A perennial problem encountered when using small molecules (drugs) to manipulate cell or protein function is to assess whether observed changes in function result from specific interactions with a desired target or from less specific off-target mechanisms. This is important in laboratory research as well as in drug development, where the goal is to identify molecules that are unlikely to be successful therapeutics early in the process, thereby avoiding costly mistakes. We pursued this challenge from the perspective that many bioactive molecules (drugs) are amphiphiles that alter lipid bilayer elastic properties, which may cause indiscriminate changes in membrane protein (and cell) function and, in turn, cytotoxicity. Such drug-induced changes in bilayer properties can be quantified as changes in the monomer↔dimer equilibrium for bilayer-spanning gramicidin channels. Using this approach, we tested whether molecules in the Pathogen Box (a library of 400 drugs and drug-like molecules with confirmed activity against tropical diseases released by Medicines for Malaria Venture to encourage the development of therapies for neglected tropical diseases) are bilayer modifiers. 32% of the molecules in the Pathogen Box were bilayer modifiers, defined as molecules that at 10 µM shifted the monomer↔dimer equilibrium toward the conducting dimers by at least 50%. Correlation analysis of the molecules’ reported HepG2 cell cytotoxicity to bilayer-modifying potency, quantified as the shift in the gramicidin monomer↔dimer equilibrium, revealed that molecules producing <25% change in the equilibrium had significantly lower probability of being cytotoxic than molecules producing >50% change. Neither cytotoxicity nor bilayer-modifying potency (quantified as the shift in the gramicidin monomer↔dimer equilibrium) was well predicted by conventional physico-chemical descriptors (hydrophobicity, polar surface area, etc.). We conclude that drug-induced changes in lipid bilayer properties are robust predictors of the likelihood of membrane-mediated off-target effects, including cytotoxicity.

Many biologically active (bioactive) molecules, including drugs, are amphiphiles that partition into the lipid bilayer component of cellular membranes thereby altering bilayer physical properties like elasticity, curvature, and thickness (Seddon, 1990; Evans et al., 1995; Zhelev, 1998; Chen et al., 2003; Lundbaek et al., 2005; Marsh, 2008), which will alter the bilayer contribution to the energetic cost of membrane protein conformational changes that involve the proteins’ bilayer-spanning domains and, in turn, membrane protein function (Lundbaek et al., 2005; Lundbaek et al., 2010a; Rusinova et al., 2011; Ingólfsson et al., 2014). The functional consequences of such changes in bilayer properties (changes in the bilayer contribution to the energetics of a conformational change) can be quantified in studies on well-defined reporter proteins (Gruner, 1991; Brown, 1994; Lundbaek et al., 2010a; Ingólfsson et al., 2014), which show that an amphiphile may alter the activity of functionally and structurally diverse membrane proteins at similar concentrations (Ingólfsson et al., 2014). Thus, if one membrane protein is modulated by a bioactive amphiphile (at some concentration), then many other membrane proteins will be modulated at similar concentrations. These indiscriminate changes in membrane protein function are likely to compromise cellular homeostasis and, if the changes in function are of sufficient magnitude, cause cytotoxicity.

We explored this question in the context of drug development against neglected tropical diseases (NTDs), a group of communicable diseases that are prevalent in tropical and subtropical countries (Fürst et al., 2017; World Health Organization, 2017). To promote the development of effective/inexpensive treatments for the most common NTDs, the Medicines for Malaria Venture (MMV) compiled and released the Pathogen Box, an open source project containing 400 drugs and drug-like molecules (drugs for short) with confirmed activity against NTDs, to catalyze a collaborative environment for drug discovery and development (MMV, 2014), which led to the identification of numerous leads for treating NTDs, e.g., Veale, 2019.

All drugs in the Box have been characterized in terms of their biological activity and deemed to be suitable for an initial drug discovery program. MMV also provides extensive cheminformatics plus information about biological activity, including cytotoxicity, on the drugs in the Box (MMV, 2017). Among the selection criteria for including drugs in the Pathogen Box was whether they were deemed to have appropriate physicochemical properties (Hughes et al., 2008; Price et al., 2009; Waring et al., 2015), including the calculated octanol/water partition coefficient and polar surface area.

Understanding how a molecule’s physicochemical properties may relate to cytotoxicity is likely to improve drug design and development (Leeson and Springthorpe, 2007; Leeson, 2012); yet, predicting the likelihood that a drug candidate may be (cyto)toxic based on its physicochemical characteristics remains a challenge (Waring et al., 2015). It is in this context important that many drugs are amphiphiles that partition into the lipid bilayer component of cellular membranes, where they alter bilayer physical properties and thereby membrane protein function (Lundbaek et al., 1996; Lundbaek et al., 2004; Lundbaek et al., 2005; Artigas et al., 2006; Rusinova et al., 2011; Ingólfsson et al., 2014).

This bilayer-mediated regulation of membrane protein function arises because the hydrophobic adaptation between membrane proteins and their host bilayer causes membrane proteins to be energetically coupled to their host bilayer (Gruner, 1991). Conformational changes (from, say, state I to state II) that involve the proteins’ bilayer-spanning domains (Fig. 1), therefore, will alter the organization of the adjacent lipids, which has an associated energetic cost (Huang, 1986; Gruner, 1991; Lundbaek et al., 2010a; Rusinova et al., 2011): $ΔGbilayerI→II=ΔGdefII−ΔGdefI,$ where $ΔGdefI$ and $ΔGdefII$ denote the energetic cost of the local, protein-induced bilayer deformations associated with each state. Norimatsu et al. (2017) and Wang and Boudker (2020) show the existence of conformational changes in integral membrane proteins and their impact on the adjacent bilayer; Zhou et al. (2019) estimate the associated changes in $ΔGbilayerI→II,$ which may be 6–7 kcal/mole. The total energetic cost of a conformational change from state I to state II $ΔGtotalI→II$ thus will be the sum of contributions due to structural rearrangements within the protein $ΔGproteinI→II$ and rearrangements within the bilayer: $ΔGtotalI→II=ΔGproteinI→II+ΔGbilayerI→II$ plus other contributions (Rusinova et al., 2021). See also Section 1 in the supplemental text at the end of the PDF.

The functional consequences of changes in bilayer properties (changes in $ΔGbilayerI→II)$ can be quantified using well-defined reporter proteins (Gruner, 1991; Brown, 1994; Lundbaek et al., 2010a; Ingólfsson et al., 2014). One such reporter is the gramicidin channel, which forms by transmembrane dimerization of two non-conducting subunits (Bamberg and Läuger, 1973; Zingsheim and Neher, 1974; O’Connell et al., 1990). Gramicidin channels are known to be modulated by changes in lipid bilayer properties, whether induced by changes in lipid composition (Kolb and Bamberg, 1977; Elliott et al., 1983) or by bioactive molecules (Haydon et al., 1977; Pope et al., 1982; Haydon and Urban, 1983; Elliott et al., 1985; Hwang et al., 2003; Lundbaek et al., 2004; Lundbaek et al., 2005; Artigas et al., 2006; Bruno et al., 2007; Ingólfsson and Andersen, 2010; Rusinova et al., 2011; Herold et al., 2014; Rusinova et al., 2015).

Changes in bilayer properties (elasticity, thickness, and intrinsic curvature) will alter the gramicidin monomer↔dimer equilibrium (Fig. S1), which can be measured as changes in the number of conducting channels per unit membrane area: changes in appearance rates and lifetimes of bilayer-spanning gramicidin channels (Sawyer et al., 1989; Lundbaek et al., 2010a); or changes in the time course of fluorescence quenching in fluorophore-loaded large unilamellar phospholipid vesicles (LUVs) that have been doped with gramicidin and mixed with a gramicidin channel-permeant quencher, Tl+ (Ingólfsson and Andersen, 2010); see also Section 1 at the end of the PDF. Bilayer-perturbing effects can be generalized to other lipid compositions (Bruno et al., 2007; Rusinova et al., 2011; Rusinova et al., 2015; Herold et al., 2017), cells (Lin and Will, 2012), and systems (Hughes et al., 2008). Changes in bilayer properties, as evaluated using gramicidin channels, can also be used to predict changes in function of other membrane proteins (Lundbaek et al., 2005; Rusinova et al., 2011; Ingólfsson et al., 2014; Herold et al., 2017) demonstrating the generality of this mechanism.

We show that drug-induced changes in fluorescence quench rate correlate with the drugs’ likely cytotoxicity, reported by MMV (2017) as the concentration that produces 20% inhibition of HepG2 cell growth (HepG2 CC20). Drugs with higher quench rates tend to have lower CC20 values, meaning they are more likely to be cytotoxic. Drugs may, of course, be cytotoxic for reasons not associated with the membrane, but bilayer-modifying potency is a promising predictor of off-target effects and cytotoxicity. Although physico-chemical properties allow for predicting drug-likeliness (e.g., Bickerton et al., 2012), we found they are less effective in predicting off-target effects and cytotoxicity. Our results, taken together with earlier studies demonstrating that that drug-induced changes in ion channel function correlate with changes in gramicidin channel function (summarized in the previous paragraph), suggest that testing for bilayer-modifying potency will help identify membrane-mediated off-target effects when using amphiphiles (including drugs) to manipulate membrane protein or cell function.

### Materials

1,2–Dierucoyl–sn–glycero–3–phosphocholine (DC22:1PC) in chloroform (25 mg/ml) was >99% pure from Avanti Polar Lipids. Methanol ≥99.8% was from VWR. Thallium nitrate (TlNO3) ≥99.9%, sodium nitrate (NaNO3) ≥99%, HEPES ≥99.5%, bovine serum albumin (BSA) ≥98%, 3-[(3-cholamidopropyl)-dimethylammonium]-1-propanesulfonate (CHAPS) ≥98%, and gramicidin from Bacillus aneurinolyticus (Bacillus brevis) ≥95% were from Sigma–Aldrich Co. The di-sodium salt of 8–aminonaphthalene–1,3,6–trisulfonate (ANTS) was from Invitrogen. The Pathogen Box was a gift from the Medicines for Malaria Venture: https://www.mmv.org/mmv–open/pathogen–box. The drugs were provided as 10 µl aliquots of 10 mM drug dissolved in DMSO and used as supplied.

Stock solutions of buffers and quenchers were prepared ahead of the experiment and, unless otherwise noted, were dissolved in deionized water, and adjusted to pH 7 with sodium hydroxide (NaOH) and nitric acid (HNO3). Na–ANTS buffer was 25 mM ANTS, 100 mM NaNO3, and 10 mM HEPES; it was stored shielded from light. Na buffer was 140 mM NaOH and 10 mM HEPES. Tl quench buffer was 50 mM TlNO3, 94 mM NaNO3, and 10 mM HEPES. All buffer and quencher stock solutions were stored at 12.5°C; the DC22:1PC was stored at −40°C.

### Methods

#### Gramicidin channels

The naturally occurring mixture of the linear gramicidins from Bacillus brevis has historically been called gramicidin D (gD), after R. Dubos, who discovered the gramicidins (Dubos, 1939); it contains 80–85% [Val1] gramicidin A (gA), 6–7% gramicidin B (gB), [Val1, Phe11]gA, and 5–14% gramicidin C (gC), [Val1, Tyr11]gA (Abo-Riziq et al., 2006). We used the gD as a 500 μg/ml (265 µM) solution in methanol, which was stored at −40°C.

Gramicidin channels are formed by transmembrane dimerization of two non-conducting gramicidin subunits (O’Connell et al., 1990; Lum et al., 2017; Fig. S1). gA, gB, and gC form structurally equivalent anti-parallel, dimeric channels with very similar properties (Sawyer et al., 1990), meaning that approximately two-thirds of the measured ion flux will be through symmetric gA/gA homodimeric channels, approximately one-fifth will be through asymmetric gA/gB, or gA/gC heterodimeric channels; the remaining will be through symmetric gB/gB and gC/gC homodimeric channels and asymmetric gB/gC heterodimeric channels. Experiments using the readily available gD give similar results as experiments with purified gA (Sun et al., 2020). Because the channels’ hydrophobic length is less than the host bilayer’s hydrophobic thickness, channel formation produces a local bilayer thinning (Fig. S1), which incurs an energetic cost (Huang, 1986; Lundbaek et al, 2010a).

Gramicidin was incorporated into LUVs that encapsulate the aqueous fluorophore ANTS (Ingólfsson and Andersen, 2010), which is quenched by the gramicidin channel-permeant heavy monovalent cation thallium (Tl+). When a drug is added and allowed to equilibrate with the LUVs, the drug will partition into the vesicle bilayer, which will alter bilayer properties, usually decreasing the bilayer stiffness (ease of deformation). Such drug-induced bilayer softening will decrease the energetic cost of dimerization and shift the monomer↔dimer equilibrium toward the conducting dimers, and drug-induced stiffening of the bilayer will increase the energetic cost of dimerization and shift the monomer↔dimer equilibrium toward the non-conducting monomers. These shifts in the monomer↔dimer equilibrium can be evaluated using stopped-flow spectrofluorometry (see below).

