Chromatographic Determination of Permeability-Relevant Lipophilicity Facilitates Rapid Analysis of Macrocyclic Peptide Scaffolds

Abstract

Hydrocarbon-determined shake-flask measurements have demonstrated great utility for optimizing lipophilicity during early drug discovery. Alternatively, chromatographic methods confer reduced experimental error and improved handling of complex mixtures. In this study, we developed a chromatographic approach for estimating hydrocarbon–water shake-flask partition coefficients for a variety of macrocyclic peptides and other bRo5 molecules including PROTACs. The model accurately predicts experimental shake-flask measurements with high reproducibility across a wide range of lipophilicities. The chromatographic retention times revealed subtle conformational effects and correlated with the ability to sequester hydrogen bond donors in low dielectric media. Estimations of shake-flask lipophilicity from our model also accurately predicted trends in MDCK passive cell permeability for a variety of thioether-cyclized decapeptides. This method provides a convenient, high-throughput approach for measuring lipophilic permeability efficiency and predicting passive cell permeability in bRo5 compounds that is suitable for multiplexing pure compounds or investigating the properties of complex library mixtures.

Introduction

Membrane permeability is a prerequisite for accessing intracellular targets and is a desirable property in drug candidates intended for oral delivery. Most oral drugs are thought to passively diffuse across the cell membrane, a process driven by physicochemical properties such as molecular size and lipophilicity.^1,2 In general, increasing lipophilicity leads to enhanced permeability; however, for larger molecules in particular, the lipophilicity required for achieving permeability tends to result in poor aqueous solubility.³ Thus, medicinal chemists require accurate, high-throughput assessments of lipophilicity as one of the key properties to monitor in early drug discovery programs.

The pursuit of challenging, previously “undruggable” targets has expanded lead discovery efforts toward larger molecules that often violate Lipinski’s Rule of 5 for predicting oral bioavailability. Macrocyclic peptides are prototypical molecules in this “beyond Rule of 5″ (bRo5) chemical space, offering promising leads against challenging targets enabled by advancements in encoded library technologies such as mRNA display and DNA-encoded libraries.⁴⁻⁶ Despite their potent binding affinities, these leads are often impermeable and require significant optimization due to the inherent polarity of the peptide backbone and all but a few of the natural amino acid side chains.⁷

In contrast, many cyclic peptide natural products demonstrate passive cell permeability without exceeding lipophilicity limits by sequestering hydrogen bond donors, particularly backbone NH groups, in the membrane’s low dielectric environment through steric occlusion or intramolecular hydrogen bonding (IMHB).⁸ This observation has led to a growing number of empirical^9,10 and computational screening^11,12 approaches for discovering novel backbone geometries, or scaffolds, that exploit intramolecular interactions to achieve membrane permeability. Since lipophilicity is conformation-dependent and can be difficult to predict computationally, especially for larger molecules, these approaches can benefit from experimental assessments of lipophilicity and membrane permeability that are robust, reproducible, and relatively high-throughput.

A compound’s lipophilicity can be assessed empirically by its partition coefficient in a biphasic shake-flask system, such as n-octanol/water, expressed as the logarithm of the partition coefficient (Log P_oct/w) or the distribution coefficient at a given pH (Log D^7.4_oct/w). Although algorithms for predicting Log P_oct/w, such as cLogP and ALogP, have gained widespread use as calculated 2-dimensional lipophilicity descriptors, in terms of membrane permeability, the octanol–water system under-penalizes solvent-exposed hydrogen bond donors (HBD) and thus often overestimates permeability.¹³ Instead, purely hydrocarbon solvents (e.g., 1,9-decadiene, hexadecane, and cyclohexane) have been noted for their ability to better capture the desolvation penalty associated with exposed HBD.¹⁴ Since this desolvation penalty can be highly dependent on a compound’s low dielectric conformation, experimental hydrocarbon–water partition coefficients are preferable to algorithmically derived 2-dimensional descriptors. This is especially true for larger compounds and macrocycles in particular, which exhibit complex conformations.

Although shake-flask methods are generally considered the gold standard measurement of lipophilicity and are amenable to some degree of miniaturization and cassette-sampling, chromatographic lipophilicity methods enable even greater throughput with simple automation.¹⁵ Most established chromatographic tools for determining lipophilicity, such as ElogD¹⁶ and ChromLogD,^17,18 as well as more recently published methods such as AlphaLogD,¹⁹ focus on correlating retention times of small molecule drugs to their shake-flask octanol–water partition coefficients. Other chromatographic systems have been developed to provide more biomimetic models of membrane transport, including the use of immobilized artificial membranes (IAM)²⁰ for the stationary phase and the use of multiple isocratic conditions to investigate environment-dependent conformational effects (Chamelogk).²¹ Recent progress has been made to develop conditions to replicate hydrocarbon-derived shake flask lipophilicity, including a study by Caron et al. which demonstrated that an isocratic method using a column with a pure polystyrene-divinylbenzene matrix can produce capacity factors strongly correlated with algorithmically determined toluene-water partition coefficients (Log P_tol/w).²² Additionally, Jensen et al. estimated hydrocarbon–water partition coefficients using a variety of columns and both isocratic and gradient techniques.²³ These studies primarily focused on classic Ro5 small molecule drugs which, in general, lack the conformational heterogeneity exhibited by macrocycles and other bRo5 compounds.

