Choosing an Expression System: E. coli, Yeast, Insect, or Mammalian

You have a sequence in hand and a deadline. The next decision — how to choose an expression system for your recombinant protein — sets your timeline, your budget, and whether the protein you get back is the protein you actually wanted. Pick the wrong host and you spend three weeks discovering your eukaryotic protein is sitting in inclusion bodies; pick the right host and you have purified material by Friday.

This post walks through the four decision factors that drive expression system choice — post-translational modifications, solubility, yield, and infrastructure — and gives you a path through E. coli, yeast, insect (Sf9), HEK293, and CHO. It assumes you already know what your protein is and what you plan to do with it; if you are picking between a transient transfection and a stable cell line, that is a separate downstream decision.

Factor 1: Does the protein need post-translational modifications?

This is the question that drives the choice most strongly, and the one beginners skip. Post-translational modifications (PTMs) — disulfide bonds, glycosylation, phosphorylation, proteolytic processing — happen during and after translation, and different hosts perform them differently. The PTM requirement collapses an open-ended choice into a short list.

• No PTMs needed (cytosolic enzymes, small soluble domains, peptides for crystallography): E. coli is almost always the right call. Cheapest, fastest, highest yields.
• Disulfide bonds only: E. coli can handle a small number of disulfides in specialized strains (SHuffle, Origami) that have an oxidizing cytoplasm, or by secreting the protein to the periplasm. Beyond two or three disulfides, the failure rate climbs.
• Glycosylation, complex folding, multiple disulfides: you need a eukaryote. Yeast handles N-linked glycosylation but the glycan structure is hyper-mannosylated and not human-like. Insect cells (Sf9) produce simpler paucimannose glycans. HEK293 and CHO produce mammalian-style complex glycans.
• Human-like glycosylation for a therapeutic: CHO is the industry default. HEK293 is fine for research-grade material but the glycan profile differs subtly between the two lines, and regulators expect CHO for biologics.

Factor 2: How much protein do you actually need?

Yield ranges differ by more than an order of magnitude between systems. Map your scale before you map your host:

• Crystallography, biochemistry, biophysics: typically 10–100 mg of pure protein per batch. E. coli at 1–100 mg/L culture is the easy answer for anything that folds in bacteria.
• Cell-based assays, characterization runs: typically 1–10 mg. HEK293 transient transfection (~5–50 mg/L for a well-behaved construct, often less) is workable. Adherent or suspension — suspension scales better.
• Antibody discovery, large-scale biochemistry: 50–500 mg. CHO stable lines (typically 1–5 g/L for optimized lines, less for first-pass research-grade) or Expi293 transient suspension.
• Gram-scale for clinical or process development: CHO stable, possibly with cell line development.

The numbers above are typical ranges, not guarantees. A poorly-behaved construct in any system will under-perform; a high-expressing construct in HEK can rival a moderate CHO line. Use the ranges to size the decision, not to predict your specific outcome.

Factor 3: What infrastructure does your lab actually have?

An expression system is not just a plasmid — it is the entire setup needed to grow, harvest, and purify from that host. Software dropdowns offer all five systems with equal weight; your lab does not.

• E. coli: a shaker, a centrifuge, IPTG, and basic media. Most labs already have everything. Setup cost is functionally zero.
• Yeast (S. cerevisiae, P. pastoris): similar to E. coli in equipment, but P. pastoris methanol-fed protocols need fume-hood access and the cultures take longer to reach density. Less common than it used to be — many labs have lost the practical knowledge.
• Insect (Sf9 with baculovirus): tissue culture hood, 27°C incubator with shaker, ability to generate and titer baculovirus stocks. The baculovirus generation step adds 2–3 weeks to the first run; subsequent runs from the same stock are fast.
• HEK293 transient: tissue culture hood, CO₂ incubator, polyethylenimine (PEI) or commercial transfection reagent. Reasonable to stand up in a lab that already does mammalian culture.
• CHO stable: requires cell line development capacity — weeks of selection, single-cell cloning, and clonal screening. Most academic labs outsource this to a CRO.