#### LUVs

ANTS-loaded LUVs incorporating gD (Ingólfsson and Andersen, 2010) were prepared using gD and DC22:1PC (molar ratio 1:2,000), which were mixed in a 50-ml round-bottom flask, dried to a thin film under nitrogen to remove the chloroform and methanol, then further dried under a vacuum overnight to remove any remaining solvent. The lipid film was rehydrated in Na–ANTS buffer to give a 10 mM lipid suspension, which was thoroughly vortexed (the round-bottom flask was covered and protected from light for the duration of the LUV preparation and experiment). The suspension was incubated at room temperature for at least 3 h, followed by sonication for 1 min at low power. The resulting suspension then was subjected to six freeze–thaw cycles using dry ice (10 min) and 45–55°C water (5 min); after each cycle, the sample was thoroughly vortexed. At the end of the sixth cycle, the resulting multilamellar vesicle (MLV) suspension was extruded 20 times (passes through the filter) at room temperature through a 10 ml LIPEX Extruder (Northern Lipids Inc) with a 0.1 μm polycarbonate filter and a 25-mm polyester drain disc. The resulting LUV suspensions were stored at −40°C and could be used for at least 1 mo. Before use, extravesicular ANTS was removed using a PD–10 desalting column (GE Healthcare); these LUVs (5 mM lipid suspension) were stored at 12.5°C and used within 7 d.

#### Dynamic light scattering

The LUV size distribution was determined using a Litesizer 500 dynamic light scattering instrument with the Kalliope software (Anton Paar). Using disposable cuvettes with 1.0 ml sample volume and a lipid concentration of 50 μM in Na buffer, the transmittance was ≈88%. The refractive index and viscosity of the Na buffer were set to 1.3318 and 0.9064 cP, respectively, the default settings in Kalliope. The default correlation function and fitting curve were used to calculate the diffusion coefficient, mean hydrodynamic diameter (dLUV), and the polydispersity index (PDI), defined as $(σ/dLUV)2,$ where σ2 denotes the variance of the size distribution (e.g., Clayton et al., 2016). Each sample was tested 1 d after extrusion with three independent measurements and at least seven repeats in each measurement. There was only one discernable peak (Fig. S2), with dLUV = 130 ± 5 nm and a PDI of 0.09 ± 0.04 (n = 9). A PDI < 0.1 is considered to indicate a monodisperse sample (Clayton et al., 2016). For an LUV sample with dLUV = 130 nm and a PDI = 0.06, 10% of the LUVs will have a diameter <87 nm and 10% of the LUVs will have a diameter >160 nm.

#### Physicochemical parameters and biological activity

Most of the physicochemical molecule properties used in the analysis of the results were estimated using the Schrödinger Suite (Schrödinger). Polar surface area (PSA), defined as the Van der Waals surface area of polar nitrogen and oxygen atoms, was calculated using QikProp. Hydrophobicity was estimated using ALogP, which was calculated using fragmental methods (Ghose et al., 1998). ALogP and the aromatic ring count were evaluated using Canvas. Data for the in vitro fraction unbound in mouse plasma (fumouse), or human microsomal protein (fumic), were from (MMV, 2017). Toxicity information was provided by MMV, as the drug concentration that causes death of 20% of cells (CC20) in the human liver cancer cell line HepG2.

#### Stopped-flow spectrofluorometry

The time course of ANTS fluorescence quench was measured at 25°C using an SX–20 stopped-flow spectrofluorometer (Applied Photophysics) in the single mixing mode. The excitation wavelength was 352 nm and the emission above 450 nm was recorded using a high-pass filter and a sampling rate of 5,000 points/s. The measured instrument deadtime was ∼1.2 ms. Samples were prepared by diluting the ANTS-LUV stock suspension 100-fold with Na buffer to 50 μM lipid; for each sample, an aliquot of the drug in question was added to a final concentration of 10 μM and allowed to equilibrate for 10 min before testing (the final DMSO concentration in the mixture was 12.8 mM, or 0.1%, a concentration at which DMSO has no effect on bilayer properties; Ingólfsson and Andersen, 2010). For each sample, 8 1-s control mixing reactions were recorded by mixing the LUV suspension with Na buffer (no Tl+), followed by 10 1-s mixing reactions with the Tl quench buffer. “Bad” traces, which may occur for reasons such as air bubbles, were removed based on visual inspection. Each drug was evaluated using two independently prepared LUV preparations; the quality of each batch was evaluated using the control rate, in the presence of 0.1% DMSO with no added drug (negative control), and the increase in quench rate observed with 5% ethanol (positive control).

To evaluate the possible effects of protein binding, some drugs were retested in the presence of an additional 60 µM BSA (corresponding to the BSA concentration in cell culture media supplemented with 10% fetal calf serum).

#### Data analysis

The rate of Tl+ influx was quantified from the time course of fluorescence quench using the Stern–Volmer relationship for dynamic quenching (Moore and Raftery, 1980; Ingólfsson and Andersen, 2010). Due to the unavoidable variations in LUV size (surface area and volume) and surface density of conducting channels in the LUV membranes, the volume-averaged change in [Tl+]i cannot be described by a single exponential function. [Tl+]i will increase faster in the smaller LUVs and slower in the larger LUVs (Ingólfsson and Andersen, 2010), and the increase in the volume-averaged [Tl+]i will be a weighted sum of exponential functions, which can be expressed in a mathematically convenient form by a so-called modified stretched exponential function (Berberan-Santos et al., 2005); for details, see Section 2 at the end of the PDF.

The time course of fluorescence quench can be expressed as (Eq. S17):
$F(t)F(0)=11+KSV⋅[Tl+]i(t)=11+KSV⋅[Tl+]e⋅(1−exp{1−(1+t/τ0)β}),$
(1)
where F(t) denotes the fluorescence intensity as function of time, t; F(0) is the fluorescence intensity at time 0, before any quench has occurred; KSV is the Stern–Volmer coefficient (60 M–1 for Tl+; Ingólfsson and Andersen, 2010); [Tl+]i(t) is the intravesicular Tl+ concentration (as function of time); and [Tl+]e is the extravesicular Tl+ concentration. τ00 > 0) is a parameter with units of time and β (0 < β ≤ 1) is a parameter reflecting the dispersity of vesicle volumes, areas, and surface density of conducting gramicidin channels (β = 1 for a homogenous population of LUVs). The initial quench rate at t = 0, Rate(0) is given by (cf. Eq. S18):
$Rate0=ddtFtF0|0=−KSV⋅Tl+e⋅βτ0⋅exp1−1+t/τ0β1+KSV⋅Tl+e⋅1−exp1−1+t/τ0β2|=−KSV⋅Tl+e⋅βτ0,$
(2)
where $[Tl+]e⋅β/τ0$ is the initial rate of Tl+ influx (cf. Eqs. S10 and S19).

In practice, there are three separate ANTS-containing (and fluorescent) compartments: first, the ANTS in the extravesicular solution, which will be quenched rapidly, within the time resolution of the instrument; second, LUVs with conducting channels that can be quenched; third, vesicles that cannot be quenched because they either are LUVs without any conducting gramicidin channels (during the 1 s measurement), or multilamellar vesicles (e.g., Scott et al., 2019). Tl+ can cross the LUV membrane, as TlNO3 ion pairs (Martinus and Vincent, 1976) and, maybe, through transient membrane defects (Paula et al., 1996), which gives rise to slow quench of the ANTS fluorescence (Ingólfsson and Andersen, 2010); this has no significance for quench rates measured over the first 1 s of mixing and is not considered further.

The fluorescence signal recorded when mixing the LUVs with Na buffer, F(t = 0,b), where nothing is quenched, is the sum of the initial fluorescence from the three compartments:
$F(t=0,b)=F(extravesicular)+F(intravesicular)+F(unquenchable).$
(3)

Only the second group of LUVs with conducting channels is of interest; the other two groups contribute to the signal, however, and it becomes convenient to explicitly consider the three compartments and their contributions to the fluorescence signal: F(extravesicular), the fluorescence signal from the extravesicular ANTS; F(intravesicular), the fluorescence signal from the quenchable LUVs; and F(unquenchable), the fluorescence signal from the population of unquenched vesicles. These three contributions can be quantified as follows:

First, the fluorescence signal recorded immediately after mixing the LUV suspension with Tl quench buffer, F(t = 0,q), will be less than F(t = 0,b) because the extravesicular ANTS will be quenched “instantly,” within the dead time of the instrument:
$F(extravesicular)=[F(0,b)−F(0,q)]1+KSV⋅[Tl+]eKSV⋅[Tl+]e.$
(4)
Second, the fluorescence recorded “long” after mixing the LUV suspension with the Tl+ quencher, F(t = ∞,q) will be the sum of the quenched signal from the extravesicular ANTS and the population of quenchable vesicles, plus the signal from the unquenched vesicles (and any other non-quenching elements):
$Ft=∞,q=Fextravesicular+Fintravesicular1+KSV⋅Tl+e+Funquenchable,$
(5)
because [Tl+]i(∞) in the quenchable vesicles will be equal to [Tl+]e.
Combining Eqs. 3, 4, and 5, F(extravesicular), F(intravesicular), and F(unquenchable) can be expressed in terms of the experimental observables, F(t = 0,b), F(t = 0,q), and F(t = ∞,q):
$F(extravesicular)=[F(0,b)−F(0,q)]⋅1+KSV⋅[Tl+]eKSV⋅[Tl+]e,$
(6)
$F(intravesicular)=[F(0,q)−F(∞,q)]⋅1+KSV⋅[Tl+]eKSV⋅[Tl+]e,$
(7)
and
$F(unquenchable)=F(∞,q)⋅(1+KSV⋅[Tl+]e)−F(0,b)KSV⋅[Tl+]e.$
(8)
The time course of fluorescence quench thus can be expressed by
$F(t)=F(extravesicular)1+KSV⋅[Tl+]e+F(intravesicular)1+KSV⋅[Tl+]i(t)+F(unquenchable)$
(9)
or
$F(t)=F(∞,q)⋅(1+KSV⋅[Tl+]e)−F(0,q)KSV⋅[Tl+]e+1+KSV⋅[Tl+]eKSV⋅[Tl+]e⋅F(0,q)−F(∞,q)1+KSV⋅[Tl+]i(t).$
(10)
Inserting KSV = 60 M–1 (Ingólfsson and Andersen, 2010) and [Tl+]e = 25 mM, Eq. 10 reduces to:
$Ft=2.5⋅F∞,q−F0,q1.5+2.51.5⋅F0,q−F∞,q1+1.5⋅1−exp1−(1+tτ0)β,$
(11)
which was fitted to the fluorescence quench traces between 2 ms to 1 s using the non-linear least squares curve fitting method in MATLAB (The MathWorks). Using the resulting values of F(0,q), F(∞,q), β, and τ0, initial rate of fluorescence quench (Tl+ influx) could be determined from Eqs. 2 and 11:
$Rate0=ddtFtF0,q−F∞,q|0=−2.5⋅βτ0⋅exp1−1+t/τ0β1+KSV⋅Tl+e⋅1−exp1−1+t/τ0β2|=−2.5⋅βτ0.$
(12)
The drug-induced change in quench rate: the quench rate normalized to control, NormRate, was determined as follows:
$NormRate=RateRatecntrl=Ratedrug(0)Ratecntrl(0).$
(13)
where the subscripts cntrl and drug denote the rates in the absence and presence of the drug, respectively. For display, e.g., Fig. 2, the traces are normalized to F(0,b).
The drug-induced changes in bilayer deformation energy $ΔΔGbilayer|drugM→D=ΔGbilayer,drugM→D-ΔGbilayer,cntrlM→D$ were evaluated as (Artigas et al., 2006; Sun et al., 2019; Sun et al., 2020):
$ΔΔGbilayerM→D|drug=−RT⋅lnKdrugM→DKcntrlM→D=−RT⋅lnDdrug/[M]drug2Dcntrl/[M]cntrl2≈−RT⋅lnDdrugDcntrl≈−RT⋅lnNormRate,$
(14)
where R is the gas constant; T is the temperature in Kelvin; $KcntrlM→D$ and $KdrugM→D$ are the dimerization constants in the absence and presence of drug; and [D]cntrl, [D]drug, [M]cntrl, and [M]drug are the concentrations of dimer (D) and monomer (M), respectively, in the absence or presence of the drug. (Eq. 14 is valid only when [M]drug ≈ [M]cntrl, when the monomer↔dimer equilibrium is biased toward the non-conducting monomers in both the absence and presence of the drug in question, as will be the case in the thick DC22:1PC bilayers used in this study.)
Some experiments were performed in the presence of 60 µM BSA, as a proxy for the 10% fetal calf serum that often is added to cell culture media. These experiments were performed as described above, except that the LUVs were incubated for 10 min with either 60 µM BSA or 60 µM BSA plus 10 µM of the drug before testing for activity. For the analysis, the drug-induced changes in rates were evaluated as:
$NormRateBSA=Ratedrug+BSA(0)RateBSA(0).$
(15)

#### Pan assay interference compounds (PAINS)

The 400 drugs were evaluated for pan assay interference (Baell and Holloway, 2010) using Badapple (http://pasilla.health.unm.edu/tomcat/badapple/badapple; Yang et al., 2016). Badapple detects patterns of promiscuity, assay interference in high-throughput screens, associated with different molecular scaffolds (Yang et al., 2016) and assigns a promiscuity score (pScore), which is a measure of the risk of promiscuity. A molecule may incorporate several different scaffolds, which each may yield a different pScore, and we report the largest value over all scaffolds. A pScore < 100 denotes that pan assay interference is unlikely; 100 ≤ pScore < 300 denotes weak likelihood of promiscuity; and 300 ≤ pScore indicates high likelihood of promiscuity (Yang et al., 2016).

