BioByte 079: advances in brain delivery of gene therapies, RNA delivery heats up cold tumours, AF3 guide, polyextremophile engineering launch, and an LLM for therapeutics
Welcome to Decoding Bio’s BioByte: each week our writing collective highlight notable news—from the latest scientific papers to the latest funding rounds—and everything in between. All in one place.
What we read
A big step forward for brain delivery of gene therapies [Eric Minikel, CureFFI.org]
Delivery to targeted organs remains one of the biggest challenge across therapeutics, and particularly in the realm of gene therapy (think AAV). The brain is particularly challenging given its distinct subpopulations of neurons and that many disease processes affect just certain parts of the brain. Reporting back from ASGCT in Baltimore, Eric describes advances in the field as it relates to AAV-mediated brain delivery.
Previous efforts in this field had focused on making libraries of AAV9 (as example) with peptide sequences randomly mutated before injecting them in mice and using sequencing to understand which mutants got into the brain. Although results were initially promising, several of these ended up binding mouse-specific receptors. Many groups started panning for better AAVs in monkeys, but these mutants were only slightly better than unengineered AAV9; additionally some engineered AAV would work in only one species of monkey and not another, which rightfully raised concerns about whether this approach would prove fruitful in clinical applications.
Ben Deverman’s lab took a new angle to the approach - start with the receptor of interest, purify it, attach it to beads, and pan for viruses that bind said beads. This approach seems to be working - they announced development of a vector that binds the human transferrin receptor (TfR), a well-known target in CNS-mediated drug delivery. They discovered an AAV that bound to TfR and penetrates the brain in mice with human TfR almost 50x more than AAV9.
This study – and evolution of framework – is critical, as engineering and optimization of AAV is one of the biggest challenges facing the field. However, it is important to note there are other difficulties facing the field: immunogenicity, manufacturing, and more play a key role in bringing AAV to patients. That said, the field has made significant strides in development, and brain-specific gene therapy is much closer to becoming a reality.
A single infusion of engineered long-lived and multifunctional T cells confers durable remission of asthma in mice [Nature Immunology, May 2024]
It will be clear to many hayfever sufferers that the seasonal allergy period is increasing around the world (an average increase of 20 days). A key contributing factor is global warming, increasing pollen production in plants earlier in the season. This is a particular problem for asthma sufferers, an analysis of such patients from Maryland saw an increase by 17% in hospitalisations due to an earlier-onset pollen season. With an incidence of already 300 million patients per year, novel treatments are of high importance.
Current treatment options for asthma have varied effectiveness depending on the immune complement of the patient, however many still do not promote long term remission, can take years of therapy to be truly effective or have adverse events.
Jin et al working at Tsinghua University, designed CAR-T cells that kill IL-5R eosinophils (a key immune cell involved in the allergic response) and express an IL-4 cytokine mutein that could induce longer-term remission from asthma. The CAR-T mechanism mimics the action of benralizumab, a cytotoxic antibody that kills eosinophils in asthma sufferers, and dupilumab, one of the biggest allergy blockbusters.
The authors infused their so-called 5TIF4 cells into immunocompetent allergy models of mice and this led to lung inflammation suppression as well as asthmatic symptom relief. This treatment mechanism is promising for allergy sufferers as it would only entail one single dose for long term relief. Whilst promising, this study serves as a proof-of-concept for an interesting new MOA in allergy, and significant work is still needed to translate these findings to humans.
RNA delivery heats up cold tumours [M. Teresa Villanueva, Nature Reviews Drug Discovery, 2024]
Immunotherapy is still a challenge when treating cold tumours. In this paper, the authors designed lipid particle aggregates (LPAs) containing mRNA antigens that activate the specific RIG-I immune response that activates immunity in the periphery within the tumour microenvironment.
The authors designed an onion-like delivery system, where mRNA acts as a molecular bridge between liposomes, forming a series of self-assembling concentric membrane layers.
In vivo, the RNA-LPA aggregates loaded with whole-tumour-derived mRNA resulted in extended survival when injected in mice with sarcoma and melanoma-derived tumours. Similar results were seen in mice carrying diffuse midline glioma and gliomas overexpressing pp65 antigen, when injected with antigen specific RNA-LPA encoding H3K27M and PP65.
The authors demonstrated that the aggregates activate the pattern recognition receptors response, recruiting PBMCs and TILs in the TME.
The safety was tested in a first in human trial of RNA-LPAs encoding PP65 and whole tumor antigens in four patients with glioblastoma. All patients showed a rapid rise in pro-inflammatory cytokines/chemokines and showed increased immune cell infiltration in the biopsy, as well as absence of viable tumor cells.
An Opinionated AlphaFold3 Field Guide [Simon Barnett, Dimension Research, June 2024]
Simon Barnett put together a fantastic overview of the inner workings on AlphaFold3, as well as some takes on the state (and future) of the field. Some highlights:
Proprietary data generation will be key to unlocking the next significant improvement in ML for molecules. The era of vast improvements in structure generation using open-access data repositories is coming to an end, and companies capable of paying for immense data generation will drive the development of state-of-the-art models in the future.
