Spatial configuration of charge and hydrophobicity tune particle transport through mucus

Abstract

Mucus is a selectively permeable hydrogel that protects wet epithelia from pathogen invasion and poses a barrier to drug delivery. Determining the parameters of a particle that promote or prevent passage through mucus is critical, as it will enable predictions about the mucosal passage of pathogens and inform the design of therapeutics. The effect of particle net charge and size on mucosal transport has been characterized using simple model particles; however, predictions of mucosal passage remain challenging. Here, we utilize rationally designed peptides to examine the integrated contributions of charge, hydrophobicity, and spatial configuration on mucosal transport. We find that net charge does not entirely predict transport. Specifically, for cationic peptides, the inclusion of hydrophobic residues and the position of charged and hydrophobic residues within the peptide impact mucosal transport. We have developed a simple model of mucosal transport that predicts how previously unexplored amino acid sequences achieve slow versus fast passage through mucus. This model may be used as a basis to predict transport behavior of natural peptide-based particles, such as antimicrobial peptides or viruses, and assist in the engineering of synthetic sequences with desired transport properties.

Significance

Mucus is a selective hydrogel that covers all wet epithelia. Mucus selectivity is protective, restricting the passage of pathogens; however, mucus selectivity can also restrict therapeutics, challenging drug delivery. Determining how the nanoscale surface chemistry of a particle influences mucosal passage would enable predictions about the ability of pathogens to overcome the mucus barrier and would inform the design of effective therapeutics. Here, we examine the contributions of particle charge, hydrophobicity, and spatial configuration on selective passage through mucus. We find that net charge alone is a weak predictor for mucosal transport. Using our experimental findings, we develop a simple predictive model of mucosal transport that allows us to predict the transport behavior of untested peptide sequences.

Introduction

Mucus is a hydrogel that coats all wet epithelia in the human body, including the eyes and the respiratory, gastrointestinal, and urogenital tracts (1). The selective permeability of mucus protects epithelial surfaces by restricting the passage of pathogens while permitting solutes that are essential for cell survival. Mucus selectivity can also restrict the total uptake and penetration of therapeutic small molecules, peptides, and nanoparticles into the mucus layer; consequently, mucus is considered to pose a substantial obstacle to drug delivery (2,3). Thus, there is a need to understand the surface properties that determine whether a particle is permitted or restricted by mucus. This knowledge would facilitate predictions about a pathogen's ability to surmount our natural defenses and inform the design of drugs that have improved penetration or total uptake into the mucus layer.

Mucus is composed of a mixture of lipids, salts, and proteins, including the mucin glycoproteins—the main gel-forming building blocks of mucus. Each mucin is composed of an extended protein backbone that protrudes dense arrays of O-linked glycans. Mucins are considered to be negatively charged due to carboxyl and sulfate groups present in the mucin-associated glycans (1). Moreover, mucins possess hydrophobic domains that appear important for self-assembly of the gel. The charged and hydrophobic features of the mucin protein backbone and associated glycans present a plethora of potential interaction sites for diffusing particles (1).

Several groups, including our own, have examined how biochemical and physical properties of a particle, including net charge, hydrophobicity, and size impact permeability in mucus (4, 5, 6, 7, 8, 9, 10, 11), establishing the relevance of these parameters for transport. However, still missing are experiments that examine how charge and hydrophobicity impact transport when combined and displayed in distinct spatial configurations. Systematic displays of charge and hydrophobicity are interesting because they could allow us to identify, and ultimately leverage, strategies in natural mucus-interacting particles that may have evolved to enhance, or avoid, interactions with mucus (Fig. 1 A).

Figure 1.

Charged and hydrophobic peptides as probes of mucus permeability. (A) Mucins, the primary gel-forming polymers within mucus are composed of an extended protein backbone (black), densely grafted with sugars. Negatively charged sugars (red) as well as hydrophobic (green) and charged regions in the backbone present possible binding sites for particles passing through mucus. Existing studies of mucus permeability have measured the transport of uniformly charged (red, blue) or hydrophobic (green) model particles in mucus; however, mucus-interacting substrates such as viruses, bacteria, and therapeutics have surfaces composed of charged and hydrophobic regions, distributed heterogeneously. (B) Charged and hydrophobic peptides are used as probes of mucus permeability that better represent the biochemical complexity of natural mucus-interacting substances and therapeutics. The inclusion of lysine (K, blue) confers positive charge, and glutamic acid (E, red) confers negative charge. Tryptophan (W, dark green) and phenylalanine (F, green) are included to create hydrophobic regions, and asparagine (N) (gray) creates hydrophilic regions. By changing the ordered sequence of amino acids, probes with identical biochemistry but differing spatial configurations were created. (C) Schematic overview of microfluidic system used to measure the transport of peptides into mucin. Mucus is loaded into a channel, and at t = 0, fluorescently labeled peptide is added. Peptide concentration profiles in the mucin layer are monitored using fluorescence microscopy over time. (D) From profiles at 10 m, two parameters are calculated to compare the behavior of different peptides: accumulation of the peptide in the gel, which is the average peptide concentration in the mucin gel (area under the curve divided by the width of the curve), and the penetration distance, which is how far the peptide diffuses into the gel from the mucin interface (interface indicated by the dotted line).

