The work was carried out by Na-Yeong Heo, Mi-Jung Kang, and Yeon-Gu Kim at the Biotherapeutics Translational Research Center of the Korea Research Institute of Bioscience and Biotechnology, with Heo and Kim also affiliated with the KRIBB School of Biotechnology at the Korea University of Science and Technology. Their central insight addresses a long-standing frustration in industrial biomanufacturing: codon optimization has become a routine strategy for boosting recombinant protein expression in mammalian systems, yet improvements in codon-usage-based computational metrics do not always translate into enhanced production performance. In other words, an algorithm that scores a gene as well adapted to a host cell’s codon preferences may still deliver a disappointing yield at the bioreactor scale.

To understand why this matters, it helps to consider the biology of the genetic code. The code is degenerate: most amino acids are specified by multiple three-letter codons, and different organisms favor different synonyms. Since the early days of recombinant DNA technology, scientists have exploited this redundancy, rewriting therapeutic genes to match the codon preferences of the production host in the hope of speeding translation. But codon usage influences far more than translation speed. Synonymous changes can affect mRNA secondary structure, mRNA stability and abundance, the kinetics of translation initiation, co-translational folding of the nascent protein, and even the recruitment of regulatory RNA-binding proteins. Because these effects are intertwined and context-dependent, no single computational metric captures the full picture. That is precisely the gap the Korean team set out to close with an empirical, cell-based approach.

The production host at the heart of the study is the Chinese hamster ovary cell, or CHO cell, the workhorse of the biopharmaceutical industry. CHO cells are used to manufacture a large share of the world’s therapeutic proteins because they can perform complex post-translational modifications—particularly N-linked glycosylation—that are essential for the safety and efficacy of many biologics. Etanercept is a prime example: it is a fusion protein linking the extracellular domain of the human p75 tumor necrosis factor receptor to the Fc portion of human immunoglobulin G1, and it requires proper N-terminal processing and N-linked glycosylation to function correctly. Any engineering strategy that raises yield but compromises these quality attributes would be unacceptable, which is why the researchers built quality assessment directly into their screening pipeline.

Methodologically, the study focused on the 5′ coding region of the etanercept gene, the stretch of DNA immediately downstream of the start codon. This region is known to exert outsized influence on translation initiation and early elongation, making it a strategic target for synonymous redesign. The team constructed a library of synonymous codon variants concentrated in this 5′ region and introduced the variants into Flp-In CHO cells, a specialized line engineered for Flp/FRT-mediated site-specific integration. This system allows an incoming gene to recombine at a defined genomic locus, ensuring that every variant in the library is inserted at the same chromosomal address. The importance of this design choice cannot be overstated: in conventional random integration, each cell clone carries the transgene at a different genomic position, and the resulting position effects—local chromatin environment, enhancer proximity, epigenetic silencing—often swamp any true sequence-dependent differences in expression. By pinning every variant to the same locus, the researchers eliminated this confounding variable and made a fair, apples-to-apples comparison of synonymous sequences possible.

With the library in place, the next challenge was finding the rare high performers among many mediocre ones. The researchers employed a fluorescence-activated cell sorting-based antibody detection assay to enrich variants that produced more etanercept on their cell surfaces or in their secreted product. FACS allows millions of cells to be sorted individually at high speed based on a fluorescent signal proportional to protein production, effectively turning the entire cell population into a searchable library. After enrichment, the top candidates were evaluated under stable gene expression conditions, the gold standard for assessing whether a production advantage is durable rather than a transient artifact of transfection or selection. Three variants emerged from this process as clear winners: ETN-Syn-4, ETN-Syn-14, and ETN-Syn-25.

The productivity gains were striking. Compared with cells carrying the native etanercept sequence, the three top variants yielded 2.08-fold, 2.75-fold, and 2.08-fold higher specific protein productivity, a parameter denoted qp that measures the amount of product generated per cell over time. A nearly threefold improvement in specific productivity is a meaningful advance in an industry where even incremental gains in titer or cell-specific productivity can translate into substantial cost savings across thousands of liters of fermentation capacity. Perhaps the most intriguing finding, however, was what did not change dramatically: the mRNA abundance of the three top-producing variants showed only modest differences from the native sequence. This suggests that the productivity boost was not primarily driven by increased transcription or mRNA stability, pointing instead toward translational mechanisms—more efficient initiation, smoother ribosome transit, or improved co-translational folding—that allow the same amount of messenger RNA to yield more finished protein.

The selected variants also displayed distinct sequence characteristics, differing from one another and from the native gene in their codon-usage profiles. This heterogeneity among the winners is itself informative: it implies there is no single optimal codon recipe, but rather multiple synonymous solutions that converge on high productivity through different combinations of sequence features. Such diversity complicates purely rational design but validates the screening approach, which can discover effective variants that no current algorithm would necessarily rank at the top. The findings reinforce a growing consensus in the field that synonymous codon choice is a rich, underexploited design dimension for bioprocess engineering, one whose effects are best mapped empirically in the actual production host.

Critical to any real-world application is the question of product quality, and here the results were reassuring. The researchers compared etanercept derived from the synonymous variants with native-sequence etanercept across several key quality attributes: N-terminal processing, N-linked glycosylation, size heterogeneity, and binding activity. In each case, the variant-derived protein was comparable to the native product. This means the codon changes boosted quantity without altering the molecular identity or functional integrity of the drug—a crucial distinction, because regulatory agencies demand thorough characterization of glycosylation patterns and other critical quality attributes for every biologic, and even subtle shifts can trigger lengthy revalidation. A silent mutation strategy that preserves product quality while raising yield is therefore far more attractive than approaches that trade quality for quantity.

The broader implications extend well beyond etanercept. The screening framework established in this study—combining site-specific integration, focused synonymous libraries, FACS-based enrichment, and rigorous quality control—offers a generalizable template for optimizing any therapeutic protein produced in CHO cells. As the biopharmaceutical industry faces relentless pressure to reduce manufacturing costs and expand access to biologics, tools that squeeze more product from each cell without new cell lines or process overhauls are increasingly valuable. The work, supported by the National Research Foundation of Korea through the KRIBB Research Initiative Program and by Korea’s Ministry of Trade, Industry and Energy, signals a shift in how codon optimization is practiced: away from purely in silico design and toward empirical, cell-based selection that lets the biology itself reveal which silent spellings speak loudest. For a field built on the assumption that the genetic code’s redundancy is just that—redundant—this study is a reminder that synonymous DNA is anything but silent when it comes to making medicines.

Subject of Research: Empirical screening of synonymous codon variants to enhance therapeutic protein production in Chinese hamster ovary cells

Article Title: Synonymous codon variants identified by in vitro screening enhance etanercept production in CHO cells

Article References: Heo, N.-Y., Kang, M.-J., & Kim, Y.-G. (2026). Synonymous codon variants identified by in vitro screening enhance etanercept production in CHO cells. Applied Microbiology and Biotechnology. https://doi.org/10.1007/s00253-026-14057-9

Image Credits: AI Generated

DOI: 10.1007/s00253-026-14057-9

Keywords: codon optimization, synonymous codon variants, CHO cells, etanercept, Fc-fusion protein, recombinant protein expression, Flp-In site-specific integration, FACS screening, specific productivity, N-linked glycosylation, bioprocess engineering, therapeutic glycoprotein