Mammalian Protein Expression: From Cell Lines to Scale-Up

Mammalian protein expression is the process of using mammalian cells as tiny biological factories to produce proteins that are difficult or impossible to make any other way. The approach dominates the biopharmaceutical industry because mammalian cells can fold complex proteins correctly and attach sugar molecules in patterns that the human body recognizes as safe, something bacterial or yeast systems struggle to replicate. Most therapeutic antibodies, many vaccines, and an expanding list of gene therapies rely on mammalian cells grown in carefully controlled tanks to churn out their active ingredients. The technology has matured enormously over the past two decades, but getting high yields of a properly made protein still involves juggling a surprising number of variables, from the choice of cell line and the DNA construct you hand it, all the way through to the temperature of the culture and the nutrients dissolved in the liquid surrounding the cells.

Why Mammalian Cells and Not Something Simpler

Bacteria grow faster and cost less to maintain. Yeast is easy to scale. Insect cells can handle moderately complex proteins. So why bother with mammalian cells at all? The short answer is sugar. Human proteins are commonly decorated with branching chains of sugar molecules, a process called glycosylation, and those sugar patterns affect almost everything about how a drug works in the body. Glycosylation influences a protein’s biological activity, how long it circulates in the bloodstream, and whether the immune system mounts an unwanted response against it.¹ Mammalian cells assemble sugar chains that closely match what human cells produce, making the finished product more likely to behave the way doctors need it to.

Beyond sugars, mammalian cells handle other finishing steps well. Proteins entering a cell’s internal secretory pathway get their signal peptide clipped off, form the right disulfide bonds to hold their three-dimensional shape, and pass through quality-control checkpoints that reject misfolded copies.² Heavily sialylated sugar chains, the kind mammalian cells specialize in, give therapeutic proteins longer half-lives in the blood by protecting them from being scooped up and destroyed by liver receptors.³ Non-human sugar patterns, by contrast, can trigger allergic reactions or cause the drug to be cleared from the bloodstream too quickly. That practical reality is why glycosylation is treated as one of the most important quality attributes for any protein-based drug.⁴

The Two Workhorse Cell Lines

If you are making a therapeutic protein at commercial scale, the odds are very high you are using Chinese hamster ovary (CHO) cells. CHO cells have been the most commonly used mammalian host for large-scale production of therapeutic proteins, and advances in culture technology have pushed titers past 10 grams per liter.⁵ That number matters because higher concentration in the culture fluid means less processing to purify the final drug, which lowers cost. CHO cells also have a long regulatory track record. Agencies like the FDA have approved dozens of CHO-derived products, so the safety data package for a CHO-produced drug is well understood.

Human embryonic kidney 293 (HEK293) cells fill a different niche. They are popular for research-scale work, early-stage drug development, and situations where you need protein fast rather than in massive quantity. A variant called HEK293-6E, engineered to carry a piece of the Epstein-Barr virus that helps maintain the DNA you introduce, can produce more than 450 milligrams per liter of an antibody fusion protein after vector and process optimization.⁶ HEK293 cells are also the go-to host for manufacturing adeno-associated virus (AAV) vectors used in gene therapy, a use case where the cell line’s human origin and ability to support viral replication are genuine advantages.⁷

Designing the DNA Construct

Before a mammalian cell can produce your protein, you need to hand it a piece of DNA, called a vector, carrying the gene of interest along with a set of regulatory elements that tell the cell’s machinery when, where, and how aggressively to read that gene. The promoter is the single biggest lever. The CMV promoter, borrowed from a common virus, is often the first choice because it drives very high expression in some cell types. But CMV is also the most unpredictable: it can be extremely strong in one cell line and surprisingly weak in another, and it is prone to being silenced over time.⁸ In one comparison, stable clones driven by the CMV promoter had only about 60 percent of cells expressing the reporter protein, with wide variation between individual cells, whereas the EF-1α promoter drove homogeneous expression in over 97 percent of cells.⁹

EF-1α and CAGG are now widely regarded as the most consistently strong promoters across diverse mammalian cell types.¹⁰ When long-term stable expression is the goal, the resistance of EF-1α to epigenetic silencing gives it a practical edge over CMV, even if CMV can occasionally deliver a higher peak output in a specific cell line.

Another underappreciated piece of the vector is the signal peptide sequence at the front of the gene. If your protein needs to be secreted out of the cell, the signal peptide is what directs it into the secretory pathway. Different signal peptides perform differently depending on the protein they are attached to and the cell line being used. A modified human serum albumin signal peptide has shown reliable performance across CHO, HEK293, and NS0 cells for driving secretion of diverse proteins.¹¹ Recent work has even applied machine learning to predict how specific signal peptide sequences will perform for a given protein, aiming to take some of the trial-and-error out of the process.¹²

Transient Versus Stable Expression

There are two fundamentally different strategies for getting a mammalian cell to produce your protein, and the choice between them shapes the entire workflow.

