Orthogonal high-throughput screening for translocating peptides
We selected peptides from a combinatorial library based, specifically, on their ability to spontaneously translocate across synthetic lipid bilayer membranes. This was accomplished by using bilayer membrane vesicles as the basis for a high throughput screen that identifies only peptides that translocate across membranes spontaneously (
). We use an “orthogonal” high-throughput approach to select for multiple properties simultaneously and independently. In particular, we selected independently for aqueous peptide solubility, lack of membrane disruption, and rapid membrane translocation. Solubility pre-selection is accomplished by incubating library peptides in buffer for an hour before screening with lipid vesicles, such that insoluble library members are inactivated by precipitation15
. Screening for membrane disruption is done with a vesicle-based screen described elsewhere12,15
that utilizes liposome-entrapped terbium(III) with an aromatic chelator, dipicolinic acid (DPA), in the external solution, so that a quantifiable luminescent metal-chelate complex forms only upon membrane disruption. Here we describe a third orthogonal selection criterion for translocation. To detect peptide translocation into terbium-containing vesicles, we also entrapped the protease chymotrypsin and added a protease inhibitor outside the vesicles to quench any external or released enzyme (
). A fluorescent chymotrypsin-sensitive aminomethylcoumarin (amc) moiety was attached to the carboxyl terminus of the peptides in the combinatorial library to report on peptide translocation (
). Using this orthogonal strategy, membrane disruption and peptide translocation can be measured independently in the same vesicles in a multiwell plate fluorescence format. Control experiments with membrane-impermeant (
) and membrane-permeant (
) chymotrypsin-sensitive probes were used to validate the design of the screen.
Figure 1 Orthogonal High-throughput screen design. A: Library architecture: Peptide library members were attached to polystyrene resin microbeads through the side chain of a glutamate using a UV cleavable linker12, which releases an amidated glutamine upon irradiation. (more ...)
To test the sequence requirements for true spontaneous membrane translocation, a 12-residue peptide library was designed with nine variable residues to encompass the amino acid compositions of known cell penetrating peptides, ranging from fully cationic to fully hydrophobic (
). An important practical consideration was to have a library with fewer than about 12,000 sequences so that it could be screened in a reasonable amount of time. Thus, each of the nine combinatorial sites contained only 2-4 possible amino acids. We designed the library so that each site had at least one basic and at least one hydrophobic residue, with the default residues being arginine and leucine. But we also designed the library to contain as many known cell penetrating peptides as possible from multiple classes, including polycationic, hydrophobic and proline-rich peptides. Therefore several positions contained lysine as an alternate basic residue, and some leucines were either augmented or replaced with other hydrophobes (P,I,Y,F) or other residues (e.g. K in position 3 and Q in position 6 derived from the HIV tat
peptide) when that was needed to improve the match to a known CPP. The final library contained 10,368 members, and those members included the tat
peptide sequence (RRKRRQRRR) the tat homolog Arg9 (RRRRRRRRR), and sequences that closely resembled the proline-rich cell penetrating peptides that are based on fragments of the antimicrobial peptide bactenecin 7. These sequences include Bac71-7
(RRIRPRP) and Bac715-24
(PRPLPFPRP) which have potent cell penetrating activity17
. The library was also designed to have an N-terminal RRGY motif which we had identified as important for membrane activity in screening for antimicrobial peptides15
. The three C-terminal amino acids were fixed, as we describe below. Glycine in position 10 constitutes a flexible linker. The side chain of a glutamate in positon 11 is used to link the peptides to the synthesis resin beads by a photolabile linker, which releases a glutamine upon cleavage. The C-terminal Phe has an aminomethylcoumarin (AMC) group for detection of chymotrypsin cleavage.
Figure 2 Library design. The library screened in this work contained a rationally designed 9-residue combinatorial segment, followed by a fixed GQF(amc) (). Each varied (Oi) position could contain one of the residues shown vertically below. Possible cationic (more ...)
