Defence Science and Technology Laboratory

Abstract A definitive correlate of protection (CoP) for SARS-CoV-2 has yet to be formally established. Previously, using data from a series of non-human primate vaccine challenge studies, we reported that neutralising antibodies (NAbs) are the strongest candidate for clinical protection against COVID-19 and that spike binding antibody is the strongest candidate CoP for viral burden post-challenge. In this study, we further characterised the protective binding antibody profile by analysing spike antibody-dependent complement deposition, FcγR binding, isotype and antibody glycosylation. Using the machine learning platform SIMON, we demonstrate that antibody-dependent complement deposition (ADCD) and FcγR binding are strong candidate co-correlates for each of the post-challenge outcomes; viral load and lung pathology. We found that spike antibody sialylation closely followed by FcγR2A, was the spike antibody feature with the strongest negative correlation with histopathology score. Spike antibody ADCD, FcγR binding, isotype and glycosylation significantly differed by immunisation regimen and sex, which demonstrates the heterogeneity of immune mechanisms induced by different immunisation platforms. We conclude that spike binding antibody, with the protective functional characteristics described herein, is a candidate CoP that captures both protection from severe clinical disease and protection against a high viral burden. These findings should be taken into consideration for future SARS-CoV-2 vaccine development.

Human activity recognition has been a prominent research focus for several decades. While computer vision-based passive sensing is desirable for many applications, traditional RGB-based methods face significant limitations, including high computational cost, sensitivity to ambient lighting conditions, and privacy concerns. Single-photon LiDAR (light detection and ranging) is emerging as a robust alternative, offering efficient and high-resolution 3D imaging over long ranges and in difficult conditions while preserving privacy. In this study, we combine an eye-safe single-photon LiDAR system with a deep learning pipeline to achieve fast, long-range human activity recognition. To address the issue of limited available data for training, we generate synthetic datasets by combining real motion capture data with virtual models in a 3D modeling environment. A state-of-the-art recurrent neural network is trained on short-duration depth-image sequences of six activities. We also contribute two long-range, video frame rate single-photon LiDAR datasets for human activity recognition, recorded at distances of 325 m and 1.4 km, which exhibit markedly different noise levels. When evaluated on these data, the network maintains more than 80% accuracy even in the most challenging scenario, while supporting continuous and fast inference.

Abstract Ticks are of medical and veterinary concern due to their role as vectors of a wide range of pathogens, including viruses, protozoa, and bacteria. Accurate species identification is essential during tick surveillance and disease control programmes, as certain pathogens are associated with particular tick species. In this study, genetic variation of a partial sequence of the cytochrome c oxidase I ( COI ) gene was used for molecular identification of tick species to corroborate morphological identification. This also enabled investigation of cryptic diversity within tick species. Thirty species belonging to the genera Amblyomma, Argas, Carios, Dermacentor, Haemaphysalis, Hyalomma, Ixodes, Ornithodoros , and Rhipicephalu s were assessed. Tree-topology analysis confirmed discrete clustering of specimens according to species for the majority of taxa. Intraspecific genetic divergence ranged from 0 to 6.08%. In taxa where species complexes are known, separation of discrete groups were found. Several species yielded >2% intraspecific genetic divergence when compared to other taxa, suggesting potential cryptic diversity. DNA barcoding was an effective approach for the morphological identification of UK and non-native ticks, and for the detection of cryptic diversity within species.

