Biologics Brief: E3: In Vivo Antibody Discovery with Humanized Transgenic Mice

Published: August 17, 2026 | Category: Insights, Biologics Discovery

---

Humanized transgenic mice have been at the center of therapeutic antibody discovery for over 25 years, and the platforms have come a long way since the first generation. In Episode 3 of The Biologics Brief, Mosaic’s Chief Strategy Officer Tracey Mullen leads a panel conversation on what modern in vivo discovery actually looks like, how to choose the right platform for a difficult target, and what a real CD22 campaign revealed about the Atlas Full Human Diversity Mouse from AbTherx.

Joining Tracey are Dan Rohrer, Chief Technology Officer at AbTherx, along with Mosaic’s Eric Furfine (Chief Scientific Officer), Maria Lo (Head of In Vivo Antibody Discovery), and Stacy Capehart (Director of Data Sciences).

---

The Evolution of Transgenic Mouse Platforms: First, Second, and Third Generation

Transgenic mouse platforms have evolved significantly since their introduction. First-generation models were transformational for therapeutic antibody development, platforms from that era have since contributed to 13 approved therapeutics, still generating over $30 billion annually. But first-generation models had real limitations in immune performance and the breadth of human repertoire they captured.

Second-generation platforms, represented by companies like Regeneron and Kymab, significantly improved immune performance and captured a much broader diversity of human variable regions. Third-generation platforms like the Atlas Mouse build further on that foundation, streamlining the genetic engineering process to a single-step knock-in technology, removing pseudogenes and non-functional variable elements to make more compact and efficient transgenes, and introducing entirely new model types to address modern therapeutic needs, including a bispecific fixed light chain model and a long CDR3 model designed specifically for difficult targets like GPCRs and ion channels.

---

What “Full Human Diversity” Actually Means

The human immunoglobulin gene locus is one of the most genetically diverse regions of the human genome, and that diversity varies across ethnicities, geographies, and pathogen exposure histories. Building a truly representative transgenic model requires surveying published repertoire studies to understand both commonalities and differences across populations.

The Atlas Mouse was designed to capture the most common V genes while excluding rare variants that might be immunogenic in certain populations. As a benchmark: the Thera-SAbDAb database hosted by the University of Oxford catalogs 42 distinct VH genes represented across approved fully human therapeutic antibodies. The Atlas Mouse expresses 40 of those 42, a strong indicator of the breadth of diversity available for therapeutic discovery campaigns. That sequence diversity translates directly to epitopic diversity, which matters enormously when you need coverage across a difficult target landscape.

---

A Frequently Overlooked Consideration: Freedom to Operate

One topic that rarely comes up early enough in discovery discussions is freedom to operate. Transgenic mouse technology is a mature field with a substantial intellectual property landscape. The sequences your campaign generates matter, but so does the method used to discover them. If a candidate or the discovery approach infringes on existing IP, it can completely stall a program regardless of how scientifically strong it is.

The AbTherx team has built IP considerations into their platform philosophy from the start, prioritizing freedom to operate as a first principle rather than an afterthought. For bench scientists evaluating platforms, it is worth asking this question before a campaign begins, not after a lead has been identified.

---

How Mosaic Decides Which Platform Fits a Target

Mosaic operates across multiple discovery modalities, phage display, yeast display, single B-cell screening, hybridoma, and in vivo transgenic approaches. No single platform is the default. The decision starts with the target biology.

For targets that are highly homologous between mouse and human, in vivo methods face tolerance challenges that make in vitro approaches more attractive. For difficult targets, GPCRs, ion channels, multipass transmembrane proteins, in vivo discovery using a fully human transgenic model often delivers better results than in vitro display alone, because the immune system handles the complexity that is difficult to replicate in a tube.

Platform flexibility also enables blending approaches. An immune library generated from a transgenic mouse campaign can be ported into a yeast display platform to access even greater diversity. Single B-cell screening can be applied downstream of in vivo immunization for function-forward triage. The goal is always probability of success for a specific target, not loyalty to a single platform.

---

A Real Campaign: CD22 Discovery with the Atlas Mouse

To make the platform discussion concrete, the panel walks through a CD22 discovery campaign run jointly by AbTherx and Mosaic.

Immunization and hit identification: CD22 was selected as a therapeutically relevant, well-characterized target that allowed for meaningful benchmarking. Following a five-to-six-week subcutaneous immunization protocol, lymph nodes were harvested and CD138-high cells extracted. Using single B-cell screening in partnership with Single Cell Technologies, 6,000 IgG-secreting B cells were identified, of which 541 were confirmed CD22-specific and carried forward as unique sequences.

Sequence triage and expression: From 541 sequences, Mosaic selected approximately 20 candidates representing unique clonotypes and CDR3 length diversity, filtering out sequences with known liabilities, oxidation, deamidation, glycosylation sites, and unpaired cysteines in the CDRs. We expressed those candidates as human IgG1s and purified them for characterization.

Binding affinity: Full kinetic characterization by SPR revealed KD values ranging from micromolar to sub-nanomolar, indicating meaningful diversity across the hit panel.

Epitope binning: Of 17 antibodies taken into epitope binning, eight distinct bins were identified, indicating strong epitopic coverage from a relatively small expressed panel.

Developability profiling: Candidates were assessed across a suite of developability assays including hydrophobic interaction chromatography, AC-SINS for self-interaction, BVP ELISA for polyspecificity, and thermal stability by TM-TAG. The overall developability profile was favorable across the panel, not just one or two outlier candidates.

AI-assisted hit picking: Two independent AI and machine learning methods were applied to the sequence panel to select diverse candidates using different prioritization criteria. While there was some overlap, most selections were distinct. Regardless of which selection approach was used, Stacy’s bioinformatic triage or either of the AI methods, good antibodies were consistently recovered. The quality of what the Atlas Mouse generates means the hit-picking strategy matters less than it typically does with other platform types.

---

Managing Large Hit Panels Without Losing Good Candidates

One practical challenge of high-quality in vivo campaigns is that they generate more hits than any team can express and validate. The panel discusses several strategies for working through large panels without leaving valuable candidates behind.

The first step is in silico triage, filtering for known sequence liabilities before committing to expression. The second is clonotypic family analysis: surveying representative members of each family, then drilling deeper within families where interesting properties emerge. The third layer is having a clear target product profile defined before the campaign begins. Knowing what characteristics matter most, affinity threshold, functional readout, cross-reactivity, format compatibility, allows teams to build the right screens upfront and manage panel size rationally rather than reactively.

---

Where AI Fits Alongside In Vivo Discovery

The humoral immune response has been evolving for approximately 500 million years. It simultaneously optimizes for affinity, expression, self-reactivity, and a range of other parameters in a way that is genuinely difficult to replicate computationally today. That is the enduring argument for in vivo discovery.

At the same time, AI tools are already adding real value in hit picking, developability prediction, clonotype prioritization, and sequence diversity analysis. The most productive framing is not in vivo versus AI, it is in vivo plus AI, with each contributing where it is strongest. For targets that are highly conserved between mouse and human, where tolerance makes in vivo approaches challenging, AI-generated libraries may become an increasingly important complement or alternative. The landscape is evolving, and the right answer will keep shifting.

Reach out to the Mosaic team if you are evaluating an in vivo campaign or want to talk through the right platform strategy for a difficult target.

---

View and listen to more episodes on our podcast: The Biologics Brief.

---