Here we designed a small peptide library with systematically varied combinations and spatial configurations of charged and hydrophobic amino acids to identify the integrated contributions of charge, hydrophobicity, and spatial configuration on mucosal passage. Peptides were orders of magnitude smaller than the pore size of mucus, such that we interrogated the influence of the biochemical properties and spatial configuration, rather than size, on particle transport in mucus (Fig. 1 B). Our data show that incorporating hydrophobic amino acids (specifically, tryptophan or phenylalanine) into cationic peptides significantly reduces their penetration and accumulation in mucus. Our data furthermore show that the precise positioning of hydrophobic residues further fine-tuned transport of cationic peptides. In contrast, inserting the same hydrophobic amino acids into anionic peptides showed no measurable change in transport. We used our findings to develop a model that can accurately predict mucosal penetration of mixed charged and hydrophobic peptides, including certain antimicrobial therapeutics. Taken together, this work advances our understanding of mucus selectivity and provides a framework for designing therapeutics with tunable transport in mucus.

Materials and methods

Preparation of purified mucin, carboxymethylcellulose, and poly-L-lysine

MUC5AC was purified from porcine gastric scrapings as previously described (12). Briefly, mucus was harvested on ice and solubilized in sodium chloride with the addition of compounds to prevent bacterial growth and protease degradation. Solubilized mucus was then ultracentrifuged to remove insoluble material and mucins were purified using gel-filtration chromatography. Lyophilized mucins were resuspended at 1.5% (w/v) at three different salt concentrations (20 mM, 80 mM, or 140 mM NaCl, 20 mM HEPES, pH 7), with gentle shaking for 48 h at  4°C. Carboxymethylcellulose (CMC) sodium salt with an average molecular weight of 90 kDa (Acros Organics, CAS: 9004-32-4) and poly-L-lysine (PLL) with molecular weights of 30 kDa or greater (Sigma Aldrich, P9404-25MG) were dialyzed with water against a 10 kDa molecular weight cut-off and then lyophilized. Lyophilized carboxymethylcellulose and PLL were stored at −80°C until use. Lyophilized CMC and PLL were resuspended at 1.5% (w/v) (80 mM NaCl, 20 mM HEPES), with gentle shaking for 48 h at 4°C.

Preparation of fluorescent peptides

All peptides were synthesized, labeled, and purified by the Koch Institute Biopolymers and Proteomics Facility at MIT (Cambridge, MA). Peptides were synthesized by 9H-fluoren9-ylmethoxycarbonyl (FMOC) synthesis, and labeled with 6-carboxyfluorescein at the N-terminus. Singly fluorophore labeled peptides were purified using reverse-phase high-performance liquid chromatography, and matrix-assisted laser desorption ionization mass spectrometry was used to confirm identity and purity. A complete list of peptide sequences can be found in Table S1. Peptides were lyophilized and stored at −80°C. Peptides were reconstituted in buffer immediately before use in transport experiments. Lyophilized peptides were reconstituted in identical buffer to that used to reconstitute the mucin/CMC/PLL/buffer in any given experiment, so there was no salt concentration gradient between the gel and the peptide solution, which could impact transport of peptides into the gel. Peptides were used at a final concentration of 15 μM in all experiments.

Microfluidic device fabrication, experiments, and analysis

Microfluidic polydimethylsiloxane (PDMS) devices were fabricated as described in previous publications from our group (7,13). Briefly, two silicon masters were fabricated: a “channels” master to create channels for the mucin and peptides to flow through, and a “valve” master to create a valve to control mucus flow. To create the device, RTV prepolymer and curing agent (RTV515, Momentive Performance Materials, Albany, NY) were mixed (1:5 ratio (w/w) for channel wafer, 1:20 ratio (w/w) for valve wafer). The degassed RTV mixture was poured to a thickness of ∼1 cm on the channels master, and applied using a spin coater to create a thin layer on the valve master. The two wafers were cured separately at 95°C for ∼1 h, at which time the PDMS on the channels master was removed from the master, and placed on top of the thin PDMS layer on the valve wafer, aligning the valves appropriately with the channels. This was then cured at 95°C for 72 h, joining the valve PDMS layer with the channels PDMS layer. Access holes to load mucin/CMC/PLL/buffer, peptide, and to close the valve were created using biopsy punches.

This combined PDMS device was bonded to glass microscope slides using oxygen plasma treatment. Immediately after bonding, channels were washed with isopropanol, and then filled with 1 mg/mL PLL-PEG, to passivate the PDMS surface, reducing nonspecific binding and the potential effect of the charge and hydrophobicity of PDMS on the transport of peptides (14). The valve was filled with water. Before use, the PLL-PEG was washed out of the channels using buffer, and mucin/CMC/PLL/buffer was loaded into the channel. The valve, which was positioned at the bottom of the channel, >600 microns away from the mucin/CMC/PLL/buffer-peptide solution interface (so it would not appear in the fluorescent images), was closed by applying gentle pressure with a 3-mL syringe connected to the valve with polytetrafluoroethylene tubing. Fluorescent peptide solution was then loaded into the top of each channel, with the flow rate determined by gravity.

Peptide diffusion within devices was monitored for 10 m with images taken every 10 s using an inverted epifluorescence microscope (Zeiss AxioObserver Z1) with a 5× objective. To prevent photobleaching, the light source was turned off between images. All experiments were carried out at room temperature. Peptide diffusion profiles (concentration in mucin/CMC/PLL/buffer as a function of distance) in Figs. 2, 3, 4, 5, and 6 were generated in MATLAB (MathWorks, Natick, Massachusetts), by analyzing fluorescent images, assuming linearity between fluorescence intensity and peptide concentration. The fluorescence intensity within mucin/CMC/PLL/buffer in the channel was normalized to the intensity in the peptide solution. From these peptide diffusion profiles, a penetration value was calculated as the distance from the mucin/CMC/PLL/buffer interface to the point in the gel where the peptide concentration decreased to 20% of the maximum concentration of peptide in the gel. Accumulation was calculated as the area under the curve divided by the width of the curve, to give average concentration of peptide in the gel. Profiles at 10 m were used for calculation of penetration and accumulation values for all figures except Fig. 5, where the profile at 5 m was used.

