A Multispecies, Modality-Agnostic Scalable In Vivo Mosaic Screening Platform for Therapeutic Target Discovery

This paper presents a scalable, modality-agnostic AAV-based in vivo screening platform coupled with a human disease signature-matching analysis framework to identify and validate therapeutic targets across diverse species and organ systems, successfully demonstrating its efficacy in murine pulmonary fibrosis and equine osteoarthritis models.

The Gist

Imagine you are trying to fix a broken, complex machine, like a vintage car with a thousand tiny parts. For years, scientists have tried to figure out which part is broken by taking the engine apart and testing each piece on a workbench (in a petri dish). But the problem is, a car engine doesn't work the same way when it's sitting on a table as it does when it's running inside a moving car. The vibrations, the heat, and the way the parts talk to each other are all different.

This paper introduces a revolutionary new way to test medicines: The "Mosiac Screening" Platform.

Here is how it works, broken down into simple concepts:

1. The Problem: The "Workbench" vs. The "Real World"

Traditionally, scientists test new drugs on cells in a lab dish. It's like testing a new spark plug on a workbench. It tells you if the spark plug works, but it doesn't tell you if it will fix the car while it's driving down a bumpy road.

• The Issue: Diseases like lung fibrosis (scarring of the lungs) or arthritis happen in a messy, crowded, living environment. Cells in a dish are lonely; cells in a body are part of a bustling city.
• The Old Way: Scientists had to test one drug on one mouse at a time. It was slow, expensive, and often gave misleading results.

2. The Solution: The "Mosiac" Approach

The authors created a system that lets them test hundreds of different "fixes" simultaneously inside a living animal.

Think of the animal's organ (like a lung or a joint) as a giant, dark room.

• The "Mosiac": Instead of turning on one light to see one corner, they use a special virus (AAV) that acts like a glow-in-the-dark paintbrush.
• The Paint: They mix hundreds of different "paints" (each paint represents a different gene therapy or drug target) into a bucket.
• The Application: They spray this bucket into the animal. Because the paint is diluted, only a few cells in the whole room get painted.
• The Result: The room now looks like a mosaic. One cell is glowing blue (testing Gene A), the next is glowing red (testing Gene B), and another is glowing green (testing Gene C). All of them are living in the exact same real-world environment, feeling the same pain and stress of the disease.

3. The "Magic Decoder Ring" (The Analysis)

Once they have this mosaic of glowing cells, they need to know which color fixed the problem.

• The Old Way: They would look at the genes and say, "Oh, Gene A changed 50 other genes." That's a lot of data, but it's hard to know if that's good or bad.
• The New Way: The team built a "Disease Scorecard." Imagine a checklist of symptoms that a human patient has (like "stiff joints" or "scarred lungs").
  • They take the data from the glowing cells and ask: "Did this cell's behavior match the checklist of a healthy human, or a sick human?"
  • If a cell treated with "Gene A" suddenly looks like a healthy human cell, that's a winner!
  • If it looks like a sick human cell, that's a loser.

4. Testing on Real Animals (The "Horse" Factor)

Most science is done on mice. But mice are small, and their joints are tiny. They don't get arthritis the same way humans do.

• The Breakthrough: The researchers took this "mosaic" system and tried it on horses.
• Why Horses? Horses get arthritis naturally, just like humans. Their joints are big, they carry weight, and their bodies are more similar to ours.
• The Result: They injected the "mosaic paint" into a horse's knee. They found that certain genes (like SOCS1) acted like a "fire extinguisher" for inflammation, while others (like IL13) were tricky—they put out the fire but accidentally started a new one elsewhere. This kind of nuance is impossible to see in a mouse.

5. The Final Check: The "Human Taste Test"

Before a drug goes to humans, you need to be sure it works on us.

• The team took the best candidates from their mouse and horse tests and tried them on human lung slices and joint tissues growing in a lab.
• The Verdict: The "mosaic" predictions were spot on. The genes that looked promising in the horse knee actually repaired human cartilage. The genes that looked good in the mouse lung actually reduced human lung scarring.

