Stability Oracle: a structure-based graph-transformer framework for identifying stabilizing mutations.

Engineering stabilized proteins is a fundamental challenge in the development of industrial and pharmaceutical biotechnologies. We present Stability Oracle: a structure-based graph-transformer framework that achieves SOTA performance on accurately identifying thermodynamically stabilizing mutations. Our framework introduces several innovations to overcome well-known challenges in data scarcity and bias, generalization, and computation time, such as: Thermodynamic Permutations for data augmentation, structural amino acid embeddings to model a mutation with a single structure, a protein structure-specific attention-bias mechanism that makes transformers a viable alternative to graph neural networks. We provide training/test splits that mitigate data leakage and ensure proper model evaluation. Furthermore, to examine our data engineering contributions, we fine-tune ESM2 representations (Prostata-IFML) and achieve SOTA for sequence-based models. Notably, Stability Oracle outperforms Prostata-IFML even though it was pretrained on 2000X less proteins and has 548X less parameters. Our framework establishes a path for fine-tuning structure-based transformers to virtually any phenotype, a necessary task for accelerating the development of protein-based biotechnologies.

Biosensor and machine learning-aided engineering of an amaryllidaceae enzyme.

A major challenge to achieving industry-scale biomanufacturing of therapeutic alkaloids is the slow process of biocatalyst engineering. Amaryllidaceae alkaloids, such as the Alzheimer's medication galantamine, are complex plant secondary metabolites with recognized therapeutic value. Due to their difficult synthesis they are regularly sourced by extraction and purification from the low-yielding daffodil Narcissus pseudonarcissus. Here, we propose an efficient biosensor-machine learning technology stack for biocatalyst development, which we apply to engineer an Amaryllidaceae enzyme in Escherichia coli. Directed evolution is used to develop a highly sensitive (EC50 = 20 µM) and specific biosensor for the key Amaryllidaceae alkaloid branchpoint 4'-O-methylnorbelladine. A structure-based residual neural network (MutComputeX) is subsequently developed and used to generate activity-enriched variants of a plant methyltransferase, which are rapidly screened with the biosensor. Functional enzyme variants are identified that yield a 60% improvement in product titer, 2-fold higher catalytic activity, and 3-fold lower off-product regioisomer formation. A solved crystal structure elucidates the mechanism behind key beneficial mutations.

Substrate-embedded metal meshes for ITO-free organic light emitting diodes.

Organic light-emitting diodes (OLEDs) have great potential for use in large-area display and lighting applications, but their widespread adoption for large areas is hindered by the high cost and insufficient performance of indium tin oxide (ITO) anodes. In this study, we introduce an alternative anode material - a silver mesh embedded in glass - to facilitate production of large-area OLEDs. We present a facile, scalable manufacturing technique to create high aspect ratio micromeshes embedded in glass to provide the planar geometry needed for OLED layers. Our phosphorescent green OLEDs achieve a current efficiency of 51.4 cd/A at 1000 cd/m2 and reach a slightly higher external quantum efficiency compared to a standard ITO/glass reference sample. Notably, these advancements are achieved without any impact on the viewing angle of the OLEDs. These findings represent a promising advancement towards ITO-free, high-efficiency OLEDs for various high performance, large-area applications, such as lighting and displays.

Identification and characterization of a MAPT-targeting locked nucleic acid antisense oligonucleotide therapeutic for tauopathies.

Tau is a microtubule-associated protein (MAPT, tau) implicated in the pathogenesis of tauopathies, a spectrum of neurodegenerative disorders characterized by accumulation of hyperphosphorylated, aggregated tau. Because tau pathology can be distinct across diseases, a pragmatic therapeutic approach may be to intervene at the level of the tau transcript, as it makes no assumptions to mechanisms of tau toxicity. Here we performed a large library screen of locked-nucleic-acid (LNA)-modified antisense oligonucleotides (ASOs), where careful tiling of the MAPT locus resulted in the identification of hot spots for activity in the 3' UTR. Further modifications to the LNA design resulted in the generation of ASO-001933, which selectively and potently reduces tau in primary cultures from hTau mice, monkey, and human neurons. ASO-001933 was well tolerated and produced a robust, long-lasting reduction in tau protein in both mouse and cynomolgus monkey brain. In monkey, tau protein reduction was maintained in brain for 20 weeks post injection and corresponded with tau protein reduction in the cerebrospinal fluid (CSF). Our results demonstrate that LNA-ASOs exhibit excellent drug-like properties and sustained efficacy likely translating to infrequent, intrathecal dosing in patients. These data further support the development of LNA-ASOs against tau for the treatment of tauopathies.

Application of Pharmacokinetic/Pharmacodynamic Modeling to Bridge Mouse Antitumor Efficacy and Monkey Toxicology Data for Determining the Therapeutic Index of an Interleukin-10 Fc Fusion Protein.

