A peptide is called cell-penetrating because an assay said so, and the assay is the whole story. The label has been applied to sequences that were measured for cell association, for total internalization, for endosomal accumulation and, much more rarely, for arrival in the cytosol. A 2019 critical evaluation of these methods argues the point directly: measurements of "cell penetration" that do not adequately distinguish total cellular uptake from cytosolic localization often lead to inaccurate interpretations of a molecule's cytosolic localization [1]. Designing a cell-penetrating peptide is therefore two jobs. One is chemistry. The other is building a control set strong enough that the number you report means what its name says.

Two design families, and what separates them

The working taxonomy is physicochemical. A 2012 review proposes a classification of cell-penetrating peptides based on their physical-chemical properties and origin, and offers it as a framework for understanding and discovering new ones [2]. The two families that dominate practice are the cationic and the amphipathic.

Cationic sequences carry a high density of arginine or lysine and little else that is structurally load-bearing: oligoarginines, the Tat basic region, and their many analogues. Their first interaction with a cell is electrostatic, with the polyanionic sugars of the cell surface. That interaction is not merely a smear of charge. One group reported that syndecan-4, an isoform of the syndecan family of transmembrane proteoglycans, binds and mediates transport into cells of the three most frequently used cationic peptides — penetratin, octaarginine and Tat — with quantitative uptake studies and mutational analyses indicating that attachment is mediated by specific interactions between the heparan sulfate chains of syndecan-4 and the peptides, and with protein kinase C alpha also heavily involved; those authors described this as the first direct evidence of receptor-mediated uptake of cationic peptides [3].

Amphipathic sequences instead segregate hydrophobic and polar faces, either along a helix or across a primary sequence, so that the molecule can partition into a bilayer rather than only adsorb to its surface. Chimeric and model amphipathic designs, and secondary amphipaths whose faces only resolve on folding, are conventionally grouped here. The practical consequence for design is that the two families have different levers. A cationic peptide is tuned by charge count and spacing; an amphipathic one is tuned by hydrophobic moment, and its behavior depends on whether the conformation that creates the amphipathic face actually forms under your buffer, temperature and lipid conditions. Neither family has a design rule that predicts cytosolic arrival from sequence, and the assays below are the reason we can say that.

Cargo is part of the molecule, not a passenger

A common design error is treating the peptide as a property that transfers to whatever is attached to it. It does not. Using green fluorescent protein as a model cargo, one study compared five sequences — penetratin, R8, Tat, transportan and Xentry — plus cyclic derivatives across four cell lines the authors describe as representing different tissues. At 10 µM, most fusion proteins localized to endosomes by confocal microscopy. Flow cytometry showed that uptake depended on both cell type (10T1/2 > HepG2 > HeLa > HEK) and sequence (transportan > R8 > penetratin, approximately equal to Tat, > Xentry). Cyclization or addition of an HA sequence increased cellular uptake, but the fluorescence remained contained in vesicles with no evidence of endosomal escape [4]. Read that last clause carefully. Two modifications improved the measured quantity and did not improve the quantity anyone actually wanted.

Cargo size can also gate the mechanism itself, not just the efficiency. Investigating a route by which cationic peptides trigger ceramide formation, one group reported that adding sphingomyelinase lowered the threshold for direct translocation and enhanced the translocation of R9 coupled to low-molecular-weight cargos, but not to high-molecular-weight cargos [5]. That is a specific result for R9 conjugates under exogenous sphingomyelinase, not a general law. But it illustrates the design consequence: a linker, a fluorophore and a protein are not neutral additions, and a peptide validated with a small cargo has not been validated with a large one.

The label is part of the molecule too

Cell-penetrating peptide characterization leans heavily on fluorescence readouts, which means the thing actually tested is often a labeled conjugate rather than the bare sequence. An array of seven fluorophores from distinct structural classes was conjugated to penetratin and examined for effects on physicochemical properties and uptake-related behavior. Exposing several mammalian cell types to the conjugates revealed a strong structure-dependent reduction in viability, with 1.5- to 20-fold lower IC50 values than non-labeled penetratin — that is, the labeled conjugates were more cytotoxic, across that set of seven fluorophores on that one peptide backbone. Effects on membrane integrity and intracellular distribution patterns also differed among conjugates, and the authors concluded that selection of the appropriate fluorophore is critical [6].