Figure 2.

Both peptide hydrophobicity and net charge are important for predictions of mucosal transport. The transport of net cationic peptides K, KE.1, and KN.1, as well as cationic and hydrophobic peptides KW.1 and KF.1 into mucin gels (1.5% w/v) was monitored using fluorescence microscopy. (A) Representative fluorescent micrographs (inset) and peptide transport profiles for each peptide at 10 m are shown. All cationic peptide transport profiles display a peak, consistent with electrostatic interactions between the cationic peptides and anionic mucins; however, the transport of each peptide into mucin is different. The difference between the transport profiles of K (blue), KN.1 (gray), and KE.1 (red) confirm previous studies that assert that particle charge is important for predicting mucosal transport; however despite the same net charge, KN.1 (gray), KF.1 (medium green), and KW.1 (dark green) have substantially different transport profiles in mucin, suggesting that particle net charge alone is not sufficient for predicting mucosal passage, and that particle hydrophobicity (Table S1) is an important consideration. (B) Representative fluorescence micrographs (inset) and peptide transport profiles for net anionic peptides E, EW.1, EF.1, and EN.1 are shown. The transport profiles linearly decrease, consistent with a lack of interactions between the anionic peptides and mucin. The profiles for all four anionic peptides are nearly indistinguishable, suggesting that anionic peptides are less sensitive to proximal biochemistry than cationic peptides. From peptide transport profiles at 10 m, the (C) penetration depth and (D) total accumulation of K, KW.1, KF.1, KN. 1, and KE.1 were determined. From peptide transport profiles at 10 m, the (E) penetration depth and (F) total accumulation of E, EW.1, EF.1, and EN. 1 were determined. In (C)–(F), all replicate measurements are shown, with the average indicated by the height of the bar. All error bars represent standard deviation of replicates (n ≥ 6). (^∗) indicates a significant difference as determined by a two-sided t-test (P < 0.05, Bonferroni-corrected for multiple comparisons).

Figure 3.

Spatial configuration of charged and hydrophobic residues differentiates the transport of peptides with the same biochemical composition. Transport of the peptides with varying spatial configuration of charged and hydrophobic residues into mucin gels (1.5% w/v). (A) Representative fluorescence micrographs (inset) and peptide transport profiles for KF.1, KF.2, KF.3 at 10 m. Despite identical composition, the transport profiles of each peptide into mucin are different, suggesting that spatial configuration of charged and hydrophobic residues, in addition to net charge, is a relevant parameter for predictions of mucosal passage. (B) Representative fluorescence micrographs (inset) and peptide transport profiles for KN.1, KN.2, and KN.3 at 10 m. Rearrangement of the residues resulted in different transport behavior. (C) Representative fluorescence micrographs (inset) and peptide transport profiles for KE.1, KE.2, KE.3 at 10 m. Despite equivalent net charge, the three peptides have dramatically different concentration profiles, with a peak indicative of mucin interaction observed for KE.1, but linearly decreasing curves indicative of minimal or no interaction for KE.2, KE.3, underscoring the relevance of considering spatial configuration as well as net charge for predictions of mucosal passage. (D) Penetration depth of peptides from (A) to (C), calculated from transport profiles at 10 m. Data for KF.1, KN.1, and KE.1 are repeated from Fig. 2, for ease of comparison. (E) Total accumulation of peptides from (A) to (C), calculated from transport profiles at 10 m. Data for KF.1, KN.1, and KE.1 are repeated from Fig. 2, for ease of comparison. All error bars represent standard deviation of replicates (n ≥ 12). (^∗) indicates a significant difference as determined by a two-sided t-test (P < 0.05, Bonferroni-corrected for multiple comparisons).

Figure 4.

Ionic strength impacts cationic peptide passage through mucus. Transport of a representative subset of charged and hydrophobic peptides (KF.1, KN.1, EF.1, and EN.1) was measured in mucin at three different ionic strengths. (A) Representative transport profiles of anionic, hydrophobic peptide EF.1 (dotted line) and anionic, hydrophilic EN.1 (solid line) are indistinguishable at 20 mM NaCl (light gray), 80 mM NaCl (medium gray), and 140 mM NaCl (dark gray), suggesting that the transport of anionic peptides in mucin is not sensitive to ionic strength. (B) Total accumulation (left) and penetration depth (right) within mucin for EF.1 and EN.1 are similar at each ionic strength tested. (C) Representative transport profiles of cationic hydrophobic KF.1 (dotted line) and cationic, hydrophilic KN.1 (solid line) in mucin at 20 mM NaCl (light gray), 80 mM NaCl (medium gray), and 140 mM NaCl (dark gray). At each salt concentration, there is a substantial difference between the transport profile of hydrophobic KF.1 and hydrophilic KN.1, which is reflected in the (D) accumulation (left) and penetration depth (right). Error bars represent standard deviation of replicates (n ≥ 5) (^∗) indicates a significant difference as determined by a two-sided t-test (P < 0.05, Bonferroni-corrected for multiple comparisons).

Figure 5.

