	p1	a: 7.3 b: 8.9 γ: 105°	a: 6.9 b: 8.3 γ: 106° ²²	Neg. stain projection map
Desulfurococcus mucosus	E8R795	904	p4	17.5	18.0 ²³	Neg. stain tilt series
Aeromonas salmonicida	P35823	502	p4	12.5	12.1 ²⁴ 12.5 ²⁵	Neg. stain tilt series
Sporosarcina ureae	Q0VJW4	1097	p4	14.0	12.9 ²⁶	Neg. stain tilt series
Aneurinibacillus thermoaerophilus	Q6TL21	738	p4	10.6	10 ²⁷	Freeze etch
Viridibacillus arvi	A0A0K2Z0V7	1016	p4	13.2	13.3 ²⁸	AFM
Corynebacterium glutamicum	Q2VRQ3	491	p6	17.1	16.0 ²⁹	AFM
Methanococcus vannielii	A6URZ5	567	p6	10.8	12.0 ³⁰	Crystal structure of M. acetivorans SLP
Brevibacillus brevis	P06546	1053	p6	17.0	17.8 ³¹	Neg. stain projection map
Thermoanaerobacter kivui	P22258	762	p6	18.1	19 ³²	Neg. stain tilt series
Bacillus licheniformis	P49052	874	p1	a: 7.5 b: 9.3 γ: 82°		No experimental data
Vibrio aerogenes	A0A1M5ZCF8	646	p4	12.7		No experimental data
Vibrio quintilis	A0A1M7YYM3	642	p4	12.2		No experimental data
Paenibacillus naphtalenovorans	A0A0U2M877	1053	p4	12.4		No experimental data
Pyrococcus abyssi	Q9V0N3	604	p6	18.2		No experimental data
Methanococcus voltae	Q50833	576	p6	10.1		No experimental data
Thermococcus camini	A0A7G2D8K1	489	p6	14.4		No experimental data
Thermococcus thioreducens	A0A0Q2M111	607	p6	18.5		No experimental data
Phocaeicola vulgatus	A0A5P3AY64	1100	p6	25.9		No experimental data

The unit cell parameters extracted by SymProFold of our assembled models correlate with literature values (Table 1). If available, we also compared the experimental microscopy data with our models (Fig. 4). Overall, the domain arrangement and assembly of our predictions agree well with the published data, showing average differences of 5% regarding cell parameters (Table 1). The SymProFold predicted S-layer assemblies reveal detailed structural insights where experimental studies provide limited and ambiguous information. For S-layers from Vibrio aerogenes, Paenibacillus naphtalenovorans, Pyrococcus abyssi, Methanococcus voltae, Thermococcus camini, Thermococcus thioreducens, and Phocaeicola vulgatus, no experimental data on the unit cell parameters, structure, or symmetry are reported. Still, SymProFold predictions of the mentioned S-layers exhibit high output scores, indicating possible symmetry and structural architecture of these S-layers (Fig. 3).

Exploring Novel Features Revealed by Predicted S-Layer Assemblies

To date, the structures of only a few SLPs have been reported (Table S1). The assembly data acquired for novel S-layers provides a foundation for the functional analysis of distinct S-layers, as well as for delineating novel features within this protein family (Figure S24). As observed for D. mucosus, pore sizes can be enormous, indicating that in this individual case, the flexibility of the layer is more critical than the barrier function. Most of the other predicted S-layer assemblies form a tight network, acting as an efficient filter selectively allowing molecules to pass in and out. Moreover, depending on the environmental conditions ^34–36 either flexibility or mechanical stability of the S-layer is of greater importance. The length of the structural elements linking the surface exposed region to the membrane provides additional valuable information about the thickness of the periplasmic-like space in bacteria and archaea.

The prediction of the C. glutamicum S-layer reveals a single domain alpha-helical protein with p6 symmetry. The C-terminus ends in a predicted transmembrane helix resembling the anchoring mechanism of archaeal S-layers. The SymProFold prediction of S-layers from P. abyssi, M.voltae, T. camini, T. thioreducens and M.vannielii, even though different in sequence, size, and domain structure, show an analog domain arrangement as the crystal structure of M. acetivorans SLP (Figure S25) and reveal an undescribed dimeric anchor, which has not been described previously (Figure S24A-D). For the assembly prediction of S-layers from B. brevis and V. arvi (Slp1), an additional region above the core, responsible for the self-assembly, is present. The domains and residues in this additional layer might be important for interaction with the environment, including the host.

