Controlling membrane deformations

Facilitating the study of membrane remodeling processes

The plasticity of membranes plays an important functional role in cells, cell components and micelles, where bending, budding and remodeling implement numerous recognition and communication processes. Comparatively, molecular simulation methods to induce, control and quantitatively characterize such deformations remain scarce. We introduce a novel collective coordinate associated with membrane bending, which strives to combine realism (by preserving the notion of local atomic curvatures) and low computational cost (allowing its evaluation at every time step of a molecular dynamics simulation). Enhanced sampling simulations along this conformational coordinate provide convenient access to the underlying bending free energy landscape. We validate the method on three state-of-the-art problems: the determination of the bending free energy landscape of a POPE bilayer, the formation of a POPE liposome, and the study of the influence of the Pseudomonas quinolone signal on the budding of Gram-negative bacterial outer membranes.

• Reference:
• B. Bouvier, Journal of Chemical Theory and Computation 15, 6551, 2019.
• Categories: membrane curvature, free energy calculations, enhanced sampling simulations, method development

Deep learning protein interfaces

Predicting structure and function from protein-protein interface shape

To power the specific recognition and binding of protein partners into functional complexes, a wealth of information about the structure and function of the partners is necessarily encoded into the global shape of protein-protein interfaces and their local topological features. To identify whether this is the case, this study uses convolutional deep learning methods (typically leveraged for 2D image recognition) on 3D voxel representations of protein-protein interfaces colored by burial depth. A novel two-stage network, fed with voxelizations of each interface at two distinct resolutions, achieves balance between performance and computational cost. From the shape of the interfaces, the network tries to predict the presence of secondary structure motifs at the interface and the molecular function of the corresponding complex. Secondary structure and certain classes of function are found to be very well predicted, validating the hypothesis of interface shape as a conveyor of higher-level information. Interface patterns triggering the recognition of specific classes are also identified and described.

• Reference:
• B. Bouvier, Journal of Chemical Information and Modeling 61, 3292, 2021.
• Categories: protein-protein recognition, protein-protein interfaces, deep learning, computational geometry, method development

Oligosaccharides as protein interface mimics

Using deep learning to tap the potential for novel drugs

Protein-protein interfaces are areas of contact between the partners of the protein-protein complexes that implement cellular life. They are widely recognized as promising drug targets, much less prone to resistance issues than ligand binding sites on which most therapeutic molecules are based today. However, targeting protein-protein interfaces is notoriously difficult: unlike protein-ligand interfaces which are small and have specific shapes, protein-protein interfaces are often large and flat, and less than 5% of the interface aminoacids (called hotspots) actually have large binding affinities to the other protein partner. In this project, we design polysaccharide scaffolds obtained from renewable sources and functionalize them with the relevant chemical functions to mimic the interaction of proteins with the hotspot residues of their partners. This can prevent the corresponding native protein-protein complexes from forming, inhibiting their function; in addition, sugar-based scaffolds have advantages over other types of protein-protein interface mimics (reduced self-aggregation, proteolysis and immune response).

We perform microsecond-timescale enhanced sampling simulations of several hundreds of oligosaccharide mimics; then, we use recurrent deep neural networks to analyze and model the resulting conformational spaces. The trained networks are able to categorize any mimic conformation as a free energy minimum, and suggest stable conformations of mimics of given substituent sequences. This can be used, for instance, to suggest potential candidates for high-throughput protein-mimic docking simulations within a drug design pipeline.

• Reference:
• B. Bouvier, Journal of Chemical Information and Modeling, doi:10.1021/acs.jcim.3c00179, 2023.
• Categories: drug design, protein-protein recognition, oligosaccharides, chemobiology, deep learning, free energy calculations, enhanced sampling simulations, method development

Ionic liquids

From green fuels to batteries and biomass extraction

Ionic liquids are organic salts with low melting points. Their applications range from the extraction of pollutants during soil remediation, to increasing ionic conductivity in novel hybrid batteries, through the extraction of polymers from biomass for the generation of biofuels... To help organic chemists in the lab design novel ionic liquids obtained from renewable sources, we simulate them in their intended roles. Examples of such projects include (i) checking that ionic liquids do not harm soil or biomass microorganisms (yeast, bacteria...) by destroying their membranes; (ii) making sure that they facilitate ionic conduction in the electrolytes of next-generation rechargeable batteries; (iii) evaluating their performance for extracting chitosan (a valuable building block for bio/nanotechnology applications) from the natural chitin polymer found in crustacean and insect shells.

