Protein Nanoparticles for mRNA Delivery: A Lipid-Free Alternative

This explainer is for anyone who's wondered whether mRNA medicines have to rely on lipid nanoparticles. It walks through a Nature Biotechnology paper describing a protein-based carrier, what the experiments showed, and where the evidence stops. Throughout, we separate what was demonstrated, what the authors infer, and what still needs clinical confirmation.

Who did the study and where was it published?

The paper is Self-assembling protein nanoparticles for cytosolic delivery of nucleic acids and proteins by Eweje, Ibrahim, Shajii, Walsh, Ahmad, Alrefai, Miyasato, Davis, Ham, Li, Roehrl, Haller, Liu, Chen and Chaikof. It appeared online in Nature Biotechnology on 15 May 2025 and in the April 2026 print issue (volume 44, pages 602–615).

The lead group was Feyisayo Eweje, Jiaxuan Chen and Elliot L. Chaikof in the Department of Surgery at Beth Israel Deaconess Medical Center, Harvard Medical School. Collaborators came from MIT's Health Science and Technology division, the Wyss Institute and Harvard's Department of Chemistry and Chemical Biology. That includes David R. Liu of HHMI and the Broad Institute, who developed base editing and prime editing. The NIH-funded Somatic Cell Genome Editing consortium supported the work.

Why look for an alternative to lipid nanoparticles?

mRNA vaccines, the siRNA drug patisiran and in vivo CRISPR editing all rely on the same delivery technology: lipid nanoparticles (LNPs). LNPs are the dominant carrier for nucleic acid medicines. They're also at the center of many questions about mRNA vaccines, including biodistribution, organ accumulation, persistence and immune responses to the lipid components.

This team built a carrier that contains no lipids at all. It's a recombinant protein, made in E. coli, that self-assembles in water into nanoparticles and carries cargo into cells. Its lead version delivered mRNA, DNA, CRISPR–Cas9, base editors and siRNAs about as well as, or better than, the commercial lipid-based transfection reagents used as benchmarks.

It's preclinical and hasn't been tested in humans. Still, it's one of the more credible protein-based alternatives to LNPs published in a major journal so far.

Why does this paper matter?

The paper sits where three lines of research meet.

1. The limits of LNPs. Lipid nanoparticles are very good at getting nucleic acids into liver cells after intravenous injection, and less precise elsewhere. Reviews of LNP chemistry describe accumulation in the liver and spleen, innate immune and complement activation, and anti-PEG antibodies in some recipients. Questions about where LNP cargo goes and how long it lasts also feed into the debate over spike protein persistence after vaccination. Whatever your view of the COVID-19 vaccines, the field has been looking for non-lipid carriers for years.

2. Better protein engineering. Designing recombinant proteins that fold predictably, self-assemble and respond to their environment has matured a lot over the past decade. Elastin-like polypeptides (ELPs), built from the repeating unit Val–Pro–Gly–Xaa–Gly found in human elastin, are among the most studied. They're biodegradable, can be made in bacteria and can be tuned to form micelles at body temperature.

3. The gene editing pipeline. CRISPR–Cas9 is now an approved therapy (Casgevy, for sickle cell disease and β-thalassemia), and in vivo CRISPR has been tested in people with transthyretin amyloidosis. Base editing and prime editing, both from David Liu's lab, are moving toward the clinic. Each of these needs a delivery vehicle, and today that's almost always an LNP or a viral vector.

The open question has been whether a fully protein-based platform can match LNP-class performance across the whole range of payloads, not just one carefully chosen cargo. This paper is a serious attempt at an answer.

How do protein nanoparticles carry mRNA into cells?

What is an elastin-like polypeptide?

Elastin-like polypeptides (ELPs) are engineered proteins made of a repeating five-amino-acid unit, Val–Pro–Gly–Xaa–Gly. The Xaa position is the tuning knob: changing it adjusts the polymer's hydrophobicity, charge and the temperature at which it changes phase.

ELPs have a transition temperature (technically a lower critical solution temperature). Below it, the polymer stays dissolved in water; above it, the chain collapses into a compact, water-repelling state. Join two ELP blocks with different transition temperatures, one that stays soluble at body temperature and one that collapses, and you get an amphiphilic diblock that self-assembles into a micelle, much like soap does in water.

In plain terms:

• Picture a sausage made of two halves, one that likes water and one that doesn't
• Drop lots of them into warm water
• The water-avoiding halves clump in the middle and the water-loving halves face outward
• You've made a tiny ball (a micelle), and that ball can carry cargo

What is an endosomal escape peptide?

Cells swallow nanoparticles into bubbles called endosomes, which usually head for the lysosome, where contents get broken down. To work, the cargo has to reach the cytosol first: that's where mRNA is translated, where CRISPR reaches the DNA and where siRNA engages the RISC machinery.

An endosomal escape peptide (EEP) is a short, usually α-helical and often positively charged sequence that disrupts the endosome membrane as the endosome turns acidic. That lets the cargo slip out into the cytosol.

Most EEPs in the literature are repurposed cell-penetrating peptides (TAT, penetratin) or membrane-active sequences from antimicrobial peptides, and many are inefficient or nonspecific. This paper didn't just reuse one. It went looking for a better one.

What did the researchers build?

Iterative ELP design

The team didn't reach their lead candidate in one step. They worked through four generations of ELP, progressively improving:

• the size and stability of the self-assembled nanoparticle
• the strength of cargo binding
• the responsiveness to endosomal pH
• the efficiency of cytosolic release

The fourth-generation ELP became the chassis for the platform they call ENTER (Elastin-based Nanoparticles for Therapeutic dElivery).

In silico EEP discovery

Rather than testing existing EEPs one by one, the authors built a computational pipeline:

1. Compiled a database of α-helical peptides from public sources (the Antimicrobial Peptide Database, DBAASP and a structural motif database)
2. Filtered for properties expected to favor pH-triggered membrane disruption: net charge, helical propensity, amphipathicity and hydrophobic moment
3. Trained predictive models on existing peptide-delivery data to score candidates
4. Synthesized and tested only the top-scoring handful

One peptide from that screen, labelled EEP13, stood out. The authors report that it improved protein delivery efficiency by 48% compared with the benchmark peptide it was tested against. Fusing it to the ELP chassis gives the lead candidate the rest of the paper is built around: ELP–EEP13.

What did ELP–EEP13 do in cell experiments?

The first test used Cre recombinase, a standard tool in mouse genetics. Reporter cells switch on a fluorescent protein as soon as Cre reaches the nucleus, so Cre is a sensitive readout of whether a carrier really got its cargo into the cytosol.

The team supplied Cre in three forms:

In all three formats, ELP–EEP13 produced editing similar to or better than the commercial lipid-based transfection reagents used as positive controls. The protein format mattered most, because ribonucleoprotein gene editors have to arrive as a protein and RNA complex rather than as a message the cell reads.

Can it deliver CRISPR and base editors?

The authors then tested ELP–EEP13 with two payloads from David Liu's gene editing toolkit:

1. SpCas9 ribonucleoprotein (RNP), Cas9 protein pre-loaded with its guide RNA, the format used in the approved sickle cell therapy
2. Adenine base editor (ABE), a Cas9-deaminase fusion that converts A·T base pairs to G·C without cutting both DNA strands

Across multiple cell lines and primary cell types, ELP–EEP13:

• Delivered functional Cas9 RNP, producing measurable DNA edits at target sites
• Delivered base editors with editing efficiency comparable to lipid-based delivery controls
• Did so without the cytotoxicity typical of cationic transfection reagents at higher doses

So the ENTER platform isn't tied to one cargo class. It handled mRNA, DNA, ribonucleoprotein and base-editor payloads, the same range LNPs and electroporation are used for today.

Does it work for siRNA drugs?

Short interfering RNAs (siRNAs) silence specific messenger RNAs in the cytosol. That's the drug class behind patisiran, the approved treatment for hereditary transthyretin amyloidosis, which is delivered in an LNP by intravenous infusion every three weeks.

ELP–EEP13 delivered siRNAs in cell experiments and produced target gene knockdown. The authors didn't run a head-to-head comparison against patisiran's LNP, which would take a separate study, so the in vitro data show only that the platform is compatible with this payload class.

That's a pointer toward future siRNA therapies for amyloid diseases that might not need lipid carriers, not evidence that any such therapy works.

Clinical next step

Concerned about lipid nanoparticles, spike persistence, or post-vaccination symptoms?

Explore screening options and discuss findings with a qualified clinician.

Start screening

What happened in mice?

Cell experiments are necessary but not sufficient. The harder test is whether a carrier works in a living animal.

The authors gave ELP–EEP13 carrying Cre protein intranasally to Cre reporter mice (the Ai14 strain, which switches on the red fluorescent protein tdTomato wherever Cre acts on DNA). Histology showed tdTomato-positive cells in the lung epithelium, spread across multiple lung regions, with no obvious tissue toxicity in the limited evaluation reported.

A single nasal dose producing measurable editing in lung tissue is a proof of platform, not a treatment for any disease. Put simply: the nanoparticle reached living lung cells and delivered its cargo intact, and the lungs lit up where the molecular switch had flipped.

How do protein carriers compare with lipid nanoparticles?

What LNPs do well, and where they fall short

Decades of LNP chemistry, especially work on ionizable lipids, made the rapid rollout of mRNA COVID-19 vaccines possible. The same literature is candid about the trade-offs:

• Biodistribution is hard to control. After injection, LNPs spread beyond the injection site and concentrate in the liver and spleen. Biodistribution studies, including those filed with regulators, document organ-level accumulation.

• Innate immune activation can be significant. Both the ionizable lipid and the PEG-lipid can activate complement and innate immunity in some recipients, and anti-PEG antibodies are well documented.

• Manufacturing is demanding. It needs lipid synthesis and microfluidic mixing, and cold-chain shelf life is short.

• Repeat dosing is complicated. Some people develop reactivity to LNP components after a first exposure.

These aren't fringe concerns. They're discussed openly in the gene therapy literature and have driven steady investment in alternative carriers.

What protein carriers offer instead

Protein-based delivery systems offer a different set of tradeoffs:

• They break down into amino acids, not into synthetic lipids whose long-term metabolism is still being characterized.
• They can be engineered for tissue specificity by fusing targeting domains (antibody fragments, peptide ligands) onto the protein chain.
• They're made by bacterial fermentation, the same mature process used for insulin and many biologics.
• They avoid PEGylated lipids and the anti-PEG antibody problem that comes with them.

The challenges are real too. Recombinant proteins have to be purified carefully to remove bacterial endotoxin, yields have to be economic, and any new platform needs its own immunogenicity and safety profiling in humans. What's meaningful here is that the chemistry is genuinely different, not that it's already proven safer.

What could this mean for amyloid diseases?

Several of the busiest areas of nucleic acid medicine involve amyloid proteins. Note that these are the systemic and brain amyloidoses, which are different from the fibrin amyloid microclots described in Long COVID research.

• Hereditary transthyretin amyloidosis (hATTR), treated with the LNP-delivered siRNA patisiran and, in trials, with in vivo CRISPR editing also delivered by LNP.
• Cerebral amyloid angiopathy and Alzheimer's disease, targets for siRNA, antisense oligonucleotides and gene editing, where delivery into brain tissue is still unsolved.
• Light-chain amyloidosis (AL), where CRISPR and antibody approaches are in development.