Prediction of Viral Capsids

In the case of viral capsids, a curved surface favors a 5-fold rotational symmetry (Figure S26). SymProFold predicted an icosahedral viral capsid from Odonata-associated circular virus 21 (T = 1) belonging to the Smacoviridae ³⁷. Analog to the regular SymProFold, two symmetry complexes were calculated and superimposed to one fully assembled tile (Fig. 4A-C). The curvature is present in the calculated tile (Fig. 4D, E) that, when superimposed on each other, generates the full viral capsid (Fig. 4F).

Prediction of symmetrical Oligomers with only one Symmetry Axis

SymProFold was further tested on proteins that do not form symmetrical 2D arrays yet are known, through literature or existing crystal structure data, to possess a symmetry axis. The circadian clock protein KaiC (Uniprot: Q8GGL1) forms symmetrical ring-like shaped hexamers ³⁸. SymProFold can reliably predict the 6-fold axis with high ipTM + pTM scores. Another 2-fold axis was found with extremely low scores just above 0.2 (Figure S27). At the superposition step, it was not possible to generate an assembly in a plane without obtaining severe clashes. Therefore, it was automatically filtered out. The crystal structure of YabJ (PDB 5Y6U ³⁹) forms a homotrimer (Figure S28A). SymProFold identified a 3-fold symmetry complex with good scores (Figure S29). A 4-fold symmetry complex was predicted with much lower scores, showing the same main binding interfaces as the 3-fold prediction, and therefore was clustered to the same axis. A superposition is not possible. As a third example, we tested the sequence of N9 neuraminidase from the Influenza A virus. The crystal structure (PDB 6MCX ⁴⁰) shows a symmetrical 4-fold axis (Figure S28B). For this protein, SymProFold predicts a single cluster with a 4-fold rotational symmetry axis (Figure S30). Since there is only one axis of rotational symmetry, spanning a 2D layer is not possible.

SymProFold can predict the structural organization of higher symmetrical assemblies, such as S-layers, as present in their native state in assembled form at the cell surfaces. As only sequential information is required as an input, this approach enables numerous avenues for investigating symmetrical formations, even those that pose significant experimental challenges, such as high toxicity, pathogenicity, extreme growth conditions, and substantial costs for the experimental investigation. Furthermore, the self-assembly property and significant sequence variation of S-layers presents a challenge for structural characterization using techniques such as X-ray crystallography and electron microscopy.

Due to the considerable size diversity of SLPs, some outreach 1000 amino acids ⁶, predictions of tetramers or hexamers can easily exceed the computing power needed to get highly confident models with AlphaFold. Nevertheless, the rapid improvements of AlphaFold in predicting large protein complexes will increase the size limit of SymProFold calculations. Calculated models can potentially be improved by manual pre-processing of input sequences according to prior knowledge. Regions like signal sequences or domains not involved in the assembly, such as cell wall anchors, can be removed before the calculation, thereby reducing the size of the input sequence. The SymProFold method has several stringent restrictions that lead to early termination of the pipeline when sequences of non-assembling proteins are used. In this case, it would be highly unlikely to find more than one defined symmetry complex. A superposition and assembly are highly unlikely even if two symmetry complexes are found due to a prediction artifact.

In case SymProFold fails to detect more than one strong intermolecular interaction, the construction of a fully assembled layer is not possible and results in a partially assembled model. This limitation causes a loss of information regarding weak but critical interactions, which may be required for a complete assembly. S-layers and viral capsids, which consist of multiple proteins, also represent a challenge for SymProFold predictions. In such cases, prior knowledge may be needed as the identification of the interactions important for the initiation of the assembly could be difficult via the automatic pipeline. Nevertheless, manual intervention and/or an extension of the software could overcome this limitation and enable the prediction of assemblies composed of more than one protein. Further adaptations in SymProFold are also needed for viral capsid automated predictions to address potential challenges like pseudosymmetry, higher triangulation number and variations in assembly curvature.