The library was synthesized by a split-and-recombine method as described previously12,15
in which members were linked to 300 μm diameter polystyrene beads. In this library each bead contained about 1 nmol of a single peptide sequence. High throughput screening was performed in 96-well plate format, with each well containing about 0.5-1.0 nmol of a single peptide sequence, extracted from a single photocleaved bead, along with 100 nmol total lipid in the form of 100 nm diameter single bilayer vesicles (
). At this stringency most library members are inactive, allowing the identification of extraordinarily active peptides. Upon addition of screening vesicles to a plate, the rate of AMC cleavage was monitored simultaneously in all 96 wells (
). When the cleavage rate was not zero, it was linear with time. The cleavage rates in wells containing library members were compared to control wells containing ~1 nmol peptide from a pool of library peptides added to lysed vesicles either with or without protease inhibitor. These wells serve as negative and positive cleavage controls, respectively. After 30 min of incubation, the amount of terbium leakage in each well was measured in the same plate (
) and was compared to control wells with either intact or detergent-lysed vesicles serving as zero and 100% leakage controls.
Figure 3 A represented screening plate. Results for an example 96-well plate screened with the orthogonal high throughput screen. Each well contains about 5 μM of a single library peptide extracted from a single microbead, plus 1 mM lipid in the form of (more ...)
About 24,000 randomly selected sequences from the library were screened with this orthogonal high throughput approach. Because the library contains only 10,368 members, any particular sequence in the library could have been tested from once to several times, overall. Most library members (>99%) were inactive in both translocation and leakage assays (). A small number of peptides (<0.1%) had high leakage without protease cleavage, suggesting that they form large pores through which both terbium and the protease chymotrypsin can be released. An small number of peptides (<0.1%) showed high rates of AMC cleavage with little or no Tb3+/DPA leakage ( and ). The median relative cleavage rate of library members was very close to zero and the average was ~0.03. Thus the majority of peptides are not cleaved by entrapped chymotrypsin at all. As shown in , the most active translocating peptides had cleavage rates at least 100 fold higher than the mean of the library. As we described above in methods, control experiments showed that the inherent cleavage rates were constant within the library, thus these rapidly cleaved peptides have to be translocating across the membranes into the vesicles in order to be cleaved by entrapped chymotrypsin. Eighteen such spontaneously translocating, non-permeabilizing library members were identified and sequenced (). Every translocating peptide identified in the screen was successfully sequenced. Several sequences were independently found multiple times within the 18 translocating sequences, demonstrating that the translocation-active peptides are a highly unique family of peptides.
Figure 4 Screening results. 24,000 library members were screened using the orthogonal high throughput screen. Inset: Histogram of library screening results in the two axes of the orthogonal screen, leakage and chymotrypsin cleavage. More than 99% of all library (more ...)
Figure 5 Analysis of translocating sequences. A. The amino acid sequences of eighteen selected translocating peptide sequences are shown. For visualization, the identified sequences are separated by dashes into a 4-residue and 5-residue segments. The fixed C-terminal (more ...)
None of the selected membrane translocating sequences resembled any of the known cell penetrating peptides contained in the library. Instead, the selected translocating peptides contained a novel, conserved sequence motif (). Ten of the 18 sequences contained a PLIL-XXXXX-GQF or PLIY-XXXXX-GQF sequence and 10 of the 18 contained an XXXX-LRLLR-GQF motif. Two nearly identical sequences, PLIL-LRLLR-GQF and PLIY-LRLLR-GQF, comprising 7 of the 18 translocating peptides, contained both motifs. Compared to the highly cationic cell penetrating peptides such as tat or Arg9, which were present in the library but were not selected, the spontaneously translocating peptides are much less polar, with only two or three cationic residues (range = 1 to 4, most common = 2), plus the N-terminal charge, in an otherwise hydrophobic sequence. Compared to the abundance in the library overall, the basic residues arginine and lysine were significantly underrepresented in the translocating sequences while hydrophobic residues leucine and isoleucine were significantly overrepresented (). An important exception occurred at positions 6 and 9, part of the XXXX-LRLLR-GQF motif, where cationic arginine residues were found more often than expected by chance, and hydrophobes were found less often than expected by chance. The orthogonal high-throughput screen thus selected specifically for a novel, conserved membrane translocation motif that is mostly hydrophobic, with a few cationic residues at conserved positions.
