Speakers

Integrative modeling

Structure-based discovery of conformation-specific small molecule modulators of proteins

Avner Schlessinger*

* Department of Pharmacological Sciences, Icahn School of Medicine at Mount Sinai, New York, New York 10029

* Department of Artificial Intelligence and Human Health, Icahn School of Medicine at Mount Sinai, New York, New York 10029

Small molecules can control the functions of proteins through various mechanisms. For instance, different small molecule ligands, targeting distinct conformations and binding sites, can induce unique conformational changes within the same target, leading to diverse downstream signaling effects. Here, we develop and apply a range of computational methods that combine homology modeling, AlphaFold2, ligand docking, and MD simulations to devise strategies for modulating challenging protein targets such as Solute Carrier (SLC) transporters and protein kinases. Initial hit compounds are subsequently refined through iteration of molecular docking, estimation of free energy of binding, and medicinal chemistry, complemented by biochemical and structural approaches. Our results provide useful chemical tools to characterize reprogrammed disease networks, and establish a framework for developing efficacious lead compounds against emerging therapeutic targets.

Bayesian metamodeling across representations and scales: from nucleocytoplasmic transport to T-cell activation

Bayesian metamodeling offers a versatile framework for integrative modeling of complex systems, by seamlessly integrating diverse input models. These models can leverage various data types and mathematical representations, enabling them to describe different facets of the system at hand across multiple scales and levels of granularity. The metamodeling process proceeds in three steps. First, input models are converted into a standardized statistical representation, underpinned by the foundations of probabilistic graphical models. Next, the models are coupled through correlated variables that are selected by meticulously considering their interrelations in the physical world. Finally, the original input models are harmonized by considering the posterior probabilities of their parameter values. In this presentation, I will delve into some of the benefits of metamodeling, illustrating its efficacy through the lens of two biological systems: nucleocytoplasmic transport and T-cell activation, each operating at distinct scales. For the case of nucleocytoplasmic transport, we used metamodeling to construct a unifying depiction of transport phenomena, bridging the gap between individual nuclear pore complexes and an entire cell and resolving perplexing disparities between experimental data and computational predictions. In the context of T-cell activation, we used metamodeling to unravel the dynamic nanoscale organization of molecules at the immune synapse during the initial stages of T-cell contact formation, as discerned from high-resolution imaging data. By embracing a decentralized approach to the integrative modeling process, metamodeling yields more comprehensive, precise, and accurate descriptions, effectively contextualizing input models, resolving conflicting information, and, crucially, fostering collaborative scientific inquiry by providing a robust framework for sharing expertise, resources, data, and models among researchers.

Extending the value of biomolecular models and simulations: directly influencing clinical decisions

With the rapid increase in sophistication of computational technologies and algorithms, careful applications of biomolecular modelling and simulations at multiple scales have provided extensive insights into the mechanisms that underpin a variety of biomolecular processes. These have resulted in (a) mechanisms that have been experimentally validated, (b) successful design of molecules that have served as leads for drug discovery (c) discovery of new mechanisms among others. Parallel developments in sequencing and associated technologies have generated vast amounts of data that has been mined to generate Alphafold like approaches that are speeding the process of discovery in an unprecedented manner. The final phase of this process is to query whether these methods can directly influence clinical decisions which ultimately is the end goal of national “Precision Medicine” efforts. We explore some recent examples which demonstrate this potential and support the beginning of a promising era in the clinic. 

Microbiology's platypus and the Tree of Life

The relationship between the three domains of life is still one of the most important unanswered question in Biology. Our study of the eukaryotic endomembrane system lead to the proto-coatomer hypothesis for its origin, stating that all non-endosymbiontic eukaryotic compartments have a common origin. Proteins typical of this organization can also be found in peculiar bacteria from the Planctomycetes-Verrucomicrobia-Chlamydiae (PCV) superphylum. Members of this group display features that are rarely observed in bacteria, some of which are usually more associated with eukaryotes or archaea. I will show how the study of these bacteria participate in our exploration of the biodiversity and can redefine our understanding of the Tree of Life

Molecular architecture and interaction dynamics of SARS-CoV-2 envelope via integrative modeling

Despite tremendous efforts, the exact structure of SARS-CoV-2 and related betacoronaviruses remains elusive. SARS-CoV-2 envelope is a key structural component of the virion that encapsulates viral RNA. It is composed of three structural proteins, spike, membrane (M), and envelope, which interact with each other and with the lipids acquired from the host membranes. In my talk, I will present a model of the envelope structure of SARS-CoV-2 with near atomistic detail. In particular our study focuses on the complex interaction network of the virus' most abundant, but largely understudied, M protein. Our results are in good agreement with current experimental data, demonstrating a generic and versatile approach to model the structure of a virus de novo.