Many biological functions of S-layers depend on the completeness of the cell coverage as well as the structural and physicochemical repetitive uniformity, down to the subnanometer scale. Obtained structural data of assembled S-layers reveal properties of the exposed surface regions (Figure S31), internal pores, and anchoring domains essential for cell attachment and highlight interaction interfaces within the layer. Understanding these aspects is crucial for elucidation of the specific function of a particular S-layer, depending on the microorganism. This is especially important in pathogenic bacteria since S-layers play a role in surface adhesion and in interactions with the host's immune system. Furthermore, numerous studies on the in vivo and in vitro morphogenesis of S-layers demonstrated that lattice growth on growing cells is a highly dynamic process ^{6,7,11,41–45}. Approximately 500 subunits per second must be synthesized at high growth rates, translocated to the cell surface, and incorporated in a defined orientation to the existing lattice while maintaining an equilibrium of the lowest free energy ^6,44–47. The adaptable lattice of S-layers on growing and dividing bacterial and archaeal cells represents an advanced evolutionary stage in morphogenesis ⁴⁴. At these specific sites, bonds must swiftly open and re-form. The dynamics is also present in the assembly of the capsid during the viral life cycle, including disassembly upon infection and reassembly during viral packaging. To gain insights into these dynamic processes, the SymProFold predicted assemblies can serve as a starting point to identify the required interactions within the lattice and perform molecular dynamic studies.

Structural insights into the S-layer assembly offer a vast opportunity for developing new technologies that can mimic the unique properties of these versatile and adaptable structures. Rational engineering of artificial structures with different properties could include developing new materials with advanced self-assembly properties or creating new drug delivery systems and biosensors ⁴⁸. SymProFold predictions enable the development of computational models to simulate S-layers' behavior under different conditions. Since the S-layers of pathogens serve as favorable regions for drug targeting ^34,49, understanding of anchoring mechanisms and/or interactions within the self-assembled S-layer provides the ideal basis for rational drug design and opens new strategies for weakening S-layer protective function. Furthermore, structural information now allows for analyzing the unique antifouling properties of S-layer lattices⁶ and offers a foundation for mimicking these structures in polymer technologies ⁵⁰. The precise biophysical characterization of the pores allows for custom alterations of S-layer permeability, which was previously possible only through chemical modifications ¹¹. S-layers have also shown great potential as a platform for drug delivery because of their biocompatibility, stability, and regular pore structure. Lipidic nanoformulations such as liposomes, solid lipid nanoparticles (SLNs), or emulsomes as drug delivery systems show better resistance to oxidative stress and membrane damage when covered with a crystalline S-layer ^51,52. Uptake studies with emulsomes coated with S-layer proteins did not show significant cytotoxicity by human liver carcinoma cells (HepG2). The capacity to recrystallize S-layer proteins in lipidic formulations allows extremely precise targeted delivery of specific antibodies in high concentrations^53,54 or drug-loaded particles, especially poorly water-soluble targets such as antimicrobial peptides or easily degradable biologicals such as enzymes used for the enzyme replacement therapy ^6,51–53,55. New functionalities, such as cellular targeting and generation of fusion proteins with specific properties⁵⁶ can be introduced. The S-layers of bacteria and archaea in the human microbiome are perfect candidates for these biotechnological applications. Altogether, this opens a new chapter in rational engineering and adaptation of many S-layers for applications in medicine and diagnostics, like treating fungal and viral infections, dermal conditions, cancer, immune deficiency, or rare genetic disorders.

Pre-filtering of possible Prediction Candidates

SymProFold uses sequence files in fasta format as input and consequently checks their predictability by performing homodimer predictions with AlphaFold-Multimer ¹⁸. Homodimer predictions are scored according to the ipTM + pTM score, a model confidence score for complex predictions, which ranges from 0–1 and is calculated using 80% of the ipTM score and 20% of the pTM score as described by Evans et al. 2022 ¹⁸. Only homodimer predictions with a minimal score of 0.3 ipTM + pTM are further processed, excluding proteins that are unlikely to dimerize. The sequences of 19 annotated SLPs, 3 non-SLPs, and viral capsid from Odonata-associated circular virus 21 were used as input. Table 1 includes a list of proteins for which SymProFold S-layer predictions were performed. We used the annotated S-layer protein for each species, as reported in the Uniprot databank, for all of the presented S-layer predictions and calculations.