#### Quantitative estimate of drug-likeness (QED)

The QED (Bickerton et al., 2012) provides a measure of a molecule’s drug-likeness based on eight molecular properties: molecular mass; ALogP; PSA; number of hydrogen bond donors; number of hydrogen bond acceptors; number of rotatable bonds; number of aromatic rings; and number of structural alerts, or unwanted chemical groups (Brenk et al., 2008). We estimated the drugs’ QED score using the RDKit chemoinformatics software (http://www.rdkit.org/) with average descriptor weights (Bickerton et al., 2012).

#### Cluster analysis

To evaluate how different groups of molecules (grouped based on NormRate, or CC20) clustered, we used silhouette analysis (Rousseeuw, 1987). Silhouette plots visualize how close the objects in a cluster are grouped compared to neighboring clusters. For each molecule in a cluster, a silhouette score is calculated based on the average dissimilarity (distance to other molecules) in the cluster compared to the dissimilarity to molecules in other clusters. The silhouette score ranges between 1 (good separation among clusters) and −1 (poor/non-existent separation among clusters). The scores for all molecules in a cluster are averaged; the higher the average score, the better this cluster is defined.

#### Statistics

The quality of the fits of Eq. 11 to the fluorescence quench traces was judged by the regression coefficient for the fits, (average 0.996 ± 0.015; six samples had regression coefficients below 0.95%, all were >0.70).

All experiments were performed in duplicate, and the results are reported as mean ± range/2. The average range/(2⋅mean) was 0.07 ± 0.07 (mean ± SD); six drugs had range/(2⋅mean) >0.3, they were tested in triplicate and the results are reported as mean ± SD. Comparisons among different groups of drugs were carried out using the two-tailed Mann–Whitney test using the Bonferroni correction for multiple comparisons when needed.

When results are presented as box plots (Fig. S4, S6, S7, S8, S10, and S11), the lengths of the bottom and top whiskers were calculated using the MatLab boxplot function as 1.5⋅(q3 – q1), where q1 and q3 denote the first and third quartiles (in case of a normal distribution, this provides 99.3% coverage of the data between the top and bottom whiskers); values below the bottom and above of the top whisker are defined as outliers and denoted by +, and the whiskers end with the largest (smallest) value that fall within the range defined by the whiskers. CC20 values in the MMV database were truncated at 80 µM, and some plots may not have a top whisker.

### Online supplemental material

Fig. S1 shows gramicidin channel structure and function. Fig. S2 shows the distribution of LUV diameters for vesicles used in the fluorescence quench experiments. Fig. S3 shows distribution of NormRates and HepG2 CC20 among drugs in the Pathogen Box. Fig. S4 shows distribution of drugs in the Pathogen Box by their intended disease, quench rates, and HepG2 CC20. Fig. S5 shows the distribution of HepG2 CC20 values (from MMV) as function of NormRate. Fig. S6 shows scatter plots of the distributions of HepG2 CC20 and NormRate vs. ALogP and PSA. Fig. S7 shows bilayer-modifying potency and cytotoxicity vs. ALogP and PSA. Fig. S8 shows distribution of aromatic rings per drug vs. NormRate. Fig. S9 shows scatter plot of the distribution of QED vs. NormRate. Fig. S10 shows box plots of the distribution of NormRates and HepG2 CC20 as function of QED. Fig. S11 shows box plots of the distribution of NormRates and HepG2 CC20 as function of pScore. Table S1 provides detailed information about the 400 drugs in the Pathogen Box, which lists HepG2 CC20 values (for 397 drugs), information on protein binding (fumouse and fumic), NormRates and Range or SD, estimates of the drug concentrations in the aqueous and membrane phases, QED (along with the eight properties used to calculate it), and the pScore. Tables S2, S3, S4, and S5 provide effect of albumin on bilayer-modifying potency; average ALogP and PSA for drugs with low, intermediate, or high bilayer-modifying potency or cytotoxicity; odds for drugs being bilayer-modifying or cytotoxic vs. ALogP; and chemically similar drugs in the Pathogen Box, respectively. Three text sections are provided at the end of the PDF.

We first present results on the bilayer-modifying effects of the drugs in the Pathogen Box, as evaluated using stopped-flow fluorescence quench experiments to quantify their effects on the gramicidin monomer↔dimer equilibrium. We then consider the bilayer-mediated regulation of membrane protein function, emphasizing that the drugs tested here alter lipid bilayer properties as opposed to disrupting the membrane barrier properties. Next, we consider the question of the drug concentrations in the membrane required to cause the changes in quench rate and cytotoxicity (evaluated as changes in HepG2 CC20) and the relation(s) between molecular structure and bilayer-modifying potency. Then, we consider the possible relationship between bilayer-modifying drugs and PAINS. Finally, we discuss the implications for the use of small molecules to manipulate biological function including drug development.

### Stopped-flow fluorescence quench experiments reveal correlation between bilayer-modifying potency and cytotoxicity

The molecules’ bilayer-modifying potency was examined with 10 µM drug in the system (aqueous plus membrane phase) using stopped-flow spectrofluorometry.

Fig. 2 A shows fluorescence quench traces recorded with fluorophore-loaded, gramicidin-doped LUVs that had been equilibrated for 10 min in the absence or presence of drug and then mixed with either buffer or Tl+, a gramicidin channel-permeant fluorescence quencher. The quench rate varies with the number of open channels and therefore reflects shifts in the gramicidin monomer↔dimer equilibrium resulting from changes in bilayer properties. For drugs with low bilayer-modifying potency (meaning they produce little change in the monomer↔dimer equilibrium), such as MMV011229, the fluorescence quench traces will be similar to the control traces (absence of drug). For drugs with high bilayer-modifying potency (large shift in the gramicidin monomer↔dimer equilibrium), the fluorescence quench traces will differ from the control traces. MMV0689244, for example, increases the quench rate, which means that Tl+ enters the LUVs faster than in the absence of drug because there are more conducting channels in the LUV membrane (the monomer↔dimer equilibrium is shifted toward the conducting dimers).

Fig. 2 B shows the distribution of quench traces recorded in the presence of drugs. Different drugs increase the number of dimers (NormRate ≥ 1) to varying extents reflecting their bilayer modifying potency: green, orange, and red traces denote drugs with low (1 ≤ NormRate < 1.25), moderate (1.25 ≤ NormRate < 1.5), and high (1.5 ≤ NormRate) bilayer-modifying potencies. Importantly, the tested drugs did not compromise bilayer integrity—increase leakage of intravesicular contents during the 10 min incubation, which would result in instantaneous quench when drug-treated LUVs were exposed to the Tl+ quencher. This was not observed for any drug. Leakage of the trivalent ANTS out of the LUVs is slow, meaning undetectable after 24 h at 25°C: the initial drop in fluorescence, evaluated as F(0,b), is 0.02 ± 0.03 after 10 min and 0.08 ± 0.05 after 24 h under control conditions. In the presence of 5% ethanol, the positive control used in these studies, it was 0.16 ± 0.03 after 10 min and 0.17 ± 0.03 after 24 h; in the presence of 200 µM CHAPS, it was 0.09 ± 0.03 after 10 min and 0.09 ± 0.01 after 24 h.

The quench traces were fit with Eq. 11, the initial rates of Tl+ influx at t = 0 (Rate(0)) were calculated using Eq. 12, and the bilayer-mediated shift in the monomer↔dimer equilibrium was estimated as the NormRate, the quench rate in the presence of the drug normalized to the rate in the absence of drug, Eq. 13. The resulting NormRates are listed in Table S1, which also includes detailed information about the 400 drugs in the Pathogen Box.

Fig. 3 shows the distribution of NormRates binned by increasing NormRates for the 400 drugs in the Pathogen Box (for drugs with NormRate < 1, the results are plotted as 1/NormRate); see also Fig. S3 A and Fig. S4 A. Fig. S4 B shows the distribution of NormRates among drugs with different intended target diseases; there is little difference among the groups.

Comparing the drugs’ bilayer-modifying potency to their likely cytotoxicity, reported by MMV as HepG2 CC20 values (MMV, 2017), there was weak correlation between the two (r2 < 0.06) when analyzed using a simple scatter plot (Fig. S5 A). Drugs with high NormRates tend to have low CC20 values, but drugs with low NormRates do not necessarily have high CC20 values.

Binning the results by NormRate (Fig. 4), we found that more potent bilayer modifiers (those producing larger changes in NormRate) tend to have lower CC20 values (more likely to be cytotoxic; e.g., Greene et al., 2010).

Fig. 4 A shows the distribution of CC20 values binned by increasing NormRates. By inspection, drugs with 1 ≤ NormRate < 1.25 tend to have higher HepG2 CC20 values than drugs with 1.5 ≤ NormRate. Comparing the HepG2 CC20 distributions for the 199 drugs with 1 ≤ NormRate < 1.25 and the 124 drugs with 1.5 ≤ NormRate, the difference is significant, P = 2.1 · 10–6 (Fig. 4 B). As expected from Fig. 4 A, the probability a drug has a CC20 <50 µM (and therefore is likely to be cytotoxic; Greene et al., 2010) increases with increasing NormRate (Fig. 4 C); 90% of drugs with 1.5 ≤ NormRate have CC20 < 50 µM (100% of drugs with NormRate ≥ 4), whereas only 60% of drugs with NormRate < 1.25 have CC20 < 50 µM. Changes in quench rate therefore do not provide a rule to determine whether or not a molecule will be cytotoxic, rather they provide a measure of the probability that a molecule may be cytotoxic.

When comparing the distributions of NormRates for different ranges of HepG2 CC20 values (Fig. S5 B), there is a significant difference between drugs with 50 µM ≤ CC20 and drugs with CC20 < 50 µM (P = 6.8 ⋅ 10–8). Overall, Figs. 4 and S5 show that high quench rates tend to be associated with low CC20 values, meaning that drugs that are potent bilayer modifiers tend to be cytotoxic. The opposite need not be true, drugs with low quench rates may have a significant probability of being cytotoxic because drugs may be cytotoxic for reasons that are unrelated to the membrane. It is in this context relevant that the drugs in the Pathogen Box by design are likely to be cytotoxic (at least, for their intended target) through mechanisms that may not involve the bilayer.

Some drugs with 1.5 ≤ NormRate have large CC20 values (some are marked as outliers, denoted by +, in Fig. 4). This could be due to binding to proteins in the cell culture medium used in cytotoxicity assays or to metabolism, which would reduce the free concentrations (and likelihood of cytotoxicity). We retested these drugs in the presence of 60 µM BSA (Table S2). In all cases, NormRate in the presence of BSA (NormRateBSA) was less than NormRate in the absence of BSA, suggesting that these drugs (except, maybe, MMV688330) indeed bind to albumin. This was confirmed by independent information about the fraction of unbound drugs (fu) in the presence of fumic, fumouse (MMV, 2017). Other tested drugs may also bind to protein; we do not consider this further.

### Potent bilayer modifiers tend to alter membrane protein and cell function

The results in Figs. 4 and S5 show that the extent of drug-induced changes in bilayer properties (quantified as changes in NormRate) allow for predicting the likelihood a drug or drug-lead will be cytotoxic: 90% of drugs with 1.5 ≤ NormRate have CC20 < 50 µM (100% of drugs with 4 ≤ NormRate). Changes in cell membrane composition and lipid bilayer physical properties have long been known to alter membrane protein function (Seeman, 1972; Sandermann et al., 1978; Spector and Yorek, 1985; Bienvenüe and Marie, 1994; Andersen, 2007), which in turn will alter cell function (Spector and Yorek, 1985) and, when the changes in cell function are of sufficient magnitude, may cause cytotoxicity. Changes in protein function arise because the conformational equilibria of transmembrane transporters, channels, and receptors are sensitive to changes in their lipid bilayer environment. In some cases, e.g., the phosphoinositides (Hilgemann et al., 2018; Thompson and Baenziger, 2020; Cheng et al., 2022), specific lipid molecules function as direct or allosteric modulators of membrane protein function. In other cases, the regulation is due to the changes in lipid bilayer physical properties like thickness, intrinsic curvature, and the associated elastic moduli (Brown, 1994; Andersen, 2007).

