Parallel screening using the chloroalkane penetration assay reveals structure-penetration relationships

Abstract

The efficiency with which polycationic peptides penetrate the cytosol depends on the number and overall patterning of arginine residues. While general trends and unusually penetrant patterns of arginine residues have been discovered, prior work has not effectively leveraged high-throughput screens to measure cytosolic penetration, rather than total cell uptake. In this work, we adapted the chloroalkane penetration assay, which exclusively measures cytosolic penetration, to screen peptide libraries in a high-throughput, quantitative, and automation-ready manner. We demonstrate the usefulness of the screening platform by efficiently exploring how the number, patterning, and stereochemistry of arginine residues affect the cytosolic penetration of a model 10-residue peptide.

TOC Graphic

Introduction

Some of the most exciting recent developments in drug discovery focus on large-molecule therapies such as peptides, proteins, oligonucleotides, and gene-editing complexes. For large-molecule drugs with intracellular targets, cytosolic penetration is often the greatest barrier.¹ Large molecules with poor intrinsic cell penetration can be transported into the cytosol using a cell-penetrating peptide (CPP). The most common CPPs are short, arginine-rich peptide sequences 5-30 amino acids in length.² CPPs have become a common method for preclinical evaluation of large-molecule therapeutics due to the ease of synthesis and cargo attachment. Over two decades of research have explored mechanisms of CPP-mediated transport.³

A growing body of work is focused on applying modifications to bioactive peptides to increase their intrinsic cell penetration, without the use of a discrete CPP. Modifications that are used for improving cytosolic penetration include masking backbone amides with N-methyl groups, peptide stapling, amphipathic patterning, and patterning of guanidinium groups.^4–7 However, the process of discovering which modifications will increase cell penetration when substituted in which positions is not straightforward, and is typically accomplished through trial-and-error. A handful of screening strategies have emerged to more systematically study how different modifications at different positions, or combinations thereof, affect cell penetration. In a landmark effort, Kodadek and co-workers employed a high-throughput method to screen combinatorial libraries of peptides, cyclic peptides, and peptoids using a steroid label and a luciferase reporter assay.⁸ Schepartz and co-workers recently adapted a related assay for a large-scale RNAi screen.⁹ In another study, a total of 46 dye-labeled, hydrocarbon-stapled peptides were screened for cellular uptake by Walensky and co-workers using high-content fluorescence microscopy.⁶ Inspired by these library screens, in this work we adapt the previously described chloroalkane penetration assay, which has unique advantages for measuring cytosolic penetration, to screen combinatorial libraries of peptides.

The chloroalkane penetration assay (CAPA) has been used by our group and others to test the cytosolic penetration of a large variety of peptides, peptidomimetics, and small molecules.^10–18 CAPA uses a simple pulse-chase format with HeLa cells that express HaloTag protein in the cytosol.^19,20 When a chloroalkane-tagged molecule is applied in the pulse step, it can penetrate into the cytosol and covalently bind HaloTag. After wash steps, the cells are chased with a cell-permeable, chloroalkane-tagged dye. Subsequent measurements by flow cytometry determine the amount of HaloTag-bound dye in the cells, a signal that is inversely proportional to the amount of chloroalkane-tagged molecule that accessed the cytosol during the pulse step.

Dose-dependent CAPA data can be fit to an IC₅₀ curve and an IC₅₀-like value can be extracted, called a CP₅₀ (concentration at which 50% HaloTag occupancy is observed).²¹ Dose-dependent CAPA data and associated CP₅₀ values have provided quantitative comparisons of cytosolic penetration for small collections of peptides and small molecules.^10–16 However, using CAPA as a single-point assay for screening larger numbers of molecules, such as a split-and-pool combinatorial library, has not been reported. Here, we adapt CAPA as an efficient method for screening a library of arrayed peptides in parallel. This method was used to explore structure-penetration relationships of a model 10-mer peptide with respect to patterning L-arginine and D-arginine residues within its sequence. The CAPA screen is generalizable and readily amenable to automation, opening the door to wider exploration of sequence-penetration relationships for large-molecule therapeutics.

Results and Discussion

First, we optimized library synthesis and cleavage protocols for compatibility with CAPA in an arrayed library format. The optimized protocol used a high-loading 300 μm polystyrene aminomethyl hydroxymethylbenzoic acid resin. After sidechain deprotection, single beads were arrayed into individual tubes and peptides were cleaved using an adapted vapor cleavage protocol with ammonium hydroxide.²² This protocol allowed for the recovery of chloroalkane-tagged peptide from a single resin bead, without extraction steps, and resolubilization in an aqueous solution that could be used directly in CAPA (Figure 1A; see Supporting Information for detailed methods).

