Theoretical and Data-Driven Approaches for Biomolecular Condensates.

Biomolecular condensation processes are increasingly recognized as a fundamental mechanism that living cells use to organize biomolecules in time and space. These processes can lead to the formation of membraneless organelles that enable cells to perform distinct biochemical processes in controlled local environments, thereby supplying them with an additional degree of spatial control relative to that achieved by membrane-bound organelles. This fundamental importance of biomolecular condensation has motivated a quest to discover and understand the molecular mechanisms and determinants that drive and control this process. Within this molecular viewpoint, computational methods can provide a unique angle to studying biomolecular condensation processes by contributing the resolution and scale that are challenging to reach with experimental techniques alone. In this Review, we focus on three types of dry-lab approaches: theoretical methods, physics-driven simulations and data-driven machine learning methods. We review recent progress in using these tools for probing biomolecular condensation across all three fields and outline the key advantages and limitations of each of the approaches. We further discuss some of the key outstanding challenges that we foresee the community addressing next in order to develop a more complete picture of the molecular driving forces behind biomolecular condensation processes and their biological roles in health and disease.

Learning the molecular grammar of protein condensates from sequence determinants and embeddings.

Intracellular phase separation of proteins into biomolecular condensates is increasingly recognized as a process with a key role in cellular compartmentalization and regulation. Different hypotheses about the parameters that determine the tendency of proteins to form condensates have been proposed, with some of them probed experimentally through the use of constructs generated by sequence alterations. To broaden the scope of these observations, we established an in silico strategy for understanding on a global level the associations between protein sequence and phase behavior and further constructed machine-learning models for predicting protein liquid-liquid phase separation (LLPS). Our analysis highlighted that LLPS-prone proteins are more disordered, less hydrophobic, and of lower Shannon entropy than sequences in the Protein Data Bank or the Swiss-Prot database and that they show a fine balance in their relative content of polar and hydrophobic residues. To further learn in a hypothesis-free manner the sequence features underpinning LLPS, we trained a neural network-based language model and found that a classifier constructed on such embeddings learned the underlying principles of phase behavior at a comparable accuracy to a classifier that used knowledge-based features. By combining knowledge-based features with unsupervised embeddings, we generated an integrated model that distinguished LLPS-prone sequences both from structured proteins and from unstructured proteins with a lower LLPS propensity and further identified such sequences from the human proteome at a high accuracy. These results provide a platform rooted in molecular principles for understanding protein phase behavior. The predictor, termed DeePhase, is accessible from https://deephase.ch.cam.ac.uk/.

Catalysts from synthetic genetic polymers.

The emergence of catalysis in early genetic polymers such as RNA is considered a key transition in the origin of life, pre-dating the appearance of protein enzymes. DNA also demonstrates the capacity to fold into three-dimensional structures and form catalysts in vitro. However, to what degree these natural biopolymers comprise functionally privileged chemical scaffolds for folding or the evolution of catalysis is not known. The ability of synthetic genetic polymers (XNAs) with alternative backbone chemistries not found in nature to fold into defined structures and bind ligands raises the possibility that these too might be capable of forming catalysts (XNAzymes). Here we report the discovery of such XNAzymes, elaborated in four different chemistries (arabino nucleic acids, ANA; 2'-fluoroarabino nucleic acids, FANA; hexitol nucleic acids, HNA; and cyclohexene nucleic acids, CeNA) directly from random XNA oligomer pools, exhibiting in trans RNA endonuclease and ligase activities. We also describe an XNA-XNA ligase metalloenzyme in the FANA framework, establishing catalysis in an entirely synthetic system and enabling the synthesis of FANA oligomers and an active RNA endonuclease FANAzyme from its constituent parts. These results extend catalysis beyond biopolymers and establish technologies for the discovery of catalysts in a wide range of polymer scaffolds not found in nature. Evolution of catalysis independent of any natural polymer has implications for the definition of chemical boundary conditions for the emergence of life on Earth and elsewhere in the Universe.

Protein Condensate Atlas from predictive models of heteromolecular condensate composition.