As suggested by Spector and Yorek (1985), using different language, the changes in protein function are likely to reflect, at least in part, changes in the energetic cost of the bilayer adaptation/deformation to membrane protein conformational changes, which is the bilayer contribution (ΔGbilayer) to the free energy cost of the conformational changes (ΔGtotal) that underlie protein function (see Section 1 at the end of the PDF). The changes in cell function, in turn, reflect the aggregate result of indiscriminate changes in the function of many different membrane proteins that support membrane transport and cell signaling functions. The structure of, and conformational changes in, membrane proteins are, of course, different from the transmembrane dimerization of (nearly cylindrical) subunits, and the bilayer-mediated regulation of integral membrane proteins is likely to have features not observed with gramicidin channels, though conformational changes in transporters involve changes in the local lipid packing (Norimatsu et al., 2017; Wang and Boudker, 2020) that may be associated with substantial changes in ΔGbilayer (Zhou et al., 2019). Fig. 5 shows a highly schematized membrane protein (ion channel) highlighting different non-exclusive mechanisms for drug modulation of membrane protein function (see also Payandeh and Volgraf, 2021).

The mechanisms range from binding to the pore/active site to block ion movement/catalytic activity (Site 1), over allosteric regulation arising from (specific) binding to the target protein (Sites 2 and 3), to allosteric regulation that arises from (less specific) drug-induced changes in lipid structure and dynamics in the lipid shells adjacent to the protein (Sites 3 and 4), and in bulk bilayer material properties (Site 5). For any protein, $ΔGdefI,$$ΔGdefII,$ and $ΔGbilayerI→II$ vary with bilayer thickness, intrinsic lipid curvature and the associated elastic moduli, which are determined by intermolecular interactions among the membrane lipids (Helfrich, 1981; Venable et al., 2015); ΔGdef also depends on protein “shape,” protein-bilayer hydrophobic mismatch that reflect interactions between embedded proteins and their adjacent lipids (Cantor, 1997; Dan and Safran, 1998; Mondal et al., 2011; Corradi et al., 2019; Zhou et al., 2019; Obi and Natesan, 2022).

The lipid bilayer-dependent parameters pertain to all membrane proteins, whether gramicidin channels or integral membrane proteins, meaning that changes in gramicidin channel function will predict changes in integral membrane protein function (Lundbaek et al., 2005; Rusinova et al., 2011; see also Lundbaek et al., 2010a), though the magnitude of the drug-induced changes in channel function (NormRate) varies with bilayer composition (Elliott et al., 1985; Bruno et al., 2007; Rusinova et al., 2011; Rusinova et al., 2015; Herold et al., 2017; Sun et al., 2020), which could reflect changes in drug partition coefficients (e.g., Haydon et al., 1977; Mason et al., 1992), or that the bilayer contribution to the free energy of dimer formation varies with lipid composition (Sun et al., 2020). The protein-specific parameters depend on the protein in question, meaning that the bilayer-mediated regulation of protein function by small molecules may include contributions that are not detected by gramicidin channels (e.g., Rusinova et al., 2021).

In the case of orally administered drugs, drug-target engagement (Hughes et al., 2011; Simon et al., 2013; Stefaniak and Huber, 2020) requires that drugs cross one or more membranes, whether by solubility-diffusion through the membranes’ lipid bilayer component or by protein-catalyzed mechanisms (Sugano et al., 2010; Smith et al., 2014; Basore et al., 2015). Irrespective of the mechanism by which they cross cell membranes, many drugs and drug leads are sufficiently hydrophobic that they partition into the membranes’ bilayer/solution interface (e.g., Seeman, 1972; Avdeef et al., 1998; Rusinova et al., 2011; Kapoor et al., 2019; Bennett et al., 2020), where they will alter many, if not all, bilayer properties including thickness, intrinsic curvature, acyl chain order, elasticity, fluidity, phase transition temperature, and others (e.g., Seddon et al., 2009). Changes in any of these properties may impact membrane protein function through their aggregate effect on the bilayer contribution, $ΔGbilayerI→II,$ to the free energy cost of membrane protein conformational changes, with the dominant mechanism likely to be the thermodynamic membrane softening caused by the reversible partitioning of drugs into the bilayer/solution interface (Evans et al., 1995; Zhelev, 1998; Lundbaek et al., 2010b; Rusinova et al., 2011; Kapoor et al., 2019). Though the term “fluidity” often is invoked, A. G. Lee showed long ago (Lee, 1991) that changes in fluidity do not serve as a primary mechanism for regulating membrane protein function. The accumulation of drugs into the bilayer/solution interface may also alter protein function by mechanisms that involve more direct, if non-specific, interactions (Andersen, 2008; Payandeh and Volgraf, 2021; Rusinova et al., 2021; Cheng et al., 2022), which also would alter the free energy cost of protein conformational changes, or cause even more complex changes such as phospholipid hydrolysis (Baciu et al., 2006), phospholipidosis (Tummino et al., 2021), or changes in membrane domain organization (Fricke et al., 2022).

Importantly, the changes in $ΔGbilayerI→II$ (and the ensuing changes in protein and cell function that may cause cytotoxicity) occur at drug concentrations that do not cause a breakdown in membrane barrier properties (as evident from the fluorescence quench traces in Fig. 2). Rather, the drugs cause more subtle changes in bilayer properties that corrupt normal membrane protein function, and thereby cell homeostasis and signaling.

### Drug concentrations in the membrane is a factor in cytotoxicity

The drugs in the Pathogen Box have been tested for cytotoxicity on HEPG2 cells and reported as HepG2 CC20 values (MMV, 2017), which we used for our evaluation of the relation between bilayer-modifying potency and likely cytotoxicity. Previous studies have shown that ∼95% of molecules tested for cytotoxicity have similar effects on different cell lines (Lin and Will, 2012; Chiaravalli and Glickman, 2017; Lee et al., 2020).

We tested for bilayer-modifying effects at 10 µM, a concentration commonly used to determine a molecule’s cytotoxicity (Chiaravalli and Glickman, 2017). Our results can be extrapolated to other concentrations because a bilayer-modifying molecule’s effect on bilayer properties, as estimated using the gramicidin monomer↔dimer equilibrium, varies as an approximately linear function of the aqueous drug concentration (Ingólfsson and Andersen, 2010; Ingólfsson and Andersen, 2011; Alejo et al., 2013; Kapoor et al., 2019), which allows for estimating the bilayer-modifying effect (change in NormRate) at other drug concentrations (see Section 3 and Eq. S2 at the end of the PDF).

The relevant concentrations in this context are the actual drug concentrations in the membrane ([Drug]m) and the aqueous phase ([Drug]a), where [Drug]a may be less than the nominal concentration in the system, [Drug]nom, because drugs will equilibrate between the aqueous and membrane phases (Bruno et al., 2007; Rusinova et al., 2011; Heerklotz and Keller, 2013; Kapoor et al., 2019). Knowing a drug’s partition coefficient, K1, defined as K1 = [Drug]m/[Drug]a, [Drug]a and [Drug]m can be expressed as:
$[Drug]a=[Drug]nom1+K1⋅Vlip/Vaq;[Drug]m=K1⋅[Drug]nom1+K1⋅Vlip/Vaq,$
(16)
where Vlip and Vaq denote the volumes of the lipid (acyl chains plus head group) and aqueous phases, respectively (Vlip/Vaq = 3.6 ⋅ 10–5 in our experiments; see Section 3 and Eqs. S22 and S23 at the end of the PDF). Approximating K1 by ALogP, we can estimate [Drug]a and [Drug]m at [Drug]nom = 10 µM. These [Drug]a and [Drug]m estimates are summarized in Table S1; the [Drug]a estimates vary between 0.5 nM (for the most hydrophobic drugs) and 10 µM, the associated [Drug]m estimates vary between sub-pM (for the least hydrophobic drugs) and 280 mM. In any case, it will be important to evaluate a drug candidate’s bilayer-modifying potency at the actual, unbound concentrations where it has its desired effects. Drugs that exert their on-target effects at low nanomolar concentrations may well be bilayer modifying (and cytotoxic) at 10 μM, but that may not to be important at the concentrations where the drugs exert their desired effects.

### Physicochemical properties are weak predictors of bilayer-modifying potency and cytotoxicity (HepG2 CC20)

Drug candidates’ physicochemical properties are important for their successful development (e.g., Lipinski et al., 2001; Leeson and Springthorpe, 2007; Hughes et al., 2008; Bickerton et al., 2012). Physicochemical properties determine not only a drug’s pharmacokinetic profile (Lipinski et al., 2001) but also its promiscuity (Leeson and Springthorpe, 2007), toxicity (Hughes et al., 2008), and overall drug-likeness (Bickerton et al., 2012). Yet, standard physico-chemical properties (ALogP and PSA) do not allow for robust prediction of cytotoxicity.

Fig. 4, plus Fig. S4 A and Fig. S5, shows that the tested drugs’ bilayer-modifying potency is correlated with their cytotoxicity (as quantified by their HepG2 CC20). In contrast, there was little/no correlation between cytotoxicity and physicochemical parameters used in drug design (Hughes et al., 2008; Price et al., 2009) such as ALogP (r2 < 0.05) and PSA (r2 < 1.8⋅10–3; Fig. S6). Fig. S6, A and B, shows the distribution of HepG2 CC20 values vs. ALogP and PSA, respectively. Fig. S6, C and D, show the corresponding distribution of NormRates vs. ALogP and PSA. There is a weak correlation between the drugs’ NormRate and ALogP (r2 < 0.1), but no correlation with PSA (r2 < 2.6⋅10–5); Table S3 summarizes information about the average ALogP and PSA values for the six groups (low, medium, and high NormRate or CC20). Focusing on ALogP, comparing drugs with ALogP ≤ 3 and ALogP > 3, the odds ratios for a drug being bilayer-modifying or cytotoxic are 3.8 and 2.0, respectively, (95% CI: 2.4–5.9) and (1.3–3.1); Table S4.

Yet, both ALogP and PSA are likely to be important for a drug’s partitioning in the bilayer/solution interface. Combining ALogP and PSA did not yield strong correlation with our experimental results (Fig. S7). But following Hughes et al. (2008) (see also Price et al., 2009), who found that molecules with CLogP > 3 with a relatively small total polar surface area (TPSA < 75 Å2) were more likely to be promiscuous and cytotoxic, we calculated the odds for a drug being bilayer-modifying (NormRate ≥ 1.25) after grouping the drugs into four groups: ALogP ≤ 3 (low ALogP) and > 3 (high ALogP); PSA ≤ 75 Å2 (low PSA) and > 75 Å2 (high PSA; Table 1, top). The bottom of Table 1 shows the corresponding information for drugs having CC20 < 50 µM.

Bilayer-modifying potency and the likelihood that a drug is cytotoxic (has a CC20 < 50 µM) increase with increasing ALogP and, for a given ALogP range, the bilayer-modifying potency and cytotoxicity (though to a lesser extent) increase with increasing PSA. Whereas neither ALogP nor PSA are robust predictors of bilayer-modifying potency or likely cytotoxicity (Table S3), the combination of a high ALogP (≥3) and PSA (≥75 Å2) is associated with increased bilayer-modifying potency: comparing drugs having low ALogP and PSA with drugs having high ALogP and PSA, the odds ratio for drugs having a NormRate ≥1.25 is 11-fold higher for the latter group (Table 1).

Drugs with high ALogP and PSA, which will tend to localize near the bilayer/solution interface, thus tend to be more potent bilayer modifiers than drugs with low ALogP and PSA. Yet, despite the odds ratios, ALogP together with PSA do not allow for predicting a drug’s bilayer-modifying potency because of overlap among the groups (Fig. S7). For cytotoxicity: comparing drugs having ALogP ≤ 3 and PSA ≤ 75 Å2 to drugs having ALogP > 3 and PSA > 75 Å2, the odds ratio for a drug having a HepG2 CC20 < 50 µM is only 2.5-fold higher for the latter group, which may be due to the high number of cytotoxic drugs with modest bilayer-modifying potency (drugs may be toxic for reasons that have nothing to do with the membrane). In contrast to Hughes et al. (2008), however, for drugs with ALogP > 3, those with PSA > 75 Å2 are as likely to be cytotoxic (have CC20 < 50 µM) as those with PSA ≤ 75 Å2.

We also explored other parameters commonly considered including molecular mass, number of heavy atoms, number of hydrogen bond donors or acceptors, and number of Rule of 5 violations (Lipinski et al., 2001); none showed meaningful correlation to NormRate (r2 < 0.1, r2 < 0.1, r2 < 8.8⋅10–6, r2 < 3.9⋅10–3, and r2 < 0.1, respectively) or HepG2 CC20 (r2 < 6.3⋅10–3, r2 < 6.8⋅10–3, r2 < 4.6⋅10–5, r2 < 2.1⋅10–3, and r2 < 4.6⋅10–3, respectively). There was a correlation with the number of aromatic rings in the drugs (Fig. S8); drugs with more aromatic rings tended to have higher NormRates, which may reflect that molecules with more aromatic rings are likely to be more hydrophobic (e.g., Ritchie and Macdonald, 2009); the average ALogP of drugs with ≤2 aromatic rings was 3.0, whereas the average ALogP of drugs with ≥3 aromatic rings was 3.7.