Identification of Domains and Subchains

As a next step, domains of the protein need to be defined in a fasta file with domains separated by line breaks. Domain boundaries can be set manually, or an automated domain identification can be used. Both start with predicting the full-length protein structure as a monomer using AlphaFold. For automated domain identification, we created a ‘Domain_Separator’ tool that can be included in the SymProFold pipeline to prepare the fasta file. Domain_Separator is a Python script that uses a coordinate file as input, identifies structural domains, and creates a fasta file with domain sequences separated by line breaks. Additionally, a ChimeraX file with separately colored domains can be generated. In an iterative process, small domain subsections are merged until they describe a complete domain. The initial domain subsections are chain ranges created through local crosslinking of secondary structures. Using relations between contact areas, surface areas, and subsection sequence lengths, neighboring domain subsections are then iteratively merged into even larger domain subsections. This iterative process ends when a saturation of merging is reached. In a postprocessing step, for each linker between domain sections, a cropping point is identified that minimizes the contacts between both domains. An example of Domain_Separator utilization is found in the SI and Figure S1. The generated fasta file can be used as direct input for SymProFold. Both methods lead to comparable results for identifying domains in our presented examples.

Once the domains are defined in the fasta file, a set of five different subchains of the protein is generated. These subchains represent truncations of the protein according to a scheme described in Table S2, which includes the full-length sequence, a subchain without the N-terminus, a subchain without the C-terminus, the first third of the domains, and the last third of the domains. Minimum or maximum lengths were defined (Table S2) to mitigate the effects of very large or small domains.

Prediction of Oligomer Sets and Filtering

For each subchain, the algorithm starts different oligomer predictions (dimers, trimers, tetramers, and hexamers). Five models are calculated for each prediction, and the resulting symmetry complexes are evaluated and further processed.

All predictions are evaluated to check if there is a rotational symmetry axis that can align all the subchain components within the oligomer prediction to each other, satisfying the symmetry requirement. Therefore, the algorithm checks whether the occurring angles between the monomers correspond to a 2-, 3-, 4- or 6-fold rotational symmetry and whether the associated rotational symmetry axis is uniform within a tolerance range. Models with an ipTM + pTM score of ≥ 0.20 and existing interfaces between the monomers are treated as symmetry complexes. The order of the rotational symmetry axis (k-fold) of each complex is deduced from its symmetry angle (2-, 3-, 4-, or 6-fold rotational symmetry axis). For example, a trimer prediction of a given subchain of an SLP that forms a p6 S-layer could result in a 3-fold axis (∆φ = 120°) or 6-fold axis (∆φ = 60°). In the latter case, the property of a 6-fold symmetry axis is described by only three predicted subunits instead of 6. The symmetry complexes are further filtered by criteria to discard predictions of low quality. With respect to the complete protein chain, relaxed models with at least 3.0 clashes per 100aa (and 60.0 per 100aa for unrelaxed) are excluded. Additionally, in all possible sequence sections of length 200aa, only 6.0 clashes per 100aa for relaxed models (and 120.0 per 100aa for unrelaxed) are allowed. The number of clashes is calculated using the corresponding method in ChimeraX ⁵⁷. Models in which a part of the protein chain passes incorrectly through the neighboring molecule are excluded.

Clustering via Interfaces

The symmetry complexes are clustered according to the agreement between their binding interfaces, which are compared via interface matrices (distograms). Each cluster can be viewed as a candidate for a rotational symmetry axis of the assembled S-layer. All symmetry complexes are represented by a nodes in a network graph, connected by edges. Every edge weight is proportional to a correlation coefficient between the interface matrices of the two symmetry complexes (nodes) connected by it. A more significant agreement between both interface matrices leads to a larger correlation coefficient (see SI for details). This network graph is partitioned using the Louvain ⁵⁸ method.

Typically, each cluster contains differently cropped subchains, resulting in the same symmetry axis. For each cluster, high ipTM + pTM scores and, where present, several different molecule counts leading to the same order of symmetry are indicators for the fold of the respective rotational symmetry axis.

Superposition of Symmetry Complexes

Pairs of symmetry complexes from 2 different clusters (symmetry axes) are tested on the possibility of repetitive 2D assembly formation by the superposition of overlapping regions. At least one domain must constitute the overlapping region. The pairs of symmetry complexes to be tested are selected according to indicators of highest scores in a cluster. In case no clear result is found, more pairs of symmetry complexes with lower ipTM + pTM scores from different clusters can be tested. From a pair of symmetry complexes (A, B), the symmetry complex (A) with the higher order of rotational symmetry is aligned in z direction and then complemented by the corresponding number of copies of the other (B) by overlapping superposition. The symmetry complexes are superposed using the matchmaker method of ChimeraX ⁵⁷. Subsequently, the rotational symmetry axes of symmetry complexes B are aligned in z direction using linkers between complexes A and B as pivot points. Then the symmetry complexes B are complemented by copies of the symmetry complex A by overlapping superposition and the rotational symmetry axes are aligned likewise.

