Professor Johan Rockberg

Building Biology: Directed Evolution, Cell Factories and the Realization of Curative Therapies
 

KTH, Stockholm

Abstract: 

Background: Path to PI
This talk traces a single argument across twenty years of research: the quality of a biological prediction is always determined by the quality of the data that trained it. The career began as the chemical engineer and bioinformatics student at KTH and the University of Sydney, returned back to Stockholm to pursue a PhD, leading to antigen selection algorithms for the Human Protein Atlas (2004). The lack of quality data forced the PhD student into the lab developing high-resolution epitope mapping by bacterial surface display (Nature Methods 2008) and high throughput ultrahigh density peptide arrays (MCP 2012) for generation of antibody epitope training data. The antibody portfolio generated by the Atlas was commercialised through the co-founding of Atlas Antibodies AB (2006), today holding over 22,000 validated antibodies. An observation of synergistic HER2 antibody combinations during epitope mapping, led to co-founding Atlas Therapeutics (Stockholm) and Abclon Inc. Seoul (2010), followed by a directorship at Alligator Bioscience, translating this epitope engineering into clinical assets now represented by Abclon’s AC101/HLX22 in global Phase 3 for HER2-positive gastric and breast cancer (FDA and EC Orphan Drug Designations 2025) and the anti-CD19 CAR-T Nespe-cel, with 94% objective response rate and 68% complete remission in Phase 2. 

Scientific focus: engineering the molecule and the factory 
Current research addresses two interdependent engineering problems: designing drug candidates and engineering the mammalian cell lines that produce them. On the molecule side, combinatorial libraries and automated screening platforms drive bispecific antibody development, including a CD40 agonist forming non-covalent conjugates with patient-specific neoantigen peptides for precision cancer immunotherapy (Nature Communications 2024), the platform underlying StrikePharma AB. Within the GeneNova centre (114 MSEK, Vinnova), AAV capsids are engineered for improved manufacturability and selective receptor-dependent transduction, with results including 10x productivity improvement, elimination of hepatic uptake, and targeted delivery to pancreatic and neural tissue. On the factory side, transcriptomic analysis of HEK293 and CHO cells has identified secretory pathway differences rescuing one third of difficult-to-produce proteins by host cell switching (Metabolic Eng. 2022), reduced antibody aggregation through regulatory elements controlling ribosomal speed (Nucleic acids Res 2022), and improved therapeutic enzyme specific activity 150-fold through secretional machinery tuning (Metabolic Eng. 2024). More recent HEK293 cells are boosted to increase AAV titres 60x with viral helper genes enhancing capsid synthesis. At the bioprocess level, HEK293 perfusion culture sustains ≥80 × 10⁶ cells/mL with EPO productivity of 0.6 g/L/day (Journal of Biotechnology 2020), shear stress transcriptomics inform perfusion device selection (iScience 2020), and high-cell-density rAAV production in stirred-tank bioreactors matches per-cell yields from conventional small-scale conditions (Biotechnology Journal 2025). 

Future perspective: programmable biology for manufacturing
Looking ahead, three developments appear likely to shape how biology is engineered for manufacturing. The first is the systematic curation of biological datasets. Experience from the Human Protein Atlas, GeneNova and AdBIOPRO suggests that well-annotated data at scale enables improved prediction and design. Interpreting high-dimensional, sparse and noisy datasets such as bioprocess data, remains a challenge. Promising tools include the Gromov-Wasserstein framework developed within GeneNova (Ryner M., NeurIPS 2023, co-founder Auweai AB), which aligns unstructured data without a shared coordinate system. First applied to AAV capsid annotation from electron microscopy images, the approach may generalise to patient registry data, selection experiment output, and bioprocess. More interpretable data leads to sharper hypotheses and better-designed experiments. The second is a more systematic exploration of the sequence space of binding proteins. Deep sequencing of the full selection output maps the fitness landscape and preserves diversity, for generation of maturation libraries with targeted diversity to rescue diversity. This 'survival of the less fit' principle (Karlander thesis KTH, 2024 Rockberg group), has shown to deliver more diverse binders, and offers a richer starting point for combining experimental selection with protein language models. A principle and method attractive also for targeted evolutionary cell factory experiments. The third is using large language models to lower the barrier to laboratory automation, sensor incorporation and meaningful down-scale assays. If natural-language descriptions can be reliably converted into executable robot and bioreactor protocols with results feeding back into mechanistic and LLM cellular model updates, the design-build-test-learn will be faster accomplished and allowing for more efficient cell factory and protein engineering.

 

 

´

Short Bio: 

Johan Rockberg is Professor of Antibody Engineering and Directed Evolution at KTH - Royal Institute of Technology in Stockholm, Sweden. He holds a Master of Science in Chemical Engineering from KTH with studies in Bioinformatics and Biotechnology at the University of Sydney, and completed his doctoral degree in 2008 at KTH with a thesis on bioinformatical and combinatorial methods for the generation and characterization of monospecific antibodies.

 

Since 2004, he has been a contributor to the Human Protein Atlas - aimed at cataloguing all human proteins across organs, tissues and disease conditions. Today, his research group works at the intersection of synthetic and systems biology, with a particular focus on protein and cell line engineering for novel curative gene therapies and their scalable bioproduction, including AAV-based vectors, protein vaccine, bispecific therapeutics for cancer and autoimmune diseases. Routed in protein engineering and synthetic biology his lab leverage engineering of both the drug candidate, using in-house combinatorial libraries, deep sequencing and automated downscale screening platforms, as well as the mammalian cell line for their production using directed and evolutionary methods.

Rockberg is Director of GeneNova, a Vinnova-funded innovation milieu in precision health with a budget of 114 MSEK, and is PI in several major national competence centers including CellNova and AdBIOPRO focused on advanced continuous manufacturing. He is co-founder of 7 spinn-out companies — Atlas Antibodies, Atlas Therapeutics, Abclon Inc. (Seoul), Labrock AB, StrikePharma AB, Auweai AB and AareBio AB— spanning antibody reagents, cancer immunotherapy, CAR-T gene therapy and Precision Patient Segmentation through machine learning.

He is a co-founder of the Korean unicorn company Abclon, listed on the KOSDAQ exchange, currently developing CAR-T and antibody therapies, including an antibody combination in global Phase 3 trials for
gastric and breast cancer. He is an elected member of the Royal Swedish Academy of Engineering Sciences and recipient of the 2024 Biotech Builders Award and the 2021 SwedenBIO Life Science Star Award. With over 25,900 citations and publications in journals including Science, Nature Methods, Nature Communications and PNAS, Rockberg represents a combination of scientific depth, institutional leadership and entrepreneurship.