The above parameters (plus information about structural alerts; Brenk et al., 2008) have been combined into the QED score (Bickerton et al., 2012), which is a weighted score based on the following descriptors: molecular mass, ALogP, PSA, number of hydrogen bond donors, number of hydrogen bond acceptors, number of rotatable bonds, number of aromatic rings, and number of structural alert. Although QED was not developed to predict toxicity per se, successful drugs cannot possess unacceptable toxicity, meaning that QED scores implicitly include an assessment of toxicity. Fig. S9 shows the distribution of QED vs. NormRate (r2 < 0.08).

Fig. 6 shows the binned distribution of QED values as function of NormRate. QED tends to decrease with increasing NormRate (Fig. 6 A). Comparing drugs with NormRate < 1.25 and 1.5 ≤ NormRate, (Fig. 6 B) high NormRates tend to be correlated with low QED scores (P = 4.2⋅10–7). Fig. S10 shows the relation between QED and NormRate or CC20 and QED; there is no systematic trend in either distribution or the probability that a drug has a CC20 < 50 µM does not vary significantly between molecules with 0 ≤ QED < 0.2 and 0.8 ≤ QED ≤ 1 (Fig. 6 C).

Combining NormRate and QED, however, provides improved prediction of drugs having CC20 < 50 µM: 82% of drugs with NormRate ≥ 1.25 have CC20 < 50 µM (n = 198); 79% of drugs with QED < 0.5 have CC20 < 50 (n = 71), whereas 89% of drugs with NormRate ≥ 1.25 and QED < 0.5 have CC20 < 50 (n = 47).

### Bilayer-modifying potency does not predict and pan assay interference

A perennial problem in drug development is the so-called PAINS (Baell and Holloway, 2010) or nuisance compounds (Dahlin et al., 2021), and some PAINS are potent bilayer modifiers (Baell and Walters, 2014; Aldrich et al., 2017). Yet, the molecular properties that cause a molecule to be bilayer-active (hydrophobicity, sufficient polarity to localize to the bilayer/solution interface) do not involve chemical reactivity, and bilayer-modifying potency per se may not be sufficient to cause a molecule to be promiscuous and masquerade as a hit in high-throughput protein-based screens. We, therefore, explored the relationship between a drug’s bilayer-modifying potency and its promiscuity index (pScore) using Badapple (http://datascience.unm.edu/public-biocomputing-apps; Yang et al., 2016). The results (Fig. 7, A and B; and Fig. S11 A) show that there is no correlation between bilayer-modifying potency, quantified as NormRate, and pScore, although drugs with high pScores are more likely to have low CC20 values than drugs with low pScores (Fig. S11 B).

Bilayer-active molecules, however, are likely to masquerade as hits in cell-based screens because changes in bilayer properties, measured as changes in the gramicidin monomer↔dimer equilibrium, will produce changes in the function of many, diverse membrane proteins (Lundbaek et al., 2010a; Ingólfsson et al., 2014), which may alter overall cell function, thereby making the molecule appear to be a hit. Bilayer-active molecules thus can be considered as cell-based assay interference compounds (CAINS; see also Sun et al., 2020) or membrane PAINS (Magalhães et al., 2022).

### Drug-induced changes in bilayer properties, implications for drug development and laboratory research?

The motivation for this study was to explore whether it would be possible to use simple and fast biophysical measurements to gain information about the likelihood that a drug candidate will be cytotoxic. Our results show that drug-induced changes in bilayer properties (quantified as NormRate) predict the probability a molecule will be cytotoxic, as quantified by its HepG2 CC20 (Fig. 4). This does not, however, provide a rule for determining whether a molecule is cytotoxic because molecules may be cytotoxic for many reasons that do not involve the bilayer, rather it provides an estimate of the probability a molecule may be cytotoxic, information that will be valuable when interpreting the results of cell physiological experiments of selecting molecules for drug development. This strategy is almost uniquely suited for such a complementary analysis because it is robust and fast: it can be completed (drug equilibration with fluorophore-loaded LUVs, conducting the stopped-flow experiments, and analyzing the results) within 30 min.

The correlation between drug-induced changes in the gramicidin monomer↔dimer equilibrium (lipid bilayer properties) and the changes in HepG2 CC20, as a measure of cytotoxicity (Fig. 4), show that drug-induced changes in bilayer properties will perturb membrane protein and cell function, which may cause off-target effects and, if the changes in function are large, cytotoxicity. It cannot be excluded, however, that subtle, bilayer-mediated changes in the function of many proteins could produce desired changes in system function (e.g., Eger et al., 2008; Rusinova et al., 2015). In any case, this provides a guide for drug development because chemical modifications that reduce the likelihood of bilayer perturbation, while leaving desired therapeutic effects intact, may produce candidates for development where measurable bilayer perturbation at concentrations much higher than the desired effects (Rusinova et al., 2011; Rusinova et al., 2015)—assuming the desired effect is not due to bilayer-mediated regulation. Substituting the naphthalene (octanol/water partition coefficient = 2.2⋅103; Leo et al., 1971) in propranolol with indole (octanol/water partition coefficient = 1.8⋅102; Leo et al., 1971) in pindolol, for example, reduces the bilayer-modifying potency by an order of magnitude (Rusinova et al., 2015), and the bilayer-modifying potency for short-chain n-alcohols scales with their octanol/water partition coefficients (Ingólfsson and Andersen, 2011). These and other studies (Rusinova et al., 2011; Howery et al., 2012; Dockendorff et al., 2018; Zhang et al., 2018; Kapoor et al., 2019) have shown how seemingly modest alterations in a molecule’s structure—including cis-trans isomerization (Howery et al., 2012)—may produce large changes in its bilayer-modifying potency and effect on membrane protein function. Not surprisingly, therefore, it is possible to modify a potent bilayer-modifying molecule (e.g., resveratrol; Ingólfsson et al., 2014), to minimize its bilayer-perturbing effects while maintaining its desired biological effects (Bosquesi et al., 2020). To our knowledge, this is the first example of a drug candidate that was developed with explicit consideration of its bilayer-modifying properties; a similar strategy has been used to minimize the off-target effects of photostabilizers used to protect fluorescent reporter groups (Grenier et al., 2022).

Table S5 lists pairs of drugs in the Pathogen Box that are chemically similar. Table S5 A lists drugs with different bilayer-modifying potencies; Table S5 B lists drugs with comparable bilayer-modifying potencies. In some cases, the different bilayer-modifying potencies may reflect the drugs’ different hydrophobicity (ALogP) and partitioning into the bilayer; in other cases, e.g., MMV676269 and MMV676270, the differences may reflect drug-induced alterations in the acyl chain dynamics. These results, together with previous studies summarized above, suggest that it may be possible to use a drug-lead’s bilayer-modifying potency as a guide to synthesize analogs that retain the desired biological effects but with less bilayer-modifying potency (see also Payandeh and Volgraf, 2021).

We finally note that the experimental strategy used here also can be used to determine whether changes in membrane protein, or cell, function caused by a bioactive molecule might be due to drug-induced changes in bilayer properties, as opposed to specific, on-target interactions. Platelet-activating factor (1-O-alkyl-2-acetyl-sn-glycero-3-phosphocholine; PAF), for example, binds to a GPCR (Honda et al., 1991) and activates pathways involved in coagulation and inflammation at low nanomolar concentrations (Demopoulos et al., 1979). At high nanomolar concentrations, PAF has additional effects, such as inducing differentiation of cultured neurons, and it becomes cytotoxic at low micromolar concentrations (Kornecki and Ehrlich, 1988). At these concentrations, PAF also alters gramicidin channel function and disrupts bilayer properties (Sawyer and Andersen, 1989) suggesting that bilayer-mediated mechanisms may be involved. Amiodarone is widely used to treat complex cardiac arrythmias (Mujović et al., 2020), but has a complex therapeutic profile and exerts its effects through mechanisms that involve diverse ion channels, transporters, and receptors (Heijman et al., 2013) with no well-defined primary target, suggesting a bilayer-mediated mechanism. Indeed, amiodarone alters bilayer properties at concentrations where it exerts its clinical effects (Rusinova et al., 2015), which may provide a mechanism for its poly-pharmacology (see also Lundbaek et al., 2010a).

At the other extreme, a bilayer-mediated mechanism can be largely excluded if a drug’s desired effects occur at concentrations where it does not appear to alter bilayer properties, as reflected in changes in $ΔGbilayerM→D,$ e.g., in the case of general anesthetics (Herold et al., 2014; Herold et al., 2017) or alkylphenol propofol analogs (Tibbs et al., 2013), or there is no correlation between a drug’s effect on its target and its bilayer-modifying potency, e.g., in the case of the marine toxin 6-bromo-2-marcaptotryptamine dimer (BrMT; Dockendorff et al., 2018). It is in this context important that ΔGbilayer is the difference between two ΔGdefs: $ΔGbilayerM→D=ΔGdefD−2⋅ΔGdefM,$ and $ΔGbilayerI→II=ΔGdefII−ΔGdefI.$ Drug-induced changes in ΔGbilayer thus may be 0 even though the drug changes the underlying ΔGdef s; that is, $ΔGbilayerI→II=0$ when the changes in $ΔGdefII$ are equal to the changes in $ΔGdefI.$

### Conclusions

We have shown that drug-induced changes in the transmembrane gramicidin monomer↔dimer equilibrium in a model membrane composed of a single lipid species correlate with changes in cell function, specifically the drugs’ cytotoxicity, estimated as HepG2 CC20. Even modest changes in bilayer properties (evaluated as changes in the fluorescence quench rate, which reflect shifts in the gramicidin monomer↔dimer equilibrium) are associated with a reduction in CC20, indicative of increased risk of cytotoxicity (Fig. 4). Although gramicidin monomer↔dimer transitions differ from conformational transitions in integral membrane proteins, both involve rearrangements within the bilayer hydrophobic core (e.g., Lundbæk et al., 2010a). Drug-induced changes in lipid bilayer properties therefore will alter the distribution among membrane protein conformations, and drug-induced changes in gramicidin channel function can be related to changes in membrane protein function (Lundbaek et al., 2005; Søgaard et al., 2006; Chisari et al., 2010; Lundbaek et al., 2010a; Rusinova et al., 2011; Herold et al., 2014; Ingólfsson et al., 2014; Herold et al., 2017). Because drugs that alter bilayer properties will produce indiscriminate changes in membrane protein and, in turn, cell function, potent bilayer-modifying drugs would be expected to be cytotoxic, at some concentration. The ability to predict changes in cell function based on a simple biophysical measurement shows that the cell membrane’s lipid bilayer moiety is a target for bioactive molecule. It further provides information about the concentrations where a drug can be used to manipulate membrane protein and cell function with minimal risk of bilayer-mediated regulation. Drugs (drug candidates) that exert their desired effects only at concentrations where they modify lipid bilayer properties are unlikely to be successful. The ability to identify such molecules therefore opens up for new, mechanism-based approaches to guide drug development, which may help reduce the cost of developing therapeutics, including drugs for NTD.

Christopher J. Lingle served as editor.

We thank Medicines for Malaria Venture for providing the Pathogen Box; and Sara Bahri, Kelsey Brown, Helgi I Ingólfsson, Radda Rusinova, R. Lea Sanford, Daniel S. Stor for discussions and comments on earlier versions of the manuscript. We also thank three anonymous reviewers for their insightful comments, which helped us improve the presentation.

This work was supported by a grant from the National Institutes of Health (GM021342-45) to O.S. Andersen.

The authors declare no competing financial interests.

Author contributions: T.A. Peyear and O.S. Andersen developed the hypothesis to be tested. O.S. Andersen requested the Pathogen Box from Medicines for Malaria Venture. T.A. Peyear performed the experiments and the analysis. T.A. Peyear and O.S. Andersen jointly wrote the manuscript.

