The science of delivering cures straight to your cells | Eric Kelsic

Key Concepts

• Gene Therapy: A therapeutic approach that aims to treat diseases by correcting the underlying genetic cause through the introduction of DNA molecules into cells.
• Gene Delivery: The process of getting the therapeutic DNA payload into the target cells effectively and safely.
• Capsids: Protein shells derived from viruses, specifically adeno-associated viruses (AAVs), used as delivery vehicles for gene therapy.
• Directed Evolution: A technique used to engineer proteins by mimicking natural evolution in a laboratory setting to achieve specific functional improvements.
• DNA Multiplexing Technology: A method for creating and testing very large libraries of engineered capsids by programming their DNA sequences.
• Non-human Primates (NHPs): Animals, particularly cynomolgus monkeys, used in preclinical studies to assess the safety and efficacy of gene therapies due to their physiological similarities to humans.
• AI-Guided Design: The application of machine learning and artificial intelligence to analyze vast datasets from experiments, identify patterns, and predict improved capsid designs.
• In Silico Testing: Computational simulation and testing of potential capsid sequences using AI models before conducting physical experiments.
• Wright's Law: An economic principle stating that with every doubling of production, there is a percentage decrease in cost, applicable to the potential cost reduction in gene therapy.
• Genetic Agency: The concept of individuals having greater control and choice over their genetic makeup and health through advanced genetic technologies.

Gene Therapy: The Grand Challenge of Delivery

The fundamental human desire for a life filled with good experiences, time with loved ones, and the ability to love is often hindered by genetic illnesses or the natural breakdown of the body. Our bodies are essentially genetic machines, and for many diseases, the root cause lies in mutations within the genome. Gene therapy, a vision that has existed for over 50 years, offers the potential for a one-time treatment that can address these root causes by delivering a DNA molecule that can persist for the lifetime of a cell. However, the current reality is that the genome we are born with is largely the genome we die with, as access to this molecular level is out of reach for most. The primary obstacle has been the challenge of effectively delivering the genetic payload into the target cells.

Dyno Therapeutics and the Quest for Mainstream Gene Therapy

Eric Kelsic, CEO and co-founder of Dyno Therapeutics, has dedicated the past decade to solving the "grand challenge of gene delivery" to make gene therapy a mainstream medicine. This involves ensuring that therapeutic payloads can reach every organ or cell where they can benefit patient health. Dyno's approach focuses on engineering protein shells derived from viruses, specifically adeno-associated viruses (AAVs).

Adeno-Associated Viruses (AAVs) as Delivery Vehicles

AAVs are small viruses that naturally do not cause disease. Their small size makes them attractive for gene therapy as they can access many parts of the body. While our understanding of AAVs' natural functions is still evolving, Dyno's focus is on adapting them as therapeutic technologies by engineering their capsid sequences to improve their delivery capabilities.

Natural Capsid Functionality: Capsids are naturally evolved to perform complex tasks:

1. Circulation: Travel through the body via the bloodstream.
2. Cellular Entry: Find and enter target cells.
3. Cytoplasmic Release: Release their contents into the cell's cytoplasm.
4. Nuclear Translocation: Navigate through the nuclear pore into the nucleus.
5. Genome Release: Break open and release the therapeutic DNA.
6. Expression: The released DNA then expresses the therapeutic gene.

When therapeutic genes are expressed in the nucleus, they can potentially treat cells for a patient's entire lifetime, offering a curative one-time treatment.

Limitations of Natural Capsids and the Evolution of Engineering

Despite their natural capabilities, natural capsids are often not efficient enough for most therapeutic purposes. For the past 28 years, protein engineers have attempted to modify capsids using directed evolution. This process involves randomly altering capsid sequences to create vast libraries (millions to billions of variants) in the hope of finding improved versions. However, this "needle in a haystack" approach is fraught with difficulties:

• Essential Functions: Capsids have many critical functions, and even a single mutation can break an essential one, rendering the variant useless.
• High Failure Rate: Approximately 80% of single amino acid changes to the capsid sequence disrupt its assembly and genome packaging capabilities. This means four out of five random mutations are detrimental.
• Compounding Mutations: Achieving improved function often requires multiple, even hundreds, of changes. With each detrimental mutation, the viability of the capsid decreases, making it extremely difficult to find a library of beneficial mutations.
• Scalability and Stability: Beyond functional improvements, engineered capsids must be producible and purifiable at scale, stable at various temperatures (including body temperature), and capable of reaching specific target cells.

The Challenge of Targeted Delivery and Off-Target Effects

A significant unmet need in gene therapy is delivery to the brain, which is hindered by the blood-brain barrier. Current AAVs can only reach a very small percentage (around 0.1%) of neurons in the brain, which is insufficient for treating many neurological diseases. Furthermore, most AAVs tend to deliver their payload to the liver, and high doses can lead to toxicity. Therefore, improving the efficiency and specificity of delivery to target cells is crucial. Decades of using traditional directed evolution have yielded insufficient improvements or variants optimized for all necessary functions.

Dyno's Innovative Approach: DNA Multiplexing and AI-Guided Design

Dyno Therapeutics is leveraging a new wave of technologies to revolutionize protein engineering. Their approach begins with DNA multiplexing technology, enabling the creation of very large, programmed capsid libraries.