• References:
• C. Ferreira Funes, B. Bouvier, C. Cézard, C. Fuentealba, A. Jamali, M. Courty, C. Hadad, A. Nguyen Van Nhien, Journal of Molecular Liquids 375, 121350, 2023.
• J. Bidal, C. Cézard, B. Bouvier, C. Hadad, A. Nguyen Van Nhien, M. Becuwe, Journal of Colloid and Interface Science 625, 734, 2022.
• G. Huet, M. Araya-Farias, R. Alayoubi, S. Laclef, B. Bouvier, I. Gosselin, C. Cézard, R. Roulard, M. Courty, C. Hadad, E. Husson, C. Sarazin, A. Nguyen Van Nhien, Green Chemistry 22, 2935, 2020.
• Categories: ionic liquids, green chemistry, biomass extraction, batteries, membranes

Ion mobility in batteries

Exploiting the affinities of LiTFSI and silica to enhance conductivity

This work aims at providing a fundamental insight into the interaction between a lithium salt and an inorganic filler in a perspective of lithium mobility. Through a synergistic approach, coupling experimental results and molecular dynamics simulations, the influence of the surface chemical state of the nanosilica Stöber-type on the dissociation of LiTFSI and its impact on the lithium conduction properties are studied. For this purpose, the surface modification of silica nanoparticles was performed by different methods such as calcination, lithiation and capping with organosilane. The impact of the surface modification on the dissociation of the lithium salt is further investigated by electrochemical impedance spectroscopy after impregnation of the material with a defined amount of lithium salt. The combined experimental and in silico analyses of the results, performed for the first time on such systems, allow a detailed understanding of the interaction between the salt and the support and should prove itself useful for the future design of hybrid polymer electrolytes in new generation batteries.

• Reference:
• J. Bidal, C. Cézard, B. Bouvier, C. Hadad, A. Nguyen Van Nhien, M. Becuwe, Journal of Colloid and Interface Science 625, 734, 2022.
• Categories: batteries, ionic liquids, green chemistry, conductivity

Multivalent antibiotics

Preventing pathogenic bacteria from recognizing their host

Bacterial lectins are nonenzymatic, sugar-binding proteins involved in the formation of biofilms and the onset of virulence. The weakness of individual sugar-lectin interactions is compensated by the potentially large number of simultaneous copies of such contacts, resulting in high overall sugar-lectin affinities and marked specificities. Therapeutic compounds functionalized with sugar residues can only compete with host glycans for the binding to lectins if they are able to take profit of this multivalent binding mechanism. Glycopeptide dendrimers, featuring tree-like topologies with sugar moieties at their leaves, have already shown great promise in this regard. However, optimizing the dendrimers’ aminoacid sequence is necessary to match the dynamics of the lectin active sites with that of the multivalent ligands.

This work combines long-timescale, coarse-grained simulations of dendrimers and lectins with a reasoned exploration of the dendrimer sequence space, in an attempt to suggest sequences that could maximize multivalent binding to the galactose-specific bacterial lectin LecA. These candidates are validated by simulations of mixed dendrimer/lectin solutions, and the effects ofthe dendrimers on lectin dynamics are discussed. This approach is an attractive first step in the conception of therapeutic compounds based on the dendrimer scaffold, and contributes to the understanding of the various classes of multivalency which underpin the ubiquitous ‘sugar code’.