The fixation problem compounds this. A 2003 reevaluation of Tat 48–60 and nonaarginine demonstrated that cell fixation, even under mild conditions, leads to artifactual uptake of these peptides, and that flow cytometry analysis cannot be used validly to evaluate cellular uptake unless a step of trypsin digestion of the cell-membrane-adsorbed peptide is included in the protocol [7]. Any sequence characterized before that control became routine deserves a second look, and any sequence characterized after it should report the trypsin or heparin step by concentration and duration rather than by name. That reporting convention is bench practice, not a published finding.

The scrambled control: what shuffling does and does not test

A scrambled control reorders the same residues, holding composition, net charge and mass constant while destroying sequence order and, with it, any folded amphipathic face. Its purpose is to ask whether the activity is a property of the sequence or of the amino acid inventory. This is convention rather than a result, and it is worth being clear about what it can decide.

For an amphipathic peptide the scrambled control is informative, because shuffling usually abolishes the segregated faces. For a cationic peptide it is weak, and the reason is worth spelling out. Shuffling leaves residue composition and net charge untouched, while the surface attachment reported for penetratin, octaarginine and Tat is described as a specific interaction between the heparan sulfate chains of syndecan-4 and those peptides [3]. Those authors did not test scrambled sequences. A scramble varies the one property — residue order — that their account of attachment is not stated in terms of, so a null result from it says little about that step. That last sentence is reasoning about what the control can decide, not a finding those authors reported.

Order does matter within the cationic family, but not in the coarse way a single scramble tests. The background framing of one high-throughput study is that the efficiency with which polycationic peptides penetrate the cytosol depends on the number and overall patterning of arginine residues, and those authors used a cytosol-specific screen to explore how number, patterning and stereochemistry of arginines affect cytosolic penetration of a model 10-residue peptide [8]. A graded series of arginine arrangements answers a question that one scrambled control cannot.

The D-enantiomer control is sharper than it looks — and blunter

Replacing every residue with its mirror image inverts the molecule's chirality while preserving composition, charge and, in mirror form, its fold. Historically it was the decisive experiment for receptor independence. Reporting on the 16-residue third helix of the Antennapedia homeodomain, one group found that a reverse helix and a helix composed of D-enantiomers still translocate across biological membranes at 4 and 37 °C, and that introducing one or three prolines did not hamper internalization, demonstrating that the alpha-helical structure was not necessary; they read the stereochemical result as strongly suggesting a receptor-independent mechanism, compatible with direct interaction with membrane phospholipids [9]. Note the caveat that applies to any pre-2003 uptake dataset: that work predates the fixation and trypsinization controls established later [7], so the mechanistic inference should be weighted accordingly.

There is a second, structural limit, and it is logic rather than a citation: a stereochemical control only excludes recognition that is itself stereospecific. The cell-surface attachment reported for penetratin, octaarginine and Tat runs through the heparan sulfate chains of syndecan-4 [3], and that report does not establish whether the interaction depends on the peptide's chirality. An unchanged D-peptide result therefore does not, by itself, rule out a cell-surface partner.

What the D-enantiomer does deliver reliably is proteolytic stability, and that turns it into an analytical tool. Testing mirror-image sequences carrying a phosphorodiamidate morpholino oligomer-biotin cargo, one group set out to test whether mirror-image D-peptides would match the L-peptides that this delivery work had until then been restricted to, and found that several enantiomeric sequences delivered the cargo with activities similar to that comparator while remaining stable against serum proteolysis; the biotin allowed affinity capture of fully intact conjugates from whole-cell and cytosolic lysates, and profiling a mixture of the constructs in cells gave their relative intracellular concentrations [10]. The stability is what makes the quantification honest — an L-peptide partly degraded during the experiment reports a concentration for a molecule that no longer exists.