Parameter Extraction

The superposition of the assembled S-layer is scored based on the average intermolecular clashes per residue and the bending score of the assembly. Assembly bending is assessed by the axial tilt between the rotational symmetry axes of symmetry complexes A and B. The bending score is the RMSD of the angles between the axes of the central symmetry complex A and the nearest neighbor symmetry complexes B. The score is normalized so that an angle of 0 rad (0°) corresponds to a bending score of 0, and an angle of π/2 rad (90°) corresponds to a bending score of 1. A combined quality score is calculated from the average clashes per residue and the bending score.

$$\text{scor}{\text{e}}_{\text{clash}} =\frac{{N}_{\text{clas}\text{hes}}}{{N}_{\text{res}}}$$

$$\text{scor}{\text{e}}_{\text{bend}}=\frac{2}{\pi }\sqrt{\frac{1}{n}{\sum }_{i=1}^{n}{\left({\phi }_{i}-{\phi }_{0}\right)}^{2}}$$

$$\text{scor}{\text{e}}_{\text{quality}}=\text{scor}{\text{e}}_{\text{clash}}+\text{scor}{\text{e}}_{\text{bend}}$$

Unit cell parameters are determined using averaged distances calculated from vector differences between the noncentral symmetry complexes A with their central symmetry mate. Function command clashes in ChimeraX is used to calculate the clash score.

Assembly of S-Layer Unit Cell

A mathematically exact unit cell is created using the determined unit cell parameters. To obtain the best possible model, the quality score can be optimized by variation of the pairs of symmetry clusters. The final output is a mmcif file with the primitive unit cell, the unit cell parameters, and symmetry operations to generate the symmetry mates to get a fully assembled S-layer. Furthermore, in the output folder files from previous steps of SymProFold for a subchain definition, symmetry complexes (SymPlot) and interface network graph are available.

Prediction of p1 S-Layer

If SymProFold terminates during the workflow and cannot create a model of a primitive unit cell, it might be due to a p1 symmetry of the S-layer. Within a p1 S-layer, no rotational symmetry axes of order 2 or higher occur, therefore SymProFold needs some adjustments to predict the assembly successfully. If a p1 symmetry is known or assumed, an alternative SymProFold method for the p1 S-layer can be manually applied. In the p1 pipeline, no set of subchains is generated, but a full-length prediction is split either manually or with Domain_Separator into all possible single domains. Sets of heterodimers of the full-length protein and each single domain are calculated. If two or more strong interactions are found, the full-length model is mapped to the individual domains, and the fully assembled layer in both directions is generated. Automated superposition of axis objects can still be applied to generate the fully assembled S-layer.

Prediction of the Viral Capsid

As a test case we selected the viral capsid from Odonata-associated circular virus 21 (Uniprot: A0A0B4UH63). Allowing for the possibility of a 5-fold rotational symmetry axis in the SymProFold workflow basically allows the prediction of some viral capsids. For this, oligomer predictions are performed for dimers, trimers, tetramers, pentamers and hexamers. In our test case, the predictions is made for only one subchain. In the filtering step, the oligomer predictions are tested for the presence of rotational symmetry axes that can map subchains in the oligomer prediction to one another and fulfill rotational symmetry operations. Therefore, the algorithm checks whether the occurring angles between the monomers correspond to 2-, 3-, 4-, 5- or 6-fold rotational symmetries and whether the associated rotational symmetry axes are uniform within a tolerance range. Since the surface of viral capsids is not a flat 2D plane, the rotational symmetry axes of symmetry complexes B are not aligned parallel to symmetry complex A. After manual parameter setting, the rotational symmetry axes can be snapped into an orientation that results in the icosahedral symmetry of a fully assembled capsid.

Resources

For SymProFold, we present an implementation in Python 3, which uses function libraries of ChimeraX 1.6 ⁵⁷. The Domain_Separator tool is written in Python and uses ChimeraX function libraries and dssp ⁵⁹ to determine secondary structures. Contacts between residues are determined using the function contacts of ChimeraX and the functions surface and measure to calculate solvent-excluded surface areas. Domain_Separator uses the python library networkx.

Computing requirements

An AlphaFold-Multimer 2.3 installation in standard configuration with complete databases and v3 model weights was used for all complex calculations. The predictions were calculated at the VSC-5 Vienna Scientific Cluster (Vienna, Austria) on a GPU (NVIDIA A100, 40GB of vram).