Abo-Riziq
,
A.
,
B.O.
Crews
,
M.P.
Callahan
,
L.
Grace
, and
M.S.
de Vries
.
2006
.
Spectroscopy of isolated gramicidin peptides
.
Angew. Chem. Int. Ed.
45
:
5166
5169
.
Aldrich
,
C.
,
C.
Bertozzi
,
G.I.
Georg
,
L.
Kiessling
,
C.
Lindsley
,
D.
Liotta
,
K.M.
Merz
Jr
,
A.
Schepartz
, and
S.
Wang
.
2017
.
The ecstasy and agony of assay interference compounds
.
J. Med. Chem.
60
:
2165
2168
.
Alejo
,
J.L.
,
S.C.
Blanchard
, and
O.S.
Andersen
.
2013
.
Small-molecule photostabilizing agents are modifiers of lipid bilayer properties
.
Biophys. J.
104
:
2410
2418
.
Allen
,
T.W.
,
O.S.
Andersen
, and
B.
Roux
.
2003
.
The structure of gramicidin A in a lipid bilayer environment determined using molecular dynamics simulations and solid-state NMR data
.
J. Am. Chem. Soc
.
125
:
9868
9878
.
Andersen
,
O.S.
2008
.
Perspectives on how to drug an ion channel
.
J. Gen. Physiol.
131
:
395
397
.
Andersen
,
O.S.
, and
R.E.
Koeppe
II
.
2007
.
Bilayer thickness and membrane protein function: An energetic perspective
.
Annu. Rev. Biophys. Biomol. Struct.
36
:
107
130
.
Andersen
,
O.S.
,
D.B.
Sawyer
, and
R.E.
Koeppe
, II
.
1992
. Modulation of Channel Function by the Host Bilayer. In
Biomembrane Structure and Function
.
K.R.K.
Easwaran
, and
B.
Geber
, editors.
,
. pp.
227
244
.
Andersen
,
O.S.
,
M.J.
Bruno
,
H.
Sun
, and
R.E.
Koeppe
, II
.
2007
.
Single-molecule methods for monitoring changes in bilayer elastic properties
.
Methods Mol. Biol
.
400
:
543
570
.
Arseniev
,
A.S.
,
I.L.
Barsukov
, and
V.F.
Bystrov
.
1986
a.
Conformation of gramicidin A in solution and micelles: two-dimensional proton NMR study
.
Chem. Pept. Proteins
.
3
:
127
158
.
Arseniev
,
A.S.
,
I.L.
Barsukov
,
V.F.
Bystrov
, and
Y.A.
Ovchinnikov
.
1986
b.
Spatial structure of a gramicidin A transmembrane ion channel. NMR analysis in micelles
.
Biol. Membr
.
3
:
437
462
.
Arsinev
,
A.S.
,
I.L.
Barsukov
,
V.F.
Bystrov
,
A.L.
Lomize
, and
Y.A.
Ovchinnikov
.
1985
.
1H-NMR study of gramicidin A transmembrane ion channel. Head-to-head right-handed, single-stranded helices
.
FEBS Lett
.
186
:
168
174
.
Artigas
,
P.
,
S.J.
Al’aref
,
E.A.
Hobart
,
L.F.
Díaz
,
M.
Sakaguchi
,
S.
Straw
, and
O.S.
Andersen
.
2006
.
2,3-butanedione monoxime affects cystic fibrosis transmembrane conductance regulator channel function through phosphorylation-dependent and phosphorylation-independent mechanisms: The role of bilayer material properties
.
Mol. Pharmacol.
70
:
2015
2026
.
Avdeef
,
A.
,
K.J.
Box
,
J.E.
Comer
,
C.
Hibbert
, and
K.Y.
Tam
.
1998
.
pH-metric logP 10. Determination of liposomal membrane-water partition coefficients of ionizable drugs
.
Pharm. Res.
15
:
209
215
.
Baciu
,
M.
,
S.C.
Sebai
,
O.
Ces
,
X.
Mulet
,
J.A.
Clarke
,
G.C.
Shearman
,
R.V.
Law
,
R.H.
Templer
,
C.
Plisson
,
C.A.
Parker
, and
A.
Gee
.
2006
.
Degradative transport of cationic amphiphilic drugs across phospholipid bilayers
.
Philos. Trans. R. Soc. Math. Phys. Eng. Sci.
364
:
2597
2614
.
Baell
,
J.
, and
M.A.
Walters
.
2014
.
Chemistry: Chemical con artists foil drug discovery
.
Nature
.
513
:
481
483
.
Baell
,
J.B.
, and
G.A.
Holloway
.
2010
.
New substructure filters for removal of pan assay interference compounds (PAINS) from screening libraries and for their exclusion in bioassays
.
J. Med. Chem.
53
:
2719
2740
.
Bamberg
,
E.
, and
P.
Läuger
.
1973
.
Channel formation kinetics of gramicidin A in lipid bilayer membranes
.
J. Membr. Biol.
11
:
177
194
.
Basore
,
K.
,
Y.
Cheng
,
A.K.
Kushwaha
,
S.T.
Nguyen
, and
S.A.
Desai
.
2015
.
How do antimalarial drugs reach their intracellular targets?
.
Front. Pharmacol.
6
:
91
.
Beaven
,
A.H.
,
A.M.
Maer
,
A.J.
Sodt
,
H.
Rui
,
R.W.
Pastor
,
O.S.
Andersen
, and
W.
Im
.
2017
.
Gramicidin A channel formation induces local lipid redistribution I: Experiment and simulation
.
Biophys. J
.
112
:
1185
1197
.
Bennett
,
W.F.D.
,
S.
He
,
C.L.
Bilodeau
,
D.
Jones
,
D.
Sun
,
H.
Kim
,
J.E.
Allen
,
F.C.
Lightstone
, and
H.I.
Ingólfsson
.
2020
.
Predicting small molecule transfer free energies by combining molecular dynamics simulations and deep learning
.
J. Chem. Inf. Model.
60
:
5375
5381
.
Berberan-Santos
,
M.N.
,
E.N.
Bodunov
, and
B.
Valeur
.
2005
.
Mathematical functions for the analysis of luminescence decays with underlying distributions 1. Kohlrausch decay function (stretched exponential)
.
Chem. Phys.
315
:
171
182
.
Bickerton
,
G.R.
,
G.V.
Paolini
,
J.
Besnard
,
S.
Muresan
, and
A.L.
Hopkins
.
2012
.
Quantifying the chemical beauty of drugs
.
Nat. Chem.
4
:
90
98
.
Bienvenüe
,
A.
, and
J.S.
Marie
.
1994
.
Modulation of protein function by lipids
.
Curr. Top. Membr.
40
:
319
354
.
Bosquesi
,
P.L.
,
A.C.B.
Melchior
,
A.R.
Pavan
,
C.
Lanaro
,
C.M.
de Souza
,
R.
Rusinova
,
R.C.
Chelucci
,
K.P.
Barbieri
,
G.F.D.S.
Fernandes
,
I.Z.
Carlos
, et al
.
2020
.
Synthesis and evaluation of resveratrol derivatives as fetal hemoglobin inducers
.
Bioorg. Chem.
100
:
103948
.
Brenk
,
R.
,
A.
Schipani
,
D.
James
,
A.
Krasowski
,
I.H.
Gilbert
,
J.
Frearson
, and
P.G.
Wyatt
.
2008
.
Lessons learnt from assembling screening libraries for drug discovery for neglected diseases
.
ChemMedChem
.
3
:
435
444
.
Brown
,
M.F.
1994
.
Modulation of rhodopsin function by properties of the membrane bilayer
.
Chem. Phys. Lipids.
73
:
159
180
.
Bruno
,
M.J.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2007
.
Docosahexaenoic acid alters bilayer elastic properties
.
.
104
:
9638
9643
.
Cantor
,
R.S.
1997
.
Lateral pressures in cell membranes: A mechanism for modulation of protein function
.
J. Phys. Chem. B.
101
:
1723
1725
.
Cantor
,
R.S.
1999
.
The influence of membrane lateral pressures on simple geometric models of protein conformational equilibria
.
Chem. Phys. Lipids
.
101
:
45
56
.
Chen
,
F.Y.
,
M.T.
Lee
, and
H.W.
Huang
.
2003
.
Evidence for membrane thinning effect as the mechanism for peptide-induced pore formation
.
Biophys. J.
84
:
3751
3758
.
Cheng
,
W.W.L.
,
M.J.
Arcario
, and
J.T.
Petroff
II
.
2022
.
Druggable lipid binding sites in pentameric ligand-gated ion channels and transient receptor potential channels
.
Front. Physiol.
12
:
798102
.
Chiaravalli
,
J.
, and
J.F.
Glickman
.
2017
.
A high-content live-cell viability assay and its validation on a diverse 12K compound screen
.
SLAS Discov.
22
:
1120
1130
.
Chisari
,
M.
,
H.J.
Shu
,
A.
Taylor
,
J.H.
Steinbach
,
C.F.
Zorumski
, and
S.
Mennerick
.
2010
.
Structurally diverse amphiphiles exhibit biphasic modulation of GABAA receptors: Similarities and differences with neurosteroid actions
.
Br. J. Pharmacol.
160
:
130
141
.
Clayton
,
K.N.
,
J.W.
Salameh
,
S.T.
Wereley
, and
T.L.
Kinzer-Ursem
.
2016
.
Physical characterization of nanoparticle size and surface modification using particle scattering diffusometry
.
Biomicrofluidics
.
10
:
054107
.
,
V.
,
B.I.
Sejdiu
,
H.
Mesa-Galloso
,
H.
,
S.Y.
Noskov
,
S.J.
Marrink
, and
D.P.
Tieleman
.
2019
.
Emerging diversity in lipid-protein interactions
.
Chem. Rev.
119
:
5775
5848
.
Dahlin
,
J.L.
,
D.S.
Auld
,
I.
Rothenaigner
,
S.
Haney
,
J.Z.
Sexton
,
J.W.M.
Nissink
,
J.
Walsh
,
J.A.
Lee
,
J.M.
Strelow
,
F.S.
Willard
, et al
.
2021
.
Nuisance compounds in cellular assays
.
Cell Chem. Biol.
28
:
356
370
.
Dan
,
N.
, and
S.A.
Safran
.
1998
.
Effect of lipid characteristics on the structure of transmembrane proteins
.
Biophys. J.
75
:
1410
1414
.
Demopoulos
,
C.A.
,
R.N.
Pinckard
, and
D.J.
Hanahan
.
1979
.
Platelet-activating factor. Evidence for 1-O-alkyl-2-acetyl-sn-glyceryl-3-phosphorylcholine as the active component (a new class of lipid chemical mediators)
.
J. Biol. Chem.
254
:
9355
9358
.
Dockendorff
,
C.
,
D.M.
Gandhi
,
I.H.
Kimball
,
K.S.
Eum
,
R.
Rusinova
,
H.I.
Ingólfsson
,
R.
Kapoor
,
T.
Peyear
,
M.W.
Dodge
,
S.F.
Martin
, et al
.
2018
.
Synthetic analogs of the snail toxin 6-bromo-2-mercaptotryptamine dimer (BrMT) reveal that lipid bilayer perturbation does not underlie its modulation of voltage-gated potassium channel
.
Biochemistry
.
57
:
2733
2743
.
Dubos
,
R.J.
1939
.
Studies on a bactericidal agent extracted from a soil bacillus: I. Preparation of the agent. Its activity in vitro
.
J. Exp. Med.
70
:
1
10
.
Eger
,
E.I.
, II
,
D.E.
Raines
,
S.L.
Shafer
,
H.C.
Hemmings
Jr
, and
J.M.
Sonner
.
2008
.
Is a new paradigm needed to explain how inhaled anesthetics produce immobility?
.
Anesth. Analg.
107
:
832
848
.
Elliott
,
J.R.
,
D.
Needham
,
J.P.
Dilger
,
O.
Brandt
, and
D.A.
Haydon
.
1985
.
A quantitative explanation of the effects of some alcohols on gramicidin single-channel lifetime
.
Biochim. Biophys. Acta
.
814
:
401
404
.
Elliott
,
J.R.
,
D.
Needham
,
J.P.
Dilger
, and
D.A.
Haydon
.
1983
.
The effects of bilayer thickness and tension on gramicidin single-channel lifetime
.
Biochim. Biophys. Acta
.
735
:
95
103
.
Evans
,
E.
,
W.
Rawicz
, and
A.F.
Hofmann
.
1995
.
Lipid bilayer expansion and mechanical disruption in solutions of water-soluble bile acid
. In
Bile Acids in Gastroenterology: Basic and Clinical Advances
.
A.F.
Hofmann
,
G.
Paumgartner
, and
A.
Stiehl
, editors.
,
Dordrecht
.
59
68
.
Fattal
,
D.R.
, and
A.
Ben-Shaul
.
1993
.
A molecular model for lipid-protein interaction in membranes: the role of hydrophobic mismatch
.
Biophys. J
.
65
:
1795
1809
. 10.1016/S0006-3495(93)81249-9.
Fricke
,
N.
,
K.
Raghunathan
,
A.
Tiwari
,
K.M.
Stefanski
,
M.
Balakrishnan
,
A.G.
Waterson
,
R.
Capone
,
H.
Huang
,
C.R.
Sanders
,
J.A.
Bauer
, and
A.K.
Kenworthy
.
2022
.
High-content imaging platform to discover chemical modulators of plasma membrane rafts
.
ACS Cent. Sci.
8
:
370
378
.
Fürst
,
T.
,
P.
Salari
,
L.M.
Llamas
,
P.
Steinmann
,
C.
Fitzpatrick
, and
F.
Tediosi
.
2017
.