Abstract Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), remains the leading cause of death from infection globally, yet the contribution of non-classical T-cell pathways to human immunity remains poorly defined. CD1c-autoreactive T-cells, which recognise self-lipids presented by the antigen-presenting molecule CD1c, are frequent in human blood, but their role during infection is unclear. Here, we investigate how CD1c-expressing antigen-presenting cells (APCs) and Mtb infection shape CD1c-autoreactive T-cell responses using engineered human APC systems, complemented by single-cell transcriptomic profiling to define the ex vivo phenotypic landscape of these T-cells. CD1c is present within human TB granulomas, whereas Mtb down-modulates CD1c expression on infected APCs, consistent with an immune evasion strategy. CD1c-autoreactive T-cells respond more strongly to Mtb-infected CD1c+ APCs than to uninfected cells, exhibiting enhanced activation, cytotoxicity, and diverse cytokine secretion via CD1c-dependent recognition. Under in vitro conditions, these T-cells reduce relative Mtb burden in infected phagocytes. Single-cell RNA-sequencing reveals cytotoxic effector-memory programmes and expression of antimicrobial molecules, providing a mechanistic basis for these responses. Together, these findings define a human CD1c-restricted T-cell response to Mtb-infected APCs and identify autoreactive CD1c-restricted T-cells as a candidate cellular axis for lipid-directed immunity in TB.

Extraction and purification of membrane proteins has for a long time represented a significant challenge. Polymer‐based extraction methods, like those using styrene maleic acid co‐polymers have provided a fertile approach to generate samples that include the local lipid environment surrounding the protein. However, the wide variety of different polymers now available provides a challenge to identify the optimal solution. In this study we develop and demonstrate a novel high‐throughput screening approach for rapid optimization of polymer solubilization agents and chromatography resins for membrane protein purification. Using this approach, we explore whether there are standard conditions that perform well for a range of membrane protein morphologies, sources and functions. These data show that no such standard conditions exist for either polymer solubilization agent or chromatography resin and that some combinations are rarely suitable for membrane protein purifications under these conditions, such as the use of TALON resin at a pH of 7.5 or SMALP300 in the Synthetic Nanodisc Screening Kit MINI kit. Instead, the use of the screening approach developed in this work is the best route to an optimal membrane protein preparation protocol.

On 4 June 2025, the MHRA published ‘Regulatory considerations for therapeutic use of bacteriophages in the UK’. This was in response to recommendations made by the House of Commons Science, Innovation and Technology Committee Inquiry into the ‘The Antimicrobial Potential of Bacteriophages’. The MHRA Regulatory Considerations for phage therapeutic products (PTPs) outlines the relevant regulatory route and requirements to use PTPs as licensed or unlicensed medicines. While this guidance provides the necessary information, it is recognized that regulatory information can be inaccessible to academic and small- to medium-sized enterprise developers who are often unfamiliar with the language, terminology and location of such information. The MHRA, in consultation with the Innovate UK Phage Innovation Network, has therefore developed this interpretation to help PTP developers understand what the guidance is saying, and what evidence is required for regulatory assessment of a marketing authorization application. Examples have been included throughout to provide context and as an aid to understanding.

Internet of Battlespace Things (IoBT) deployments in adversarial environments face critical challenges in network resource management, where rapid environmental changes and dynamic threats render traditional optimization approaches inadequate. The inherently volatile nature of these environments, where network conditions can change within seconds, necessitates algorithms that prioritize speed over perfect optimization to maintain operational effectiveness. This paper presents a novel Graph Neural Network (GNN) based Deep Reinforcement Learning (DRL) framework specifically designed for combinatorial task admission and routing optimization in dynamic IoBT networks. Our approach integrates Graph Attention Networks (GATs) for capturing network topology dependencies, Deep Sets encoders for permutation-invariant task processing, Adaptive Path GNNs for learning path representations, and statistical feature encoders that complement learned embeddings with interpretable routing heuristics. The system formulates network optimization as a Markov Decision Process, enabling real-time decision-making that maximizes task utility while respecting capacity constraints and adapting to topology changes. Comprehensive experimental evaluation across multiple network scenarios demonstrates that our Deep Q-Network (DQN) agent consistently outperforms greedy baselines by 5-58%, achieving near-optimal utility with sub-second inference times compared to Mixed Integer Programming solvers that require hundreds of seconds. The framework shows strong generalization capabilities, with agents trained on smaller task sets effectively scaling to larger workloads, and exhibits superior performance when trained under dynamic conditions rather than static environments.