Pharmacokinetic/pharmacodynamic (PK/PD) modeling was performed to quantitatively integrate preclinical pharmacology and toxicology data for determining the therapeutic index (TI) of an interleukin-10 (IL-10) fragment crystallizable (Fc) fusion protein. Mouse Fc fused with mouse IL-10 (mFc-mIL-10) was studied in mice for antitumor efficacy, and the elevation of interleukin-18 (IL-18) was examined as a PD biomarker. The in vivo mFc-mIL-10 EC50 for the IL-18 induction was estimated to be 2.4 nM, similar to the in vitro receptor binding affinity (Kd) of 3.2 nM. The IL-18 induction was further evaluated in cynomolgus monkeys, where the in vivo induction EC50 by a human IL-10 human Fc-fusion protein (hFc-hIL-10) was 0.08 nM vs. 0.3 nM measured as the in vitro Kd. The extent of the IL-18 induction correlated with mouse antitumor efficacy and was used to connect mouse efficacy to that in monkeys. The PD-based efficacious dose projected in monkeys was comparable to the results obtained using a PK-based method in which mouse efficacious exposure was targeted and corrected for affinity differences between the species. Furthermore, PK/PD relationships were developed for anemia and thrombocytopenia in monkeys treated with hFc-hIL-10, with thrombocytopenia predicted to be dose-limiting toxicity. Using quantitative pharmacology and toxicology information obtained through modeling work in the same species, the TI of hFc-hIL-10 in monkeys was determined to be 2.4 (vs. PD-based efficacy) and 1.2-3 (vs. PK-based efficacy), indicating a narrow safety margin. The model-based approaches were proven valuable to the developability assessment of the IL-10 Fc-fusion protein.

Target-Mediated Drug Disposition Affects the Pharmacokinetics of Interleukin-10 Fragment Crystallizable Fusion Proteins at Pharmacologically Active Doses.

Fragment crystallizable (Fc) fusion is commonly used for extending the half-life of biotherapeutics such as cytokines. In this work, we studied the pharmacokinetics of Fc-fused interleukin-10 (IL-10) proteins that exhibited potent antitumor activity in mouse syngeneic tumor models. At pharmacologically active doses of ≥0.1 mg/kg, both mouse Fc-mouse IL-10 and human Fc-human IL-10, constructed as the C terminus of the Fc domain fused with IL-10 via a glycine-serine polypeptide linker, exhibited nonlinear pharmacokinetics after intravenous administration to mice at the doses of 0.05, 0.5, and 5 mg/kg. With a nominal dose ratio of 1:10:100; the ratio of the area under the curve for mouse Fc-mouse IL-10 and human Fc-human IL-10 was 1:181:1830 and 1:75:633, respectively. In contrast, recombinant mouse or human IL-10 proteins exhibited linear pharmacokinetics in mice. Compartmental analysis, using the Michaelis-Menten equation with the in vitro IL-10 receptor alpha binding affinity inputted as the Km, unified the pharmacokinetic data across the dose range. Additionally, nontarget-mediated clearance estimated for fusion proteins was ∼200-fold slower than that for cytokines, causing the manifestation of target-mediated drug disposition (TMDD) in the fusion protein pharmacokinetics. The experimental data generated with a mouse IL-10 receptor alpha-blocking antibody and a human Fc-human IL-10 mutant with a reduced receptor binding affinity showed significant improvements in pharmacokinetics, supporting TMDD as the cause of nonlinearity. Target expression and its effect on pharmacokinetics must be determined when considering using Fc as a half-life extension strategy, and pharmacokinetic evaluations need to be performed at a range of doses covering pharmacological activity. SIGNIFICANCE STATEMENT: Target-mediated drug disposition can manifest to affect the pharmacokinetics of a fragment crystallizable (Fc)-fused cytokine when the nontarget-mediated clearance of the cytokine is decreased due to neonatal Fc receptor-mediated recycling and molecular weight increases that reduce the renal clearance. The phenomenon was demonstrated with interleukin-10 Fc-fusion proteins in mice at pharmacologically active doses. Future drug designs using Fc as a half-life extension approach for cytokines need to consider target expression and its effect on pharmacokinetics at relevant doses.

Acute Neurotoxicity of Antisense Oligonucleotides After Intracerebroventricular Injection Into Mouse Brain Can Be Predicted from Sequence Features.

