AstraZeneca (Australia)

The first use of sulfinate salts for the late-stage modification of peptidic disulfide bonds is reported. While the majority of cysteine-based peptide modifications rely on the nucleophilicity of the side-chain thiol functionality, umpolung approaches-exploiting instead the electrophilicity of the cystine disulfide-are underexplored. Using structurallydiverse sulfinate salts, we have optimized a mild, photochemical strategy for the generation and coupling of carbon-centered radicals with both symmetrical and electronically-distinct, unsymmetrical cystine disulfides using high-throughput experimentation techniques. A library of modified peptides was accessible, as confirmed by qualitative and quantitative analytical data, providing valuable insights into the matched reactivity of specific radical/disulfide substrate pairings. The method was broadly compatible with a range of unprotected amino acids, including His, Trp and Tyr, and can be used for the functionalization of biologically relevant peptides, as exemplified by the late-stage modification of a semaglutide analogue and the preparation of high-value macrocyclic peptides.

Abstract Background & Aims Liver cancer primarily develops in patients with chronic liver disease (CLD), yet most cases are diagnosed at an advanced stage with poor prognosis. While clinical surveillance of patients with CLD generates extensive longitudinal data, its unstructured free-text nature hinders large-scale research. To unlock this real-world evidence, we developed a scalable framework using open-source Large Language Models (LLMs) to transform unstructured clinical text into structured data. Methods We conducted a multi-stage evaluation of LLM-based extraction from multi-source clinical documentation of liver transplant recipients. A calibration set comprising 507 reports (414 radiology, 65 pathology, and 28 liver transplant assessment reports) from 30 patients was manually annotated to benchmark four open-source LLMs (Llama 3.1 8B, Llama 3.3 70B, Open-BioLLM 70B, DeepSeek R1 8B) against a regular expression baseline across 73 tasks. To ensure structured outputs, we compared constrained decoding (Guidance and Ollama packages) against unconstrained prompting across 5,590 prompt–output pairs. The finalised pipeline was then applied to the full cohort of 835 patients transplanted in our centre over the past decade. Results Among the models tested, Llama 3.3 70B performed best, exceeding 90% accuracy on 59/73 tasks, outperforming both a medically fine-tuned model (OpenBioLLM 70B) and a smaller variant (Llama 3.1 8B). Constrained decoding achieved > 99.9% format adherence, far surpassing unconstrained prompting (87.4%). Applied to the full cohort, the pipeline successfully analysed 22,493 reports to generate 37,125 datapoints (45 variables, 835 patients) without manual annotation. Further analysis confirmed known liver cancer risk factors (male sex, viral hepatitis, smoking, diabetes), and allowed for reconstruction of longitudinal disease timelines. Conclusions This work provides a scalable blueprint for transforming real-world clinical free-text into structured formats, paving the way for accelerated, data-driven research into complex pre-cancerous diseases like CLD.

Abstract The spontaneous deamidation of asparagine residues remains a major obstacle to the stability and efficacy of protein therapeutics. Currently available models in the literature for predicting deamidation liabilities can suffer from limited generalizability, likely due to biases such as sequence similarity within datasets. In this study, we built machine learning models using protein language models (e.g., ESM2) and graph neural networks (GNNs), trained on a comprehensive dataset of 591 asparagine sites from over 105 protein molecules. To address the critical issue of data leakage, we implemented a peptide grouping strategy yielding more accurate estimates of model performance for novel deamidation sites. Our analysis shows that, when sequence similarity bias is controlled, protein language models match traditional feature-based models that use amino acid composition, k-mers, PSSMs, and predicted secondary structure/solvent accessibility, while offering substantial computational advantages. Additionally, our GNN-based pipeline further increases prediction accuracy by up to 8% compared to language model-only tools and delivers a 15–25% improvement over motif-based approaches. This methodological framework enables more reliable and rapid in-silico prediction of deamidation liabilities, potentially reducing costly late-stage interventions in protein therapeutic development and is generalizable to the modeling of additional protein post-translational modifications.

CKD is a global health issue exacerbated by the rising prevalence of diabetes and obesity. Renal fibrosis, characterized by the accumulation of extracellular matrix proteins, is a critical factor in CKD progression. Transforming growth factor β (TGF-β), a key profibrotic cytokine, plays a pivotal role in this process. However, systemic inhibition of TGF-β has been limited by associated toxicities. This study explores the role of αvβ8 integrin in renal fibrosis and its potential as a therapeutic target in CKD. We used various preclinical models of CKD, including humanized αvβ8 mice and the db/db uninephrectomy model, to investigate the role of αvβ8 integrin in renal fibrosis. Gene Set Variation Analysis (GSVA) was employed to assess fibrotic gene signatures in human kidney biopsies. The therapeutic potential of MEDI8367, a monoclonal antibody targeting αvβ8 integrin, was evaluated in vitro and in vivo. Our findings demonstrate that αvβ8 integrin is upregulated in the tubulointerstitium of CKD kidneys, particularly in diabetic kidney disease (DKD), and correlates with TGF-β activation and renal function decline. MEDI8367 effectively inhibited αvβ8-mediated TGF-β activation in vitro and attenuated murine unilateral ureteral obstruction (UUO)-induced renal fibrosis. Notably, inhibition of αvβ8 reduced kidney damage and improved kidney function in models of DKD and hypertensive nephropathy. Our study highlights the potential of MEDI8367 to mitigate renal fibrosis and improve kidney function, offering a novel approach to CKD treatment that complements existing therapies.