Global health policy and neglected tropical diseases: Then, now, and in the years to come
.
PLoS Negl. Trop. Dis.
11
:e0005759.
Ghose
,
A.K.
,
V.N.
, and
J.J.
Wendoloski
.
1998
.
Prediction of hydrophobic (lipophilic) properties of small organic molecules using fragmental methods: An analysis of ALOGP and CLOGP methods
.
J. Phys. Chem. A.
102
:
3762
3772
.
Greene
,
N.
,
M.D.
Aleo
,
S.
Louise-May
,
D.A.
Price
, and
Y.
Will
.
2010
.
Using an in vitro cytotoxicity assay to aid in compound selection for in vivo safety studies
.
Bioorg. Med. Chem. Lett.
20
:
5308
5312
.
Grenier
,
V.
,
K.N.
Martinez
,
B.R.
Benlian
,
D.M.
García-Almedina
,
B.K.
Raliski
,
S.C.
Boggess
,
J.C.
Maza
,
S.J.
Yang
,
A.M.M.
Gest
, and
E.W.
Miller
.
2022
.
Molecular prosthetics for long-term functional imaging with fluorescent reporters
.
ACS Cent. Sci.
8
:
118
121
.
Gruner
,
S.M.
1991
.
Lipid membrane curvature elasticity and protein function
. In
Biologically Inspired Physics
.
L.
Peliti
, editior
Plenum Press
,
New York
.
127
135
.
Haydon
,
D.A.
,
B.M.
Hendry
,
S.R.
Levinson
, and
J.
Requena
.
1977
.
Anaesthesia by the n-alkanes. A comparative study of nerve impulse blockage and the properties of black lipid bilayer membranes
.
Biochim. Biophys. Acta.
470
:
17
34
.
Haydon
,
D.A.
, and
B.W.
Urban
.
1983
.
The action of alcohols and other non-ionic surface active substances on the sodium current of the squid giant axon
.
J. Physiol.
341
:
411
427
.
Heerklotz
,
H.
, and
S.
Keller
.
2013
.
How membrane partitioning modulates receptor activation: Parallel versus serial effects of hydrophobic ligands
.
Biophys. J.
105
:
2607
2610
.
Heijman
,
J.
,
N.
Voigt
, and
D.
Dobrev
.
2013
.
New directions in antiarrhythmic drug therapy for atrial fibrillation
.
Future Cardiol.
9
:
71
88
.
Helfrich
,
W.
1981
.
. In
Physique des défauts (Physics of defects)
.
R.
Balian
,
M.
Kléman
, and
J.-P.
Poirier
, editors
North-Holland Publishing Company
,
New York
.
716
755
.
Herold
,
K.F.
,
R.L.
Sanford
,
W.
Lee
,
O.S.
Andersen
, and
H.C.
Hemmings
Jr
.
2017
.
Clinical concentrations of chemically diverse general anesthetics minimally affect lipid bilayer properties
.
.
114
:
3109
3114
.
Herold
,
K.F.
,
R.L.
Sanford
,
W.
Lee
,
M.F.
Schultz
,
H.I.
Ingólfsson
,
O.S.
Andersen
, and
H.C.
Hemmings
Jr
.
2014
.
Volatile anesthetics inhibit sodium channels without altering bulk lipid bilayer properties
.
J. Gen. Physiol.
144
:
545
560
.
Hilgemann
,
D.W.
,
G.
Dai
,
A.
Collins
,
V.
Lariccia
,
S.
Magi
,
C.
Deisl
, and
M.
Fine
.
2018
.
Lipid signaling to membrane proteins: From second messengers to membrane domains and adapter-free endocytosis
.
J. Gen. Physiol.
150
:
211
224
.
,
S.B.
, and
D.A.
Haydon
.
1972
.
Ion transfer across lipid membranes in the presence of gramicidin A. I. Studies of the unit conductance channel
.
Biochim. Biophys. Acta
.
274
:
294
312
.
Honda
,
Z.
,
M.
Nakamura
,
I.
Miki
,
M.
Minami
,
T.
Watanabe
,
Y.
Seyama
,
H.
,
H.
Toh
,
K.
Ito
,
T.
Miyamoto
, and
T.
Shimizu
.
1991
.
Cloning by functional expression of platelet-activating factor receptor from guinea-pig lung
.
Nature
.
349
:
342
346
.
Howery
,
A.E.
,
S.
Elvington
,
S.J.
Abraham
,
K.H.
Choi
,
S.
Dworschak-Simpson
,
S.
Phillips
,
C.M.
Ryan
,
R.L.
Sanford
,
J.
Almqvist
,
K.
Tran
, et al
.
2012
.
A designed inhibitor of a CLC antiporter blocks function through a unique binding mode
.
Chem. Biol.
19
:
1460
1470
.
Huang
,
H.W.
1986
.
Deformation free energy of bilayer membrane and its effect on gramicidin channel lifetime
.
Biophys. J.
50
:
1061
1070
.
Hughes
,
J.D.
,
J.
Blagg
,
D.A.
Price
,
S.
Bailey
,
G.A.
Decrescenzo
,
R.V.
Devraj
,
E.
Ellsworth
,
Y.M.
Fobian
,
M.E.
Gibbs
,
R.W.
Gilles
, et al
.
2008
.
Physiochemical drug properties associated with in vivo toxicological outcomes
.
Bioorg. Med. Chem. Lett.
18
:
4872
4875
.
Hughes
,
J.P.
,
S.
Rees
,
S.B.
Kalindjian
, and
K.L.
Philpott
.
2011
.
Principles of early drug discovery
.
Br. J. Pharmacol.
162
:
1239
1249
.
Hwang
,
T.C.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2003
.
Genistein can modulate channel function by a phosphorylation-independent mechanism: Importance of hydrophobic mismatch and bilayer mechanics
.
Biochemistry
.
42
:
13646
13658
.
Ingólfsson
,
H.I.
, and
O.S.
Andersen
.
2010
.
Screening for small molecules’ bilayer-modifying potential using a gramicidin-based fluorescence assay
.
Assay Drug Dev. Technol.
8
:
427
436
.
Ingólfsson
,
H.I.
, and
O.S.
Andersen
.
2011
.
Alcohol’s effects on lipid bilayer properties
.
Biophys. J.
101
:
847
855
.
Ingólfsson
,
H.I.
,
P.
Thakur
,
K.F.
Herold
,
E.A.
Hobart
,
N.B.
Ramsey
,
X.
Periole
,
D.H.
de Jong
,
M.
Zwama
,
D.
Yilmaz
,
K.
Hall
, et al
.
2014
.
Phytochemicals perturb membranes and promiscuously alter protein function
.
ACS Chem. Biol.
9
:
1788
1798
.
Jackson
,
M.B.
1989
.
Perfection of a synaptic receptor: Kinetics and energetics of the acetylcholine receptor
.
.
86
:
2199
2203
.
Kapoor
,
R.
,
T.A.
Peyear
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2019
.
Antidepressants are modifiers of lipid bilayer properties
.
J. Gen. Physiol.
151
:
342
356
.
Ketcham
,
R.
,
B.
Roux
, and
T.
Cross
.
1997
.
High-resolution polypeptide structure in a lamellar phase lipid environment from solid state NMR derived orientational constraints
.
Structure
.
5
:
1655
1669
.
Kim
,
T.
,
K.I.
Lee
,
P.
Morris
,
R.W.
Pastor
,
O.S.
Andersen
, and
W.
Im
.
2012
.
Influence of hydrophobic mismatch on structures and dynamics of gramicidin A and lipid bilayers
.
Biophys. J
.
102
:
1551
1560
.
Kolb
,
H.A.
, and
E.
Bamberg
.
1977
.
Influence of membrane thickness and ion concentration on the properties of the gramicidin a channel. Autocorrelation, spectral power density, relaxation and single-channel studies
.
Biochim. Biophys. Acta
.
464
:
127
141
.
Kornecki
,
E.
, and
Y.H.
Ehrlich
.
1988
.
Neuroregulatory and neuropathological actions of the ether-phospholipid platelet-activating factor
.
Science
.
240
:
1792
1794
.
Lakowicz
,
J.R.
2006
.
Principles of Fluorescence Spectroscopy
.
Springer
,
New York
.
Lee
,
A.G.
1991
.
Lipids and their effects on membrane proteins: Evidence against a role for fluidity
.
Prog. Lipid Res.
30
:
323
348
.
Lee
,
O.W.
,
S.
Austin
,
M.
Gamma
,
D.M.
Cheff
,
T.D.
Lee
,
K.M.
Wilson
,
J.
Johnson
,
J.
Travers
,
J.C.
Braisted
,
R.
Guha
, et al
.
2020
.
Cytotoxic profiling of annotated and diverse chemical libraries using quantitative high-throughput screening
.
SLAS Discov.
25
:
9
20
.
Leeson
,
P
.
2012
.
Drug discovery: Chemical beauty contest
.
Nature
.
481
:
455
456
.
Leeson
,
P.D.
, and
B.
Springthorpe
.
2007
.
The influence of drug-like concepts on decision-making in medicinal chemistry
.
Nat. Rev. Drug Discov.
6
:
881
890
.
Leo
,
A.
,
C.
Hansch
, and
D.
Elkins
.
1971
.
Partition coefficients and their uses
.
Chem. Rev.
71
:
525
616
.
Lin
,
Z.
, and
Y.
Will
.
2012
.
Evaluation of drugs with specific organ toxicities in organ-specific cell lines
.
Toxicol. Sci.
126
:
114
127
.
Lipinski
,
C.A.
,
F.
Lombardo
,
B.W.
Dominy
, and
P.J.
Feeney
.
2001
.
Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings
.
46
:
3
26
.
Lomize
,
A.L.
,
V.I.
Orekhov
, and
A.S.
Arsen’ev
.
1992
.
Refinement of the spatial structure of the gramicidin A ion channel
.
Bioorg. Khim
.
18
:
182
200
.
Lum
,
K.
,
H.I.
Ingólfsson
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2017
.
Exchange of gramicidin between lipid bilayers: Implications for the mechanism of channel formation
.
Biophys. J.
113
:
1757
1767
.
Lundbaek
,
J.A.
,
P.
Birn
,
J.
Girshman
,
A.J.
Hansen
, and
O.S.
Andersen
.
1996
.
Membrane stiffness and channel function
.
Biochemistry
.
35
:
3825
3830
.
Lundbaek
,
J.A.
,
P.
Birn
,
A.J.
Hansen
,
R.
Søgaard
,
C.
Nielsen
,
J.
Girshman
,
M.J.
Bruno
,
S.E.
Tape
,
J.
Egebjerg
,
D.V.
Greathouse
, et al
.
2004
.
Regulation of sodium channel function by bilayer elasticity: The importance of hydrophobic coupling. Effects of micelle-forming amphiphiles and cholesterol
.
J. Gen. Physiol.
123
:
599
621
.
Lundbaek
,
J.A.
,
P.
Birn
,
S.E.
Tape
,
G.E.
Toombes
,
R.
Søgaard
,
R.E.
Koeppe
II
,
S.M.
Gruner
,
A.J.
Hansen
, and
O.S.
Andersen
.
2005
.
Capsaicin regulates voltage-dependent sodium channels by altering lipid bilayer elasticity
.
Mol. Pharmacol.
68
:
680
689
.
Lundbaek
,
J.A.
,
S.A.
Collingwood
,
H.I.
Ingólfsson
,
R.
Kapoor
, and
O.S.
Andersen
.
2010a
.
Lipid bilayer regulation of membrane protein function: Gramicidin channels as molecular force probes
.
J. R. Soc. Interf.
7
:
373
395
.
Lundbaek
,
J.A.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2010b
.
Amphiphile regulation of ion channel function by changes in the bilayer spring constant
.
.
107
:
15427
15430
.
Magalhães
,
P.R.
,
P.B.P.S.
Reis
,
D.
Vila-Viçosa
,
M.
Machuqueiro
, and
B.L.
Victor
.
2022
.
Optimization of an in silico protocol using probe permeabilities to identify membrane pan-assay interference compounds
.
J. Chem. Inf. Model.
62
:
3034
3042
.
Marsh
,
D.
2007
.
Lateral pressure profile, spontaneous curvature frustration, and the incorporation and conformation of proteins in membranes
.
Biophys. J
.
93
:
3884
3899
.
Marsh
,
D.
2008
.
Energetics of hydrophobic matching in lipid-protein interactions
.
Biophys. J.
94
:
3996
4013
.
Martinus
,
N.
, and
C.A.
Vincent
.
1976
.
Viscosity of aqueous solutions of TlNO3, Tl2SO4 and TlOH at 25°C
.
J. Chem. Soc. Faraday Trans. I.
72
:
2505
2511
.
Mason
,
R.P.
,
D.M.
Moisey
, and
L.
Shajenko
.
1992
.
Cholesterol alters the binding of Ca2+ channel blockers to the membrane lipid bilayer
.
Mol. Pharmacol.
41
:
315
321
MMV
.
2014
.
Exploiting the pathogen box: An international open source collaboration to accelerate drug development in addressing diseases of poverty Medicines