Figure 1. Library design, preparation, and CAPA screen reproducibility.

(A) Schematic of CAPA screen workflow. (B) Individual sequences for the homochiral Library 1 and the heterochiral Library 2. Library members are numbered from 1 to 64 with an increasing number of arginines and patterning shifting from the C to N terminus. All peptides had a chloroalkane tag on their N-terminus. (C) Reproducibility of technical replicates was examined by screening 108 single beads from Library 1 on two separate days. Reproducibility of sequence replicates was examined by comparing 82 sequences from both libraries that were observed multiple times among the arrayed library members.

We chose a model peptide, L1, that derives from the Beclin-1 protein.¹⁰ When fused to the CPP Tat, the resulting “Tat Beclin-1” peptide induces autophagy when applied to cultured cells at 10 μM.^10,23 Tat Beclin-1 has been used in many applications and remains an important reagent for studies of the therapeutic benefits of autophagy induction.^23–25 In prior work, we showed that a 10-mer truncated version of the Beclin-1 peptide retained some autophagy-inducing activity at 50 μM.¹⁰ This 10-mer sequence corresponds to chloroalkane-tagged peptide L1 in this work, which we used as a parent peptide to address sequence-penetration relationships independent of biological function.

First, we verified that unpurified L1 cleaved from a single bead produced CAPA data that matched data from bulk synthesized and purified L1 (Figure S1). This provided confidence that reliable CAPA data can be obtained from unpurified material derived from a single bead. Next, we prepared two libraries compatible with arrayed CAPA screening. Peptide libraries were designed to keep residues Trp2, Phe6, and Ile7 of L1 constant. These residues are important for the biological activity of Beclin-derived peptides,^10,23 but we also retained them because hydrophobic residues are known to be critical for the cytosolic penetration of poly-arginine peptides.^26,27 The C-terminal Histidine was also not varied to allow for consistent first-residue loading, and thus more consistent overall yields among library members. Thus, six positions were varied in each library. Two libraries were prepared. Library 1 is a homochiral library which represents all possible permutations of the six positions substituted with L-arginine (L1 through L64, Figure 1B). Library 2 is a heterochiral library which represents all possible permutations of the same six positions substituted with D-arginine (het1 through het64, Figure 1B). Libraries 1 and 2 were designed to provide an investigation into how the number and patterning of L- and D-arginines within an all-L-amino-acid peptide would affect cytosolic penetration.

After 384 individual beads were selected, cleaved, and arrayed, we used mass spectrometry to identify 58 unique sequences from Library 1 and 57 unique sequences from Library 2. Of the 115 library members isolated, 85% of the parent ion masses were non-redundant, allowing for direct identification using the parent ion mass using MALDI-MS (Tables S1–S3). The remaining 15% were unambiguously identified using tandem mass spectrometry (Figure S2). It is unknown whether the 13 peptides that were not detected had poor synthetic yields or were undetected for another reason. Quality was assessed on a subset of library members using LC-MS, and individual purities ranged from 60 to 85% (examples are shown in Figure S2). The primary byproduct in most syntheses was peptide lacking the N-terminal chloroalkane, which was expected to introduce a small bias into concentration determination but not CAPA signal. The 115 library members were arrayed in 96-well format with controls (Figure S3), in stocks that were concentration-normalized by UV spectroscopy. Before the full library screen, test screens were performed on up to eight single-bead samples from each library to find the optimal screening concentration for that library. These tests indicated that the optimal dynamic range for each library would be achieved by testing Library 1 at a nominal concentration of 0.125 μM and Library 2 at a nominal concentration of 2.5 μM (Figure S4).

Many of the sequences were identified in multiple single-bead samples. This allowed us to compare both technical replicates (same single-bead stock tested in independent trials, Figure 1C) as well as sequence replicates (different single-bead stocks of the same sequence tested in the same trial, Figure 1C). Both controls demonstrated excellent reproducibility, with R² values of 0.89 and 0.87 for the technical replicates and the sequence replicates, respectively.

