Toward generalizable prediction of antibody thermostability using machine learning on sequence and structure features.

Over the last three decades, the appeal for monoclonal antibodies (mAbs) as therapeutics has been steadily increasing as evident with FDA's recent landmark approval of the 100th mAb. Unlike mAbs that bind to single targets, multispecific biologics (msAbs) have garnered particular interest owing to the advantage of engaging distinct targets. One important modular component of msAbs is the single-chain variable fragment (scFv). Despite the exquisite specificity and affinity of these scFv modules, their relatively poor thermostability often hampers their development as a potential therapeutic drug. In recent years, engineering antibody sequences to enhance their stability by mutations has gained considerable momentum. As experimental methods for antibody engineering are time-intensive, laborious and expensive, computational methods serve as a fast and inexpensive alternative to conventional routes. In this work, we show two machine learning approaches - one with pre-trained language models (PTLM) capturing functional effects of sequence variation, and second, a supervised convolutional neural network (CNN) trained with Rosetta energetic features - to better classify thermostable scFv variants from sequence. Both of these models are trained over temperature-specific data (TS50 measurements) derived from multiple libraries of scFv sequences. On out-of-distribution (refers to the fact that the out-of-distribution sequnes are blind to the algorithm) sequences, we show that a sufficiently simple CNN model performs better than general pre-trained language models trained on diverse protein sequences (average Spearman correlation coefficient, ρ, of 0.4 as opposed to 0.15). On the other hand, an antibody-specific language model performs comparatively better than the CNN model on the same task (ρ= 0.52). Further, we demonstrate that for an independent mAb with available thermal melting temperatures for 20 experimentally characterized thermostable mutations, these models trained on TS50 data could identify 18 residue positions and 5 identical amino-acid mutations showing remarkable generalizability. Our results suggest that such models can be broadly applicable for improving the biological characteristics of antibodies. Further, transferring such models for alternative physicochemical properties of scFvs can have potential applications in optimizing large-scale production and delivery of mAbs or bsAbs.

Light chain subunit of a poorly soluble human IgG2λ crystallizes in physiological pH environment both in cellulo and in vitro.

Prominent inclusion bodies can develop in the endoplasmic reticulum (ER) when overexpressed antibodies possess intrinsically high condensation propensities. These observations suggest that antibodies deemed to show notable solubility problems may reveal such characteristics preemptively in the form of ER-associated inclusion bodies during antibody overexpression. To define the relationships between solubility problems and inclusion body phenotypes, we investigated the biosynthesis of a model human IgG2λ that shows severe opalescence in an acidic formulation buffer yet retains high solubility at physiological pH. Consistent with the pH-dependent solubility characteristics, the model antibody did not induce notable inclusion body in the physiological pH environment of the ER lumen. However, when individual subunit chains of the antibody were expressed separately, the light chain (LC) spontaneously induced notable crystal-like inclusion bodies in the ER. The LC crystallization event was readily reproducible in vitro by simply concentrating the purified LC protein at physiological pH. Two independent structural determinants for the LC crystallization were identified through rational mutagenesis approach by monitoring the effect of amino acid substitutions on intracellular LC crystallogenesis. The effect of mutations on crystallization was also recapitulated in vitro using purified LC proteins. Importantly, when introduced directly into the model antibody, a mutation that prevents the LC crystallization remediated the antibody's solubility problem without compromising the secretory output or antigen binding. These results illustrate that the ER can serve as a "physiological test tube" that not only reports secretory cargo's high condensation propensity at physiological pH, but also provides an orthogonal method that guides antibody engineering strategy.

Computational design of closely related proteins that adopt two well-defined but structurally divergent folds.

The plasticity of naturally occurring protein structures, which can change shape considerably in response to changes in environmental conditions, is critical to biological function. While computational methods have been used for de novo design of proteins that fold to a single state with a deep free-energy minimum [P.-S. Huang, S. E. Boyken, D. Baker, Nature 537, 320-327 (2016)], and to reengineer natural proteins to alter their dynamics [J. A. Davey, A. M. Damry, N. K. Goto, R. A. Chica, Nat. Chem. Biol. 13, 1280-1285 (2017)] or fold [P. A. Alexander, Y. He, Y. Chen, J. Orban, P. N. Bryan, Proc. Natl. Acad. Sci. U.S.A. 106, 21149-21154 (2009)], the de novo design of closely related sequences which adopt well-defined but structurally divergent structures remains an outstanding challenge. We designed closely related sequences (over 94% identity) that can adopt two very different homotrimeric helical bundle conformations-one short (∼66 Šheight) and the other long (∼100 Šheight)-reminiscent of the conformational transition of viral fusion proteins. Crystallographic and NMR spectroscopic characterization shows that both the short- and long-state sequences fold as designed. We sought to design bistable sequences for which both states are accessible, and obtained a single designed protein sequence that populates either the short state or the long state depending on the measurement conditions. The design of sequences which are poised to adopt two very different conformations sets the stage for creating large-scale conformational switches between structurally divergent forms.

De novo design of a homo-trimeric amantadine-binding protein.