The lack of code and constrained usage server make AF3's claims hard to replicate or verify. There is a vigorous debate surrounding the limited access to AF3's code and the requirement to use a web portal for inference, but Isomorphic has been clear about their intentions to become a vertical biotech company with a pipeline, and therefore going closed source isn’t entirely unexpected.
The removal of equivariance from the model (if the input molecule is rotated or translated, the predicted structure should be rotated in the same way) has agitated the debate around the best path forward for capturing physical invariances. While some argue that removing equivariance is confusing, AF3 still incorporates equivariance through data augmentation techniques.
We are still a long way from replacing experimental methods with ML in the life sciences. Despite the impressive progress made by AF3, the field is still early. ML models have limitations in handling certain types of proteins and novel small molecule scaffolds. Like with everything else in biology and drug discovery, more data will drive the next monumental improvement in the field.
Tx-LLM: A Large Language Model for Therapeutics [Zambrano Chaves et al., arXiv, 2024]
The paper from DeepMind introduces Tx-LLM, a large language model fine-tuned from PaLM-2, designed to handle a diverse array of tasks across the therapeutic development pipeline. Tx-LLM is trained using a comprehensive collection of 709 datasets targeting 66 different tasks, encompassing stages from drug discovery to clinical trials. The model utilizes a unified set of weights to process a variety of entities, such as small molecules, proteins, nucleic acids, and cell lines, combined with free-text instructions. This enables Tx-LLM to predict a broad range of properties with competitive or superior performance compared to state-of-the-art (SOTA) models in many tasks.
Tx-LLM achieves SOTA or near-SOTA performance on 43 out of 66 tasks. Notably, it excels in tasks involving molecular SMILES representations and textual data, outperforming existing models on average for these tasks. Evidence of positive transfer between tasks involving different drug types was observed, indicating the model's ability to generalize across various datasets.
Tx-LLM represents a significant advancement towards integrating AI into the therapeutic development process. It shows potential as an end-to-end tool capable of assisting in various stages, from target discovery to clinical trial predictions. The generalist nature of Tx-LLM allows it to address multiple tasks with a single model, making it a versatile and powerful tool for researchers.
Folding the human proteome using BioNeMo: A fused dataset of structural models for machine learning purposes [Hetmann et al., Nature Data Descriptors, June 2024]
Researchers from Innophore and NVIDIA have introduced a groundbreaking dataset of predicted protein structures for 42,042 human proteins, including splicing variants. Derived from the UniProt reference proteome, this dataset was generated using state-of-the-art modeling tools within NVIDIA’s BioNeMo platform, combined with homology modeling from Innophore’s CavitomiX platform. The dataset, which includes both unedited and refined structures, is poised to become a valuable resource for various research applications, particularly in drug design and protein interaction studies.
The comprehensive dataset encompasses 99.72% of the human protein sequences from the UniProt reference proteome, totaling 122,907 novel protein structures and was developed utilizing advanced AI-guided tools, such as AlphaFold 2, OpenFold, and ESMFold, to ensure high-quality, full-sequence modeling. Homology modeling was also employed for proteins lacking suitable templates, enhancing the dataset's structural diversity. The refined dataset, which includes energy-minimized structures and predicted ligands, is tailored to meet diverse research needs, from structure-based drug design to the prediction of protein function and interactions.
Notable Deals
SR One raises an additional $200M - after closing a $600M fund in only March 2023, the venture capital firm, once the corporate venture arm of GSK, has cemented its growth opportunities fund enabling it double down on critical investments.
Pioneer Labs launches to engineer microbes for Mars - the start-up is aiming to design more robust ‘polyextremophile’ microbes (that can survive in environments with many extreme properties). Pioneer’s aim is two-fold: short-term biomanufacturing and long term terraforming.
Seres, indebted and in need of cash, agrees to sell microbiome pill to Nestle - the microbiome start-up will sell full rights of Vowst, raising money to fund its earlier experimental therapies.
Ochre Bio announces multi-year data license agreement with GSK - the $37.5M value partnership will give GSK access to Ochre Bio’s proprietary platforms to generate bespoke datasets.
GSK acquires tiny RNA startup Elsie for $50M, boosting its grand plans for developing RNA drugs
In case you missed it
Strand Therapeutics announces first patient dosed with programmable mRNA therapy [May 2024]
Strand Therapeutics has dosed the first patient in its Phase 1 trial for STX-001, a programmable mRNA therapy aimed at treating solid tumors. This innovative therapy uses synthetic self-replicating mRNA to express the IL-12 cytokine directly in the tumor microenvironment, potentially enhancing immunotherapy effectiveness. Initial animal studies showed promising results, including cancer cell death and immune cell activation. Strand will present the trial details at the ASCO Annual Meeting, highlighting the therapy's design and early findings. This marks a significant step in the validation of the platform biotech model and is an advancement in drugging the tumor microenvironment with precision immunotherapy.
Events
AI x Bio Summit 2024 @ NYSE: Sign up here for the reception. Space is very limited!
Field Trip
Did we miss anything? Would you like to contribute to Decoding Bio by writing a guest post? Drop us a note here or chat with us on Twitter: @ameekapadia @ketanyerneni @morgancheatham @pablolubroth @patricksmalone