Overall charge of mucin polymer contributes to mucus selectivity. To understand the features of mucin that contribute to its selectivity, we compared the transport of (A) a subset of charged and hydrophobic peptides, KF.1, KN.1, EF.1, and EN.1 in (B) mucin and (C) carboxymethylcellulose (CMC), a polymer which like mucin is negatively charged and hydrophobic, but which does not have mucin-specific ligands for particle binding such as the glycans and peptide backbone. (B) Representative transport profiles and fluorescence micrographs for KF.1, KN.1 (left) and EF.1, EN.1 (right) in mucin. (C) Representative transport profiles and fluorescence micrographs for KF.1, KN.1 (left) and EF.1, EN.1 (right) in CMC. Profiles in mucin qualitatively agree with profiles in CMC, suggesting that general physicochemical properties contribute to mucin selectivity. (D) Representative transport profiles of KF.1, KN.1 (left) and EF.1 and EN.1 (right) in polycation poly-L-lysine (PLL) are reverse of what is observed in mucin and CMC, with little to no binding observed for cationic peptides (left) and binding for anionic peptides, with a reduction in accumulation and penetration for hydrophobic EF.1 relative to hydrophilic EN.1 (right).

Figure 6.

Charge, hydrophobicity, and spatial configuration can tune transport in mucin. A multiple linear regression model was created to predict peptide penetration depth and fit to the average penetration depth of the 15 peptides in Figs. 2 and 3. The model was used to predict the penetration of four additional peptides: (A) KF.4, which had identical amino acid composition to KF.1, KF.2, and KF.3, but a different spatial configuration, (B) KV.1, which had the same net charge and spatial configuration as peptides KF.1, KW.1, and KN.1, but a different hydrophobic amino acid, and (C) F17 and F17-6K, two antimicrobial peptides that are charged and hydrophobic. (D) Average experimentally measured penetration for each peptide compared with penetration predicted by the model. Predictions for peptides that were used to generate the model (K, KW.1, KF.1, KF.2, KF.3, KE.1, KE.2, KE.3, KN.1, KN.2, KN.3, E, EF.1, EW.1, EN.1) are in gray. Predictions for the four new peptides are in black: KF.4 (diamond), KV.1 (hexagon), F17 (triangle), F17-6K (star). (E) Schematic representation of peptide transport in mucin, where arrow length represents the penetration depth of a given peptide into a mucin layer, and arrow area represents the total accumulation. (F) Representative transport profiles in mucin for all peptides tested (profiles are the same as those shown in Figs. 2 and 3, but repeated here in the same plot to allow for easy comparison) demonstrate that net charge, net hydrophobicity, and spatial configuration of charge and hydrophobicity can be used to tune the penetration and total uptake of a peptide of interest within mucus, delivering a range of transport behaviors from the extremes of those displayed by an entirely cationic peptide (K; maximally interacting with mucus) and those exhibited by an entirely anionic peptide (E, minimally interacting with mucus).

Model to predict penetration

We fit a linear model to the average penetration values measured for each peptide tested in Figs. 2 C, E, and 3 D (K, KW.1, KF.1, KF.2, KF.3, KE.1, KE.2, KE.3, KN.1, KN.2, KN.3, EF.1, EN.1.EW.1, E)

$$y={\beta }_0+{\beta }_{1}c+{\beta }_{2}h+{\beta }_3{v}_p+{\beta }_4{v}_n+{\beta }_5{v}_{unch}$$

where:

• y is the average penetration depth of the peptide

• c is the net charge of the peptide

• h is the average peptide hydrophobicity (calculated using the Wimley-White scale (15))

• v_p is the positive charge variance, calculated as the variance of the indices of R or K residues within each peptide. For example, for the peptide “KANANARANAK,” v_p = var(1,7,11) = 25.3, while for the peptide “NKRKNN,” v_p = var(2,3,4) = 1.

• v_n is the negative charge variance calculated as the variance of the indices of E or D residues within each peptide.

• v_unch is the complex uncharged residue variance calculated as the variance of N, F, W, or V residues within the peptide.

We noted that although penetration depth did not have an obvious lower bound, it was bounded above by the value for a freely diffusing peptide. Instead of a simple regression model, we therefore trained a censored regression model, in which measurements are bounded on one side, using the censReg package in R. Specifically, we set the upper bound slightly lower than the average penetration depth of peptides in PBS.

Statistical analysis

For each peptide, n ≥ 5 independent replicates were performed. Independence refers to transport evaluated in different batches of purified mucins/CMC/PLL, on different days. In the bar charts in Figs. 2, 3, 4, S2, and S3, for each peptide, all accumulation and penetration replicate measurements are shown, with the bar in each chart displaying the mean ± SD, plotted using GraphPad Prism. Comparisons between the penetration or accumulation of different peptides were performed using a two-tailed t-test. P < 0.05, adjusted with the Bonferroni correction to account for the large number of comparisons, was considered statistically significant.

Results

Peptides as probes to identify parameters that tune mucus permeability

To create probes that were positively or negatively charged, we synthesized peptides that include cationic lysine (K) residues or anionic glutamic acid (E) residues, respectively. To investigate the influence of hydrophobicity on transport, we inserted the hydrophobic residues phenylalanine (F) and tryptophan (W) into our peptides. By varying the specific position of these charged and hydrophobic residues within the peptide, we evaluated the influence of biochemical spatial configuration on mucosal transport. All peptides were synthesized and fluorescently labeled with 6-carboxyfluorescein at the N-terminus.