Process Overview:

1. Design: Researchers design capsid sequences on a computer, targeting specific receptors or exploring promising regions of sequence space.
2. Synthesis and Cloning: The designed DNA sequences are synthesized and cloned into the capsid structure.
3. In Vivo Screening (Non-human Primates): These large libraries are injected into non-human primates (NHPs), particularly cynomolgus monkeys, to assess their in vivo performance. The use of NHPs is critical because their physiology is similar to humans, and their lives are precious, necessitating maximum information extraction from each experiment.
4. High-Throughput Data Collection: Each animal experiment can evaluate hundreds of thousands, or even millions, of different capsid sequences simultaneously.
5. Tissue Analysis: After the experiment, tissues from all organs are collected. Nucleic acids (DNA and RNA) are extracted and purified.
6. Sequence Reconstruction: Through DNA sequencing, researchers can identify which capsid sequences were present in the library and their relative abundance in different tissues.
7. Functional Inference: An increased abundance of a specific capsid sequence in a particular organ might indicate functional improvement for delivery to that organ. Conversely, a decreased abundance could suggest a problem or broken function.
8. Data Analysis and AI Integration: Dyno has accumulated petabytes of data from DNA sequencing. Initially, human observation of patterns in the vast string of amino acid sequences was challenging. However, the realization that these patterns are ideal for machine learning led to the development of AI-guided design.

AI-Guided Design Workflow:

• Automated Analysis: AI models automate the analysis of experimental data, identifying nuanced patterns that humans might miss.
• In Silico Prediction: Trained AI models can be queried billions of times to predict the performance of potential capsid sequences in silico (on a computer). This significantly expands the scope of testing beyond what is feasible in physical experiments.
• Model Ensemble: Dyno utilizes tens or hundreds of different AI models, each with unique insights, to compare their predictions and select the most promising candidates for further experimental validation.
• Iterative Cycle: This process forms an iterative cycle:
  • Design DNA libraries.
  • Measure their properties through experiments.
  • Build AI models to analyze and understand these properties.
  • Query the models to identify promising regions of sequence space.
  • Design new libraries based on these insights.
  • Repeat.

Human-AI Collaboration and Future Vision

While AI is a powerful tool, human judgment remains critical. Dyno aims to increase human leverage by automating routine tasks to AI agents or scripts, allowing humans to focus on higher-level judgment. The goal is to collaborate more effectively with AI, enabling instructions for automated analysis and design, and receiving expected answers. The urgency of patient needs drives the desire for rapid results and the need for a "human in the loop" to quickly identify and address any issues.

The Transformative Potential of Gene Therapy: Zolgensma as a Case Study

The power of gene therapy lies in its ability to introduce a DNA molecule that can persist for the cell's lifetime. For non-dividing cells like neurons, this can translate to a lifelong cure. Dyno's excitement stems from this potential.

Zolgensma Example:

• Disease: Spinal Muscular Atrophy (SMA), a genetic disease that was the leading cause of death in children.
• Cause: A non-functional SMN1 gene.
• Outcome without Gene Therapy: Fatal at a young age (typically 2-3 years old).
• Zolgensma's Impact: A one-time gene therapy treatment, administered early in life (first few weeks), can restore the function of the SMN1 gene, effectively curing the disease.

Addressing the Long Tail of Genetic Diseases and Cost Reduction

Despite breakthroughs like Zolgensma, only a handful of gene therapies are FDA-approved, while thousands of genetic diseases remain without effective treatments. The current focus is on diseases with larger patient populations and commercial viability. However, there's a vast "long tail" of rare and ultra-rare diseases affecting very few individuals.

Challenges and Goals:

• High Cost: Current gene therapies can cost millions of dollars per dose.
• Dyno's Goal: To reduce the cost of delivery to zero or near-zero.
• Enabling Competition: Increasing the number of gene therapies will foster competition and drive down costs.
• Economies of Scale: Dyno draws parallels to industries like semiconductors and solar power, where dramatic cost reductions and scale economies have been achieved through increased production. This phenomenon, known as Wright's Law, suggests that with every doubling of production, costs decrease by a certain percentage.
• Future Cost Projections: Dyno anticipates gene therapy costs could drop from hundreds of thousands of dollars to $10,000 or even $1,000 per treatment. This would enable non-profit efforts to fund treatments for rare diseases.

The Role of AI in Personalized and Scalable Therapies

Beyond delivery, AI can play a crucial role in designing therapies. The vision is for AI to design personalized therapies for individual patients, customized to their specific genome and health goals. This could be done "on-demand," with AI even outlining the development, production, testing, and safety assurance processes. This massively scalable approach could address the long tail of diseases, making treatments economically affordable for patients whose genetic causes are understood but currently untreatable.

Future of Gene Therapy: Genetic Agency and Upgradable Treatments

Looking ahead, the ability to "reset" or remove prior gene therapies could become important. While not an urgent priority now, this capability would empower patients with genetic agency, allowing them to upgrade their therapies as more effective treatments emerge over time. This would make gene therapy a more routine decision, akin to choosing clothing.

Shifting Perceptions: The development of these technologies suggests a future where genetics is less about an immutable part of oneself and more about a choice. Individuals will have greater control over their genetic makeup and what they become, enabling them to live their best possible lives. This represents a profound shift from the current perception of one's genome being an unchangeable aspect of identity.