Data and code availability

The source code of SymProFold and for Domain_Separator, as well as the selected test cases to validate Domain_Separator and all the models calculated with SymProFold, are available under https://github.com/symprofold.

Authors’ contributions:

Conceptualization: CB, TS, TPK

Methodology: CB, TS, CG, NG, UBS, IU, TPK

Investigation: TS, CB, CG, NG, IU, TPK

Visualization: TS, CB, CG

Funding acquisition: NG, IU, TPK

Project administration: TPK

Supervision: UBS, IU, TPK

Writing – original draft: CB, TS, CG, NG, TPK

Writing – review & editing: CB, TS, CG, NG, UBS, IU, TPK

ACKNOWLEDGMENTS

This study was supported by doc.fund projects Molecular Metabolism (DOC 50) and Biomolecular Structure and Interactions (DOC 130) funded by the Austrian Science Fund FWF, Land Steiermark, the City of Graz and Doctoral Academy Graz (BioMolStruct consortium). The computational results presented have been achieved using the Vienna Scientific Cluster (VSC), project 71272. N.G. thanks FWF for financial support (T-1239) and I.U. for Grants PGC2018-101370-B-I00 and PID2021-128751NB-I00 (Ministry of Science and Innovation / Spanish State Research Agency / European Regional Development Fund / European Union) and support from Science and Technology Facilities Council (CCP4-ARCIMBOLDO_LOW).