Peptide transport through mucus was measured using a microfluidic system previously developed in our lab (7,13) (Fig. 1 C). Briefly, the system consists of a channel loaded with a mucin gel. At t = 0, fluorescent peptide is loaded into the top of the channel, and peptide transport into the gel is monitored using time-lapse microscopy. From fluorescence images, profiles of the peptide concentration in mucin as a function of distance from the mucin interface (Fig. 1 C) are generated. The area under the profile and the width of the profile are measured and used to calculate two parameters: the total accumulation or uptake of each peptide within the mucin layer and the penetration depth into the mucin layer (Fig. 1 D). As a defined model system for the mucus barrier, we used hydrogels consisting of purified mucin polymers, which have previously been used in studies of mucus permeability (7,8,16,17). In this work, we focus on mucin MUC5AC, which is one of the two main gel-forming mucins found in the lungs, and which serves as a simplified, yet biochemically defined, environment to model the challenges faced for pulmonary delivery.

Previous studies investigated how charge of a particle impacts its diffusion through mucus, measuring the transport of cationic and anionic small molecules and nanoparticles (5, 6, 7, 8,16). To validate the use of peptides as probes of mucus permeability, we first quantified the transport of cationic (KN.1) and anionic (EN.1) 20-mer peptides in mucin (1.5% w/v reconstituted in buffer) and mucin-free buffer. Cationic KN.1 consists of seven lysine residues followed by three hydrophilic asparagine residues, separated with alanine spacers (Fig. 2 A). Anionic EN.1 is identical in sequence to KN.1, except that glutamic acid replaces lysine (Fig. 2 B). Peptides were introduced in the microfluidic channel at 15 μM. In Fig. 2 A, we see that the concentration profile of KN.1 in mucin displayed a peak just inside the interface of the peptide solution and the mucin layer (interface indicated with the dotted line) that was absent in the concentration profile of KN.1 in buffer alone (Fig. S1), suggesting electrostatic binding between the cationic peptide and the anionic mucins. In contrast, anionic peptide EN.1 exhibited a steadily decreasing peptide concentration within mucin (Fig. 2 B), with no substantial difference in transport in the presence or absence of mucin (Fig. S1). These findings suggest a lack of interaction between EN.1 and mucin that is mediated by electrostatic repulsion between the anionic mucins and anionic peptides, consistent with the findings of previous studies (5,8), including our own (6,7,13). To increase or decrease the net charge of KN.1, the terminal asparagine residues were replaced with lysine or glutamic acid, yielding peptides K and KE.1, respectively (Fig. 2 A). Both K and KE.1 exhibited peaks in mucin (Fig. 2 A) that were absent in buffer alone (Fig. S1), indicative of electrostatic interactions between the peptides and mucin. However, the penetration depth of K into the mucin layer was significantly reduced relative to KN.1 (Fig. 2 C), consistent with stronger electrostatic interactions resulting from increased net charge. KE.1 exhibited a significant increase in penetration relative to KN.1, which is consistent with weaker electrostatic interactions resulting from KE.1's lower net charge (Fig. 2 C). Taken together, these results confirmed previous reports that the net charge of a peptide influences its transport through the mucin barrier, validating the use of peptides as reporters for probing mucus permeability.

Inclusion of hydrophobic residues in cationic peptides affects their penetration and accumulation in mucus

Although net charge is clearly relevant for mucosal transport, we next asked how the inclusion of hydrophobic residues might alter transport. This is a relevant question as many natural antimicrobial peptides are amphipathic, containing both hydrophilic, charged amino acids and hydrophobic amino acids. We designed amphipathic peptides for which we exchanged the three terminal asparagine residues for hydrophobic phenylalanine and tryptophan residues, creating cationic and hydrophobic peptides KF.1 and KW.1, respectively. As with KN.1, the concentration profiles of KF.1 and KW.1 in mucin exhibited a peak (Fig. 2 A) that was absent in buffer (Fig. S1), indicating interactions between peptides and mucin. However, KF.1 and KW.1 exhibited significantly reduced penetration (Fig. 2 C) and accumulation (Fig. 2 D) in mucin relative to KN.1. These data show that hydrophobic amino acids incorporated in cationic peptides can reduce mucopenetration and total accumulation relative to a hydrophilic particle with the same number of cationic residues. Further, these data suggest that charge alone is not sufficient for predictions of mucosal transport, as KF.1, KW.1, and KN.1 were equivalent in net charge, yet had significantly different transport behaviors within mucin.

Considering the ability of hydrophobic residues to tune transport of net cationic peptides through mucin gels, we tested if a similar tunability is observable for net anionic peptides. The transport of entirely anionic E exhibited a steadily decreasing peptide concentration in the presence (Fig. 2 B) or absence (Fig. S1) of mucin, consistent with a lack of interaction between E and mucin mediated by electrostatic repulsion. In contrast to what was observed for the cationic peptides, there was no significant difference between the transport profiles of anionic E, anionic and hydrophobic EF.1 and EW.1, and anionic and hydrophilic EN.1 in mucin (Fig. 2 B), which is reflected in a lack of significant difference in penetration depth (Fig. 2 E) and total accumulation (Fig. 2 F) for these peptides. These data demonstrate that the contribution of negative charge to transport dominates over proximal hydrophobic biochemical features as measured by the presented readouts.

The spatial positioning of hydrophobic residues within cationic peptides can further fine-tune transport properties

We next tested if the detailed spatial distribution of hydrophobic amino acids affects peptide transport (Fig. 3 A–C). The residues in KF.1 were rearranged, positioning the hydrophobic residues as a continuous block in the center of the peptide (peptide KF.2) or as three smaller blocks distributed over the length of the peptide (peptide KF.3) (Fig. 3 A). Despite identical amino acid composition, KF.1, KF.2, and KF.3 had transport profiles that were different from each other (Fig. 3 A). For peptide KF.2, in which the continuous block of hydrophobic residues at the end of KF.1 were repositioned to the center of the peptide, we observed a significant increase in mucopenetration (Fig. 3 D) and accumulation (Fig. 3 E), relative to KF.1. When the hydrophobic residues were distributed from a continuous block to single residues arranged over the length of the peptide, as in peptide KF.3, mucopenetration (Fig. 3 D) and accumulation (Fig. 3 E) were further increased relative to KF.1 and KF.2, confirming that spatial configuration is an important determinant for mucosal transport.

Building on this result, we tested the effect of spatial distribution of residues on transport using peptides in which hydrophobic phenylalanine was exchanged for “guest” residues with different side chain chemistry: hydrophilic asparagine (Fig. 3 B; KN.1, KN.2, KN.3) and anionic glutamic acid (Fig. 3 C; KE.1, KE.2, KE.3). Mirroring in part what we observed for KF.1, KF.2, and KF.3, we detected a significant increase in the penetration depth (Fig. 3 D) and accumulation (Fig. 3 E) of KN.2 and KN.3 compared with KN.1. When we substituted asparagine for glutamic acid (KE.1, KE.2, and KE.3; Fig. 3 C), we again observed an increase in mucopenetration (Fig. 3 D) and total accumulation (Fig. 3 E) for KE.2 and KE.3, compared with KE.1. The difference between the transport profiles of KE.1, KE.2, and KE.3 (Fig. 3 C) was especially pronounced. Despite the same net charge, the transport profile of KE.1 exhibited a peak in mucin-containing samples that was absent in buffer, which is indicative of binding interactions, whereas the transport profiles of KE.2 and KE.3 in mucin (Fig. 3 C) and mucin-free buffer (Fig. S1) were nearly indistinguishable, suggesting a near lack of interaction between these peptides and mucin, despite net positive charge. Taken together, these data may suggest that smaller, distributed blocks of positive charge increase mucus uptake and penetration relative to a peptide in which the charged residues are configured as a longer continuous block. Furthermore, these data demonstrate that selectivity of the mucin barrier is sensitive at the nanoscale, distinguishing among spatial configurations of charged and hydrophobic residues in peptides with the same amino acid composition. This result further underscores that mucosal transport cannot be predicted well by consideration of the net charge of a particle alone.

Ionic strength modulates mucus permeability to cationic peptides

In health, the ionic strength of the mucus barrier varies with anatomical location; changes in the ionic strength of mucus are hypothesized to be a mechanism by which the body can rapidly tune selectivity (18,19). Specific niches in the body, such as cervical mucus, experience systematic, cyclical changes in ionic strength (20,21). Additionally, in the context of diseases such as bacterial infections, the ionic composition of the mucus layer can be altered away from the healthy baseline (18).

To test the effect of ionic strength on mucus permeability to charged and hydrophobic substrates, we compared the transport of a representative subset of our charged and hydrophobic peptide probes (KF.1, KN.1, EF.1, and EN.1) in mucin at three ionic strengths that span the range of ionic strengths reported for mucus layers in the literature (22, 23, 24, 25, 26, 27, 28, 29): 20 mM NaCl, 80 mM NaCl, and 140 mM NaCl. The transport profiles of anionic peptides EF.1 and EN.1 (Fig. 4 A) were nearly indistinguishable at each salt concentration tested, which was reflected in the similar total accumulation (Fig. 4 B, left) and penetration (Fig. 4 B, right) recorded for these peptides at each salt concentration evaluated. In contrast, examining the profiles of cationic KF.1 and KN.1 (Fig. 4 C), as ionic strength increased from 20 to 140 mM, total uptake and mucopenetration increased for both peptides, which is consistent with increased electrostatic screening between the peptides and mucins. At each salt concentration, KF.1 exhibited significantly less uptake (Fig. 4 D, left) than KN.1. Together, these data demonstrate that the interactions between positively charged peptides and mucins, and their resulting effects on transport (Fig. 4 C and D), are much more sensitive to ionic strength than negatively charged peptides (Fig. 4 A and B). The consistent difference between the transport of KF.1 and KN.1 at various salt concentrations suggests that particle surface biochemistry beyond net charge is an important consideration for accurately predicting mucosal transport in different mucus layers in the body as well as in various disease contexts.

Anionic character of mucin polymer contributes to mucus selectivity

To identify the features of mucin gels that contribute to their selective permeability, we compared the transport of a representative subset of our charged and hydrophobic peptide probes (KF.1, KN.1, EF.1, and EN.1) (Fig. 5 A) in mucin (1.5% w/v) (Fig. 5 B) and high molecular weight carboxymethylcellulose (CMC) (1.5% w/v) (Fig. 5 C). Like mucin, CMC is a polymer that is composed of anionic and hydrophobic domains; however, it lacks mucin-specific features such as O-linked glycans and the protein backbone. Transport profiles for the four peptides in CMC (Fig. 5 C) qualitatively recapitulated transport in mucin (Fig. 5 B), with a peak observed for both KF.1 and KN.1 (Fig. 5 B, left; Fig. 5 C, left), and reduced penetration and accumulation observed for cationic, hydrophobic KF.1 relative to hydrophilic KN.1 (Fig. S2 A). Meanwhile, there was no substantial difference between the profiles of anionic peptides EF.1 and EN.1 (Fig. 5 B right; Fig. 5 C, right). These data suggest that mucin's selectivity to charged and hydrophobic particles is driven in part by general physicochemical features of the mucin polymer as opposed to specific, receptor-ligand type interactions between mucins and nearby particles. The similarity between peptide transport in CMC and mucin (Fig. 5 B and C) further suggests that our conclusions about transport in mucin may be applicable to other polyanionic selective barriers. We were curious to understand whether our findings about the impact of charge and hydrophobicity on peptide transport could further extend to polycationic selective barriers. We hypothesized that transport of the same four peptides (KF.1, KN.1, EF.1, and EN.1) in a model polycation such as PLL (Fig. 5 D) would be the opposite of that observed in the polyanions mucin and CMC. Indeed, in PLL (1.5% w/v), we detected no difference between the transport profiles of the cationic peptides (Fig. 5 D, left) (whereas in mucin (Fig. 5 B, right) and CMC (Fig. 5 C, right), there was no difference between the profiles of the anionic peptides). Meanwhile, anionic peptides EF.1 and EN.1 had substantially different transport profiles in PLL (Fig. 5 D, right), with reduced penetration and accumulation observed for hydrophobic EF.1 relative to the hydrophilic KN.1 (Fig. S2 B). This observation parallels what was observed in mucin (Fig. 5 B, left) and CMC (Fig. 5 C, left) for cationic hydrophobic KF.1 relative to hydrophilic KN.1. These data confirmed our hypothesis and highlight the potential for applying our findings about gel selectivity broadly to polyelectrolyte contexts beyond mucin.

Peptide charge, hydrophobicity, and spatial configuration can be used to predict mucopenetration for novel peptides

Taken together, our data suggest that although peptide transport in mucin is partly determined by net charge, the spatial configuration of charge within the peptide (Fig. 6 A) and the specific biochemistry of noncharged residues in the peptide (Fig. 6 B) can modulate transport. To evaluate this conceptual framework, we generated a multiple linear regression model to predict peptide penetration depth in mucin based on net biochemical properties of a peptide as well as the configuration of residues within a peptide. Our regression model was a censored regression (also known as Tobit regression), in which prediction of penetration depth cannot go above a set level. We chose censored regression while formulating our model because penetration depth is upper bounded by the depth achieved by a freely diffusing peptide. To capture overall properties of the peptide, we included peptide net charge and average hydrophobicity as predictors. In addition to these predictors, we included two additional predictors that capture spatial configuration of residues within the peptide. These spatial configuration predictors were computed using the variance of the position of charged and uncharged residues in each peptide sequence (Materials and methods).

We then fit this model to the average penetration depth (Figs. 2 C, E, and 3 D) of the 15 cationic and anionic peptides we had previously measured (K, KW.1 KF.1, KF.2, KF.3, KN.1, KN.2, KN.3, KE.1, KE.2, KE.3, EF.1, EW.1, EN.1, E). Consistent with our finding that alteration of net charge induces substantial differences in peptide transport (Fig. 2), the fit coefficients for our model show stronger effects for net charge than for any other predictor (Table S2, standardized coefficients). However, peptide hydrophobicity and positive charge variance are also important fitting parameters in our model (Table S2), which is consistent with the qualitative observations we made from our experimental results.

To test the predictive ability of our model, we experimentally measured the penetration of four additional peptides (Figs. 6 A–C) as a test set (KF.4, KV.1, F17-6K, and F17) and compared these measurements with the model's predictions (Fig. 6 D). The sequence of KF.4 was designed by rearranging the positively charged lysine and hydrophobic phenylalanine residues in peptide KF.1, in order to test how well our model predicted novel spatial configurations of charge and hydrophobicity (Fig. 6 A). KF.1, KF.2, KF.3, and KF.4 all had the same net charge and average hydrophobicity; however the position of hydrophobic and charged residues within each peptide was different. The charge variance of KF.4 was larger than KF.1, and similar to KF.2 and KF.3 (Fig. S3 A), so we predicted that KF.4 would have increased penetration relative to KF.1, similar to what was observed for KF.2 and KF.3 relative to KF.1 (Fig. 3 A and D). Indeed, this is what we observed experimentally (Fig. S3 A) with a 6% difference between the predicted and measured penetration for KF.4 (Fig. 6 D). KV.1, which included hydrophobic residue valine, was designed to test our model's ability to extrapolate to hydrophobic residues that were not present in the set of peptides used to build the model (Fig. 6 B). KV.1 had the same net charge and spatial configuration of residues as KF.1, KW.1, and KN.1; however, the net hydrophobicity of each peptide was different. The average peptide hydrophobicity for KV.1 fell between KF.1 and KN.1 (Fig. S3 B), so we predicted that KV.1's penetration into mucin would be increased relative to KF.1. We found that this was the case (Fig. S3 B), with a 13% difference between our prediction and measurement for this peptide (Fig. 6 D). Last, we tested predictions for two cationic and hydrophobic antimicrobial peptides, F17-6K and F17 (30), because these sequences are biologically of interest as potential therapeutics, and because they have compositions that are more “natural” (Fig. 6 C), compared with the peptides we used to build the model, which all share a composition of charged and/or hydrophobic residues alternating with alanine residues. Additionally, similar to KF.1 and KF.2, F17-6K and F17 are identical in composition, but differ in spatial configuration of amino acids in the peptide (Fig. S3 C). We hypothesized that F17 would penetrate further into mucin than F17-6K, as we observed for KF.2 relative to KF.1. There was good agreement between the predicted and measured penetrations for these peptides, with a 5% and 17% percent difference for F17-6K and F17, respectively (Fig. 6 D). As predicted, we found that F17, which had multiple short stretches of cationic residues, penetrated further into mucin than F17-6K, which had one long stretch of cationic residues (Fig. S3 C). Taken together, these data support our conceptual framework that mucus permeability is determined by more than the average properties of the peptide, and in a potentially a predictable fashion, which may allow for data-driven design or modification of peptides for optimal mucus penetration.