Macromolecular complexes in the control of agricultural pests

Abstract: Agricultural pests are the cause of significant crop loss worldwide every year. Transgenic traits delivering protection against these insect pests has proven to be highly effective and has provided environmental and economic benefits where deployed. Almost all insecticidal proteins known to date are pore-forming proteins that oligomerize into macromolecular rings that punch holes in cell membranes lining the insect gut. We present here the structure-function characterization of a newly identified insecticidal protein, Mpf2Ba1, active against Western Corn Rootworm (WCR). Using a combination of x-ray crystallography and cryo-EM we elucidate the key steps of pore-formation using the structures of the soluble monomer, the pre-pore and pore conformations. We also demonstrate that gut fluid extracted from WCR larvae provides the necessary and sufficient environment for the protein activation, triggering conversion from the monomer to the pre-pore state and arming it for membrane penetration as it converts into the pore state. Understanding the structure and function of these oligomeric complexes is critical to design novel insecticidal proteins to combat emerging resistance in insects against existing commercial traits.

Toward fast and accurate prediction of enzyme mutant activity 

Abstract:  My group has been developing and applying computational methods to predict 3D structures of protein mutants, and to evaluate the effects of these mutations on protein function and dynamics, including ligand response. I will present our recent studies on cancer mutations of kinases and drug responses, and on engineering of key enzymes in synthetic biology and biocatalysis, with the aid of structural modeling, molecular dynamics simulations, and machine learning.

Evaluating the role of the nuclear microenvironment for gene function from integrative genome structure modeling

Lorenzo Boninsegna 1,2 , Yuxiang Zhan 1,2 , Francesco Musella 1,2 , Asli Yildirim 1,2 , Ye Wang 1,2 , Frank Alber 1,2 

Institute for Quantitative and Computational Biosciences, 2 Department of Microbiology, Immunology & Molecular Genetics, University of California Los Angeles, 611 Charles E. Young Drive, Los Angeles, CA 90095

Recent advances in microscopy and sequencing-based technologies have provided an opportunity to study the 3D folding of chromosomes in their nuclear environment at single cell level, which can advance our understanding of the causal connections between chromatin structure and function. Here, we discuss a method to produce a population of single cell diploid 3D genome structures by integrating such multimodal data sources. These genome structures predict chromosomal folding relative to nuclear bodies at single cell level, thus can provide a detailed description of the nuclear microenvironment of genes, defined by their subnuclear positions, local chromatin compaction, and preferences in chromatin compartmentalization. We then evaluate how the nuclear microenvironment of genes is linked to their function with respect to their gene transcriptional efficiency, replication timing, and chromatin compartmentalization. Moreover, we study how the nuclear microenvironment of genes varies during cell differentiation and disease phenotypes. We find that certain chromatin regions are distinguished by their strong preferences to a single microenvironment, due to associations to specific nuclear bodies in most cells. Other chromatin shows high structural variability, which is a strong indicator of functional heterogeneity. Moreover, we identify specialized nuclear microenvironments, which distinguish chromatin in different functional states and reveal a key role of nuclear speckles in chromosome organization. Overall, our study reveals a key role for the nuclear microenvironment in gene function.

Multi-scale characterization of a biomolecular condensate

Biomolecular condensates, membraneless assemblies of protein and nucleic acids, are now recognized as an organizing principle of cell biology. Thus, the incorporation of biomolecular condensates into cell modeling efforts is critical. Unlike traditional protein assemblies or complexes, condensates lack fixed stoichiometry, are highly dynamic, and emerge from weak, multivalent interactions. These properties present unique challenges from a modeling perspective. Here, I will present a multi-scale model of the PopZ condensate, a phase-separated microdomain at the poles of the asymmetrically dividing bacterium Caulobacter crescentus. I will further show that PopZ modulates asymmetric division by selectively sequestering regulatory signaling pathways and that the structure and emerging material properties of the PopZ condensate modulate Caulobacter fitness. Finally, I will outline our efforts to simulate PopZ condensation in the context of the entire cell.

Sex-determining 3D regulatory hubs revealed by genome spatial auto-correlation analysis

Mammalian sex is determined by opposing networks of ovarian and testicular genes that are well characterized. However, its epigenetic regulation is still largely unknown, thus limiting our understanding of a fundamental process for species propagation. Here we explore the 3D chromatin landscape of sex determination in vivo, by profiling FACS-sorted embryonic mouse gonadal populations, prior and after sex determination, in both sexes. We integrate Hi-C with ChIP-seq experiments using METALoci, a novel genome spatial auto-correlation analysis that identifies 3D enhancer hubs across the genome. We uncover a prominent rewiring of chromatin interactions during sex determination, affecting the enhancer hubs of hundreds of genes that display temporal- and sex-specific expression. Moreover, the identification of the 3D enhancer hubs allows the reconstruction of regulatory networks, revealing key transcription factors involved in sex determination. By combining predictive approaches and validations in transgenic mice we identify a novel Fgf9 regulatory hub, deletion of which results in male-to-female sex reversal with the upregulation of ovarian-specific markers and the initiation of meiosis. Thus, spatial auto-correlation analysis is an effective strategy to identify regulatory networks associated to biological processes and to further characterize the functional role of the 3D genome.