The results from screening Library 1 are shown in Figure 2A. Broad trends include a general increase in cytosolic penetration with increasing arginine content, as expected from prior work on arginine-containing peptides.^3,28 Within this homochiral library, peptides containing only one or two arginines were uniformly among the less-penetrant peptides, and peptides containing four to six arginines were among the more-penetrant peptides (Figure 2A, lower panel). Individual peptides stand out as exceptions to this trend, such as L43 which has 4 arginines but was among the less-penetrant library members tested, and L38 which has 3 arginines but was among the more-penetrant library members tested. Interestingly, a general trend of increasing cytosolic penetration with increasing arginine content was not observed for Library 2, which has heterochiral backbones (Figure 2B). These observations are consistent with our initial tests which suggested that Library 2 should be screened at a higher nominal concentration than Library 1. One possible reason for the apparent lesser cytosolic penetration of heterochiral peptides is increased susceptibility to degradation for homochiral L-peptides, which could lead to false positive CAPA signal. While we cannot rule this out completely, previous work has shown little evidence of degradation effects on CAPA data, as measured with diverse linear and stapled peptides and canonical CPPs with similar lengths and overall sequences.^{10,11,14,16,17} The poorer cytosolic penetration by heterochiral peptides is consistent with prior work showing that display of arginine side chains in a structured array, such as on one face of an α-helix, contributes to efficient cytosolic penetration and/or endosomal escape.^7,16,29 While these peptides are too short to maintain secondary structures at room temperature, optimal binding to membrane components could be promoted by a structured conformation such as an α-helix.^7,16 If this is the case, the heterochiral library would be generally expected to have poorer helical propensity and thus poorer membrane binding and/or endosomal escape. Still, within Library 2, a subset of peptides appeared more penetrant than others with the same number of D-arginines, such as het43 (Figure 2B). These data suggest that there are optimal arrangements of arginine residues within short peptides even in the context of a heterochiral backbone.

Figure 2. Results from CAPA library screens.

(A) Results from screens of Library 1 are shown in order of sequence number L1 to L64 (upper panel), and also in order of average percent HaloTag occupancy (lower panel). (B) Results from screens of Library 2 are shown in order of sequence number het1 to het64 (upper panel), and also in order of average percent HaloTag occupancy (lower panel). Open circles denote individual data points, and filled circles denote averages of two or more independent data points. Colors correspond to the number of arginines in each peptide, as denoted in Figure 1.

Library 1 and Library 2 produced interesting general trends, but also some outliers (some only tested once) that showed unusually high or low cytosolic penetration. Thus, we sought to check individual chloroalkane-tagged peptides by synthesizing representative sequences, purifying them, and testing them in full dose-dependent CAPA experiments (Figures 3 and S5).^11,21 CP₅₀ values for individually synthesized and purified chloroalkane-tagged peptides are shown in Table 1. Some peptides that were only observed once in the screen, such as L19, had CP₅₀ values that differed from the relative value predicted by the screen. Others, like L38, had CP₅₀ values more consistent with their single-point data. These results highlight the importance of screening multiple replicates of each sequence to ensure reproducibility. We note that scale-up with automated liquid handling would largely avoid these discrepancies, thus further improving the reproducibility and predictive power of the CAPA screen.

Figure 3. Dose-dependent CAPA data for individually synthesized and purified library members.

(A) Comparison of dose-dependent CAPA data show selected homochiral peptides with improved cytosolic penetration compared to parent sequence L1. (B) Comparing diastereomers revealed L54 and L62 followed the expected trend where the homochiral sequence was more cell-penetrant, and het43 represents an exception where the heterochiral peptide was more cell-penetrant. (C) Comparing peptides with the same number of arginines but different Halotag occupancy in the CAPA screens, we observed L52 is more cell-penetrant than L43 and het14 is more cell-penetrant than het12. For all dose-dependent CAPA data, individual trials are shown in Figures S5–S6. Data points are averages of three independent trials and error bars show standard error of the mean. Colors correspond the number of arginines in each peptide, as denoted in Figure 1.

Table 1.

CP₅₀ values for selected library members, individually synthesized and purified and tested in dose-dependent CAPA.

Library Member	Sequence*	CP50 (μM)
L1	VWNATFHIWH	1.2 ± 0.1
L52	RWRRTFHIRH	0.25 ± 0.01
L60	RWRARFRIRH	0.26 ± 0.01
L47	RWNRTFRIRH	0.26 ± 0.02
L62	RWRRRFHIRH	0.31 ± 0.0
L38	RWRATFRIWH	0.45 ± 0.02
L64	RWRRRFRIRH	0.51 ± 0.01
L41	RWRARFHIWH	0.54 ± 0.04
L19	RWNARFHIWH	1.0 ± 0.03
het43	VWNrrFrIrH	1.3 ± 0.4
L54	RWNRRFRIWH	1.7 ± 0.1
L43	VWNRRFRIRH	2.8 ± 0.4
het62	rWrrrFHIrH	4.0 ± 0.2
het58	VWrrrFrIrH	5.9 ± 0.6
het54	rWNrrFrIWH	6.0 ± 0.7
het14	VWNrTFrIWH	6.3 ± 0.8
het12	rWNATFHIrH	10.5 ± 0.6

^*All peptides had an N-terminal chloroalkane tag and an amidated C-terminus. CP₅₀ values are reported as the average and standard error of the mean of CP₅₀ values derived from independent curve fits to data from three independent CAPA experiments. Averaged data are shown in Figure 3, and all independent trials are shown in Figures S5–S6. r denotes D-arginine.