For over 30 years, research has highlighted a sex bias in early research, risking the validity of biological knowledge. The first step towards change is effectively challenging misconceptions, allowing researchers to perceive sex inclusive research as doable. Utilising the theory of planned behaviour, we quantified researchers' intention as a proxy measure for conducting sex inclusive research and explored attitude (value of the behaviour), subjective norm (perceived social pressure), and behavioural control (ability to conduct the behaviour). Additionally, we quantified the knowledge gap, prevalence of misconceptions, and assessed perceived benefits and barriers. We tested a workshop intervention that directly challenges the cultural embedded barriers. The data shows researchers' intentions were high, but they had weak statistical knowledge and misunderstandings leading to a perception that inclusive research is prohibitive due to cost and animal use. We demonstrate that participation in the training intervention improved knowledge, altered the perceived barriers, and cultural expectations.

In silico prediction of cell membranepermeabilityis crucial in drug discovery, since a compound’s permeationthrough membranes influences parameters such as its in vivo efficacy,bioavailability, and pharmacokinetics. This study investigates theuse of multitask graph neural networks to predict a selection of permeability-relatedendpoints. The study utilized a harmonized, single-laboratory internaldata set of over 10K compounds measured in human colorectal adenocarcinoma(Caco-2) and Madin–Darby canine kidney (MDCK) cell lines, routinelyemployed in experimental assays for drug permeability and efflux.This data set is an order of magnitude larger than comparable publiccollections, thus providing greater statistical power and a consistenterror profile for model development. A series of multitask learning(MTL) models trained on such data were benchmarked against single-taskapproaches and evaluated on an external public data set to investigatethe model’s applicability domain. The comparison between theperformance of single- and multitask models suggests that MTL canachieve higher accuracy by leveraging shared information across endpoints.MTL is also shown to perform better when augmented with molecularfeatures. In particular, the inclusion of pKa and LogD, is shown to improve the accuracy of both permeabilityand efflux endpoints. This work presents benchmarking results of modelsutilizing different data splitting strategies, accompanied by guidelinesfor optimal validation in the context of MTL.

The ability to pinpoint and predict sites of metabolism(SoMs)is essential for designing and optimizing effective and safe bioactivesmall molecules. However, the number of molecules with annotated SoMsis limited, hindering the advancement of data-driven methods suchas machine learning for metabolism prediction. Here, we provide acomprehensive characterization of SoM data obtained from the readoutsof a human hepatocyte assay conducted at AstraZeneca Gothenburg. Weexplore a new strategy for SoM annotation that accounts for uncertaintyin the experimental data, and we relate our findings to the most comprehensiveSoM data collection available to date. Our study includes entropyanalysis of SoM annotations, accompanied by representative examplesthat highlight the complexities of interpreting and working with metabolismdata. Furthermore, we demonstrate the impact and value of the newmetabolism data on SoM prediction. Importantly, a substantial portionof the data generated and analyzed as part of this work is made publiclyavailable.

Osimertinib is the first-line therapy for EGFR-mutated non-small cell lung cancer, but acquired resistance emerges in most patients and remains a major barrier for complete cure. This phenomenon is most likely associated with the drug-tolerant persister (DTP) cell phenotype, a reversible state that enables survival under treatment and leads to irreversible drug resistance. To uncover the molecular mechanism driving this distinct phenotype, we applied data-independent acquisition mass spectrometry (DIA-MS) to establish the dynamic proteomic and phosphoproteomic landscape in the osimertinib DTPs. While osimertinib initially blocks EGFR signaling, ribosome synthesis and protein translation related pathways arise in DTP phase, and resistance developed through the reactivation of EGFR downstream pathways and anti-apoptotic mechanisms such as YAP1 and mTOR-BAD hyperphosphorylation, as validated by growth combination assays. Kinase enrichment revealed elevated phosphorylation of multiple CDK1 substrates in DTP phase and pharmacological or genetic inhibition of CDK1-mediated SAMHD1 activation significantly impair DTP growth and survival. This study illuminates the dynamic landscape underlying the DTPs biology and identifies biomarker and new targets to potentially prevent or delay the onset of resistance.

<bold>Background:</bold> COPD patients exhibit epithelial remodelling characterized by goblet cell hyperplasia and excessive mucus accumulation. Tozorakimab, an anti-IL-33 mAb which has shown efficacy in COPD, inhibits both IL-33^red and IL-33^ox activity through ST2 and RAGE/EGFR signalling pathways, respectively. <bold>Objective:</bold> To understand the extent to which signalling pathways with suggested clinical benefit in COPD have an impact on epithelial remodelling and wound healing. <bold>Methods:</bold> Anti-IL-33/ST2 mAbs were evaluated for their ability to reduce COPD epithelial remodelling <italic>in vitro</italic>. In addition, effects of anti-IL-4R and anti-TSLP on MUC5AC expression were determined. <bold>Results:</bold> Tozorakimab was the only anti-IL-33/ST2 mAb, including itepekimab and astegolimab, that reduced mucus secretion in differentiated, bronchial COPD ALI cultures (<bold>Table</bold>). mAbs targeting IL-4R and TSLP signalling did not alter MUC5AC expression. An IL-33^ox/RAGE/EGFR-dependent epithelial scratch wound assay was used to compare pharmacology between mAbs, with only tozorakimab showing the ability to promote repair (<bold>Table</bold>). Epitope competition assays revealed a distinct binding epitope for tozorakimab binding to IL-33^red versus other anti-IL-33 mAbs. <bold>Conclusion:</bold> Of the anti-IL-33/ST2 mAbs tested, only tozorakimab demonstrated the ability to reduce epithelial remodelling and promote repair. Tozorakimab is under clinical investigation in COPD and other respiratory diseases. <fig><object-id>erj;66/suppl_69/OA6527/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig>