Cell-Penetrating Peptides Explained: The Artifact That Reset a Field

Peptide science

Cell-penetrating peptides are short sequences, usually thirty amino acids or fewer, that carry themselves and attached cargo across the plasma membrane. They are one of the most heavily studied ideas in peptide science — tens of thousands of papers since 1988 — and one of the least productive. Not a single cell-penetrating-peptide conjugate has been approved as a drug. The reason is worth understanding, because it involves one of the cleanest examples of a scientific field catching and correcting its own mistake.

Where cell-penetrating peptides came from

In December 1988, two groups published back to back in the same issue of Cell. Frankel and Pabo showed that purified HIV-1 Tat protein was taken up by cultured cells and went on to trans-activate the viral promoter, detectable at concentrations as low as 1 nM in the presence of chloroquine. Green and Loewenstein showed the same thing with chemically synthesised Tat and narrowed the active region considerably.

Six years later Derossi and colleagues found the same behaviour in something entirely unrelated: the third helix of the Antennapedia homeodomain, a sixteen-residue fragment now called penetratin. Crucially, they reported that it translocated at 4 degrees C — a temperature at which endocytosis is largely shut down. In 1997 Vivès, Brodin and Lebleu narrowed Tat further, showing that the alpha-helical region around residues 37 to 47 was dispensable and the basic cluster was what mattered. In 1998 Pooga and colleagues in Ülo Langel’s group described transportan, a chimera of galanin and mastoparan, and used the phrase “cell-penetrating peptide” — the earliest indexed appearance of the term.

By 1999 the concept had gone in vivo. Schwarze and colleagues reported in Science that intraperitoneal injection of a 120-kilodalton beta-galactosidase fused to the Tat transduction domain delivered biologically active protein “to all tissues in mice, including the brain.” It is still the single most-cited basis for the claim that Tat crosses the blood-brain barrier.

The correction

The trouble started in 2002, when Lundberg and Johansson noticed that positively charged proteins “adhered to the cell membrane of living cells and were not removed by extensive washing” — but after fixation they “relocated to the cell nucleus.” Their conclusion was blunt: apparent membrane permeability might be a fixation artifact, and most of the field’s images had been taken on fixed cells.

The definitive paper came from Lebleu’s own group. Richard and colleagues, publishing in the Journal of Biological Chemistry in 2003, re-examined Tat 48-60 and nonaarginine and reported that “cell fixation, even in mild conditions, leads to the artifactual uptake of these peptides.” Five minutes in 3.7 per cent formaldehyde was enough to produce the characteristic nuclear staining, while control markers such as transferrin were unaffected. In living, unfixed cells the peptide showed a punctate cytoplasmic pattern that co-localised with endocytic markers.

The mechanism of the artifact is elegant and slightly grim. These peptides are highly cationic, so they bind hard to the negatively charged outer membrane. Fixation destroys the membrane’s barrier function. The peptide then redistributes to the largest pool of negative charge available — nucleic acid — and concentrates in the nucleus of a cell that is already dead.

To the field’s credit, the authors did not overclaim in the other direction either, noting that they “cannot formally exclude the possibility that a small fraction of CPP enters cells by an endocytosis-independent but biologically relevant pathway.” That caveat still stands.

How they actually get in

The current picture is that two routes coexist, and which one dominates depends on concentration, peptide class and cargo. At physiological conditions and low concentration, endocytosis prevails. At higher concentrations, direct translocation becomes possible.

• Endocytosis. Macropinocytosis is the most commonly reported route for Tat-protein conjugates and polyarginines; clathrin- and caveolae-mediated pathways also contribute. Uptake of octaarginine is suppressed by inhibitors of sodium-proton exchange and of actin polymerisation, and depends on heparan sulfate proteoglycans at the cell surface.
• Direct translocation. Several models compete: inverted micelle formation, barrel-stave and toroidal pores, the carpet model, membrane thinning, and a counterion-driven transient water pore that has support from both molecular dynamics and cell experiments. None of them is settled, and different peptides plausibly use different ones.

Cargo changes the answer. The same peptide can enter by macropinocytosis when carrying a protein and by clathrin-mediated endocytosis when complexed with an oligonucleotide, which is one reason results from one laboratory rarely transfer cleanly to another.

The step that decides everything

Getting inside a vesicle is not getting inside the cell. A peptide sitting in an endosome is still topologically outside the cytoplasm, and if it stays there it does nothing. Endosomal escape has been called the rate-limiting step for two decades, but for most of that time the estimates of how bad it is were qualitative, because bright punctate endosomal fluorescence swamps any diffuse cytosolic signal in a microscope.

In 2021 a Monash group built a split-luciferase assay that separates cytosolic protein from total internalised protein and finally put numbers on it. Plain GFP escaped in under 2 per cent of cases in HEK293 cells and around 7 per cent in HeLa. The uncomfortable result was that none of the endosomal escape peptides they tested had higher escape efficiency than plain GFP. Positively charged peptides increased total cytosolic delivery, but they did it by increasing non-specific membrane association, not by escaping endosomes better.