. http://www.who.int/phi/implementation/MMV_documents.pdf?ua=1.
Mondal
,
S.
,
G.
Khelashvili
,
J.
Shan
,
O.S.
Andersen
, and
H.
Weinstein
.
2011
.
Quantitative modeling of membrane deformations by multihelical membrane proteins: Application to G-protein coupled receptors
.
Biophys. J.
101
:
2092
2101
.
Moore
,
H.P.
, and
M.A.
Raftery
.
1980
.
Direct spectroscopic studies of cation translocation by Torpedo acetylcholine receptor on a time scale of physiological relevance
.
.
77
:
4509
4513
.
Mujović
,
N.
,
D.
Dobrev
,
M.
Marinković
,
V.
Russo
, and
T.S.
Potpara
.
2020
.
The role of amiodarone in contemporary management of complex cardiac arrhythmias
.
Pharmacol. Res.
151
:
104521
.
Nazari
,
M.
,
M.
Kurdi
, and
H.
Heerklotz
.
2012
.
Classifying surfactants with respect to their effect on lipid membrane order
.
Biophys. J
.
102
:
498
506
.
Norimatsu
,
Y.
,
K.
Hasegawa
,
N.
Shimizu
, and
C.
Toyoshima
.
2017
.
Protein-phospholipid interplay revealed with crystals of a calcium pump
.
Nature
.
545
:
193
198
.
O’Connell
,
A.M.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
1990
.
Kinetics of gramicidin channel formation in lipid bilayers: Transmembrane monomer association
.
Science
.
250
:
1256
1259
.
Obi
,
P.
, and
S.
Natesan
.
2022
.
Membrane lipids are an integral part of transmembrane allosteric sites in GPCRs: A case study of cannabinoid CB1 receptor bound to a negative allosteric modulator, ORG27569, and analogs
.
J. Med. Chem.
65
:
12240
12255
.
Paula
,
S.
,
A.G.
Volkov
,
A.N.
Van Hoek
,
T.H.
Haines
, and
D.W.
Deamer
.
1996
.
Permeation of protons, potassium ions, and small polar molecules through phospholipid bilayers as a function of membrane thickness
.
Biophys. J.
70
:
339
348
.
Payandeh
,
J.
, and
M.
Volgraf
.
2021
.
Ligand binding at the protein-lipid interface: Strategic considerations for drug design
.
Nat. Rev. Drug Discov.
20
:
710
722
.
Pope
,
C.G.
,
B.W.
Urban
, and
D.A.
Haydon
.
1982
.
The influence of n-alkanols and cholesterol on the duration and conductance of gramicidin single channels in monoolein bilayers
.
Biochim. Biophys. Acta
.
688
:
279
283
.
Price
,
D.A.
,
J.
Blagg
,
L.
Jones
,
N.
Greene
, and
T.
Wager
.
2009
.
Physicochemical drug properties associated with in vivo toxicological outcomes: A review
.
Expert Opin. Drug Metab. Toxicol.
5
:
921
931
.
Ritchie
,
T.J.
, and
S.J.
Macdonald
.
2009
.
The impact of aromatic ring count on compound developability-are too many aromatic rings a liability in drug design?
.
Drug Discov. Today
.
14
:
1011
1020
.
Rousseeuw
,
P.J.
1987
.
Silhouettes: A graphical aid to the interpretation and validation of cluster analysis
.
J. Comput. Appl. Math.
20
:
53
65
.
Rusinova
,
R.
,
C.
He
, and
O.S.
Andersen
.
2021
.
Mechanisms underlying drug-mediated regulation of membrane protein function
.
.
118
:
118
.
Rusinova
,
R.
,
K.F.
Herold
,
R.L.
Sanford
,
D.V.
Greathouse
,
H.C.J.
Hemmings
Jr
, and
O.S.
Andersen
.
2011
.
Thiazolidinedione insulin sensitizers alter lipid bilayer properties and voltage-dependent sodium channel function: Implications for drug discovery
.
J. Gen. Physiol.
138
:
249
270
.
Rusinova
,
R.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
2015
.
A general mechanism for drug promiscuity: Studies with amiodarone and other antiarrhythmics
.
J. Gen. Physiol.
146
:
463
475
.
Sandermann
,
H.
, Jr
.
1978
.
Regulation of membrane enzymes by lipids
.
Biochim. Biophys. Acta
.
515
:
209
237
.
Sawyer
,
D.B.
, and
O.S.
Andersen
.
1989
.
Platelet-activating factor is a general membrane perturbant
.
Biochim. Biophys. Acta
.
987
:
129
132
.
Sawyer
,
D.B.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
1989
.
Induction of conductance heterogeneity in gramicidin channels
.
Biochemistry
.
28
:
6571
6583
.
Sawyer
,
D.B.
,
R.E.
Koeppe
II
, and
O.S.
Andersen
.
1990
.
Gramicidin single-channel properties show no solvent-history dependence
.
Biophys. J
57
:
515
523
.
Scott
,
H.L.
,
A.
Skinkle
,
E.G.
Kelley
,
M.N.
Waxham
,
I.
Levental
, and
F.A.
Heberle
.
2019
.
On the mechanism of bilayer separation by extrusion, or why your LUVs are not really unilamellar
.
Biophys. J.
117
:
1381
1386
.
Seddon
,
A.M.
,
D.
Casey
,
R.V.
Law
,
A.
Gee
,
R.H.
Templer
, and
O.
Ces
.
2009
.
Drug interactions with lipid membranes
.
Chem. Soc. Rev.
38
:
2509
2519
.
Seddon
,
J.M.
1990
.
Structure of the inverted hexagonal (HII) phase, and non-lamellar phase transitions of lipids
.
Biochim. Biophys. Acta
.
1031
:
1
69
.
Seeman
,
P.
1972
.
The membrane actions of anesthetics and tranquilizers
.
Pharmacol. Rev.
24
:
583
655
Simon
,
G.M.
,
M.J.
Niphakis
, and
B.F.
Cravatt
.
2013
.
Determining target engagement in living systems
.
Nat. Chem. Biol.
9
:
200
205
.
Smith
,
D.
,
P.
,
A.
Avdeef
,
L.
Di
,
G.F.
Ecker
,
B.
Faller
,
J.B.
Houston
,
M.
Kansy
,
E.H.
Kerns
,
S.D.
Krämer
, et al
.
2014
.
Passive lipoidal diffusion and carrier-mediated cell uptake are both important mechanisms of membrane permeation in drug disposition
.
Mol. Pharm.
11
:
1727
1738
.
Søgaard
,
R.
,
T.M.
Werge
,
C.
Bertelsen
,
C.
Lundbye
,
K.L.
,
C.H.
Nielsen
,
J.A.
Lundbaek
, and
J.A.
Lundbaek
.
2006
.
GABAA receptor function is regulated by lipid bilayer elasticity
.
Biochemistry
.
45
:
13118
13129
.
Spector
,
A.A.
, and
M.A.
Yorek
.
1985
.
Membrane lipid composition and cellular function
.
J. Lipid Res.
26
:
1015
1035
.
Sperotto
,
M.M.
, and
O.G.
Mouritsen
.
1993
.
Lipid enrichment and selectivity of integral membrane proteins in two-component lipid bilayers
.
Eur. Biophys. J
.
22
:
323
328
.
Stefaniak
,
J.
, and
K.V.M.
Huber
.
2020
.
Importance of quantifying drug-target engagement in cells
.
ACS Med. Chem. Lett.
11
:
403
406
.
Sugano
,
K.
,
M.
Kansy
,
P.
,
A.
Avdeef
,
S.
Bendels
,
L.
Di
,
G.F.
Ecker
,
B.
Faller
,
H.
Fischer
,
G.
Gerebtzoff
, et al
.
2010
.
Coexistence of passive and carrier-mediated processes in drug transport
.
Nat. Rev. Drug Discov.
9
:
597
614
.
Sun
,
D.
,
T.A.
Peyear
,
W.F.D.
Bennett
,
O.S.
Andersen
,
F.C.
Lightstone
, and
H.I.
Ingólfsson
.
2019
.
Molecular mechanism for gramicidin dimerization and dissociation in bilayers of different thickness
.
Biophys. J.
117
:
1831
1844
.
Sun
,
D.
,
T.A.
Peyear
,
W.F.D.
Bennett
,
M.
Holcomb
,
S.
He
,
F.
Zhu
,
F.C.
Lightstone
,
O.S.
Andersen
, and
H.I.
Ingólfsson
.
2020
.
Assessing the perturbing effects of drugs on lipid bilayers using gramicidin channel-based in silico and in vitro assays
.
J. Med. Chem.
63
:
11809
11818
.
Thompson
,
M.J.
, and
J.E.
Baenziger
.
2020
.
Ion channels as lipid sensors: From structures to mechanisms
.
Nat. Chem. Biol.
16
:
1331
1342
.
Tibbs
,
G.R.
,
T.J.
Rowley
,
R.L.
Sanford
,
K.F.
Herold
,
A.
Proekt
,
H.C.
Hemmings
Jr
,
O.S.
Andersen
,
P.A.
Goldstein
, and
P.D.
Flood
.
2013
.
HCN1 channels as targets for anesthetic and nonanesthetic propofol analogs in the amelioration of mechanical and thermal hyperalgesia in a mouse model of neuropathic pain
.
J. Pharmacol. Exp. Ther.
345
:
363
373
.
Townsley
,
L.E.
,
W.A.
Tucker
,
S.
Sham
, and
J.F.
Hinton
.
2001
.
Structures of gramicidins A, B, and C incorporated into sodium dodecyl sulfate micelles
.
Biochemistry
.
40
:
11676
11686
.
Tummino
,
T.A.
,
V.V.
Rezelj
,
B.
Fischer
,
A.
Fischer
,
M.J.
O’Meara
,
B.
Monel
,
T.
Vallet
,
K.M.
White
,
Z.
Zhang
,
A.
Alon
, et al
.
2021
.
Drug-induced phospholipidosis confounds drug repurposing for SARS-CoV-2
.
Science
.
373
:
541
547
.
Urban
,
B.W.
2002
.
Current assessment of targets and theories of anaesthesia
.
Br. J. Anaesth.
89
:
167
183
.
Veale
,
C.G.L.
2019
.
Unpacking the pathogen box-an open source tool for fighting neglected tropical disease
.
ChemMedChem
.
14
:
386
453
.
Venable
,
R.M.
,
F.L.H.
Brown
, and
R.W.
Pastor
.
2015
.
Mechanical properties of lipid bilayers from molecular dynamics simulation
.
Chem. Phys. Lipids
.
192
:
60
74
.
Wang
,
X.
, and
O.
Boudker
.
2020
.
Large domain movements through the lipid bilayer mediate substrate release and inhibition of glutamate transporters
.
eLife
.
9
. e58417.
Waring
,
M.J.
,
J.
Arrowsmith
,
A.R.
Leach
,
P.D.
Leeson
,
S.
Mandrell
,
R.M.
Owen
,
G.
Pairaudeau
,
W.D.
Pennie
,
S.D.
Pickett
,
J.
Wang
, et al
.
2015
.
An analysis of the attrition of drug candidates from four major pharmaceutical companies
.
Nat. Rev. Drug Discov.
14
:
475
486
.
White
,
S.H.
,
W.C.
Wimley
,
A.S.
, and
K.
Hristova
.
1998
.
Protein folding in membranes: determining energetics of peptide-bilayer interactions
.
Methods Enzymol
.
295
:
62
87
.
World Health Organization
.
2017
.
Integrating Neglected Tropical Diseases into Global Health and Development: Fourth WHO Report on Neglected Tropical Diseases
.
World Health Organization
,
Geneva
.
Yang
,
J.J.
,
O.
Ursu
,
C.A.
Lipinski
,
L.A.
Sklar
,
T.I.
Oprea
, and
C.G.
Bologa
.
2016
.
Badapple: Promiscuity patterns from noisy evidence
.
J. Cheminform.
8
:
29
.
Zhang
,
M.
,
T.
Peyear
,
I.
Patmanidis
,
D.V.
Greathouse
,
S.J.
Marrink
,
O.S.
Andersen
, and
H.I.
Ingólfsson
.
2018
.
Fluorinated alcohols’ effects on lipid bilayer properties
.
Biophys. J.
115
:
679
689
.
Zhelev
,
D.V.
1998
.
Material property characteristics for lipid bilayers containing lysolipid
.
Biophys. J.
75
:
321
330
.
Zhou
,
W.
,
G.
Fiorin
,
C.
Anselmi
,
H.A.
Karimi-Varzaneh
,
H.
Poblete
,
L.R.
Forrest
, and
J.D.
Faraldo-Gómez
.
2019
.
Large-scale state-dependent membrane remodeling by a transporter protein
.
eLife
.
8
. e50576.
Zingsheim
,
H.P.
, and
E.
Neher
.
1974
.
The equivalence of fluctuation analysis and chemical relaxation measurements: A kinetic study of ion pore formation in thin lipid membranes
.
Biophys. Chem.
2
:
197
207
.