Spontaneous peptide translocation across membranes
We synthesized peptide-AMC positives and measured the translocation/cleavage in the original assay compared to a random pool of library members. This experiment verified that the purified positives do, in fact, translocate under the original screen conditions, and that most library members do not. However, to fully test spontaneous membrane translocation we designed a translocation experiment with more stringent conditions than the original screen. Representative members of the translocating peptide family, including an all D-amino acid version of a translocating peptide, and a set of negative peptides ()
were synthesized. Each peptide had a C-terminal cysteine residue to which we attached a mock “cargo” moiety: 6-carboxytetramethylrhodamine (TAMRA), a zwitterionic, membrane-impermeant fluorescent dye of 430 Da molecular weight. TAMRA is larger and more polar than AMC, and thus allows us to test the peptides’ ability to deliver drug-like “cargo” across membranes. Furthermore, because TAMRA is a bright, visible-wavelength fluorophore we can use the same peptide-cargo conjugate to test translocation across both synthetic and cellular membranes by fluorescence microscopy. To conduct stringent tests of spontaneous membrane translocation we prepared multilamellar vesicles (MLV), which are 10-40 μm in diameter and have at least 10-15 concentric bilayers (
). Peptide translocation across MLVs was assessed using laser scanning confocal fluorescence microscopy. Soluble, polar probes such as free TAMRA dye and fluorescein dextran 3,000 Da (FD3) were excluded from the vesicles (
). When the selected translocating peptides and attached TAMRA cargo moiety were added to anionic multilamellar vesicles of the same lipid composition as used in the high-throughput screen (
) the peptides equilibrated across the multiple bilayers in within about 15-30 minutes ()
indicating very rapid, spontaneous translocation. In confocal images taken quickly (~10 min after addition of peptide), we could sometimes observe the last 10-20% of the transbilayer equilibration of the peptide-TAMRA conjugates suggesting a half-time of translocation on the order of 2-4 minutes (See supplemental information
for more on the translocation rate). To test the generality of the peptide translocation we also measured translocation across multibilayer vesicles composed of 100% zwitterionic phosphatidylcholine (PC). The selected peptides rapidly translocated across multibilayer PC bilayers, indicating that translocation activity is a general property of these peptides and is not dependent on specific electrostatic interactions between the peptide arginines and anionic lipids
Figure 6 Spontaneous translocation across multilamellar vesicles. A: Multilamellar vesicles were prepared by dispersing vacuum dried lipid in PBS followed by ten cycles of freezing and thawing. Vesicles were either made from 90% phosphatidylcholine (PC) + 10% (more ...)
As we show quantitatively in the translocating peptides and the covalently attached TAMRA moiety equilibrated across PC/PG bilayers and PC bilayers such that the ratio of interior to exterior fluorescence was about 1. The all D-amino acid version of TP2 (TP2D) behaved identically to TP2, indicating that there is no stereochemically specific interaction with lipids, which are chiral, in translocation. In each experiment, preformed vesicles were co-incubated with peptides and with fluorescein dextran, FD3. The dextran did not enter the vesicles measurably, giving only a background intensity ratio of about 0.1. Similarly, free TAMRA did not enter vesicles. For one negative control peptide we used the known cationic cell penetrating peptide, Arg9, which was contained in the library and was not found as a positive. We also used a randomly selected negative peptide from the screen (ONEG) and a negative peptide designed by replacing three arginine residues in the translocating peptide TP3 with aspartate (DNEG). The negative peptides did not translocate substantially across anionic or zwitterionic multibilayer vesicles even in experiments where we increased the incubation time to several hours. These translocation experiments showed that the peptides selected by the orthogonal screen have the remarkable ability to spontaneously translocate across multiple lipid bilayer membranes in just a few minutes while carrying a large polar cargo moiety.
Figure 7 Compilation of MLV translocation data. The intensity ratio is the ratio of TAMRA fluorescent intensity inside and outside multilamellar vesicles after 30 minutes of simultaneous incubation with peptide or free TAMRA plus FD3. Experiments were done at (more ...)
The confocal fluorescence experiments showed no strong peptide binding to the multilamellar vesicle membranes, which is consistent with their low predicted membrane binding18,19
. Because we observe little detectable membrane labeling, the steady state concentration of vesicle-bound peptide must be relatively low. To explain the observed overall translocation time, the rate of translocation across each bilayer must be high. We estimate that the first order translocation rate must be 0.1 sec-1
or faster. (See supplemental information
for more on translocation rates).