Biomolecular condensates help cells organise their content in space and time. Cells harbour a variety of condensate types with diverse composition and many are likely yet to be discovered. Here, we develop a methodology to predict the composition of biomolecular condensates. We first analyse available proteomics data of cellular condensates and find that the biophysical features that determine protein localisation into condensates differ from known drivers of homotypic phase separation processes, with charge mediated protein-RNA and hydrophobicity mediated protein-protein interactions playing a key role in the former process. We then develop a machine learning model that links protein sequence to its propensity to localise into heteromolecular condensates. We apply the model across the proteome and find many of the top-ranked targets outside the original training data to localise into condensates as confirmed by orthogonal immunohistochemical staining imaging. Finally, we segment the condensation-prone proteome into condensate types based on an overlap with biomolecular interaction profiles to generate a Protein Condensate Atlas. Several condensate clusters within the Atlas closely match the composition of experimentally characterised condensates or regions within them, suggesting that the Atlas can be valuable for identifying additional components within known condensate systems and discovering previously uncharacterised condensates.

Serological fingerprints link antiviral activity of therapeutic antibodies to affinity and concentration.

The effectiveness of therapeutic monoclonal antibodies (mAbs) against variants of the SARS-CoV-2 virus is highly variable. As target recognition of mAbs relies on tight binding affinity, we assessed the affinities of five therapeutic mAbs to the receptor binding domain (RBD) of wild type (A), Delta (B.1.617.2), and Omicron BA.1 SARS-CoV-2 (B.1.1.529.1) spike using microfluidic diffusional sizing (MDS). Four therapeutic mAbs showed strongly reduced affinity to Omicron BA.1 RBD, whereas one (sotrovimab) was less impacted. These affinity reductions correlate with reduced antiviral activities suggesting that affinity could serve as a rapid indicator for activity before time-consuming virus neutralization assays are performed. We also compared the same mAbs to serological fingerprints (affinity and concentration) obtained by MDS of antibodies in sera of 65 convalescent individuals. The affinities of the therapeutic mAbs to wild type and Delta RBD were similar to the serum antibody response, indicating high antiviral activities. For Omicron BA.1 RBD, only sotrovimab retained affinities within the range of the serum antibody response, in agreement with high antiviral activity. These results suggest that serological fingerprints provide a route to evaluating affinity and antiviral activity of mAb drugs and could guide the development of new therapeutics.

Both COVID-19 infection and vaccination induce high-affinity cross-clade responses to SARS-CoV-2 variants.

The B.1.1.529 (omicron) variant has rapidly supplanted most other SARS-CoV-2 variants. Using microfluidics-based antibody affinity profiling (MAAP), we have characterized affinity and IgG concentration in the plasma of 39 individuals with multiple trajectories of SARS-CoV-2 infection and/or vaccination. Antibody affinity was similar against the wild-type, delta, and omicron variants (K A ranges: 122 ± 155, 159 ± 148, 211 ± 307 µM-1, respectively), indicating a surprisingly broad and mature cross-clade immune response. Postinfectious and vaccinated subjects showed different IgG profiles, with IgG3 (p-value = 0.002) against spike being more prominent in the former group. Lastly, we found that the ELISA titers correlated linearly with measured concentrations (R = 0.72) but not with affinity (R = 0.29). These findings suggest that the wild-type and delta spike induce a polyclonal immune response capable of binding the omicron spike with similar affinity. Changes in titers were primarily driven by antibody concentration, suggesting that B-cell expansion, rather than affinity maturation, dominated the response after infection or vaccination.

Microfluidic Antibody Affinity Profiling Reveals the Role of Memory Reactivation and Cross-Reactivity in the Defense Against SARS-CoV-2.

Recent efforts in understanding the course and severity of SARS-CoV-2 infections have highlighted both potentially beneficial and detrimental effects of cross-reactive antibodies derived from memory immunity. Specifically, due to a significant degree of sequence similarity between SARS-CoV-2 and other members of the coronavirus family, memory B-cells that emerged from previous infections with endemic human coronaviruses (HCoVs) could be reactivated upon encountering the newly emerged SARS-CoV-2, thus prompting the production of cross-reactive antibodies. Determining the affinity and concentration of these potentially cross-reactive antibodies to the new SARS-CoV-2 antigens is therefore particularly important when assessing both existing immunity against common HCoVs and adverse effects like antibody-dependent enhancement (ADE) in COVID-19. However, these two fundamental parameters cannot easily be disentangled by surface-based assays like enzyme-linked immunosorbent assays (ELISAs), which are routinely used to assess cross-reactivity. Here, we have used microfluidic antibody affinity profiling (MAAP) to quantitatively evaluate the humoral immune response in COVID-19 convalescent patients by determining both antibody affinity and concentration against spike antigens of SARS-CoV-2 directly in nine convalescent COVID-19 patient and three pre-pandemic sera that were seropositive for common HCoVs. All 12 sera contained low concentrations of high-affinity antibodies against spike antigens of HCoV-NL63 and HCoV-HKU1, indicative of past exposure to these pathogens, while the affinity against the SARS-CoV-2 spike protein was lower. These results suggest that cross-reactivity as a consequence of memory reactivation upon an acute SARS-CoV-2 infection may not be a significant factor in generating immunity against SARS-CoV-2.