Transient expression means you introduce DNA into cells and harvest protein over the next few days before the DNA is lost through cell division. It is fast, flexible, and cheap, making it the standard approach for screening candidates, producing small amounts of protein for early testing, and academic research. Polyethylenimine (PEI) is the most widely used reagent for transient transfection because it is inexpensive and effective at large scale.¹³ Optimized protocols using PEI, the right vector elements, and careful tuning of DNA-to-reagent ratios have pushed transient transfection efficiencies above 76 percent while keeping cell viability over 85 percent.¹⁴ In one landmark study with HEK293-EBNA1 cells and an optimized plasmid, transient transfection produced intracellular recombinant protein at levels rivaling virus-mediated expression, representing up to 20 percent of total cellular protein.¹⁵

Stable expression takes a different path. Here the gene is integrated permanently into the cell’s genome, creating a cell line that can, in theory, produce the protein indefinitely. This is what commercial manufacturing requires, since you need consistency across months or years of production. Building a stable line takes much longer, often months, because you have to select for cells that incorporated the gene, amplify expression, and then screen individual clones to find the best producers.

Selection Systems and Gene Amplification

Once you have integrated your gene into a cell’s genome, you need a way to identify and keep only the cells that took up the DNA. The two dominant selection systems in commercial use both work by tying survival to a metabolic gene placed next to your gene of interest. The DHFR system uses the enzyme dihydrofolate reductase and is probably the more widely used of the two. It can achieve very high levels of gene amplification and expression, but the amplification process is lengthy and may take several months to yield a stable, amplified line. The GS system, based on glutamine synthetase, typically needs only a single round of selection to reach peak expression levels, making it substantially faster.¹⁶ In practice, many large biopharmaceutical companies have licensed and built their manufacturing processes around one system or the other, so the choice is sometimes as much about institutional infrastructure as it is about technical performance.

The Silencing Problem

One of the most persistent headaches in mammalian protein expression is transgene silencing: the gradual or sudden loss of protein production from a cell line that was previously working fine. The cell’s own epigenetic machinery can recognize the foreign DNA and shut it down by modifying the surrounding chromatin, essentially wrapping the gene so tightly that the cell can no longer read it.¹⁷ This problem is especially severe in primary cells and stem cells, but it affects workhorse lines too. The K562 human leukemia line, for instance, is notorious for efficiently reducing expression of stably transfected reporter genes through both epigenetic silencing and physical loss of the inserted DNA.¹⁸

Several strategies can reduce the risk. Choosing a promoter resistant to silencing, like EF-1α over CMV, helps. Flanking the transgene with insulator elements that block the spread of repressive chromatin is another approach. And increasingly, researchers are using targeted integration to place the gene into a pre-characterized “safe harbor” site in the genome where expression is known to be stable, rather than relying on random insertion.

Targeted Integration and Safe Harbors

Random integration, where the gene lands wherever it happens to insert, was the standard for decades. It works, but it means every clone is different: some land in active regions of the genome and express well, others land in quiet zones and barely produce anything, and still others disrupt an important gene and grow abnormally. Screening hundreds or thousands of clones to find the rare good producer is time-consuming and expensive.

Targeted integration using tools like CRISPR/Cas9 changes the equation. By directing the gene to a specific genomic locus known to support stable, high-level expression, you can dramatically reduce clone-to-clone variability. In HEK293 cells, master cell lines have been created by integrating a “landing pad” at the AAVS1 locus, then using recombinase-mediated cassette exchange to swap in different genes of interest quickly.¹⁹ Similar work in CHO-S cells targeting the H11 locus has shown stable expression maintained through more than 40 passages, which is a strong indicator of long-term reliability.²⁰

Culture Conditions That Make or Break Yield

Even with a perfectly designed vector in a well-characterized cell line, the culture environment has an outsized effect on how much protein you actually collect. Two variables stand out: media composition and temperature.

Historically, mammalian cells were grown in media supplemented with fetal bovine serum, a complex cocktail of growth factors, hormones, and nutrients derived from cow blood. Modern bioprocessing has moved aggressively toward chemically defined, serum-free media. The shift eliminates batch-to-batch variability from the serum, removes animal-derived components that could introduce contaminants, and gives scientists precise control over every ingredient.²¹ Adapting cells to serum-free conditions is done gradually, stepping down the serum concentration over multiple passages while monitoring growth and viability.