Path A: Pick E. coli when the protein is bacterial-friendly

Default to E. coli when all of the following are true:

• The protein has no glycosylation requirement (or you can deglycosylate downstream).
• It has zero or a small number of disulfide bonds you can handle with SHuffle or periplasmic secretion.
• The folded form is known to be soluble (or you have a fusion partner like SUMO or MBP that you trust — see purification tags and fusion partners).
• You need protein within a week.

Backbone choices for E. coli are well-trodden: pET (T7-based, IPTG-inducible) for high yield, pBAD (arabinose-inducible) when T7 over-expression is toxic, pTrc when you want a less leaky alternative to T7. Designing a plasmid for bacterial expression covers the backbone selection in more detail.

Path B: Pick mammalian (HEK293) when you need a research-grade human-like protein quickly

Default to HEK293 transient when:

• You need mammalian PTMs (glycosylation, processing of secreted proteins, proper disulfide patterns).
• You only need milligram-scale material.
• You have access to mammalian culture infrastructure already.
• You will use the material for research, not therapeutic development.

Standard backbones: pcDNA3 family (CMV promoter, neomycin resistance for selection), Lake-pancreatic-like pTT or pTT5 for high-yield transient suspension expression with EBNA-1 episomal maintenance, or pCAG when you want a stronger constitutive promoter than CMV in certain cell types.

Path C: Pick CHO when you need therapeutic-quality material or large amounts

CHO is the default once you are producing biologics, building stable cell lines, or generating gram-scale material. The cell line development overhead (4–12 weeks for a workable pool, 3–6 months for a clonal line) only makes sense at scale. Most labs that need CHO outsource cell line development and then run their own production fermentations.

Path D: Pick insect (Sf9) when HEK does not give enough and CHO is too heavy

Sf9 with baculovirus is the in-between choice that gets neglected. Use it for:

• Complex eukaryotic proteins (multi-domain, multiple disulfides) at higher yields than HEK transient.
• Membrane proteins — insect cells are often the best practical host for membrane protein structural work.
• Cases where you have time for the baculovirus generation step (2–3 weeks first time, 1 week after).

The downside is the baculovirus step and the non-human glycosylation pattern. The upside is real — many proteins that are intractable in HEK express well in Sf9.

Path E: Pick yeast for secreted proteins and budget-constrained eukaryotic expression

P. pastoris (now Komagataella phaffii) is the workhorse yeast for secreted protein production. It handles disulfide bonds, secretes correctly-folded protein into the medium (simpler purification), and reaches high cell densities. The glycosylation is hyper-mannose and unlike mammalian patterns, which matters for therapeutics but not for many research uses.

S. cerevisiae sees less use for expression these days — mostly for proteins that need specific yeast biology or for two-hybrid screens. If you're choosing yeast, default to P. pastoris unless you have a specific reason to use S. cerevisiae.

Summary: which path for which job

How the expression system choice flows back into your construct

The expression system choice propagates through the rest of the plasmid design. Your promoter has to match the host (T7 for E. coli, CMV or CAG for mammalian, AOX1 for P. pastoris induction, polyhedrin for baculovirus). Your selection marker has to be functional in your host (ampicillin or kanamycin for E. coli, puromycin or G418 for mammalian, methotrexate for CHO amplification). Your codon usage should match: codon optimization for E. coli and human vs mouse codon tables for mammalian expression cover the host-specific considerations.

Picking the host is the first decision because nearly every other design decision depends on it. Make it deliberately, with the protein's PTM needs and your lab's infrastructure both on the table — not at the bottom of a default dropdown.

If you want a design tool that propagates the expression system choice through promoter suggestions, codon tables, and design health checks automatically, PlasmidStudio uses your host selection to drive the rest of the design.