Evaluating codon optimization strategies for mammalian glycoprotein production with an open-source expression vector

This study demonstrates that while native codon sequences are generally sufficient for robust mammalian glycoprotein expression, a skewed codon strategy utilizing the most abundant codons can occasionally enhance yields, all evaluated using a newly developed open-source Golden Gate compatible vector called pTipi.

The Gist

Imagine you are a chef trying to bake the perfect cake (a protein) for a very specific customer (a mammalian cell). For decades, the baking industry has believed that to get the best cake, you need to rewrite the recipe using only the most popular, high-quality ingredients available in the kitchen. This is called "codon optimization." The idea is that by swapping out "rare" ingredients for "common" ones, the baker (the cell) will work faster and produce more cake.

However, a new study by Yang and colleagues suggests that this "rewrite the recipe" strategy might be unnecessary, and in some cases, even harmful, when baking for mammalian cells.

Here is the story of their experiment, broken down into simple concepts:

1. The New Kitchen Tool: The "pTipi" Vector

First, the researchers needed a reliable kitchen. They built a new, streamlined expression vector called pTipi. Think of this as a super-efficient, no-frills baking pan. They stripped away all the extra plastic and unnecessary parts of standard pans, leaving only the essential elements needed to bake a cake in a mammalian cell.

• The Result: They tested this pan by baking 247 different "chimeric" cakes (antibodies). It worked beautifully, producing high yields of protein. This proved their new kitchen tool was solid and ready for the main experiment.

2. The Great "Recipe" Bake-Off

The researchers wanted to settle a debate: Does rewriting the recipe actually help?
They chose 18 specific proteins from the Wnt pathway (a complex signaling system in the body, like a group of messengers passing notes). They created five different versions of the recipe for each protein:

1. The Native Recipe: The original recipe found in nature (Human DNA).
2. The "Skewed" Recipe: A recipe that uses only the single most popular ingredient for every step (e.g., if a cake needs flour, it uses only the most common brand of flour, ignoring all others).
3. The "Harmonized" Recipe: A recipe that tries to balance the ingredients to match the rhythm of the original, just spreading them out more evenly.
4. The "LinearDesign" Recipe: A recipe designed by a super-computer to make the instructions as stable and smooth as possible (focusing on the structure of the paper the recipe is written on).
5. The "Company" Recipes: Recipes generated by big commercial gene-synthesis companies (the industry standard).

3. The Small-Scale Taste Test

They baked small batches of all these recipes in a lab dish (Expi293F cells).

• The Surprise: The "Native" recipe (the original) performed just as well as, or better than, the fancy rewritten ones.
• The Failure: The "LinearDesign" recipe, which was supposed to be the most stable, actually produced the worst results. It was like trying to bake a cake with instructions written in a language that was too perfect to be understood by the baker.
• The "Skewed" Surprise: The recipe that used only the most common ingredients didn't break the cake. In fact, for some proteins, it made the cake slightly bigger! This suggests that cells can handle a repetitive diet of common ingredients just fine.

4. The Big Batch Test

To be sure, they baked three of the proteins on a large scale (100 mL cultures).

• The Verdict: The "LinearDesign" recipe failed again, producing almost no protein. The "Native" recipe remained the most consistent champion. The "Skewed" recipe was a hit-or-miss: sometimes it was the best, sometimes the worst, but it never completely failed.
• The Conclusion: You don't need to rewrite the recipe. The original "Native" code works perfectly fine. In fact, trying to "optimize" it by focusing too much on RNA stability (the LinearDesign method) actually hurts production.

5. The Golden Gate Solution

Because the researchers realized that sometimes you do want to swap in a specific ingredient (maybe to remove a restriction site or for a specific experiment), they upgraded their "pTipi" pan to be Golden Gate compatible.

• The Analogy: Imagine a baking pan with a special magnetic latch. You can snap different recipe cards (gene inserts) in and out instantly without any glue or mess. This allows scientists to easily test different "recipes" (codon strategies) without rebuilding the whole kitchen every time.

The Big Takeaway

For a long time, biotech companies have spent a lot of money and time "optimizing" genes, believing that the natural code is flawed. This paper says: Stop overthinking it.

If you are baking a cake for a mammalian cell, stick to the original recipe. The cell knows exactly how to read the natural instructions. While you can force a recipe to use only the most common ingredients (Skewing) and it might work, trying to "perfect" the stability of the instructions often backfires.

In short: Nature's code is usually the best code. Don't try to fix what isn't broken, but if you do need to change it, use a flexible tool (like their new Golden Gate vector) to test your changes easily.

Technical Summary

1. Problem Statement

The production of human glycoproteins in mammalian cell lines (e.g., HEK293, CHO) is critical for developing biologics and tool compounds. While codon optimization is a standard industry practice intended to boost protein yields by addressing "rare" codons, RNA stability, and GC content, its efficacy in homologous mammalian systems (expressing human genes in human cells) remains poorly defined.

• The Gap: Most systematic codon studies focus on E. coli or heterologous expression (e.g., human genes in bacteria). There is a lack of empirical data comparing different optimization algorithms specifically for mammalian glycoproteins.
• The Assumption: It is widely assumed that optimizing codons to match host tRNA abundance or maximizing RNA stability will universally increase yield. However, some protein families (like Wnt pathway components) are tightly regulated, suggesting that over-optimization might disrupt natural feedback mechanisms or fail to provide benefits.

