Forty Years On, Less Than 2% of the Protein Data Bank Is Membrane Proteins

In 1985, Hartmut Michel and colleagues solved the first membrane protein structure: the photosynthetic reaction centre from Rhodopseudomonas viridis. Three years later they won the Nobel Prize for it. It felt like a beginning.

Four decades on, structural biology has changed almost beyond recognition. Cryo-EM has redefined what is possible at high resolution. AlphaFold has predicted hundreds of millions of structures. The Protein Data Bank now holds over 220,000 experimentally determined entries. And yet fewer than 2% of those entries are membrane proteins. Restrict to high-resolution structures and the number creeps up to roughly 4%. That is it. After forty years.

A rounding error in the structural record

Read those numbers in the context of biology. Membrane proteins make up between 20% and 30% of every proteome ever sequenced. They account for more than 40% of FDA-approved drug targets: the receptors, channels, transporters and enzymes that mediate how cells sense, signal and respond to almost every drug we make. A field that represents a third of biology and the majority of the modern pharmacopoeia is structurally represented by what is, statistically speaking, a rounding error.

So what is going on?

The hydrophobic problem

Soluble proteins behave themselves. You can express them, purify them and crystallise them using protocols a graduate student can learn in their first year. Their surfaces love water, they tolerate buffers, and they form crystals reasonably well.

Membrane proteins are different. By definition they live in the lipid bilayer, an environment that fights aqueous crystallography at every step. Pull them out of the membrane and they unfold, aggregate or precipitate. Leave them in, and you cannot get them into a crystallographically useful environment. They are biology's hydrophobic refugees, and structural biology is a profoundly hydrophilic field.

There is a specific number that tells the story. Soluble protein crystals contain between 27% and 65% solvent. Membrane protein crystals contain between 65% and 80% solvent, mostly because of the detergent and amphiphile molecules required to keep the target stable. More solvent means weaker diffraction, lower resolution, and a much harder time extracting useful structural data from what you do manage to crystallise. This is the wall the field has been hammering at for forty years.

Three platforms reshaping what is possible

The encouraging news is that we are not stuck. The last decade has produced three powerful platforms for studying membrane proteins outside of their native lipid environment, each with distinct strengths and trade-offs worth understanding.

Detergent micelles: the workhorse

For most of structural biology's history, detergent micelles have been the default. You solubilise the membrane with a carefully chosen detergent (DDM, LMNG, OG, FOS-Choline-12, the catalogue is encyclopedic), purify the resulting protein-detergent complex, and try to crystallise it.

The strength is breadth: almost every membrane protein structure solved in the 20th century used some form of detergent micelle approach. The weakness is fidelity. Detergents are not lipids. Many proteins, especially GPCRs and complex transporters, show reduced stability, altered conformational landscapes, or outright loss of function when their native lipid environment is stripped away. Detergent micelles are still indispensable. They are not always the right choice.

Use detergent micelles when: your target is compatible with a hydrophobic shell and you need high-resolution X-ray data, particularly via lipidic cubic phase (LCP) crystallisation, or when you need a purification workhorse for a bacterial membrane protein or non-GPCR target where lipid fidelity is not the limiting experimental variable.

Nanodiscs: putting the bilayer back

Nanodiscs were a real breakthrough. Developed in Stephen Sligar's lab and refined over the last two decades, they wrap a small disc of native lipid bilayer in two copies of a membrane scaffold protein. The result is a stable, water-soluble particle that places your target in something resembling its real environment.

For functional studies this changed the game. Receptor pharmacology, channel kinetics and transporter selectivity become tractable when the protein sits in a real lipid context rather than a detergent shell. Nanodiscs also work beautifully with cryo-EM, which is increasingly displacing crystallography as the structural method of choice for membrane proteins. The trade-off is size: standard scaffold proteins make discs roughly 9 to 13 nm in diameter, so very large complexes, or proteins that need to oligomerise extensively, can be challenging.

Expression system note: BioHippo's catalog nanodisc proteins are produced using an E. coli cell-free expression system, with MSP scaffold protein and DMPC phospholipid. If your target requires mammalian expression, for example, a heavily glycosylated GPCR or a protein whose folding depends on mammalian co-chaperones, request a custom quote and we can discuss the right approach for your specific construct.

Virus-Like Particles (VLPs)

The newest and least-discussed platform is VLP-based expression. Here the target protein is presented on the surface of an engineered viral capsid, giving you a highly concentrated, near-native presentation of the protein in a real membrane context.

VLPs are particularly compelling for vaccine antigens, surface receptors that need to maintain their native oligomeric state, and cases where the protein simply refuses to fold properly in any other context. The platform also opens doors to immunisation and structural work running in parallel, which is unusual.

Platform at a glance: a three-way comparison

The three platforms are complementary, not ranked. The decision turns on your target class, your downstream method, and the biological question you are asking.

Catalog counts as of June 2026. Each collection page carries a fuller on-page comparison table.

What this means for drug discovery

The economic case is brutal in its simplicity. Membrane proteins are more than 40% of drug targets. If we are structurally blind to 96% of them, we are designing drugs without structural information for almost half the modern pharmacopoeia.

AlphaFold helps, enormously. But predicted structures are not experimental ones, they are starting points for design, not endpoints. For lead optimisation, allosteric site discovery, drug-binding confirmation, conformational selection studies, and especially for novel scaffold work, experimental structures of the actual drug-target complex remain irreplaceable.

The three platforms above are not competitors. They are complementary tools. The right question is not "which is best?" but "which is right for this specific target, at this specific stage of the project, asking this specific question?" A GPCR for early functional work? Probably nanodiscs. A bacterial transporter for high-resolution X-ray? Detergent micelles in lipidic cubic phase. A viral envelope protein for vaccine development? VLPs. A multi-pass eukaryotic protein with regulatory partners? Possibly a combination, deployed in sequence.

The honest summary

Forty years after the first membrane protein structure, we have solved tens of thousands more and barely scratched the surface of what the proteome contains. The challenge is not that we lack good tools. It is that membrane protein structural biology requires more thought, more platform diversity, and more iteration than soluble protein work. For the next decade, the labs that move fastest will be the ones that match the platform to the target, rather than defaulting to whatever they did last time.

Our Standard. Your Reproducibility.