• Reference:
• B. Bouvier, Journal of Chemical Information and Modeling 56, 1193, 2016.
• Categories: drug design, sugar-lectin recognition, coarse-grain model

Pyoverdines as antibiotics

Fighting the opportunistic bacterium Pseudomonas aeruginosa using its own iron acquisition machinery

The Gram-negative bacterium Pseudomonas aeruginosa, a ubiquitous human opportunistic pathogen, has developed resistances to multiple antibiotics. It uses its primary native siderophore, pyoverdine, to scavenge the iron essential to its growth in the outside medium and transport it back into its cytoplasm. Despite its relative simplicity, pyoverdine features remarkably flexible recognition characteristics whose origins at the atomistic level remain only partially understood. It can bind other metals than ferric iron, and the FpvA receptor on the bacterial outer membrane which recognizes and internalizes it can also bind pyoverdines from other Pseudomonads or synthetic analogues. Pyoverdine derivatives could therefore be used as vectors to deliver antibiotics into the bacterium.

We combine quantum chemistry, coarse-grained and all-atom molecular dynamics and free energy calculations to characterize the mechanisms and thermodynamics of the recognition of the native pyoverdines of P. aeruginosa and P. fluorescens by FpvA and delineate the features that pyoverdines with high affinity for FpvA should possess. In particular, we show that (i) the dynamics and interaction of the unbound pyoverdines with water should be optimized with equal care as the interface contacts in the complex with FpvA; (ii) the C-terminal extremity of the pyoverdine chain, which appears to play no role in the bound complex, is involved in the intermediate stages of recognition; and (iii) the length and cyclicity of the pyoverdine chain can be used to fine-tune the kinetics of the recognition mechanism.

Another factor behind the polymorphic recognition of pyoverdines lies in their ability to bind iron with two distinct chiralities. We examine how the stereochemistry of the iron chelating groups influences the structure and dynamics of the two aforementioned pyoverdines and impacts their recognition by the FpvA transporter. We show that conformational preferences for one metal binding chirality over the other observed in solution are canceled out by the FpvA transporter; however, FpvA discriminates between pyoverdines by altering the kinetics of stereoisomer interconversion. We present structural causes of this intriguing recognition mechanism and discuss its possible significance in the context of the competitive scavenging of iron.

Finally, using pyoverdines as a drug scaffold is made difficult by the synthetic complexity of their polycyclic and charged chromophore moiety. Fortunately, the versatility of pyoverdines, which need to accomodate multiple roles beyond iron scavenging (quorum sensing, biofilm formation...) is based on structural and dynamical tradeoffs that can be exploited (if correctly understood). Thus, we suggest analogs of the pyoverdine chromophore that are more amenable to chemical synthesis, evaluate the ability of the corresponding chimeric pyoverdines to bind iron and be recognized by FpvA, and provide guidelines for the design of effective synthetic siderophores.

• References:
• C. Cézard, P. Sonnet, B. Bouvier, Journal of Biological Inorganic Chemistry 24, 659, 2019.
• B. Bouvier, C. Cézard, Physical Chemistry Chemical Physics 19, 29498, 2017.
• B. Bouvier, C. Cézard, P. Sonnet, Physical Chemistry Chemical Physics 17, 18022, 2015.
• Categories: drug design, drug-transporter recognition, free energy calculations, enhanced sampling simulations, chirality

Enhanceosomes: cooperative DNA recognition mechanisms

Transcription factor cooperativity and DNA allostery.

Transcription factors recognize short DNA sequences found in the regulatory regions of genes. Due to the size of genomes and degeneracies linked with homology and convergence, the correct choice of gene targets has to rely on a more sophisticated mechanism than pure DNA-binding specificity. To increase gene regulation specificity, regulatory elements contain a high number of tandem transcription factor binding sites, known as enhanceosomes. Only specific combinations of proteins can bind them to control gene expression. Due to the close proximity of their binding sites within the enhanceosomes, transcription factors bind cooperatively, modifying their affinities for these particular sites and creating a transcriptional response that is both highly specific and sensitive. Moreover, their protein-protein interactions can evoke latent DNA specificities, causing them to occupy binding sites rarely bound by the isolated protein. Transcription factors can even bind cooperatively in the absence of physical interaction between them due to DNA-mediated allostery. However, the structural details behind enhanceosome assembly are still poorly understood to date.

We performed simulations of protein-DNA unbinding to study the cooperative DNA recognition by OCT4 and SOX2, key components of enhanceosomes in pluripotent cells. We found that SOX2 influences the orientation and dynamics of the DNA-bound configuration of OCT4. In addition SOX2 modifies the unbinding free energy profiles of both DNA-binding domains of OCT4, despite interacting directly only with the first. Thus, we demonstrate that the OCT4-SOX2 cooperativity is modulated by an interplay between protein-protein interactions and DNA-mediated allostery.

• Reference:
• F. Merino, B. Bouvier, V. Cojocaru, PLoS Computational Biology, e1004287, 2015.
• Categories: protein-DNA recognition, collective motion, free energy calculations, enhanced sampling simulations

Mechanisms of Ubiquitin Recognition

The cell's postal code

The protein Ubiquitin (Ubq) is the cell's postal code. Multiple ubiquitin units can assemble to form different types of polyubiquitin chains, which are then grafted onto target proteins. Each chain type tags a protein for a certain outcome. For instance, Lys48-connected polyubiquitin chains are used to mark a protein for degradation by the proteasome. Intriguingly, a single, conserved, hydrophobic interface patch ('canonical' patch) on ubiquitin is used for the recogntion of all known polyubiquitin chains (independant of length and linkage type) by protein complexes throughout the cell... What are the mechanisms behind this remarkable specificity?

Lys48-linked polyubiquitin consists of several diubiquitin (Ubq₂) motifs. The two Ubq monomers forming such a motif are connected via their canonical binding patch, which is made unavailable for recognition by other proteins such as the proteasome's ubiquitin-associated (UBA) motif. This 'closed' Ubq₂ motif is known to be very stable, but has to open somehow for the polyUbq chain to be recognized. Yet no 'open' structure of Ubq₂ has been observed until very recently, and the stability of the different forms of Ubq₂ is still the subject of ongoing debate.

Using enhanced-sampling all-atom molecular dynamics simulations, we simulate the unbinding of complexes of mono- or Lys48-linked diubiquitin bound to several UBA domains. We carefully compare and discuss the mechanisms and energetics of the three existing UBA binding modes, stressing the role of water and of correlated motion in the recognition process. Based on these results, we explain the preference of UBA for the Ubq₂ motif (which combines two of these binding modes) over monomeric Ubq. We thoroughly explore the conformational space of Ubq₂ and reveal the existence of a well-defined 'open' conformation where the canonical interfaces of both monomers are accessible. It bears striking similarity to a recently proposed NMR structure but is much less stable than closed Ubq₂. Finally, we find comparable stability for closed Ubq₂ and Ubq₂ bound to UBA, explaining their coexistence from a thermodynamical point of view.

• Reference:
• B. Bouvier, Physical Chemistry Chemical Physics, 16(1), 48, 2014.
• Categories: protein-protein recognition, free energy calculations, enhanced sampling simulations

DNA Recognition by the SRY Protein

How can transcription factors efficiently locate and bind their target DNA sequences inside a genome of several million base pairs?

Transition factors are able to locate and bind their target sequence on DNA much faster than 3D diffusion could typically allow. It has been proposed that this is accomplished via a facilitated diffusion mechanism whereby the protein slides along the DNA, exploring its sequence by quasi-1D diffusion. This mechanism would require the existence of a loosely bound, sequence-nonspecific complex between the protein and DNA, simultaneously allowing the former to scan the sequence of the latter whilst retaining low energy barriers to sliding. An example of such a complex has not been identified to date...

The SRY protein, responsible for sexual differentiation in mammals and involved in sex reversal diseases, is a convenient sandbox for exploring the subtleties of protein-DNA recognition. It binds DNA at specific sequences by intercalating an isoleucine side-chain inside the minor groove, resulting in major deformations of the DNA helix (bending, groove dilation, unwinding of the helix). This is typical of a mechanism called 'indirect recognition' whereby the protein selects its target DNA sequence by its ability to deform in a specific fashion. Starting from the structure of SRY bound to its target DNA sequence, we have simulated the dissociation of the complex and gained insight into its binding and unbinding pathways: the nature and chronology of the deformations of the two partners, the crucial role of water for the stabilization of intermediates along the pathway, and the associated free energy profile (in good agreement with the experimental value of the binding free energy).