Why uptake is the wrong endpoint

The endpoint that matters for an intracellular target is cytosolic concentration, and it is not a rescaled version of uptake. Using a split-luciferase endosomal escape quantification assay applied to a range of widely studied cell-penetrating peptides fused to a model protein, one group reported that positively charged peptides enhance cytosolic delivery as a result of increased non-specific cell membrane association, rather than increased endosomal escape efficiency [11]. That finding is arithmetically clean once the two quantities are kept apart. Cytosolic delivery per cell went up, on those authors' account because more peptide associated non-specifically with the cell membrane. Endosomal escape efficiency is a ratio rather than a total — an amount reaching the cytosol divided by an amount the cell had already engaged — and that ratio, those authors report, did not go up. Their abstract does not state the exact numerator and denominator the assay uses to compute it, so read the contrast as directional rather than as a formula. Both statements describe the same experiment, and a paper reporting only the first while implying the second is not lying so much as dividing by the wrong thing.

Assays that report the cytosol

Three families of readout make the cytosolic compartment the thing being measured, rather than something inferred from an image.

Covalent capture by a compartment-restricted enzyme

The chloroalkane penetration assay is HaloTag-based: a chloroalkane-tagged molecule that reaches the compartment where HaloTag sits is captured covalently, and the resulting loss of a subsequent fluorescent chloroalkane signal reports penetration. That pulse-chase description is how the assay is run, and is not itself drawn from the report below. Its developers described the assay as quantitative, high-throughput and compartment-specific, and demonstrated it in three distinct ways: profiling the extent of cytosolic penetration with respect to concentration, presence of serum, temperature and time; investigating structure-penetration relationships for bioactive stapled peptides and for peptides fused to cell-penetrating sequences; and, using a cell line where HaloTag is localized to the nucleus, showing quantitative measurement of nuclear penetration [12]. It has since been adapted for parallel library screening [8].

Absolute counting

A related design uses benzylguanine and biotin modifications on the vector: benzylguanine allows selective covalent capture in the cytosol by a SNAP-tag fusion protein, biotin allows quantification at picomolar sensitivity. The authors validated it by determining absolute numbers of cytosolic molecules for two vectors, the B-subunit of Shiga toxin and Tat [13]. Absolute counts are what convert a fold-change into a design constraint.

Split-reporter complementation

Split-luciferase and split-fluorescent-protein systems place one fragment in the cytosol and fuse the other to the cargo, so signal requires the two to meet [11]. These are sensitive, but they report a complementation event, and the relationship between event count and molecule count needs its own calibration.

A control set worth running

None of these is a finding. They are the comparisons without which the measurement cannot be interpreted.

  • The unlabeled peptide, characterized for purity and identity, alongside every labeled conjugate — and a second, structurally unrelated fluorophore where distribution is claimed [6].
  • A scrambled sequence, and, for cationic designs, a graded arginine series rather than a single scramble [8].
  • The all-D enantiomer, reported as a stability control by default and as a mechanism control only with the caveat above.
  • A surface-removal step with concentration and duration stated, and the no-removal condition beside it [7].
  • Live cells for any localization claim; fixed images only shown next to their live equivalent [7].
  • Viability and membrane-integrity readouts at the exact concentration, time and cell line used [6].
  • At least two cell lines, since in the five-sequence green fluorescent protein fusion comparison above, uptake at 10 µM depended on cell type as well as on sequence, those four lines ranking 10T1/2 > HepG2 > HeLa > HEK [4].
  • The intended cargo, not a surrogate, at the intended linkage [4, 5].
  • One cytosol-specific readout — covalent capture, absolute counting or complementation — for any claim containing the word cytosolic [1, 12, 13].

What is not known

There is no sequence-to-cytosol design rule for either family; screening remains empirical [8]. Those authors state that quantification of total intracellular versus cytosolic concentrations remains demanding and that determining delivery efficiency is therefore not straightforward; cross-study comparison is hampered further because the assays differ in what they count [1]. Whether the surface proteoglycan engagement described for penetratin, octaarginine and Tat [3] is a required intermediate or a parallel event is unresolved, and whether it survives a reordering of the same residues has not, as far as this survey found, been tested. Scrambled and D-enantiomer controls are near-universal convention with, again as far as this survey found, no systematic study establishing how often either changes a conclusion. And, as far as this survey found, no consensus threshold defines "penetrant" — the number depends entirely on which compartment the assay could see.