The computational design of a symmetric protein homo-oligomer that binds a symmetry-matched small molecule larger than a metal ion has not yet been achieved. We used de novo protein design to create a homo-trimeric protein that binds the C3 symmetric small molecule drug amantadine with each protein monomer making identical interactions with each face of the small molecule. Solution NMR data show that the protein has regular three-fold symmetry and undergoes localized structural changes upon ligand binding. A high-resolution X-ray structure reveals a close overall match to the design model with the exception of water molecules in the amantadine binding site not included in the Rosetta design calculations, and a neutron structure provides experimental validation of the computationally designed hydrogen-bond networks. Exploration of approaches to generate a small molecule inducible homo-trimerization system based on the design highlight challenges that must be overcome to computationally design such systems.

De novo design of tunable, pH-driven conformational changes.

The ability of naturally occurring proteins to change conformation in response to environmental changes is critical to biological function. Although there have been advances in the de novo design of stable proteins with a single, deep free-energy minimum, the design of conformational switches remains challenging. We present a general strategy to design pH-responsive protein conformational changes by precisely preorganizing histidine residues in buried hydrogen-bond networks. We design homotrimers and heterodimers that are stable above pH 6.5 but undergo cooperative, large-scale conformational changes when the pH is lowered and electrostatic and steric repulsion builds up as the network histidine residues become protonated. The transition pH and cooperativity can be controlled through the number of histidine-containing networks and the strength of the surrounding hydrophobic interactions. Upon disassembly, the designed proteins disrupt lipid membranes both in vitro and after being endocytosed in mammalian cells. Our results demonstrate that environmentally triggered conformational changes can now be programmed by de novo protein design.

Accurate computational design of multipass transmembrane proteins.

The computational design of transmembrane proteins with more than one membrane-spanning region remains a major challenge. We report the design of transmembrane monomers, homodimers, trimers, and tetramers with 76 to 215 residue subunits containing two to four membrane-spanning regions and up to 860 total residues that adopt the target oligomerization state in detergent solution. The designed proteins localize to the plasma membrane in bacteria and in mammalian cells, and magnetic tweezer unfolding experiments in the membrane indicate that they are very stable. Crystal structures of the designed dimer and tetramer-a rocket-shaped structure with a wide cytoplasmic base that funnels into eight transmembrane helices-are very close to the design models. Our results pave the way for the design of multispan membrane proteins with new functions.

Engineering dynamic cell cycle control with synthetic small molecule-responsive RNA devices.

BACKGROUND: The cell cycle plays a key role in human health and disease, including development and cancer. The ability to easily and reversibly control the mammalian cell cycle could mean improved cellular reprogramming, better tools for studying cancer, more efficient gene therapy, and improved heterologous protein production for medical or industrial applications. RESULTS: We engineered RNA-based control devices to provide specific and modular control of gene expression in response to exogenous inputs in living cells. Specifically, we identified key regulatory nodes that arrest U2-OS cells in the G0/1 or G2/M phases of the cycle. We then optimized the most promising key regulators and showed that, when these optimized regulators are placed under the control of a ribozyme switch, we can inducibly and reversibly arrest up to ~80 % of a cellular population in a chosen phase of the cell cycle. Characterization of the reliability of the final cell cycle controllers revealed that the G0/1 control device functions reproducibly over multiple experiments over several weeks. CONCLUSIONS: To our knowledge, this is the first time synthetic RNA devices have been used to control the mammalian cell cycle. This RNA platform represents a general class of synthetic biology tools for modular, dynamic, and multi-output control over mammalian cells.

Synthetic biology: advancing the design of diverse genetic systems.

A major objective of synthetic biology is to make the process of designing genetically encoded biological systems more systematic, predictable, robust, scalable, and efficient. Examples of genetic systems in the field vary widely in terms of operating hosts, compositional approaches, and network complexity, ranging from simple genetic switches to search-and-destroy systems. While significant advances in DNA synthesis capabilities support the construction of pathway- and genome-scale programs, several design challenges currently restrict the scale of systems that can be reasonably designed and implemented. Thus, while synthetic biology offers much promise in developing systems to address challenges faced in the fields of manufacturing, environment and sustainability, and health and medicine, the realization of this potential is currently limited by the diversity of available parts and effective design frameworks. As researchers make progress in bridging this design gap, advances in the field hint at ever more diverse applications for biological systems.

A yeast-based rapid prototype platform for gene control elements in mammalian cells.

Programming genetic circuits in mammalian cells requires flexible, tunable, and user-tailored gene-control systems. However, most existing control systems are either mechanistically specific for microbial organisms or must be laboriously re-engineered to function in mammalian cells. Here, we demonstrate a ribozyme-based device platform that can be directly transported from yeast to mammalian cells in a "plug-and-play" manner. Ribozyme switches previously prototyped in yeast are shown to regulate gene expression in a predictable, ligand-responsive manner in human HEK 293, HeLa, and U2OS cell lines without any change to device sequence nor further optimization. The ribozyme-based devices, which exhibit activation ratios comparable to the best RNA-based regulatory devices demonstrated in mammalian cells to-date, retain their prescribed functions (ON or OFF switch), tunability of regulatory stringency, and responsiveness to different small-molecule inputs in mammalian hosts. Furthermore, we observe strong correlations of device performance between yeast and all mammalian cell lines tested (R(2) = 0.63-0.97). Our unique device architecture can therefore act as a rapid prototyping platform (RPP) based on a yeast chassis, providing a well-developed and genetically tractable system that supports rapid and high-throughput screens for generating gene-controllers with a broad range of functions in mammalian cells. This platform will accelerate development of mammalian gene-controllers for diverse applications, including cell-based therapeutics and cell-fate reprogramming.