Further exploration of the conformational space of the SRY/DNA complex reveals two distinct structure types at intermediate interpartner distances. These correspond to two pathways. The 'binding' pathway leads to the specific SRY/DNA complex. The 'readout' pathway leads to an alternate structure where, in spite of the close proximity between partners, the DNA is unbent and protein-DNA contacts are mediated by a layer of water molecules. Along the 'binding' pathway, the protein shows a distinct preference for its specific target DNA sequence. The 'readout' pathway on the other hand is DNA sequence-agnostic; it leads to a nonspecific complex which meets all the requirements to be involved in the fast 1D readout of the DNA sequence. The choice between pathways is linked to a DNA conformational switch which is only possible when SRY's target DNA sequence is encountered.

• References:
• B. Bouvier , K. Zakrzewska, R. Lavery, Angewandte Chemie 50(29), 6516, 2011.
• B. Bouvier, R. Lavery, Journal of the American Chemical Society 131, 9864, 2009.
• K. Zakrzewska, B. Bouvier , A. Michon, C. Blanchet, R. Lavery, Physical Chemistry Chemical Physics 11, 10712, 2009.
• Categories: protein-DNA recognition, free energy calculations, enhanced sampling simulations

DNA Recognition by Anticancer Drug Daunomycin

Understanding the mechanisms of the recognition of DNA by the anticancer drug daunomycin

Daunomycin represents a class of anticancer drugs which intercalate between DNA base pairs at certain sequences. The mechanism by which the drug explores the DNA sequence and intercalates, however, are unknown. Could the facilitated diffusion mechanism known to accelerate DNQ sequence readout by transcription factors also be used by small molecules? Could this help in the design of a more selective anticancer agent than daunomycin?

Using enhanced-sampling all-atom molecular dynamics simulations, we have simulated the dissociation of daunomycin from its target DNA sequence. The corresponding free energy landscape features the intercalated state as the global minimum; however, other stable structures are visited along the binding/unbinding pathway. In particular, a stable groove-bound state where the planar 4-ring part of daunomycin follows the minor groove axis appears as a good candidate for the exploration of the DNA sequence by sliding.

We used enhanced sampling simulations again starting from this groove-bound state, this time acting on the position of daunomycin along the DNA minor groove. The sliding mechanism thus revealed is associated with free energy barriers of around 3 kcal/mol, with the target DNA sequence as a global minimum. It visits structures where the drug has partially exited the groove. It thus appears likely that the exploration of DNA sequence proceeds by short spans of sliding separated by "jumps" or phases of 3D diffusion. The stability of the groove-bound state at the target DNA sequence favors the transition from this state to the final intercalated state via a conformational transition of the DNA helix.

• Reference:
• M. Wilhelm, A. Mukherjee, B. Bouvier , K. Zakrzewska, J. T. Hynes, R. Lavery, Journal of the American Chemical Society 134, 8588, 2012.
• Categories: drug-DNA recognition, free energy calculations, enhanced sampling simulations

Predicting Active Patches at Protein-Protein Interfaces

Can the geometry of protein-protein interfaces help discriminate between essential and non-essential aminoacids?

The interaction surface (interface) between two proteins can involve contacts between many aminoacids on both partners. However, only a small fraction of these contacts (often called 'hotspots') are actually necessary for the complex to form and operate. Their identification can help in suggesting point mutations and decoy binding partners to prevent the formation of the protein-protein complex. This has however proven very difficult. Experimental mutation studies are expensive ansd time-consuming. Sequence alignement of similar proteins occuring e.g. in different origanisms can reveal aminoacids that have been conserved by evolution and are likely to be important, but a sufficiently large number of sequences is not always available. Water dynamics has also been used to define hotspots (which are generally removed from the flux of bulk water), but at the price of very costly simulations at long timescales.

Here, we use the Voronoi diagram / Delaunay triangulation duality to tesselate the interface between two proteins. We distinguish between water-mediated ('wet') and direct ('dry') contacts between the partners. By 'shelling' the interface, i. e. counting the number of Voronoi cells separating a given point from the interface rim, we obtain a simple and parameter-free descriptor of an atom's accessibility. On a representative set of 54 homo- and heterodimer complexes, we show that this measure of accessibility correlates quite well with water dynamics and residue conservation, both of which are used to define hotspots. This suggests that a straightforward topological analysis of the protein-protein interface can replace much more expensive and time-consuming approaches for the prediction of hotspots.