Discussion

The selective permeability of mucus protects wet epithelia from microbial pathogens and other harmful particles, but also presents a significant obstacle for therapeutics, which can poorly penetrate or accumulate in the mucus layer. Despite the implications that mucus permeability has for human health, this topic is poorly understood. Although a particle's net charge is a parameter commonly considered to influence transport through mucus, in this work, we sought to elucidate the fundamental biochemical parameters that govern passage through the mucus barrier, considering the contributions of charge, hydrophobicity, and spatial configuration.

By measuring the transport of a panel of peptides with systematically varied biochemical features (Fig. 1) in mucin gels, we determined that although net charge strongly influences transport through mucin, it does not entirely explain the transport of peptides in mucin. For example, we found that hydrophobic surface features proximal to charged surface features reduced the mucopenetration and total uptake of cationic peptides (KF.1 and KW.1), relative to a hydrophilic particle with the same net charge (KN.1) (Fig. 2). The significant difference in transport between KF.1 (which contained hydrophobic phenylalanine) and KW.1 (which contained hydrophobic tryptophan) (Fig. 2) suggests that mucin selectivity is sensitive to the specific chemistry of hydrophobic amino acids. When we compared the transport of positively charged particles with the same net charge and net hydrophobicity, we further determined that the spatial configuration of these features influenced transport. As the length of continuous charged stretches was reduced, penetration and accumulation increased (Fig. 3 A), although no rearrangement of charged and hydrophobic residues permitted a positively charged peptide to penetrate into mucin as far as a negatively charged peptide. This is consistent with electrostatic interactions between the positively charged peptides and the negatively charged mucins, with positively charged peptides slowed down by binding, whereas negatively charged peptides diffuse more easily with no binding. Taken together, these data suggest that although net charge is a major determinant of mucus permeability, mucus selectivity is sensitive to the nanoscale properties of a particle. The determination of which particles are restricted or permitted by the mucus layer goes beyond net biochemical properties, and is influenced by spatial configuration as well as specific chemistry. This interpretation is consistent with observations of particle transport in other selectively permeable polyelectrolytes, such as cartilage (31, 32, 33) and the nuclear pore (34).

We replicated our findings about mucin (Fig. 5B ) in CMC (a model polyanion; Fig. 5C) and detected the opposite behavior in PLL (a model polycation; Fig. 5D), suggesting that general physicochemical properties contribute to mucus' selectivity. This observation highlights the possible application of our data to other polyelectrolytes that are obstacles to drug delivery. For example, cartilage (32), the extracellular matrix of tumors (35,36), and the extrapolymeric substances of bacterial biofilms (37) are polyelectrolyte gels that constitute substantial obstacles to drug delivery; we anticipate that our findings may be applicable in these contexts. By developing a statistical inference model with net charge, average hydrophobicity, and spatial configuration of uncharged and charged residues as predictor variables, which was fit to experimentally measured penetration values reported in Figs. 2 and 3, we were able to make strong predictions about the penetration of four novel charged and hydrophobic peptides in mucus. Our observations suggest that peptide properties may dictate mucus permeability in a predictable way, which could enable rational modification of peptides for optimal mucus distribution or binding (Fig. 6 A–D).

This study expands our understanding of the parameters that control passage through mucus. This information may inform new strategies for effective drug delivery through mucus barriers. Existing strategies can be roughly divided into two categories: those that reduce interactions between therapeutics and mucus components (3,38, 39, 40) (mucoinert formulations) and those that exploit strong interactions (mucoadhesive formulations) (41,42). However, strategies to achieve a transport profile between these two extremes are more limited (43). Critically, the previous optimization of drug delivery in other biological polyelectrolytes, such as cartilage (31), suggests that transport behaviors between the extremes of strong interaction and no interaction may be important for effective drug delivery. Our measurements of the entire series of charged and hydrophobic peptides that we tested suggest that charge, hydrophobicity, and spatial configuration constitute three knobs that can be turned to engineer the penetration and total uptake of a peptide of interest within mucus, delivering a range of transport behaviors (Fig. 6 E and F) from the extremes of those displayed by an entirely cationic peptide (K; maximally interacting) and those exhibited by an entirely anionic peptide (E, minimally interacting with mucus). We anticipate that our findings will guide efforts to identify particle surface coatings that will close the gap between these two extremes. Understanding whether the principles that we determined in this work with peptides apply at larger length scales (protein, nanoparticle) as well as determining the mechanisms that underlie differences in transport between peptides represent interesting avenues of investigation for future work.

Author contributions

All authors contributed to the design of experiments, and interpretation of data. T.S. purified mucins and conducted all transport experiments and analysis. J.W. developed the linear predictive model. K.R. and A.J.G. supervised the study. T.S. wrote the manuscript; all authors helped edit the manuscript.

Editor: Samrat Mukhopadhyay.