One purpose of the high-throughput CAPA screen was to identify variants that have increased cytosolic penetration compared to the parent sequence. Testing individually prepared peptides confirmed eight different sequences with two to six arginines that have lower CP₅₀ values than that of L1 (Table 1 and Figure 3A). Also, there were some examples of peptides that have the same number of arginines but different cytosolic penetration. For example, the CAPA screen suggested L52 was more penetrant than L43 despite the fact that both peptides both have four L-arginines, two in the same position. When tested as individually synthesized, purified peptides, the CP₅₀ of L52 was roughly ten-fold lower than that of L43 (Figure 3C). One contributing factor might be that L52 has an extra His residue compared to L43, increasing its net charge. In another example, peptides het12 and het14 both have two D-arginines, one of which is in the same position, but the CAPA screen predicted that het14 was unusually penetrant. This prediction was confirmed by the CP₅₀ values for individually synthesized and purified peptides (Table 1 and Figure 3C). These comparisons reveal unusual differences in cytosolic penetration that can be attributed to the relative positions of only one or two arginine or D-arginine residues. These differences could not have been anticipated without the data from the CAPA screen.

The dose-dependent CAPA experiments also confirmed that the homochiral peptides generally had lower CP₅₀’s than their heterochiral diastereomers (compare L54 and het54, and L62 and het62, Table 1 and Figure 3B). As described above, we attribute this trend to a lower propensity to form secondary structures upon membrane binding, which may be required for maximal endosomal uptake and/or release.^7,16 One exception to this trend was heterochiral peptide het43, which was slightly more cytosolically penetrant than homochiral peptide L43 (Figure 3B). het43 was unusually cytosolically penetrant compared to other related peptides as well: het43 has four D-arginines, and yet is more cytosolically penetrant than het54 (four D-arginines), het58 (five D-arginines), and het62 (five D-arginines). It may be that het43 can more readily assume a structured conformation when bound to membranes than other heterochiral polyarginine peptides. As an initial test of this hypothesis, we used circular dichroism (CD) spectroscopy to investigate the secondary structure of L43, L54, and L62, and their heterochiral diastereomers het43, het54, and het62. In aqueous conditions, all peptides showed CD spectra indicating random coil structure, as expected for such short peptides. When spectra were acquired in 30% trifluoroethanol to promote α-helical structure, all homochiral peptides showed α-helical signatures to varying extents (Figure S7A). In 30% trifluoroethanol, heterochiral peptides het54 and het62 had CD spectra indicating predominantly random coil and beta-strand character, but het43 had a CD spectrum indicative of α-helical structure (Figure S7B). This lends evidence to the hypothesis that het43 is more cytosolically penetrant than other heterochiral peptides due to an intrinsic propensity to adopt an α-helical secondary structure, which could allow for more optimal display of guanidinium groups for interaction with biological membranes.^7,16

In conclusion, we report the adaptation of the chloroalkane penetration assay to screening spatially arrayed combinatorial libraries. The screening platform is broadly applicable for small molecules, peptides, proteins and oligonucleotides – any molecule that is amenable to attachment of a chloroalkane tag. It is also fully compatible with automated liquid handling. Day-to-day and compound-to-compound reproducibility was high, and automation would further increase reproducibility. The CAPA screen identified homochiral and heterochiral guanidinium-rich peptides with unanticipated trends in cytosolic penetration. This method will allow drug developers to efficiently map sequence-penetration relationships and identifying more cell-penetrant analogs of high-value peptides and peptidomimetics. While the work here focused on guanidinium-rich peptides, prior applications of CAPA to stapled peptides,¹⁰ peptoids,¹⁸ PROTACs,¹² and other compounds¹³ demonstrate that CAPA screens can be used to optimize cell penetration for combinatorial libraries of a large variety of compounds. In addition, it is well-documented that the nature of the “cargo” affects the efficiency of delivery by CPPs such as Tat, penetratin, polyarginine, and newer CPPs that employ structured arrays of guanidinium groups.^1–3,9,11 The CAPA screen could also be used to optimize the combinatorial problem of choosing a CPP, choosing an attachment site, and choosing a linker for optimal delivery of a specific cargo.

Methods

Methodological details are provided in the Supporting Information.