Ahnert, S. E., Marsh, J. A., Hernandez, H., Robinson, C. V. & Teichmann, S. A. Principles of assembly reveal a periodic table of protein complexes. Science (1979) 350, aaa2245–aaa2245 (2015).
Levy, E. D., Pereira-Leal, J. B., Chothia, C. & Teichmann, S. A. 3D Complex: A Structural Classification of Protein Complexes. PLoS Comput Biol 2, e155 (2006).
Muok, A. R. et al. Atypical chemoreceptor arrays accommodate high membrane curvature. Nat Commun 11, 5763 (2020).
Pum, D., Breitwieser, A. & Sleytr, U. B. Patterns in Nature—S-Layer Lattices of Bacterial and Archaeal Cells. Crystals (Basel) 11, 869 (2021).
Messner, P., Schäffer, C., Egelseer, E.-M. & Sleytr, U. B. Occurrence, Structure, Chemistry, Genetics, Morphogenesis, and Functions of S-Layers. in Prokaryotic Cell Wall Compounds 53–109 (Springer Berlin Heidelberg, 2010).
Sleytr, U. B., Schuster, B., Egelseer, E. M. & Pum, D. S-layers: Principles and applications. FEMS Microbiol Rev 38, 823–864 (2014).
Pum, D., Toca-Herrera, J. L. & Sleytr, U. B. S-Layer protein self-assembly. International Journal of Molecular Sciences vol. 14 2484–2501.
Stel, B., Cometto, F., Rad, B., De Yoreo, J. J. & Lingenfelder, M. Dynamically resolved self-assembly of S-layer proteins on solid surfaces. Chemical Communications 54, 10264–10267 (2018).
Rad, B. et al. Ion-specific control of the self-assembly dynamics of a nanostructured protein lattice. ACS Nano 9, 180–190 (2015).
Bharat, T. A. M., von Kügelgen, A. & Alva, V. Molecular Logic of Prokaryotic Surface Layer Structures. Trends in Microbiology vol. 29 405–415.
Fagan, R. P. & Fairweather, N. F. Biogenesis and functions of bacterial S-layers. Nature Reviews Microbiology vol. 12 211–222.
Stuknyte, M. et al. Lactobacillus helveticus MIMLh5-Specific Antibodies for Detection of S-Layer Protein in Grana Padano Protected-Designation-of-Origin Cheese. Appl Environ Microbiol 80, 694–703 (2014).
Schuster, B. & Sleytr, U. B. S-Layer Ultrafiltration Membranes. Membranes (Basel) 11, 275 (2021).
Schuster, B. & Sleytr, U. B. Nanotechnology with S-layer Proteins. in Methods Mol Biol. 195–218 (2020).
Johnson, J. E. & Olson, A. J. Icosahedral virus structures and the protein data bank. Journal of Biological Chemistry 296, 100554 (2021).
Baek, M. et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science (1979) 373, 871–876 (2021).
Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).
Evans, R. et al. Protein complex prediction with AlphaFold-Multimer. bioRxiv 2021.10.04.463034 (2022).
Gutnik, D., Evseev, P., Miroshnikov, K. & Shneider, M. Using AlphaFold Predictions in Viral Research. Curr Issues Mol Biol 45, 3705–3732 (2023).
Yang, Z., Zeng, X., Zhao, Y. & Chen, R. AlphaFold2 and its applications in the fields of biology and medicine. Signal Transduct Target Ther 8, 115 (2023).
Mads, J. & ProfileIngemar, A. Accurate prediction of protein assembly structure by combining AlphaFold and symmetrical docking. bioRxiv (2023).
Chateau, A., Van der Verren, S. E., Remaut, H. & Fioravanti, A. The Bacillus anthracis Cell Envelope: Composition, Physiological Role, and Clinical Relevance. Microorganisms 8, 1864 (2020).
Wildhaber, I., Santarius, U. & Baumeister, W. Three-dimensional structure of the surface protein of Desulfurococcus mobilis. J Bacteriol 169, 5563–5568 (1987).
Dooley, J. S., Engelhardt, H., Baumeister, W., Kay, W. W. & Trust, T. J. Three-dimensional structure of an open form of the surface layer from the fish pathogen Aeromonas salmonicida. J Bacteriol 171, 190–197 (1989).
Stewart, M., Beveridge, T. J. & Trust, T. J. Two patterns in the Aeromonas salmonicida A-layer may reflect a structural transformation that alters permeability. J Bacteriol 166, 120–127 (1986).
Engelhardt, H., Saxton, W. O. & Baumeister, W. Three-dimensional structure of the tetragonal surface layer of Sporosarcina ureae. J Bacteriol 168, 309–317 (1986).
Kadurugamuwa, J. L. et al. S-Layered Aneurinibacillus and Bacillus spp. Are Susceptible to the Lytic Action of Pseudomonas aeruginosa Membrane Vesicles. J Bacteriol 180, 2306–2311 (1998).
Suhr, M., Lederer, F. L., Günther, T. J., Raff, J. & Pollmann, K. Characterization of three different unusual s-layer proteins from Viridibacillus arvi jg-b58 that exhibits two super-imposed s-layer proteins. PLoS One 11, (2016).
Scheuring, S. et al. Charting and unzipping the surface layer of Corynebacterium glutamicum with the atomic force microscope. Mol Microbiol 44, 675–684 (2002).
Arbing, M. A. et al. Structure of the surface layer of the methanogenic archaean Methanosarcina acetivorans. Proc Natl Acad Sci U S A 109, 11812–11817 (2012).
Tsuboi, A. et al. In vitro reconstitution of a hexagonal array with a surface layer protein synthesized by Bacillus subtilis harboring the surface layer protein gene from Bacillus brevis 47. J Bacteriol 171, 6747–6752 (1989).
Lupas, A. et al. Domain structure of the Acetogenium kivui surface layer revealed by electron crystallography and sequence analysis. J Bacteriol 176, 1224–1233 (1994).