Protein structural ensembles by integrative computational-experimental approaches

Understanding the molecular mechanisms used by biological systems to perform their functions is often essential to rationally target associated diseases. In many cases, the determination of the 3D structure of these systems provides precious insights. However, it is often the interplay between structural and dynamical properties that determines the behavior of complex systems. While both experimental and computational methods are invaluable tools to study protein structure and dynamics, limitations in each individual technique can hamper their capabilities. On one hand, determining structural models solely from experimental data is challenging as data often come from ensemble-averaged measurements over conformationally heterogeneous states, provide sparse and sometimes ambiguous information, and are subject to random and systematic errors. On the other hand, structural models determined by computational approaches such as molecular dynamics are limited by the inaccuracies of the force fields used as well as by the challenge of exhaustively sampling the conformational landscape of complex systems. Here I will present a combined computational-experimental approach, called metainference, along with two representative applications: the characterization of structural ensembles of the intrinsically disordered amyloid-β using NMR data and the determination of the multiple conformations sampled by an inhibitor of the ASCT2 transporter in the binding pocket using cryo-EM data. Metainference is implemented in the open-source, freely available PLUMED library (www.plumed.org) and can be readily used to determine structural and dynamical properties of conformationally heterogeneous systems by integrating different types of experimental data into molecular simulations.

Modelling virus complexes: unveiling errors and ensuring structural validity in the AlphaFold era

With the emergence of Alphafold2, a breakthrough in protein structure prediction, a crucial objective is to identify errors and safeguard the integrity of the predicted models. By analyzing the limitations and challenges associated with Alphafold's predictions, we show the importance of rigorous validation methods to ensure accuracy and reliability. On one hand, various techniques, such as cryo electron microscopy or cross linking mass spectrometry, can help in error detection and refinement of the models, highlighting the significance of experimental data integration. On the other hand, structure predictions can offer fresh perspectives for interpreting experimental data. We will show examples from both scenarios, using virus complexes and CASP15 predictions.

A 3D model of the Keratin (K5-K14) intermediate filament

We present here the first near-atomic resolution integrative model of the whole Keratin K5-K14 intermediate filament of 16 K5-K14 hetero dimers. Our model is consistent with chemical cross linking studies and has 3 unit lengths of the filament with the coiled-coil regions modelled at atomic resolution. We validated the accuracy of the model in the following ways - a) used it to rationalise ~150 filament disrupting point mutations; b) the model satisfies all 19 cross links observed in the homologous keratin K1-K10 system; c) correlated with experimentally determined radial atomic distribution and d) shown that the predicted cross sectional mosaic of the bundling of filaments is consistent with data from electro micrographs. Moreover, our model is consistent with experimentally determined size and dimensions. It also helps explain why filaments could swell when in contact with lipids. This the first 3D structure of a whole IF filament and could be used as a template to model other Intermediate filaments. These models could be used for detecting possible interacting partners, creating a large near-atomic level resolution networks of the cytoskeleton and designing inhibitors and therapeutics against disease causing variants.

Mapping binding sites using saturation mutagenesis

Despite transformative recent advances in protein structure prediction, it remains challenging to predict bound structures of intrinsically disordered regions (IDRs). Using bacterial toxin-antitoxin systems as test cases, we show that both charged scanning mutagenesis, as well as Cys scanning mutagenesis coupled to chemical labeling, can provide information on binding-site residues as well as bound secondary structure of the antitoxin which has a large IDR in its free state. Computational prediction of epitopes targeted on an antigen in a polyclonal antibody response is another challenging problem with important consequences for vaccine design. We show that similar mutagenesis strategies coupled to either yeast surface display or viral infectivity can be used for this purpose, in the context of polyclonal sera elicited against important viral antigens.

Integrative structure determination of disordered proteins

A significant proportion (~50%) of protein domains constituting large macromolecular assemblies have unknown structure. Often these domains can be intrinsically disordered (IDR) and there is not enough data e.g., structural and/or interaction data, to localize these regions in the assemblies. In the first part, I will talk about a deep-learning based method to identify interface residues for binary interactions comprising an IDR and a partner protein domain.  Our method is additionally able to predict if the IDR is heterogenous in the complex. These interface residues can subsequently be used like crosslink restraints in IMP to localize these regions in an assembly more accurately. In the second part, I will talk about other recent methods my group developed in IMP, including those for improving coarse-grained representation, sampling efficiency, and analysis of integrative models.