Antisense oligonucleotides are a relatively new therapeutic modality and safety evaluation is still a developing area of research. We have observed that some oligonucleotides can produce acute, nonhybridization dependent, neurobehavioral side effects after intracerebroventricular (ICV) dosing in mice. In this study, we use a combination of in vitro, in vivo, and bioinformatics approaches to identify a sequence design algorithm, which can reduce the number of acutely toxic molecules synthesized and tested in mice. We find a cellular assay measuring spontaneous calcium oscillations in neuronal cells can predict the behavioral side effects after ICV dosing, and may provide a mechanistic explanation for these observations. We identify sequence features that are overrepresented or underrepresented among oligonucleotides causing these reductions in calcium oscillations. A weighted linear combination of the five most informative sequence features predicts the outcome of ICV dosing with >80% accuracy. From this, we develop a bioinformatics tool that allows oligonucleotide designs with acceptable acute neurotoxic potential to be identified, thereby reducing the number of toxic molecules entering drug discovery pipelines. The informative sequence features we identified also suggest areas in which to focus future medicinal chemistry efforts.

Learning the local landscape of protein structures with convolutional neural networks.

One fundamental problem of protein biochemistry is to predict protein structure from amino acid sequence. The inverse problem, predicting either entire sequences or individual mutations that are consistent with a given protein structure, has received much less attention even though it has important applications in both protein engineering and evolutionary biology. Here, we ask whether 3D convolutional neural networks (3D CNNs) can learn the local fitness landscape of protein structure to reliably predict either the wild-type amino acid or the consensus in a multiple sequence alignment from the local structural context surrounding site of interest. We find that the network can predict wild type with good accuracy, and that network confidence is a reliable measure of whether a given prediction is likely going to be correct or not. Predictions of consensus are less accurate and are primarily driven by whether or not the consensus matches the wild type. Our work suggests that high-confidence mis-predictions of the wild type may identify sites that are primed for mutation and likely targets for protein engineering.

Discovery of BMS-986144, a Third-Generation, Pan-Genotype NS3/4A Protease Inhibitor for the Treatment of Hepatitis C Virus Infection.

The discovery of a pan-genotypic hepatitis C virus (HCV) NS3/4A protease inhibitor based on a P1-P3 macrocyclic tripeptide motif is described. The all-carbon tether linking the P1-P3 subsites of 21 is functionalized with alkyl substituents, which are shown to effectively modulate both potency and absorption, distribution, metabolism, and excretion (ADME) properties. The CF3Boc-group that caps the P3 amino moiety was discovered to be an essential contributor to metabolic stability, while positioning a methyl group at the C1 position of the P1' cyclopropyl ring enhanced plasma trough values following oral administration to rats. The C7-fluoro, C6-CD3O substitution pattern of the P2* isoquinoline heterocycle of 21 was essential to securing the targeted potency, pharmacokinetic (PK), and toxicological profiles. The C6-CD3O redirected metabolism away from a problematic pathway, thereby circumventing the time-dependent cytochrome P (CYP) 450 inhibition observed with the C6-CH3O prototype.

Safety and efficacy of anti-PD-L1 therapy in the woodchuck model of HBV infection.

Immune clearance of Hepatitis B virus (HBV) is characterized by broad and robust antiviral T cell responses, while virus-specific T cells in chronic hepatitis B (CHB) are rare and exhibit immune exhaustion that includes programmed-death-1 (PD-1) expression on virus-specific T cells. Thus, an immunotherapy able to expand and activate virus-specific T cells may have therapeutic benefit for CHB patients. Like HBV-infected patients, woodchucks infected with woodchuck hepatitis virus (WHV) can have increased hepatic expression of PD-1-ligand-1 (PD-L1), increased PD-1 on CD8+ T cells, and a limited number of virus-specific T cells with substantial individual variation in these parameters. We used woodchucks infected with WHV to assess the safety and efficacy of anti-PD-L1 monoclonal antibody therapy (αPD-L1) in a variety of WHV infection states. Experimentally-infected animals lacked PD-1 or PD-L1 upregulation compared to uninfected controls, and accordingly, αPD-L1 treatment in lab-infected animals had limited antiviral effects. In contrast, animals with naturally acquired WHV infections displayed elevated PD-1 and PD-L1. In these same animals, combination therapy with αPD-L1 and entecavir (ETV) improved control of viremia and antigenemia compared to ETV treatment alone, but with efficacy restricted to a minority of animals. Pre-treatment WHV surface antigen (sAg) level was identified as a statistically significant predictor of treatment response, while PD-1 expression on peripheral CD8+ T cells, T cell production of interferon gamma (IFN-γ) upon in vitro antigen stimulation (WHV ELISPOT), and circulating levels of liver enzymes were not. To further assess the safety of this strategy, αPD-L1 was tested in acute WHV infection to model the risk of liver damage when the extent of hepatic infection and antiviral immune responses were expected to be the greatest. No significant increase in serum markers of hepatic injury was observed over those in infected, untreated control animals. These data support a positive benefit/risk assessment for blockade of the PD-1:PD-L1 pathway in CHB patients and may help to identify patient groups most likely to benefit from treatment. Furthermore, the efficacy of αPD-L1 in only a minority of animals, as observed here, suggests that additional agents may be needed to achieve a more robust and consistent response leading to full sAg loss and durable responses through anti-sAg antibody seroconversion.