Couture-Tosi, E. et al. Structural Analysis and Evidence for Dynamic Emergence of Bacillus anthracis S-Layer Networks. J Bacteriol 184, 6448–6456 (2002).
Fioravanti, A., Mathelie-Guinlet, M., Dufrêne, Y. F. & Remaut, H. The Bacillus anthracis S-layer is an exoskeleton-like structure that imparts mechanical and osmotic stabilization to the cell wall. PNAS Nexus 1, (2022).
Pandur, Ž. & Stopar, D. Evolution of mechanical stability from lipid layers to complex bacterial envelope structures. in 207–251 (2021).
Engelhardt, H. Are S-layers exoskeletons? The basic function of protein surface layers revisited. J Struct Biol 160, 115–124 (2007).
Varsani, A. & Krupovic, M. Smacoviridae: a new family of animal-associated single-stranded DNA viruses. Arch Virol 163, 2005–2015 (2018).
Mori, T. et al. Circadian clock protein KaiC forms ATP-dependent hexameric rings and binds DNA. Proceedings of the National Academy of Sciences 99, 17203–17208 (2002).
Fujimoto, Z., Hong, L. T. T., Kishine, N., Suzuki, N. & Kimura, K. Tetramer formation of Bacillus subtilis YabJ protein that belongs to YjgF/YER057c/UK114 family. Biosci Biotechnol Biochem 85, 297–306 (2021).
Streltsov, V. A., Schmidt, P. M. & McKimm-Breschkin, J. L. Structure of an Influenza A virus N9 neuraminidase with a tetrabrachion-domain stalk. Acta Crystallogr F Struct Biol Commun 75, 89–97 (2019).
Boutonnet, C. et al. Dynamic Profile of S-Layer Proteins Controls Surface Properties of Emetic Bacillus cereus AH187 Strain. Front Microbiol 13, (2022).
Comerci, C. J. et al. Topologically-guided continuous protein crystallization controls bacterial surface layer self-assembly. Nat Commun 10, 2731 (2019).
Herrmann, J. et al. A bacterial surface layer protein exploits multistep crystallization for rapid self-assembly. Proceedings of the National Academy of Sciences 117, 388–394 (2020).
Pum, D., Messner, P. & Sleytr, U. B. Role of the S layer in morphogenesis and cell division of the archaebacterium Methanocorpusculum sinense. J Bacteriol 173, 6865–6873 (1991).
Sleytr, U. B. & Plohberger, R. The Dynamic Process of Assembly of Two-Dimensional Arrays of Macromolecules on Bacteria Cell Walls. in 36–47 (1980).
Sleytr, U. B. & Glauert, A. M. Analysis of regular arrays of subunits on bacterial surfaces; evidence for a dynamic process of assembly. J Ultrastruct Res 50, 103–116 (1975).
Sleytr, U. B. Heterologous reattachment of regular arrays of glycoproteins on bacterial surfaces. Nature 257, 400–402 (1975).
Schuster, B. S-Layer Protein-Based Biosensors. Biosensors (Basel) 8, 40 (2018).
Missiakas, D. & Schneewind, O. Assembly and Function of the Bacillus anthracis S-Layer. Annual Review of Microbiology vol. 71 79–98.
Picher, M. M. et al. Nanobiotechnology advanced antifouling surfaces for the continuous electrochemical monitoring of glucose in whole blood using a lab-on-a-chip. Lab Chip 13, 1780 (2013).
Schuster, B. & Sleytr, U. B. Biomimetic interfaces based on S-layer proteins, lipid membranes and functional biomolecules. J R Soc Interface 11, 20140232 (2014).
Ucisik, M., Sleytr, U. & Schuster, B. Emulsomes Meet S-layer Proteins: An Emerging Targeted Drug Delivery System. Curr Pharm Biotechnol 16, 392–405 (2015).
Preiner, J. et al. IgGs are made for walking on bacterial and viral surfaces. Nat Commun 5, 4394 (2014).
Pérez-Herrero, E. & Fernández-Medarde, A. Advanced targeted therapies in cancer: Drug nanocarriers, the future of chemotherapy. European Journal of Pharmaceutics and Biopharmaceutics vol. 93 52–79 (2015).
Howorka, S. Rationally engineering natural protein assemblies in nanobiotechnology. Curr Opin Biotechnol 22, 485–491 (2011).
Ilk, N., Egelseer, E. M. & Sleytr, U. B. S-layer fusion proteins—construction principles and applications. Curr Opin Biotechnol 22, 824–831 (2011).
Pettersen, E. F. et al. UCSF ChimeraX: Structure visualization for researchers, educators, and developers. Protein Science 30, 70–82 (2021).
Blondel, V. D., Guillaume, J.-L., Lambiotte, R. & Lefebvre, E. Fast unfolding of communities in large networks. J. Stat. Mech. P10008 (2008).
Kabsch, W. & Sander, C. Dictionary of protein secondary structure: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers 22, 2577–637 (1983).

The authors declare no competing interests.

4689020supp8343138s4nwdb.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

SymProFold - Structural prediction of symmetrical biological assemblies

Status:

Version 1

Abstract

Figures

Introduction

Results

Validation of Models with Experimental Data

Exploring Novel Features Revealed by Predicted S-Layer Assemblies

Prediction of Viral Capsids

Prediction of symmetrical Oligomers with only one Symmetry Axis

Discussion

Methods

Pre-filtering of possible Prediction Candidates

Identification of Domains and Subchains

Prediction of Oligomer Sets and Filtering

Clustering via Interfaces

Superposition of Symmetry Complexes

Parameter Extraction

Assembly of S-Layer Unit Cell

Prediction of the Viral Capsid

Resources

Computing requirements

Data and code availability

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1