Temperature manipulation is another powerful tool. Standard mammalian cell culture runs at 37°C, which promotes fast growth. But rapid growth and high protein secretion are often at odds: cells dividing quickly may not devote enough resources to the protein you want. Shifting the temperature down to 30-33°C after an initial growth phase slows division, extends culture viability by delaying programmed cell death, and often increases per-cell productivity.²² In HEK293 cells, a drop to 33°C after transfection yielded roughly 1.5 times more protein.²³ The effect is sensitive to small differences: even a 1-1.5°C temperature change can measurably alter cell culture performance in CHO processes.²⁴

When Cells Struggle With the Protein

Not every protein is easy for a mammalian cell to produce. Complex, heavily glycosylated, or multi-subunit proteins can overwhelm the cell’s internal processing machinery. When misfolded proteins accumulate in the endoplasmic reticulum, the cell activates the unfolded protein response (UPR), a stress signaling pathway that initially tries to fix the problem by ramping up folding capacity. If the overload is not resolved, the UPR switches from adaptive to destructive and triggers programmed cell death.²⁵ ²⁶

For antibody production specifically, the bottleneck is often at folding and assembly rather than at the gene-reading step. Overexpressing individual chaperones or folding enzymes has not reliably solved the problem, likely because these helpers work as a coordinated team rather than as individual actors. Researchers have proposed that globally expanding the entire secretory pathway may be necessary to push per-cell antibody output higher.²⁷ This remains an active area of cell engineering.

Engineering the Cell Itself

Rather than just handing a cell a gene and hoping for the best, a growing trend is to engineer the host cell line to make it a better factory. One commercially important example involves knocking out the FUT8 gene in CHO cells. FUT8 encodes the enzyme responsible for adding a specific sugar (core fucose) to antibodies. Removing that sugar enhances the antibody’s ability to recruit immune cells to destroy their targets, a property called antibody-dependent cellular cytotoxicity. Antibodies produced by FUT8-knockout CHO cells showed increased binding to key immune receptors and stronger cell-killing activity compared to conventionally produced, fucosylated antibodies.²⁸ Several approved cancer antibodies now use this glycoengineering strategy to boost therapeutic potency.

Scaling Up and Intensifying Production

Monoclonal antibodies remain one of the fastest-growing product classes in the pharmaceutical industry, and the pressure to produce more protein more efficiently has driven steady advances in bioreactor technology.²⁹ Traditional fed-batch processes grow cells in a sealed tank, feed them nutrients over one to two weeks, then harvest everything at once. This remains the industry standard, but perfusion culture is gaining ground. In perfusion mode, fresh media is continuously pumped through the bioreactor while spent media (containing the secreted protein) is removed, allowing cells to reach very high densities. Modern high-density perfusion processes have achieved volumetric productivities exceeding 5 grams per liter per day over extended cultivation periods.³⁰

Perfusion is not only for CHO cells. HEK293 cells have been grown in perfusion systems exceeding 100 million cells per milliliter for the production of recombinant erythropoietin, demonstrating that human cell lines can also benefit from process intensification.³¹ For AAV vector manufacturing in the gene therapy space, where production demands can be enormous and the product is a virus particle rather than a soluble protein, scalable single-use bioreactor systems have been developed that allow the same manufacturing principles to be applied in a format meeting current regulatory expectations.³²

Cell-Free Alternatives

Sometimes you do not need, or want, living cells at all. Cell-free protein synthesis systems use extracts derived from mammalian cells, stripped of their genomes but retaining the ribosomes, enzymes, and cofactors needed to read DNA and produce protein in a test tube. These reactions run in hours rather than days and can be automated for high-throughput screening. Systems based on CHO cell extracts have produced roughly 50 micrograms per milliliter of active protein within three hours, while HeLa-derived extracts have been used to rapidly prototype genetic circuits including promoter variants and CRISPR-based repression systems.³³ ³⁴

The yields from cell-free systems are far too low for commercial manufacturing, but that is not the point. Their value lies in speed and flexibility: you can test dozens of construct designs in a single afternoon, identify the best candidates, and then move those into living cells for full-scale production. For membrane proteins, which are notoriously difficult to produce in cells because they tend to misfold or kill the host, cell-free systems offer another advantage, since the extracts contain endogenous membrane fragments that the protein can insert into directly.

Where the Field Is Headed

The convergence of genome editing, machine learning, and advanced bioprocessing is reshaping what mammalian expression platforms can do. Predictive algorithms are starting to take the guesswork out of signal peptide selection, and CRISPR-based tools are making it faster to engineer cell lines with tailored glycosylation profiles or enhanced secretory capacity. Meanwhile, the explosion of gene and cell therapies has created demand for manufacturing capabilities that go well beyond traditional antibody production, pushing mammalian cells into roles as factories for viral vectors, engineered receptors, and entirely new classes of biologics. The underlying biology is the same set of protein-folding and sugar-attaching machinery that made CHO cells the workhorse decades ago, but the toolkit for exploiting that biology is expanding faster than at any point in the field’s history.