Previously, we observed strong correlations between 1,9-decadiene-derived shake-flask lipophilicities (Log D_dd/w) and passive permeability (artificial or cell-based) for a variety of bRo5 drug and peptide model systems,²⁴⁻²⁶ highlighting the measurement’s sensitivity to the strong permeability impact of solvent-exposed H-bond donors. Experimental Log D_dd/w values, when combined with the calculated ALogP “bulk lipophilicity” descriptor, allowed derivation of a metric termed lipophilic permeability efficiency (LPE),²⁴ in which the permeability-relevant lipophilicity (Log D_dd/w) is compared to the solubility-relevant lipophilicity (ALogP) to capture the efficiency with which a particular compound utilizes its lipophilicity to achieve passive permeability. Increases in LPE for compounds with sufficient aqueous solubility yield opportunities for greater permeability in both PAMPA and cell-based permeability assays such as Caco-2 and MDCK. A LPE derivation from chromatographic measurements (cLPE) would yield improvements in convenience, throughput, and dynamic range over the shake-flask methodology, and this has already proven to be feasible in a recent study done by Watanabe et al.²⁷ Here we expand upon the work of Watanabe et al., by introducing a more general cLPE model which can be applied across multiple peptide scaffolds and other bRo5 compounds. Based on a data set of cyclic peptides and other bRo5 compounds, we conclude that Log D_dd/w and LPE can be predicted with high accuracy using chromatographic methods and that the resulting values accurately predict membrane-relevant lipophilicity for a variety of bRo5 systems.

Results and Discussion

Nonlinear Regression Model between Capacity Factor and Partition Coefficients from Diverse Peptide Macrocycles

To investigate the relationship between chromatographically determined lipophilicity and hydrocarbon shake-flask lipophilicity, we synthesized a library of diverse cyclic heptapeptides (Figure 1a, JT-7) in which both stereochemistry and side-chain lipophilicity were varied. The library was synthesized using split-and-pool solid phase peptide synthesis methods and was mass-encoded to facilitate multiplexed library screening (Figure S14, Scheme S1). JT-7 spans a wide ALogP range and contains 64 unique stereoisomeric scaffolds. Hydrocarbon shake-flask partition coefficients for each compound were determined between 1,9-decadiene and PBS (pH 7.4). From this library, we observed 188 compounds with high-fidelity Log D_dd/w data as determined by the ability to accurately quantify their presence in both the aqueous and organic phases by LCMS.

Figure 1.

Nonlinear regression between measured shake-flask lipophilicity and chromatographically determined lipophilicity. A) Structure and property data of cyclic peptide libraries. Circles in JF-10c and AF-10a-b represent either a peptoid or a N-methylated alpha amino acid. B) Relationship between shake-flask lipophilicity and chromatographic lipophilicity. Points are colored by cyclic peptide library (1a). C) Relationship between estimated shake-flask lipophilicity (Log ED_dd/w) and measured shake-flask lipophilicity (Log D_dd/w) of test set compounds.

Using JT-7, we first set out to determine the conditions that produced the strongest correlation between retention time and hydrocarbon shake-flask values. We determined chromatographic lipophilicities (LogK’) using two isocratic and two gradient methods with different C18 stationary phases (Figure S2), including traditional silica-backed columns as well as a polystyrene-backed, fully apolar C18 matrix (PRP-C18). Although silica-backed columns have been used extensively to estimate octanol–water partition coefficients, there is evidence to suggest that polystyrene-backed column matrices would afford measurements of lipophilicity that more closely resemble hydrocarbon–water partition coefficients due to the lack of silanol groups in the polymer matrix.^23,28,29

We observed a strong correlation between retention time and Log D_dd/w values for all methods and columns tested (R² > 0.88) (Table S1). Similar performance was achieved between the two silica-C18 columns and the PRP-C18 column, with the silica-C18 yielding marginally better correlations in some cases. Interestingly, at higher lipophilicities (Log D_dd/w > 1), we noted a deviation from linearity in the correlation for isocratic methods, despite maintaining a high R².

To increase the structural diversity of the training set, two cyclic decapeptide libraries (JF-10c, AF-10a-b) were included in addition to JT-7 (Figure 1a). AF-10a-b were previously synthesized and studied in our lab³⁰ whereas JF-10c was synthesized specifically for this study with traditional split-and-pool solid phase peptide synthesis methods (Scheme S2). These decapeptides feature unique cyclization patterns and are, on average, larger and more lipophilic than the heptapeptides in JT-7. Combined, the three libraries totaled 273 cyclic peptides, and 28 compounds (∼10% from each library) were left out of the model to be used as a test set (see Supporting Information for full compound information). LogK’ values were measured for all training and test set compounds using the PRP-C18 and one silica-C18 column under both isocratic and gradient conditions. Shake-flask Log D_dd/w values for JF-10c and AF-10a-b were measured in triplicate. For the training set compounds, the relationships between LogK’60 or LogK’20–100 and Log D_dd/w using either PRP-C18 (Figure S3) or silica-C18 (Figure S4) columns were fit to both exponential and linear functions. The exponential fit of the isocratic 60% method (Figure 1b) resulted in the best predictions of Log D_dd/w from LogK’ values for the test set compounds (R² = 0.97 PRP, 0.96 silica). This model predicts 1,9-decadiene/water shake-flask partition coefficients via eq 1,

1

where Log ED_dd/w is the estimated distribution coefficient derived from chromatographically determined lipophilicity. This model accurately estimates Log D_dd/w values for the test set (RMSD = 0.356) from a single chromatographic measurement (Figure 1c).

The PRP-C18 column was selected for further experiments in this study based on its affordability and relatively low back pressure, resulting in a longer column lifetime. Additionally, we pursued the isocratic 60% method because column equilibration is not required between runs, making this method more suitable for higher-throughput applications. Due to the low observed back-pressure, we investigated utilizing higher flow rates on PRP-C18 to reduce analysis times (Figure S5) and obtained an analogous exponential regression model with nearly identical predictive power of Log D_dd/w (R² = 0.96). Under higher flow rate conditions, the most lipophilic peptides in this study eluted within 30 min. Pairing higher flow rates with the analysis of mass-unique mixtures facilitated rapid assessments of lipophilicity, with more diverse mixtures resulting in over 60 Log ED_dd/w measurements per hour. As medicinal chemistry campaigns tend to be driven by pure compound investigations, we also compared our multiplexed capacity factor measurements on the long column to single compound injections on a short column under identical isocratic conditions (Figure S6). We observed very strong correlations (R² = 0.986), suggesting that this model can be adapted for facile analysis of pure compounds. While variation of these chromatographic conditions (column length, flow rate, isocratic or gradient program) results in significantly different capacity factors, transformation to Log ED_dd/w produces convergent values regardless of the methodology. This allows one to tailor the chromatographic method to the property of the analyte and produces lipophilicity values which are standardized between methods (comparable to preexisting shake-flask values).

Lipophilic Permeability Efficiency Determined by Chromatography

We next sought to determine the ability of this chromatographic technique to produce meaningful LPE scores that are both reflective of known structural differences (e.g., absence or steric occlusion of exposed HBD) and increases in passive membrane permeability. Watanabe et al. recently reported a fast-gradient method for the chromatographic determination of LPE on a library of thioether cyclized peptide macrocycles.²⁷ A good correlation was observed between cLPE and passive permeability in pig kidney epithelial cells. The cLPE values reported in their work were obtained by transforming Log D_dd/w in the original LPE equation to LogK’ via a simple linear regression of 11 structurally similar peptides with the same extent of N-methylation in the backbone. Despite the reported strong correlations between Log D_dd/w and LogK’, it is still unclear if these methods penalize exposed hydrogen donors to the same degree and if LPE and cLPE scores are comparable across diverse peptide structures.

To investigate the ability of our chromatographic system to reproduce LPE scores derived from 1,9-decadiene/water shake-flask partition coefficients, we measured the LogK’60 of the hexapeptide libraries from which the original LPE model was developed (Figure 2a).²⁴ These four hexapeptide scaffolds (MN-6a-d) vary in their number of hydrogen bond donors that are exposed in low dielectric media, with all other NH groups sequestered in IMHB or sterically occluded.^31,32 Log ED_dd/w and ALogP displayed strong linear correlations for all compounds within each scaffold set. Additionally, class separation, as reflected by the difference in y-intercept, was observed between the scaffolds which generally reflected their previously determined LPE relationships (Figure 2b).

Figure 2.

Lipophilic permeability efficiency determined by chromatography (cLPE) for bRo5 compounds. A) Structure of the cyclic hexapeptide library with exposed hydrogen bond donors colored in orange. B) Linear regressions of Log ED_dd/w and ALogP for each peptide class. Open circles have no exposed HBD (class A), squares have one side-chain HBD (class B), stars have one backbone HBD (class C), and X’s have two exposed HBD (class D). Compounds are binned by ALogP (<3, blue; 3–4, green; > 4 orange). C) Structure of PROTAC model compounds. D) cLPE plot comparing amide to ester matched pairs. Compounds are colored by scaffold (esters, green; amides, blue).

A minimization was performed as previously described²⁴ to calculate a common slope (1.01) between each of the hexapeptide classes and standardize the y-intercepts within classes. Next, the intercept of the linear regression equation for the class with the most exposed hydrogen bond donors (MN-6d, 2 HBD) was determined to produce eq 2.

2

In this equation, cLPE is defined as the lipophilic permeability efficiency determined chromatographically, with the lowest cLPE class from this data set (MN-6d) fixed to zero. The constants for this equation were consistent between the isocratic and gradient methods tested (Figure S7).

The 1,9-decadiene/water partition coefficient estimations for this data set were in agreement with measured shake-flask values for MN-6a (RMSD = 0.403) and MN-6c (RMSD = 0.522) but were less accurate for both MN-6b (RMSD = 1.153) and MN-6d (RMSD = 1.313). The Log ED_dd/w values for these poorly predicted sublibraries were consistently much higher than their respective Log D_dd/w values. This result may suggest that side-chain phenols, or hydrogen bond donor moieties in general, may incur a much higher desolvation penalty in the shake-flask experiment with 1,9-decadiene, whereas solid-phase partitioning into the C18 is more accommodated. It is unlikely that this difference could also be attributed to the relative acidity of the side-chain phenol (pK_a = ∼ 10) and the difference in buffer operating pH between shake-flask measurements (7.4) and chromatography (∼2.5). In fact, we observed no measurable difference in retention time for these compounds when we performed the chromatography at pH 7 (see Figure S8). On the contrary, hydrogen bond donors in the peptide backbone (MRN-6a and MRN-6c) seem to have a similar effect on measured lipophilicities in both assays. Despite the noted differences, our chromatographic model produces cLPE values which are both representative of differences in exposed hydrogen bond donors and strongly predictive of passive permeability in PAMPA (see Figure S9).

We next investigated a second class of bRo5 compounds, proteolysis targeting chimeras (PROTACs), whose structure-permeability relationships have been investigated previously.²⁶ This compound set (VK-P) features two series of five lipophilicity-scanning model compounds, each bearing a VH032 warhead with either an amide or ester linkage, resulting in a difference of one hydrogen bond donor between the two scaffolds. We measured the LogK’60 of each compound and transformed these values to Log ED_dd/w (Figure 2c). Encouragingly, for the matched pairs in this data set, we obtained an average difference in cLPE (ΔcLPE) of 1.5 between the esters and the amides (Figure 2d), which is identical to our previously reported LPE difference derived from shake-flask values (ΔLPE = 1.5). Additionally, this ΔcLPE is very similar to the cLPE difference observed for addition or removal of a backbone HBD in the MN6-a-d data set (ΔcLPE = 1.7). The absolute cLPE values of the PROTACs are more reflective than the shake-flask derived LPE values of the predicted solvent exposure of two HBD (the −OH of the Hyp residue and the C-terminal -NH, with the -NH of the Tle being sterically occluded as previously shown). This discrepancy could be related to the protonation of the ionizable thiazole nitrogen in our chromatographic system (pK_a = 4.06), however we observed no difference in retention time when using a buffered aqueous solvent (Figure S8).

Accurate comparisons of cLPE across diverse systems require a consistent slope between Log ED_dd/w and ALogP (m_lipo) for a given series of compounds with the same degree of HBD exposure (for the hexapeptide model system used to derive the original LPE equation, m_lipo = 1.01) (eq 2). To determine whether this relationship holds true for larger macrocycles, we calculated the m_lipo for 12 distinct peptide scaffolds and the PROTACs for various chromatographic systems and shake-flask partition coefficients (see Table S2). Encouragingly, the m_lipo was consistent across all 12 peptide scaffolds and PROTACs for both chromatographic (s.d. = 0.204) and shake-flask (s.d. = 0.160) methods. Additionally, we calculated projected Log ED_dd/w values for each of the liposcan scaffolds (except those from MN-6a-d libraries) utilizing the m_lipo from the cLPE equation (1.01) and the averaged y-intercepts of each scaffold (Figure S10). The projected Log ED_dd/w values for all 8 scaffolds were in strong agreement with their measured values (RMSD_avg = 0.178), indicating that cLPE can be accurately compared across diverse peptide scaffolds.

Log ED_dd/w Predicts PAMPA and MDCK Passive Permeability for Scaffold Diverse Thioether Decapeptides

We next investigated whether chromatographically determined Log ED_dd/w values could be used to predict passive cell permeability in a library of stereodiverse thioether decapeptides, JF-10a (Figure 3a).³³ This library consisted of 96 unique scaffolds with a liposcan element built into the peptoid position, resulting in seven stereoisomeric clusters. We were unable to obtain 1,9-decadiene/water shake-flask lipophilicity data on these compounds due to the propensity of the thioether to oxidize in the 1,9-decadiene layer, but LogK’60 data was collected on the entire library using the methods described above. As expected, we observed a wide range of lipophilicity across the seven sets of stereoisomers, identified by vertical groupings of compounds with identical ALogP values (Figure 3b). Gratifyingly, as Log ED_dd/w increased within a set of stereoisomers, PAMPA permeability generally increased. This relationship becomes nonlinear and even negative when extending the lipophilicity of the analytes outside the soluble limits of the permeability assay, which yields an inverse parabolic relationship. To investigate whether this trend was consistent across all stereoisomeric clusters, the relationship between PAMPA permeability and Log ED_dd/w was examined in closer detail (Figure S11). All seven stereoisomeric clusters had a strong linear correlation between Log ED_dd/w and PAMPA LogPe (average R² = 0.815) with an average ΔLogPe/ΔLogED of 1.13 (s.d. = 0.223). These results indicate that increases in Log ED_dd/w produce similar increases in PAMPA permeability for compounds well-within the bounds of aqueous solubility.

Figure 3.

cLPE analysis of stereodiverse thioether-cyclized decapeptides from JF-10a. A) Structure of the library. B) Relationship between Log ED_dd/w and ALogP. Compounds are colored by PAMPA LogPe in five evenly distributed bins of Lowest (−8.5 < LogPe < −7.8), Low (−7.8 < LogPe < −7.1), Medium (−7.1 < LogPe < −6.4), High (−6.4 < LogPe < −5.7), and Highest (−5.7 < LogPe < −5.1). C) Relationship between MDCK cell permeability and Log ED_dd/w. Compounds are colored by Log ED_dd/w (<2, blue; > 2, green). The linear regression for compounds with a Log ED_dd/w < 2 is displayed with a black line (R² = 0.754).

Within this data set, we identified several matched pair peptides which differ by only 1–2 stereocenters and exhibit a 50 to 100-fold increase in PAMPA permeability, which is reflected by a comparable increase in Log ED_dd/w (see Table S3). The selected matched pairs highlight the ability for cLPE to identify significant conformational changes which arise from seemingly small structural alterations, such as a stereochemical inversion of a single amino acid, which remains difficult to predict in silico.¹¹

To investigate if these trends hold true for cell-based permeability assays, we compared our Log ED_dd/w values to reported MDCK permeability values for selected compounds from the thioether library (Figure 3c). These data produced the expected reverse-parabolic relationship that has been previously observed in a variety of systems.^9,24,34 We observed an excellent correlation (R² = 0.754) between MDCK permeability and Log ED_dd/w for this set of compounds within the observed soluble regime (Log ED_dd/w < 2). The ΔLogPe/ΔLogED for these data is 1.385, which is slightly higher than the average increase we observed in PAMPA. This data set features a matched pair, which differs by inversion of a single stereocenter yet shows a nearly 10-fold difference in MDCK permeability, which is reflected by their Log ED_dd/w values. These findings underscore the strong influence of cLPE values on cell-based permeability, indicating that early cLPE assessment could greatly assist in prioritizing lead compounds during drug discovery.

PAMPA and Log ED_dd/w Trends across Diverse Peptide Macrocycles

For shake-flask measurements, we often observe a positive correlation between Log D_dd/w and PAMPA permeability for compounds residing in the soluble regime of lipophilicity. Furthermore, PAMPA and cell-based permeabilities have been shown to correlate well for a wide variety of bRo5 compounds (R² > 0.70).^2,9,25 To analyze the relationship between Log ED_dd/w and aqueous solubility constraints in PAMPA, we compared the PAMPA LogPe and Log ED_dd/w for 280 different cyclic decapeptides from four distinct peptide libraries featuring different cyclization schemes (Figure 4a). These peptides were designed based on the decapeptide scaffold introduced by Fouche et al.³⁵ and altogether comprise 104 unique scaffolds. The parabolic relationship between LogPe and Log ED_dd/w was consistent across most libraries analyzed, with the positive linear portion of the curve existing between Log ED_dd/w −2 and 2. PAMPA permeability tends to level off and start decreasing around Log ED_dd/w values of 2, presumably due to decreasing aqueous solubility or other recovery losses (Figure 4b). The triazole-containing compounds in the JF-10c library were more permeable than their thioether congeners in the JF-10a library at the same Log ED_dd/w (and Log D_dd/w), despite good correlations within each compound class. The difference in PAMPA behavior between the two chemotypes, despite their comparable lipophilicities, suggests that correlations between Log ED_dd/w and permeability may need to be determined independently for different chemical series in a medicinal chemistry program.

Figure 4.

Relationship between PAMPA permeability and Log ED_dd/w for scaffold-diverse decapeptides. A) Structures of decapeptide libraries. Circles in JF-10a-c represent either a peptoid or an N-methylated alpha amino acid. B) Relationship between PAMPA permeability and Log ED_dd/w. Compounds are colored in accordance with their library ID.

Library AF-10b maintains higher PAMPA permeability at high Log ED_dd/w values compared to its congener, Library AF-10a, consistent with the known differences in their chameleonic behavior.³⁰ Whereas both compounds exhibit a fully hydrogen-bonded, cross-beta conformation in low-dielectric solvents, the location of a peptoid residue in one of the beta strands of Library AF-10b destabilizes the folded state of this scaffold in aqueous solution, giving rise to its chameleonic behavior. In contrast, Library AF-10a, with a peptoid in one of the beta turns, is fully folded in both polar and apolar media.

To further investigate the hypothesis that Libraries AF-10a and AF-10b differ with respect to their chameleonicity, we subjected them to an adapted form of a recently reported chromatographic technique for investigating chameleonicity, Chamelogk (Figure S12).²¹ Notably, we utilized the PRP-C18 column from this study instead of the reported PLRP-S column. Library AF-10b has a higher average Chamelogk (1.081) than AF-10a (0.816), in support of our previous observations.³⁰ This additional flexibility may be the driving force behind AF-10b having a “right shifted” permeability curve, as the peak permeability for this scaffold is achieved at higher Log ED_dd/w values. These results highlight the potential for chromatographic techniques to accurately assess both permeability-relevant lipophilicity and molecular flexibility, which is critical for rational design of large peptides macrocycles.³

The Compounds in This Study Represent the Chemical Space Occupied by Known Cyclic Peptides

We wanted to determine the extent to which the compounds in this study (Table 1, Figure S1) represent the broader chemical space defined by published cyclic peptides of known membrane permeability. Therefore, we calculated seven physicochemical descriptors (MW, LogP, TPSA, number of HBD, HBA, rotatable bonds and aromatic rings) and Morgan fingerprints for the compounds herein and compared them to those of the Cyclic Peptide Membrane Permeability Database (CycPeptMPD),³⁶ which consists of 7344 cyclic peptides from 47 publications. Based on a principal component analysis of these collected features, there was very good overlap between the feature space represented by our compounds and those in the database (Figure S13), suggesting that our method should be applicable across the larger class of previously studied cyclic peptides.

Table 1. Summary of Compound Libraries Used Throughout This Study.

Libray	Subclass	Cyclization	N	ALogP	MW
AF-10a	Macrocycle	Amide	18	0.39–4.72	987–1197
AF-10b	Macrocycle	Amide	18	0.39–4.72	987–1197
JF-10a	Macrocycle	Thioether	192	1.33–2.42	967–1028
JF-10b	Macrocycle	Thioether	36	1.19–4.26	1965–1075
JF-10c	Macrocycle	Triazole	54	0.36–4.26	914–1104
JT-7	Macrocycle	Amide	188	0.22–5.04	645–1011
MN-6a-d	Macrocycle	Amide	33	0.66–6.18	578–792
VK-P	PROTAC	-	10	1.20–6.60	615–796

Log ED_dd/w Captures Permeability-Relevant Conformational Changes

To further validate that measured cLPE differences are due to differences in exposed polar surface area or exposure of hydrogen bond donors in low dielectric solvent, we conducted McMD simulations in explicit cyclohexane (a more convenient molecule for simulations with electronic character grossly similar to 1,9-decadiene) for two thioether decapeptide matched pairs (from data set JF-10a) which differ by one stereocenter. Both exhibit a cross-beta conformation and feature transannular hydrogen bonding typically observed in beta sheets. However, SB60 experiences a slight twist in the turn of the β sheet due to the difference in stereochemistry about leucine 6 (Figure 5, 6d). This twist in the backbone appears to disrupt the transannular hydrogen bond between the Leu-NH (6d) and the carbonyl of the leucine across the ring (3a). The top three populated clusters of SB60 conformations are instead predicted to contain weak, long-range hydrogen bonds from the Leu-NH (6d) to both the leucine (3a) and proline (4a) carbonyl. In contrast, across all conformational clusters, SB64 consistently exhibits a strong, short-range 6d-3a hydrogen bond. SB64 likely manages to achieve a higher effective lipophilicity by adopting conformations which consistently occlude the 6d Leu-NH in a strong, stable transannular hydrogen bond. Careful control of hydrogen bond donors is essential to optimize bRo5 peptide macrocycles and is often modulated by N-methylating amide protons in the peptide backbone. However, these backbone NH protons can play significant roles in target binding and excessive N-methylation can lead to poor solubility. Log ED_dd/w values obtained under these chromatographic conditions can identify privileged scaffolds which manage to occlude exposed backbone NHs.

Figure 5.

McMD structures of diastereotopic, thioether-cyclized matched pairs in cyclohexane. Each plot (X1-X3) represents a conformational cluster, where the intensity of the color reflects the relative occurrence of an IMHB across all structures within a cluster. The lowest RMSD frame from each X1-X3 conformational cluster is overlaid for SB60 (purple) and SB64 (orange). Dashed lines represent intramolecular hydrogen bonds which are also represented as green squares in the accompanying donor–acceptor plots.

Conclusion

Despite the simplicity of standard shake-flask methods, these chromatographic methods increase ease, throughput, reproducibility and automatability while still permitting multiplexed analysis of complex mass-encoded compound mixtures. The Log ED_dd/w values derived from our model are strongly predictive of measured Log D_dd/w values for a diverse test set of peptide macrocycles, representative of the broader chemical space generally occupied by this class of compounds (Figure S13). Additionally, we have also shown that various chromatographic methods, including isocratic and gradient approaches, can generate predictive regression models for 1,9-decadiene/water partition coefficients of structurally varied peptide macrocycles. This enables comparisons of Log ED_dd/w values across different chromatographic techniques. The ΔcLPE values produced from this model are comparable to ΔLPE values previously published which reflect known structural changes, chiefly exposure of hydrogen bond donors. As previously discussed,²⁴ the above correlation and cLPE derivation from well-understood macrocycle analytes places a convenient “flagpost” within bRo5 chemical space, but the approach herein is easily applied to other compound collections as needed. Acquiring cLPE scores early in drug discovery campaigns identifies scaffolds with opportunities for increased permeability to access intracellular targets or oral delivery.

Analyzing stereoisomer matched pair peptides from our data set effectively demonstrates the value of cLPE. SB60 and SB64 are two closely related diastereomer scaffolds in which a single stereochemical inversion results in a significant difference in both chromatographic lipophilicity and PAMPA permeability (Figure 6). McMD simulations of these diastereomers predict disruption of the β sheet in SB60, significantly reducing the strength of one of the four IMHBs. The difference in cLPE (1.47) between the scaffolds is analogous to previously observed differences for both the hexapeptide and PROTAC model systems, which feature scaffolds that differ by one exposed hydrogen bond donor.

Figure 6.

Chromatogram of a selected ion monitored spectra of diastereotopic, thioether-cyclized peptides, SB60 and SB64 (left and right, respectively).

In comparison to the shake-flask assay, chromatographic lipophilicity has significantly better precision between measurements and can be more easily reproduced between different users. Shake-flask partition coefficients are often plagued by sensitivity issues of the analyte in the less concentrated solvent (oftentimes in the low nM range),³⁷ which restricts the dynamic range of this assay without extremely sensitive instrumentation (e.g., high resolution mass spectrometry) and can lead to low precision between measurements for compounds with very high or low values.

Although we obtained similar Log ED_dd/w for isocratic and gradient methods and these values are reflective of low-dielectric conformational differences as predicted by McMD, more work must be done to elucidate the effect of the mobile phase composition on the distribution of cyclic peptide conformations while partitioning through the chromatographic system. Furthermore, the compounds in this study feature exclusively nonionizable side chains, reflecting the composition of natural products known to be passively permeable.⁸ More work is needed to evaluate the utility of our method toward the analysis of bRo5 molecules with ionizable side chains. Chromatographic lipophilicity and subsequent estimation of 1,9-decadiene/water shake-flask values represent a rapid and practical framework for assessment of lipophilicity, solubility, and potential for passive permeability in early stage drug development for compounds beyond the traditional bounds of drug-likeness. We are currently investigating the potential of applying this framework to DNA-encoded libraries.

Experimental Section

Chemistry General

Detailed methods for synthesis, assays (chromatography, shake-flask experiment, PAMPA, and McMD), and calculations are described in the Supporting Information. All compounds are ≥95% pure by HPLC analysis. Chromatographic retention times were measured either as single points, in duplicate, or in triplicate. Shake-flask partition coefficients were measured in duplicate, and PAMPA was measured in quadruplicate; both assays were quantified with UPLC-MS and selected-mass monitoring. All individual compounds in this study (MN-6a-d,²⁴ VK-P,²⁶ and a subset of JF-10a³³) have been featured in previous publications and were proven to be ≥95% pure by UV absorbance and ¹H NMR. The rest of the compounds used in this study were synthesized as purified mixtures and individual compound identities were confirmed within the mixtures with UPLC-MS and selected-mass monitoring. JF-10a, JF-10b,³³ and AF-10a-b³⁰ were featured in previous publications in our lab. JF-10c and JT-7 were both synthesized for this study.

Solid Phase Peptide Synthesis for JT-7 Library

The compounds were made in a split-and-pool fashion resulting in 16 sub libraries with 100 compounds per sub library. Preloaded 2-Chlorotrityl resin (l-Leucine loaded −0.65 mmol/g) was used. Fmoc deprotections were carried out with 20% piperidine solution in DMF for 20 min twice. Couplings were performed using Fmoc-protected amino acids (4 equiv), HATU (4 equiv), and DIPEA (8 equiv) in DMF for 1 h at room temperature. Between coupling and deprotection steps, the resin was washed with DMF (5x), DCM (3x), and DMF (5x). Coupling efficiency was evaluated by negative ninhydrin test. Linear peptides were cleaved in a 30% HFIP-DCM solution and cyclized with COMU (2 equiv) and DIPEA (4 equiv) in THF relative to the mass of cleaved peptide and average molecular weight in each sublibrary. Cyclic peptides were purified by short C-18 plug to remove excess coupling reagent and byproducts. Compound identities were determined within the mixtures via selected-mass monitoring by LCMS.

Synthesis of Boc-N-Methylpropargylamine

To a 250 mL round-bottomed flask charged with a stir bar was added methanol (60 mL) and N-methylpropargylamine (3.05 mL, 36.2 mmol). While the solution was stirring, ditert-butyl decarbonate (8.40 mL, 36.6 mmol, 1.01 equiv) was added dropwise to the solution and the reaction stirred at RT overnight. The next day, the reaction was concentrated in vacuo to yield a dark brown liquid which was placed in the vacuum desiccator overnight and used without any further purification. ¹H NMR (499 MHz, chloroform-d) δ 3.97 (s, 2H), 2.84 (s, 3H), 2.16 (s, 1H), 1.40 (s, 9H).

Solid Phase Peptide Synthesis for JF-10c Library

The JF-10c series of compounds were synthesized in multiplex fashion using routine split-pool Fmoc-SPPS, installing the 1,4-disubstituted-1,2,3-triazole moiety as the final residue in the linear synthesis before simultaneous resin-cleavage and N-terminal Boc-deprotection. Commercially purchased preloaded 2-Chlorotrityl resin (Leucine loaded −0.65 mmol/g) was used in 0.2 mmol portions. Fmoc deprotections were carried out with 2% DBU and 2% piperidine solution in DMF for 15 min. Couplings were performed using Fmoc-protected amino acids (4 equiv), HATU (3.8 equiv), HOAt (3.8 equiv) and DIPEA (6 equiv) in DMF (0.1 M with respect to amino acid) for 1 h at 50 °C or overnight at room temperature. Between coupling and deprotection steps, the resin was washed with DMF (3x), DCM (3x), and one final time with DMF. Coupling reactions were monitored by HPLC-MS and repeated until starting material was no longer observed. Peptoid residues were added in-sequence to the Fmoc-deprotected amino acid by initial bromoacetylation followed by amine substitution. Bromoacetic acid (1 M) was dissolved in DMF in a separate vessel and followed by addition of DIC (0.5 M). This solution was allowed to rock on a linear shaker for 20 min. To the preswelled Fmoc-deprotected linear sequence on-resin was added this mixture in sufficient volume to cover the resin and the suspension was allowed to rock at room temperature for 45 min. Upon completion, the resin was washed with DMF (3x), DCM (3x), and a final time with DMF. To the resin was then added a 1 M solution of the corresponding amine in sieve-dried DMF and allowed to rock overnight at room temperature. The reaction was monitored by HPLC-MS and the amine addition repeated if necessary.

Azidoacetic acid was coupled to the deprotected N-terminus of the on-resin linear peptide. In a separate vessel, azidoacetic acid (3 equiv) was dissolved in enough DMF to eventually fully swell and submerge the resin. DIC (3 equiv) was added to the solution and then allowed to rock on a linear shaker for 20 min. To the preswelled Fmoc-deprotected linear sequence on-resin was added this mixture and the suspension was allowed to rock at room temperature for 1–1.5 h. Upon completion, the resin was washed with DMF (3x), DCM (3x), and a final time with DMF. The reaction was monitored by HPLC-MS and repeated if necessary. Complete linear peptides were cleaved off resin using 10% TFA in DCM for 1 h (2x) with a DCM wash equivalent to 2 resin volumes in between repetitions. Solvent was removed under a flow of nitrogen followed by dissolution in acetone and evaporation again under nitrogen. The crude linear product was stored under vacuum.

In preparation for the CuAAC reaction, a 20% (v/v) solution of 2,6-lutidine in DMF was prepared fresh. Copper(I) iodide (5 equiv) and ascorbic acid (15 equiv) was added to a separate plastic vessel and the powders mixed thoroughly. Then the 20% (v/v) solution of 2,6-lutidine (20 mL for 0.2 mmol scale) was added to the powders and sonicated until complete dissolution (adding additional DMF if the solution was not dissolved after 10 min). This solution was then added to the preswelled resin followed by the addition of N-Boc-N-methylpropargylamine (2 equiv) and the final suspension sparged briefly (<1 min) with argon. Then, the vessel was capped and shaken at RT overnight. Upon completion, the resin was drained and washed once with DMF. Next the resin was washed with a “click-wash” solution (5% sodium diethyldithiocarbamate and 5% DIPEA in DMF; 3–5x), DMF (3x), DCM (3x), and a final time with DMF. Completion of the reaction was verified by HPLC-MS.

Unpurified linear peptides were dissolved in 10 mL of argon purged ACN with 4 equiv of DIPEA and added dropwise to a solution of argon purged ACN containing 2 equiv of COMU, for a total concentration of 0.002 M respective to the peptide. Reactions were stirred for 12–24 h under an argon atmosphere until complete cyclization was achieved as monitored by HPLC-MS. The reaction was reduced in vacuo prior to purification which was carried out via trap-and-elute method utilizing Biotage Isolute C18 column.

Glossary

Abbreviations

ΔcLPE

the difference in cLPE between two compound classes (high-low)

ΔLPE

the difference in LPE between two compound classes (high-low)

ALogP

atomistic calculated octanol/water partition coefficient

bRo5

beyond rule-of-5

cLogP

calculated octanol/water partition coefficient

cLPE

lipophilic permeability efficiency determined by chromatography

HBA

hydrogen bond acceptor

HBD

hydrogen bond donor

Hyp

hydroxyproline

Log ED_dd/w

estimated distribution coefficient at pH 7.4 with 1,9-decadiene/water

Log D

distribution coefficient at pH 7.4

Log D_dd/w

distribution coefficient at pH 7.4 with 1,9-decadiene/water

Log K

logarithm of the capacity factor

Log P

partition coefficient

LPE

lipophilic permeability efficiency

McMD

multicanonical molecular dynamics

MDCK

Madin-Darby canine kidney

PAMPA

parallel artificial membrane permeability assay Pe, effective permeability

PROTAC

proteolysis targeting chimera

RMSD

root mean squared deviation

Tle

tert-leucine

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jmedchem.4c01956.

• Supplementary figures, tables, and discussion, additional details concerning synthesis of compounds, detailed analytical procedures, and analytical spectra for JT-7 and JF-10c libraries (PDF)

• All processed compound data with Log ED_(dd/w), LogK’, Log D_(dd/w), PAMPA, and MDCK data, equations for Log ED_(dd/w) and cLPE calculations for various methods, training and test set data (XLSX)

• Molecular strings for compounds with associated lipophilicity and permeability data (CSV)

• All tabulated replicates data for all compounds and methods, code and example spreadsheets for ChamelogK and minimization calculations (ZIP)

Author Contributions

^∥ G.K. and A.E. are cofirst authors. G.K. and A.E. contributed equally and designed the project, measured all chromatographic lipophilicities, measured Log D_dd/w on all compounds but JT-7, processed all data, and wrote the paper. J.F. synthesized, characterized, and measured shake-flask lipophilicity and PAMPA of all thioether and triazole peptides. J.T. synthesized, characterized, and measured shake-flask lipophilicity of all heptapeptide compounds. S.O. performed the McMD simulations. M.N. and S.L. contributed toward writing and reviewed and edited the manuscript. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.