Why arginine, and why arginine is also the problem

Arginine is not simply a source of positive charge. Its guanidinium group forms bidentate hydrogen bonds with the sulfates, phosphates and carboxylates on the cell surface, producing hydrophobic ion pairs that lysine’s simple amine cannot. Wender and colleagues showed this directly: histidine, lysine and ornithine oligomers all performed worse than Tat(49-57) despite comparable charge, while a nonaarginine was about twentyfold more efficient than Tat(49-57), and its all-D-amino-acid counterpart more than a hundredfold. Futaki’s series across four to sixteen arginines pointed to an optimum around eight. The literature has never fully agreed on the exact number; eight or nine consecutive arginines is the honest answer.

The same property is the liability. A peptide that binds every cell surface enters every cell, which means no tissue selectivity. It also means accumulation in organs such as the liver, and membrane disruption at higher concentrations. A careful comparative toxicity study found that penetratin, Tat and pVEC caused little or no lactate dehydrogenase leakage, while the amphipathic peptides MAP and transportan-10 released 40 per cent of total LDH within ten minutes at 10 micromolar. Serum makes things worse: cytosolic entry efficiency drops substantially in the presence of serum proteins, and unmodified L-peptides are degraded by proteases — which is why D-amino-acid backbones, peptoids and cyclisation are the field’s standard defences.

What the clinical record actually looks like

The gap between the preclinical literature and the clinic is stark.

• Delcasertib (KAI-9803), a delta-PKC inhibitor peptide whose research parent was delivered by a Tat carrier, was tested in PROTECTION AMI — a phase 2b trial with 1,010 anterior STEMI patients. It “did not reduce biomarkers of myocardial injury”, with no differences in infarct size, ST-segment recovery, ejection fraction, death, heart failure or arrhythmia.
• Brimapitide (XG-102, AM-111), a Tat-conjugated JNK inhibitor, got further than anything else. Its phase 3 HEALOS trial in sudden sensorineural hearing loss enrolled 256 patients across 51 sites and did not meet its primary endpoint in the overall population, with only a post-hoc, nominally significant effect in the profound subgroup. A second phase 3 was terminated. In ophthalmology it was non-inferior to dexamethasone after ocular surgery, but needed rescue medication far more often. No approval followed in either indication.
• Vesleteplirsen (SRP-5051), a peptide-conjugated morpholino for Duchenne muscular dystrophy, is the clearest recent test of the delivery concept. Its phase 2 MOMENTUM trial was terminated in February 2025, with the registry recording only that “the study was terminated following review of safety data.”
• p28, a 28-residue fragment of the bacterial protein azurin that preferentially enters cancer cells, completed a first-in-human phase 1 in 15 patients with no dose-limiting toxicities and no measurable antibody response, with one complete response lasting 139 weeks. A subsequent paediatric CNS-tumour study produced no objective responses.
• Currently, peptide-conjugated oligonucleotide programmes from Entrada and PepGen are in phase 1 and phase 2 for Duchenne and myotonic dystrophy. PepGen’s Duchenne programme has been discontinued by the sponsor. A registry search finds nothing from this class in phase 3.

The fix that works on paper

The most elegant answer to the selectivity problem came from Roger Tsien’s laboratory in 2004. An activatable cell-penetrating peptide fuses the polycationic transporter to a polyanionic inhibitory domain through a protease-cleavable linker. In its intact form the charges neutralise each other and the peptide does not stick to cells. Where a target protease cuts the linker, the transporter is released and becomes active — cell association typically rises tenfold or more.

Applied to tumours secreting MMP-2 and MMP-9, far-red activatable peptides produced in vivo contrast ratios of two to three and a 3.1-fold increase in standardised uptake value against normal tissue and scrambled-linker controls, with comparable contrast on freshly resected human squamous cell carcinoma. It is a genuinely good idea. Twenty years on, it remains an imaging and preclinical tool.

Do they cross the blood-brain barrier?

Partly, selectively, and not in the way the 1999 result implied. A 2015 study measured unidirectional brain influx for five peptides directly and found high rates for Tat(47-57), SynB3 and especially pVEC, and negligible influx for transportan analogues. Capillary depletion showed roughly 80 per cent of the influxed peptide genuinely reached brain parenchyma rather than sticking in the endothelium, and all but pVEC showed significant efflux back out.

The authors’ own conclusion is the one to remember: “good CPP properties do not imply efficient brain influx.” The two abilities do not correlate. A further caution comes from human stem-cell-derived brain endothelial models, where Tat permeated the monolayer while penetratin did not — but both compromised barrier integrity enough to let a small inert tracer through, meaning apparent transport can partly reflect damage to the barrier rather than transport across it.

Frequently asked questions

Is any approved drug a cell-penetrating peptide?

A search of the clinical trial registry and of approved-product labelling identifies no cell-penetrating-peptide conjugate approved by the FDA or the EMA. Every programme that reached late-stage testing missed its primary endpoint or was terminated.

Was the original cell-penetrating peptide research fraudulent?

No. It was a methodological artifact, found and published by the same laboratories that produced the original results — Lebleu’s group authored both the 1997 Tat paper and the 2003 correction. That is science working, not science failing.

What is the single biggest obstacle?

Endosomal escape. Uptake is easy and reproducible; releasing the cargo from the vesicle into the cytosol, where it can act, is not, and the best quantitative data suggest current escape peptides do not improve that step at all.

Why does arginine work better than lysine?

The guanidinium group of arginine forms bidentate hydrogen bonds with sulfate, phosphate and carboxylate groups on the cell surface, creating hydrophobic ion pairs. Lysine’s primary amine cannot do this, which is why lysine oligomers underperform arginine oligomers of the same net charge.

Informational only — not medical advice · 21+