A study published in the Proceedings of the National Academy of Sciences has uncovered new clues to why some therapeutic proteins are so difficult to manufacture. The work—led by Nathan Lewis, PhD, professor of biochemistry and molecular biology at the University of Georgia, and Johan Rockberg, PhD, professor in antibody technology and directed evolution at the KTH Royal Institute of Technology in Stockholm, Sweden—offers a sweeping analysis of how Chinese hamster ovary (CHO) cells vary in their ability to produce human proteins.
CHO cells are the industry standard for producing monoclonal antibodies and other biologic drugs. Yet many human proteins remain stubbornly low-yielding, frustrating drug developers. “We wanted to understand why,” Lewis says. “Was it something inherent in the proteins themselves or something about the cells?”
The team profiled the expression of more than 1,000 human secreted proteins in CHO cells. What they found surprised them. “Transgene mRNA abundance explained less than one percent of the variability in protein yield,” Lewis says. “In contrast, protein physicochemical features, like size and cysteine content, explained about 15%. But the biggest differences were in the cells’ own gene-expression programs.”
Cells that performed poorly tended to activate stress pathways, such as endoplasmic-reticulum–associated degradation, while high-performing cells ramped up lipid metabolism and showed greater resistance to oxidative stress. “Those host-cell transcriptional differences turned out to be the most informative,” Lewis explains. “It’s not just about your recombinant gene—it’s about the cell you put it in.”
The project has deep roots in Lewis’s career. “This goes way back, about 15 years,” he says. “I was consulting for a startup and working on CHO-cell metabolism because, frankly, I had diapers to pay for.” That experience evolved into a decade-long collaboration involving genome sequencing, metabolic modeling, and a global effort to map out CHO biology. “We wanted a parts list and wiring diagram for these cells,” he says. “Now we’re finally starting to see how it all fits together.”
The researchers used machine learning to examine how hundreds of protein traits correlated with productivity. One of the simplest findings—longer proteins express poorly—was also one of the most telling. “It makes sense,” Lewis says. “Bigger proteins have more chances to misfold.”
Beyond size, the study identified amino-acid compositions and post-translational modifications that predict expression success. Those insights, Lewis points out, could guide both protein design and cell-line engineering. “If you’re in pharma trying to develop a cell line, this gives you a roadmap,” he says. “You can ask, is my protein inherently hard to express, or can I tweak the host cell to make it work better?”
Future directions include exploring how metabolic “switches” in CHO cells—particularly those involving lipid metabolism—might be tuned to enhance production. “Right now, media optimization is done more blindly,” Lewis says. “But if we understand the gene networks involved, we can start pulling the right levers.”
For an industry that relies on CHO cells to make billions of dollars’ worth of lifesaving drugs, those levers could make all the difference.