Delivery is the rate-limiting step for all of them. A carrier that handles siRNA, Cas9 RNPs and base editors, and that can reach tissue other than liver, is the kind of platform that could eventually widen these therapies. That's a reasonable expectation to hold, not a result the paper demonstrates.

What does this say about the post-vaccination conversation?

For readers worried about lipid nanoparticles in mRNA vaccines, the message here is narrow. The mainstream gene therapy field acknowledges that LNPs are imperfect and is spending NIH money and top-lab time on alternatives. That interest isn't coming only from patient advocacy.

It doesn't change the LNPs already given, and it doesn't settle any clinical question about spike protein persistence and how it's measured, fibrinolysis-resistant microclots after vaccination or amyloid cross-seeding in the brain. Those questions rest on their own evidence, and our summary of the JPandS review of reported COVID vaccine harms shows how contested that evidence base is.

If your own concern is symptoms rather than delivery chemistry, an at-home amyloid test looks for fibrin amyloid microclots in a blood sample and produces observational data you can take to a clinician. It doesn't measure spike protein or lipid nanoparticles.

What it does suggest is that future nucleic acid medicines may not use lipid carriers at all. ENTER is one candidate for what replaces them.

What the paper doesn't show

Six limits are worth keeping in view.

1. No human data. Every experiment ran in cell culture or mice. Safety and efficacy in people are untested.

2. Narrow in vivo scope. The only in vivo editing reported is mouse lung after an intranasal dose. Liver, brain, kidney and intramuscular routes aren't covered here.

3. No long-term safety or immunogenicity data. There are no extended dosing studies, no immune response to the ELP across repeated doses, and no one-month or one-year biodistribution.

4. No matched comparison with approved LNP formulations. The benchmarks were commercial transfection reagents, not the lipid formulations used in licensed vaccines or in patisiran.

5. Manufacturing isn't addressed. Protein expression in E. coli is mature, but endotoxin-free clinical-grade ELP at scale is its own development problem.

6. Commercial interests exist. A patent has been filed (PCT/US2024/038614), and David Liu has commercial relationships with gene editing companies. That doesn't invalidate the data, but it shapes how the platform reaches the clinic.

The short version

If you want it in one paragraph: most mRNA medicines, including the COVID vaccines, get into cells inside tiny fat bubbles called lipid nanoparticles. Those work, but they spread to organs nobody targeted and some people react to them. This team built a carrier made entirely of protein instead, tested it in cell dishes and sprayed it into the noses of mice, and it carried mRNA, DNA, gene-editing enzymes and gene-silencing molecules about as well as the fat bubbles did in those tests. It hasn't been given to humans.

The five claims the paper supports are specific. A protein-only nanoparticle can self-assemble, carry varied cargo and release it into the cytosol. A computationally selected escape peptide, EEP13, beat its benchmark by 48%. The lead candidate matched or exceeded commercial lipid transfection reagents across Cre (mRNA, DNA and protein), Cas9 RNPs, base editors and siRNA. One intranasal dose produced detectable editing in mouse lung. And the same payload classes are the ones used in mRNA vaccines, patisiran-type drugs and in vivo CRISPR for amyloid disease.

None of that is evidence of safety or efficacy in humans, and it would be a mistake to read it that way. What's next is clear enough: human testing, repeat dosing behaviour, manufacturing at scale, and which indication goes first.

Where to go from here

If you came to this because of questions about what vaccine or viral spike protein does in the body, the testing pillars are the place to continue. Our guide to how amyloid fibrils are detected with Thioflavin T covers the staining method itself, and the amyloid fibrin test pillar explains what fibrin amyloid microclot microscopy can and can't tell you. For the clotting mechanism specifically, see our explainer on spike protein amyloid fibrils and impaired fibrinolysis.

---

This article is educational and discusses preclinical laboratory research. It shouldn't be read as establishing safety or efficacy in humans, as endorsing any medical product, or as a substitute for care from a qualified clinician. The platform described hasn't been tested in human clinical trials.