New Frontiers for Machine Learning in Protein Science.

Protein function is fundamentally reliant on inter-molecular interactions that underpin the ability of proteins to form complexes driving biological processes in living cells. Increasingly, such interactions are recognised as being formed between proteins that exist on a broad spectrum of dynamic conformational states and levels of intrinsic disorder. Additionally, the sizes of the structures formed can range from simple binary complexes to large dynamic biomolecular condensates measuring 100 nm or more. Understanding the parameters that govern such interactions, how they form, how they lead to function and what happens when they take place in unintended manners and lead to disease, represent some of the core questions for molecular biosciences. In light of recent advances made in solving the protein folding problem by machine learning methods, we discuss here the challenges and opportunities brought by these new data-driven approaches for the next frontiers of biomolecular science.

Antibody Affinity Governs the Inhibition of SARS-CoV-2 Spike/ACE2 Binding in Patient Serum.

The humoral immune response plays a key role in suppressing the pathogenesis of SARS-CoV-2. The molecular determinants underlying the neutralization of the virus remain, however, incompletely understood. Here, we show that the ability of antibodies to disrupt the binding of the viral spike protein to the angiotensin-converting enzyme 2 (ACE2) receptor on the cell, the key molecular event initiating SARS-CoV-2 entry into host cells, is controlled by the affinity of these antibodies to the viral antigen. By using microfluidic antibody-affinity profiling, we were able to quantify the serum-antibody mediated inhibition of ACE2-spike binding in two SARS-CoV-2 seropositive individuals. Measurements to determine the affinity, concentration, and neutralization potential of antibodies were performed directly in human serum. Using this approach, we demonstrate that the level of inhibition in both samples can be quantitatively described using the dissociation constants (KD) of the binary interactions between the ACE2 receptor and the spike protein as well as the spike protein and the neutralizing antibody. These experiments represent a new type of in-solution receptor binding competition assay, which has further potential applications, ranging from decisions on donor selection for convalescent plasma therapy, to identification of lead candidates in therapeutic antibody development, and vaccine development.

Machine learning-aided protein identification from multidimensional signatures.

The ability to determine the identity of specific proteins is a critical challenge in many areas of cellular and molecular biology, and in medical diagnostics. Here, we present a macine learning aided microfluidic protein characterisation strategy that within a few minutes generates a three-dimensional fingerprint of a protein sample indicative of its amino acid composition and size and, thereby, creates a unique signature for the protein. By acquiring such multidimensional fingerprints for a set of ten proteins and using machine learning approaches to classify the fingerprints, we demonstrate that this strategy allows proteins to be classified at a high accuracy, even though classification using a single dimension is not possible. Moreover, we show that the acquired fingerprints correlate with the amino acid content of the samples, which makes it is possible to identify proteins directly from their sequence without requiring any prior knowledge about the fingerprints. These findings suggest that such a multidimensional profiling strategy can lead to the development of a novel method for protein identification in a microfluidic format.

Ribozyme-catalysed RNA synthesis using triplet building blocks.

RNA-catalyzed RNA replication is widely believed to have supported a primordial biology. However, RNA catalysis is dependent upon RNA folding, and this yields structures that can block replication of such RNAs. To address this apparent paradox, we have re-examined the building blocks used for RNA replication. We report RNA-catalysed RNA synthesis on structured templates when using trinucleotide triphosphates (triplets) as substrates, catalysed by a general and accurate triplet polymerase ribozyme that emerged from in vitro evolution as a mutualistic RNA heterodimer. The triplets cooperatively invaded and unraveled even highly stable RNA secondary structures, and support non-canonical primer-free and bidirectional modes of RNA synthesis and replication. Triplet substrates thus resolve a central incongruity of RNA replication, and here allow the ribozyme to synthesise its own catalytic subunit '+' and '-' strands in segments and assemble them into a new active ribozyme.