The Big Picture

This paper is like upgrading from testing car parts one by one on a workbench to putting hundreds of different parts into a real car, driving it around a track, and using a computer to instantly tell you which combination makes the car run smoothest.

Why does this matter?

• Speed: It finds good drug targets much faster.
• Accuracy: It tests drugs in the real "messy" environment of a living body, not a clean petri dish.
• Safety: It can spot side effects early (like the IL13 example in horses) before they reach human patients.
• Scale: It can test hundreds of ideas at once, rather than one by one.

In short, they built a high-tech, multi-species, "mosaic" testing ground that bridges the gap between lab mice and real human patients, helping us find cures for tough diseases like fibrosis and arthritis much faster and more safely.

Technical Summary

1. The Problem

The translation of novel therapeutic targets from discovery to clinical application is bottlenecked by the lack of scalable validation platforms within predictive preclinical models.

• Limitations of Current Models: Traditional high-throughput CRISPR screens are largely restricted to 2D cell cultures, which fail to recapitulate the complex in vivo microenvironment (matrix mechanics, vascular gradients, immune-stromal reciprocity).
• Constraints of Existing In Vivo Screens: Current in vivo pooled perturbation screens (e.g., Perturb-seq) face three major hurdles:
  1. Species Limitation: Most rely on Cas9-expressing transgenic mice, excluding genetically diverse or large animal models that better mimic human disease physiology.
  2. Modality Limitation: They predominantly test only Knockout (KO) strategies, limiting the evaluation of Gain-of-Function (GOF) or Knockdown (KD) therapeutic strategies.
  3. Analytical Limitation: Analysis is often restricted to standard pathway enrichment, lacking a patient-centric framework to rank targets against specific human disease molecular signatures.

2. Methodology

The authors developed a Modality-Agnostic Mosaic Screening Platform that integrates AAV-based delivery, single-cell RNA sequencing (scRNA-seq), and a curated molecular feature analysis framework.

A. Platform Architecture

• Delivery System: Uses Adeno-Associated Virus (AAV) vectors to deliver perturbations directly into diseased tissues.
  • Serotype Screening: Systematically tested 13–16 AAV serotypes to identify optimal tropism for specific cell types in fibrotic lungs (AAV6 for epithelium, AAV-DJ9 for fibroblasts) and equine joints (AAV5 for synovium).
  • Library Design:
    • Knockout (KO): Dual-gRNA CRISPR constructs (4 gRNAs per target) to ensure efficient gene deletion.
    • Gain-of-Function (GOF): Overexpression (OE) of coding sequences (CDS) with unique 5' UTR barcodes.
    • Knockdown (KD): Synthetic miRNA hairpins targeting specific transcripts.
• Experimental Workflow:
  1. Induction: Disease models are established (e.g., bleomycin-induced pulmonary fibrosis in mice; spontaneous osteoarthritis in horses).
  2. Perturbation: Pooled AAV libraries are administered in vivo.
  3. Isolation: Transduced cells (GFP/RFP+) are sorted via FACS, excluding immune and endothelial cells.
  4. Sequencing: 10x Genomics scRNA-seq with simultaneous capture of gene expression and perturbation barcodes.
  5. Assignment: A probabilistic beta-binomial model assigns perturbations to single cells based on barcode UMI counts, distinguishing single-assigned cells from background noise.

B. Analytical Framework: Molecular Features

To move beyond standard differential expression, the authors curated a set of Molecular Features:

• Source: Derived from human patient scRNA-seq datasets (e.g., IPF, OA) and clinical correlations (e.g., Forced Vital Capacity).
• Composition: Non-redundant gene sets representing specific pathophysiological states (e.g., "Lung Pro-Fibrosis Hub," "Lipofibroblast Signature," "Cartilage Matrix Degradation").
• Scoring: Single-cell activity is quantified using AUCell scoring. Perturbation effects are ranked by regressing feature scores against intervention groups, allowing for the comparison of disparate targets on a unified, disease-relevant scale.