2. Methodology

The authors employed a systematic "bake-off" approach to compare five distinct codon usage strategies across 18 human and murine Wnt pathway glycoproteins.

A. Vector Development: pTipi

To ensure a controlled environment, the team designed pTipi2.1, a minimal, open-source mammalian expression vector (Addgene #229144).

• Design: Based on pUC18, it strips away non-essential "stuffer" sequences, retaining only essential elements: CMV promoter/enhancer, a synthetic chimeric intron, Kozak sequence, human Azurocidin signal peptide, a Multiple Cloning Site (MCS), WPRE, and HSV poly-A signal.
• Validation: The vector was validated by producing various epitope-tag antibodies (Anti-V5, Anti-His, etc.) and standardized reagent antibodies, achieving yields of 30–160 mg/L, comparable to commercial vectors.
• Golden Gate Upgrade: A second version, pTipi2.2, was engineered for scarless Golden Gate cloning (using BsaI, BsmBI, and PaqCI) to facilitate high-throughput swapping of synthetic gene inserts.

B. Codon Optimization Strategies

The study compared five specific schemes for 18 targets (9 human glycoproteins and their 9 murine orthologs):

1. Native: Unmodified cDNA sequences from RefSeq.
2. Skewed: Uses the single most abundant codon for every amino acid (maximizing CAI to 1.0).
3. Harmonized: Redistributes rare codons to match the frequency of the native sequence using the Charming algorithm (maintaining native CAI).
4. LinearDesign: Optimizes for minimum free energy (MFE) of RNA structure to maximize stability, while maintaining a CAI > 0.8.
5. Commercial Algorithms: Proprietary optimizations from two major synthetic gene vendors (Company A and Company D).

C. Experimental Workflow

1. Small-Scale Screen: 18 constructs were expressed in Expi293F cells in 384-well plates. Yields were quantified via ELISA using biotinylated tags and streptavidin detection.
2. Large-Scale Validation: Top candidates (ROR1, ROR2, LRP6) were produced in 100 mL cultures, purified via Ni-NTA affinity and Size Exclusion Chromatography (SEC), and analyzed for purity and monodispersity.

3. Key Results

A. Performance of Optimization Strategies

• Native Codons are Sufficient: The Native sequence consistently provided robust expression, often ranking as the best or equal-best performer. It was never detrimental.
• RNA Stability is Detrimental: The LinearDesign strategy, which prioritizes RNA stability (lowest MFE), resulted in the lowest protein yields across the board. In large-scale production, some LinearDesign constructs (e.g., LRP6) produced only trace amounts of protein.
• Skewed Codons are Variable but Viable: The Skewed strategy (using only the most abundant codons) did not harm expression. It occasionally outperformed native sequences (e.g., for ROR1) but performed poorly for others (e.g., ROR2). This suggests that extreme codon bias is not inherently toxic but is target-dependent.
• Harmonization is Unpredictable: The Harmonized approach showed high variability, performing best for some targets and worst for others.
• Commercial Algorithms: Vendor-specific algorithms (Company A and D) clustered with the experimental schemes but did not offer a universal advantage over native sequences.

B. Structural and Quality Analysis

• Proteins produced via Native, Skewed, and Harmonized methods were of high purity, monodisperse (confirmed by Dynamic Light Scattering), and stable after freeze/thaw cycles.
• The LinearDesign failure was not due to aggregation but likely due to transcriptional/translational suppression or rapid degradation, despite the theoretical stability of the mRNA.

C. Codon Usage Trends

• Analysis of the Wnt pathway genes revealed that native sequences already possess a codon distribution typical of the human genome.
• The "Skewed" method reduced the 61 possible codons to just 20 (the most abundant), yet this did not cause tRNA depletion issues in the tested timeframe, contradicting the fear that extreme bias leads to ribosome stalling in mammalian cells.

4. Key Contributions

1. Empirical Evidence Against Universal Optimization: The study provides the first systematic data showing that for homologous mammalian expression, native codon usage is the most reliable strategy. It challenges the industry dogma that codon optimization is always necessary or beneficial.
2. Open-Source Tool (pTipi): The release of pTipi2.1 and pTipi2.2 on Addgene provides the community with a minimal, patent-free, and highly efficient vector for glycoprotein production.
3. Golden Gate Compatibility: The integration of Golden Gate cloning into pTipi2.2 enables rapid, scarless assembly of synthetic genes, facilitating future high-throughput screening of codon strategies.
4. Identification of Risks: The study highlights that strategies focusing solely on RNA stability (LinearDesign) can be counterproductive in mammalian systems, potentially leading to failed expression campaigns.

5. Significance and Conclusion

The paper concludes that for the production of human glycoproteins in mammalian cells (specifically HEK293), codon optimization is generally unnecessary and can be risky.

• Recommendation: Researchers should default to native coding sequences for robust, reproducible expression.
• Nuance: While "Skewed" codon usage can occasionally boost yields for specific targets, it is not a general rule. Strategies that aggressively alter RNA structure (LinearDesign) should be avoided.
• Impact: This work simplifies the workflow for biologics development, reducing the cost and time associated with gene synthesis and optimization, and encourages the use of open-source, minimal vectors to maximize transparency and reproducibility in protein engineering.