It might seem surprising that peptides with multiple charged residues can spontaneously translocate across bilayer membranes. However, in our recent work on antimicrobial peptides3,15,20-23
we have discussed this issue in terms of an “interfacial activity model” where interfacial activity is defined as the ability of an amphipathic peptide to partition into the membrane interface and alter the vertical segregation of the lipid polar and nonpolar groups. We have proposed that there is a broad overlap between membrane permeabilizing and membrane translocating peptides. Arginine (and perhaps also lysine) has a special role in interfacial activity because it can interact with lipid phosphate groups when inserted into membranes. As we20,21
have recently discussed, the lipid phosphates may act as chaperones for the cationic amino acids and allow their deep penetration into the bilayer, or translocation across the bilayer. Biophysical studies of the molecular mechanism of translocation are currently underway in our laboratories.
Peptide translocation into living cells
In we show that the peptides discovered here translocate rapidly across the plasma membranes of living Chinese Hamster ovary (CHO) cells at room temperature
in Dulbecco's phosphate buffered saline, which does not have a carbon source. These are conditions under which endocytosis and other active cellular processes are arrested (see below). Nonetheless, peptide translocation into cells is detectible within one minute of addition, and by ten minutes the translocating peptide and its TAMRA cargo moiety fill the cell cytoplasm with a bright, diffuse fluorescence, suggesting rapid translocation into cells. Multiple additional lines of evidence indicate that the translocating peptides described here are not entering cells by endocytosis at room temperature, and that endocytosis is inhibited under the conditions of (22°C in PBS). We know that the same peptides rapidly and spontaneously translocate across multiple synthetic lipid bilayers () on the same timescale as they enter cells, thus they have an inherent capacity for translocation that has not been demonstrated for other “cell penetrating” peptide. Furthermore, when the translocating peptides are incubated with CHO cells at 22°C in PBS, no FD3-filled endosomes are seen in the cells (). If any type of endocytosis were occurring we would have observed endosomes with entrapped FD3 which was always present in the external buffer. We also did not observe FD3 in endosomes in the absence of peptides. The classical, endocytosis-dependent5
, cell penetrating peptide Arg9 binds to cells under the conditions of (22°C in PBS), but does not enter cells at all
(see , top left), confirming that endocytosis is inhibited. Also, at 22°C in PBS, Arg9-TAMRA does not trigger the entrapment of FD3 within endosomes, showing that no endocytosis can be triggered by Arg9 under these conditions. In contrast to these observations at 22°C in PBS, we show a very different behavior for Arg9-TAMRA and FD3 incubated with cells at 37°C in full growth media (, top right). Under those conditions, Arg9-TAMRA enters cells by endocytosis, as expected, and we observe large numbers of intracellular organelles which contain co-encapsulated Arg9-TAMRA and FD3. When the translocating peptide TP1-TAMRA and FD3 were similarly incubated with CHO cells at 37°C in full growth media (, bottom right), the behavior is very different from Arg9-TAMRA and 37°C, but the same as for TP1-TAMRA at 22°C in PBS (). Even at 37°C in full growth media, the translocating peptide fills the cytoplasm with a diffuse fluorescence. No punctuate fluorescence is observed, thus there is no entrapment of peptide in endosomes. Just as importantly, there is no entrapment of FD3 in endosomes when translocating peptides and free FD3 are co-incubated with cells at 37°C in full growth media. All the translocating peptides () behave exactly like TP1-TAMRA shown in and . Furthermore, we have shown that translocation of our peptides into cells is not inhibited by metabolic poisons such as rotenone, as their behavior is identical to the behavior shown in , even when ATP synthesis is blocked (not shown). Thus, it is clear that the translocating peptides discovered here can enter cells by non-endocytotic pathways, consistent with our findings that they spontaneously translocate across synthetic lipid vesicles.
Figure 8 Example of cellular translocation. Cellular translocation was measured with living Chinese Hamster Ovary (CHO) cells that had been incubated for 30 minutes at room temperature (22°C) in phosphate buffered saline (i.e. without a carbon source) (more ...)
Figure 9 Control experiments for cellular translocation. Cellular translocation of TAMRA-labeled peptides was measured with living Chinese Hamster Ovary (CHO) cells prepared as described in and in the text. For room temperature experiments (left column, (more ...)