C. Orthogonal Validation

Hits identified in vivo were validated in human primary ex vivo models:

• Lung: Precision Cut Lung Slices (PCLS) treated with siRNA to measure soluble collagen and fibrotic gene expression.
• Joint: Human OA chondrocyte/synovial co-cultures treated with AAV-OE to measure Glycosaminoglycan (GAG) restoration.

3. Key Contributions

1. First Large Animal In Vivo Screen: Demonstrated the first application of high-throughput transcriptomic screening in a large animal model (horses with spontaneous OA), overcoming anatomical and physiological barriers of murine models.
2. Modality Agnosticism: Successfully integrated KO, GOF, and KD modalities within a single platform, expanding the therapeutic scope beyond simple gene deletion.
3. Patient-Centric Analysis: Developed a framework that maps in vivo perturbation data directly to human disease signatures, enabling quantitative ranking of therapeutic potential and safety risks.
4. Scalability: Proved the platform can scale from small screens (15 targets) to large libraries (~300 constructs targeting 96 genes) while maintaining statistical power with as few as 200–250 cells per target.

4. Key Results

A. Pulmonary Fibrosis (Mouse Model)

• Validation: Confirmed in vivo editing efficiency in fibrotic tissue using AAV-DJ9.
• KO Screen:
  • Jak1 KO: Induced a profound anti-inflammatory shift (suppressed interferon signaling) but increased fibrotic ECM organization, decoupling inflammation from fibrosis.
  • Tgfbr2 KO: Drastically reduced fibrotic programs (ECM synthesis) but triggered a compensatory interferon response.
• GOF Screen:
  • Klf15 Overexpression: Identified as a metabolic regulator driving branched-chain amino acid catabolism and fatty acid oxidation, promoting a shift from a fibrotic to a homeostatic/alveolar repair state.
  • Pparg Overexpression: Suppressed pro-fibrotic signaling in transitional AT2 cells.
• Feature Analysis: The molecular feature framework successfully detected subtle effects (e.g., Pik3ca KO inducing a lipofibroblast signature) that standard DEG analysis missed.

B. Osteoarthritis (Equine Model)

• KD Screen: Identified metabolic and inflammatory drivers (e.g., COX6C, S100A4, LGALS3) in synovial fibroblasts.
• GOF Screen:
  • IL13 Overexpression: Revealed pleiotropic liabilities. While it reduced cartilage degradation markers, it simultaneously upregulated inflammatory mediators, ER stress, and apoptosis, highlighting a complex risk profile.
  • SOCS Family (SOCS1, SOCS3): Showed targeted anti-inflammatory efficacy (suppressing JAK/STAT and NLRP3) without the broad deleterious stress signals seen with IL13.

C. Large-Scale Scaling & Validation

• Scale: A screen of 96 genes (300 constructs) in mouse lungs identified 33% of targets with significant anti-fibrotic or immunomodulatory effects.
• Human Validation:
  • PCLS: Knockdown of TGFBR1, TGFBR2, and PIK3CA significantly reduced soluble collagen, correlating with screen predictions. JAK1 knockdown did not, validating the screen's distinction between structural and immunomodulatory drivers.
  • OA Co-culture: AAV-OE of IL13 and SOCS1 significantly restored GAG content in human cartilage, confirming the screen's ability to predict functional physiological outcomes.

5. Significance

This work establishes a powerful paradigm for therapeutic target discovery by uniting highly multiplexed in vivo functional genomics with human disease signatures.

• Translational Bridge: By utilizing large animal models (horses) and validating hits in human ex vivo tissues, the platform significantly reduces the "valley of death" between preclinical discovery and clinical translation.
• Safety & Efficacy: The ability to simultaneously identify mechanistic efficacy and potential pleiotropic liabilities (as seen with IL13) allows for early risk stratification of therapeutic candidates.
• Future Impact: The platform is adaptable to diverse tissues and species, offering a scalable solution to dissect the multicellular molecular circuits driving complex chronic diseases like fibrosis and arthritis.