• Reference:
• B. Bouvier, R. Grünberg, M. Nilges, F. Cazals, Proteins: Structure, Function and Bioinformatics 76, 677, 2009.
• Categories: protein-protein recognition, interfaces, computational geometry

Slow Processes in Nucleic Acids

Large-scale and/or slow transitions in DNA are important ingredients of repair, replication and transcription but remain a methodological challenge...

We have applied enhanced-sampling molecular dynamics to explore two examples of large-scale and/or slow conformational transitions in DNA. The first is base flipping, during which a DNA base pair is flipped out of the helix like a card out of a deck to allow repair and modification enzymes to access it. Base flipping is also involved in the first steps of DNA replication. The flipping can be assisted by an enzyme but has also been shown to spontaneously occur in solution, where the lifespan of the event can reach several milliseconds. To avoid introducing bias in the flipping mechanism, we have quantified the influence of the collective motion of the DNA on the flipping of individual bases and enhanced the sampling along these collective modes rather than directly forcing the rotation of the nucleoside. We discuss the free energy profile and structural transitions thus revealed, which involve important metastable states.

The second example addresses the conformational transitions of the Holliday junction, a 4-way junction involved in homologous recombination processes such as sequence-proofing, repair or crossover (all of which occur during meiosis). The exchange of strands between DNA fragments forming the junction requires the opening of the junction (widening of the eyehole) and the rotation of the strands. By acting upon different collective modes and different atom sets, we have revealed the mechanism of the entire process.

• Reference:
• B. Bouvier , H. Grubmüller, Biophysical Journal 93, 770, 2007.
• Categories: collective motion, free energy calculations, enhanced sampling simulations

Charge and Excitation Transfer in Biomolecules and Clusters

Collective electronic states or localized charge/excitation?

Charge and excitation transfer phenomena are ubiquitous. A positive charge can migrate over long distances along a DNA chain, causing damage away from its point of insertion. During photosynthesis, excitation is transferred from light-harvesting carotenoids to where it is required to drive chemical reactions via a long chain of intermediate molecules. Understanding these mechanisms requires the calculation of the collective ionized or excited electronic states of these systems, which is impossible at the time being using traditional quantum chemistry methods. Here, we employ the effective Hamiltonian method to express these collective states on the basis of the electronic states of the consituent moieties (e.g. DNA bases for a DNA system) and the coupling between them. We first apply the method to the characterization of the ionized states of aromatic clusters, which are interesting as theoretical building blocks but are also encountered as pollutants in the atmosphere. We reveal a competition between the charge transfer effect, which tends to delocalize the charge over many moieites, and the polarization of neutral monomers by charged ones, which has the opposite effect. We propose optimal structures of the studied clusters and comparison with ionization experiments.

We then apply the method to the transfer of singlet electronic excitation in DNA strands, which typically occurs upon irradiation. We obtain a precise description of the excited states of the individual DNA bases at a high level of theory (CASSCF) which we then inject into the effective Hamiltonian framework. For static, canonical DNA strands of sequence poly-dA and poly-(dAdT), we show that the low-lying collective excited states essentially involve one monomer transition but delocalize over many consecutive bases. We then introduce dynamic disorder into the system using molecular dynamics; this tends to break the coupling between bases and localize the excitation inside specific areas of the helix depending on the index of the excited state. When nonradiative de-excitation occurs, the system samples states of decreasing indices, resulting in the transfer of the excitation between areas.

• References:
• B. Bouvier, P. Millié, M. Mons, Journal of Physical Chemistry A 108, 4254, 2004.
• B. Bouvier, J.-P. Dognon, R. Lavery, D. Markovitsi, P. Millié, D. Onidas, K. Zakrzewska, Journal of Physical Chemistry B 107, 13512, 2003.
• B. Bouvier, V. Brenner, P. Millié, J.-M. Soudan, Journal of Physical Chemistry A 106, 10326, 2002.
• B. Bouvier, T. Gustavsson, D. Markovitsi, P. Millié, Chemical Physics 275, 75, 2002.
• Categories: quantum chemistry, semiempirical methods, excited states, transfer phenomena