A selective membrane-targeting repurposed antibioticwith activity against persistent methicillin-resistantStaphylococcus aureusWooseong Kima, Guijin Zoub, Taylor P. A. Haric, Ingrid K. Wiltc, Wenpeng Zhub, Nicolas Galleb, Hammad A. Faizid,Gabriel L. Hendricksa, Katerina Toria, Wen Pana, Xiaowen Huanga,e, Andrew D. Steelec, Erika E. Csataryc,Madeline M. Dekarskec, Jake L. Rosenc, Noelly de Queiroz Ribeirof, Kiho Leea, Jenna Porta, Beth Burgwyn Fuchsa,Petia M. Vlahovskag, William M. Wuestc, Huajian Gaob, Frederick M. Ausubelh,i,1, and Eleftherios Mylonakisa,1

aDivision of Infectious Diseases, Rhode Island Hospital, Warren Alpert Medical School of Brown University, Providence, RI 02903; bSchool of Engineering,Brown University, Providence, RI 02903; cDepartment of Chemistry and Emory Antibiotic Resistance Center, Emory University, Atlanta, GA 30322;dDepartment of Mechanical Engineering, Northwestern University, Evanston, IL 60208; eDepartment of Dermatology, Nanfang Hospital, Southern MedicalUniversity, Guangzhou 510515, China; fDepartment of Microbiology, Instituto de Ciências Biológicas, Universidade Federal de Minas Gerais, Belo Horizonte,MG 31270-901, Brazil; gDepartment of Engineering Sciences and Applied Mathematics, Northwestern University, Evanston, IL 60208; hDepartment ofMolecular Biology, Massachusetts General Hospital, Boston, MA 02114; and iDepartment of Genetics, Harvard Medical School, Boston, MA 02115

Contributed by Frederick M. Ausubel, June 17, 2019 (sent for review March 22, 2019; reviewed by Dale L. Boger and Ruhong Zhou)

Treatment of Staphylococcus aureus infections is complicated bythe development of antibiotic tolerance, a consequence of theability of S. aureus to enter into a nongrowing, dormant state inwhich the organisms are referred to as persisters. We report thatthe clinically approved anthelmintic agent bithionol kills methicillin-resistant S. aureus (MRSA) persister cells, which correlates with itsability to disrupt the integrity of Gram-positive bacterial membranes.Critically, bithionol exhibits significant selectivity for bacterial com-pared with mammalian cell membranes. All-atom molecular dynam-ics (MD) simulations demonstrate that the selectivity of bithionol forbacterial membranes correlates with its ability to penetrate and em-bed in bacterial-mimic lipid bilayers, but not in cholesterol-richmammalian-mimic lipid bilayers. In addition to causing rapid mem-brane permeabilization, the insertion of bithionol increases mem-brane fluidity. By using bithionol and nTZDpa (another membrane-active antimicrobial agent), as well as analogs of these compounds,we show that the activity of membrane-active compounds againstMRSA persisters positively correlates with their ability to increasemembrane fluidity, thereby establishing an accurate biophysical in-dicator for estimating antipersister potency. Finally, we demonstratethat, in combination with gentamicin, bithionol effectively reducesbacterial burdens in a mouse model of chronic deep-seated MRSAinfection. This work highlights the potential repurposing of bithionolas an antipersister therapeutic agent.

MRSA | bacterial persister | drug repurposing | membrane-activeantimicrobials | membrane selectivity

Staphylococcus aureus is a Gram-positive opportunistic humanpathogen carried by approximately one third of the human

population. Despite antibiotic availability, S. aureus infectionsare often hard to cure and remain one of the major causes ofdeath (1), in part because of the ability of S. aureus to enter intoa nongrowing antibiotic-tolerant state, in which the organismsare referred to as persisters (2). Persisters exhibit significantlyreduced biosynthetic processes, which are the major targets formost antibiotics (2). Persisters also exist in a metabolically low-energy state (3) that prevents the energy-dependent uptake ofantibiotics such as aminoglycosides (4). S. aureus persisters arepresent in high numbers in stationary-phase suspension culturesand biofilms (4–6) and are responsible for chronic and relapsinginfections such as endocarditis, osteomyelitis, and prostheticimplant infections (3).Bacterial membranes are attractive antipersister targets be-

cause they can be disrupted independently of growth. Althoughmembrane-active agents are typically toxic to mammals becauseof low membrane selectivity (7), the clinical success of daptomycin

has sparked new interest in membrane-active antimicrobialtherapeutic agents (7). The lipophilic tail of daptomycin, a nat-ural cyclic lipopeptide synthesized by Streptomyces roseosporus,inserts into Gram-positive bacterial membranes, forming oligo-meric pores, causing membrane depolarization, potassium ionefflux, and rapid cell death (8). Despite strong antimicrobialpotency against growing bacterial cells, daptomycin has not beenreported to be effective against persisters (9, 10).Recently, our laboratory described the identification of membrane-

active synthetic retinoids that are efficacious in killing MRSA

Significance

There is a critical lack of therapeutic agents to treat infectionscaused by nongrowing persister forms of methicillin-resistantStaphylococcus aureus (MRSA). Although membrane-disruptingagents can kill persister cells, their therapeutic potential hasbeen mostly overlooked because of low selectivity for bacterialversus mammalian membranes. We report that the clinicallyapproved anthelmintic drug bithionol kills MRSA persisters bydisrupting membrane lipid bilayers at concentrations that ex-hibit low levels of toxicity to mammalian cells. The selectivityof bithionol results from the presence of cholesterol in mam-malian but not in bacterial membranes. We also show that theantipersister potency of membrane-active antimicrobial agentscorrelates with their ability to increase membrane fluidity. Ourresults significantly enhance our understanding of bacterialmembrane disruption and membrane selectivity.

Author contributions: W.K., G.Z., T.P.A.H., I.K.W., W.Z., N.G., H.A.F., G.L.H., W.M.W., H.G.,F.M.A., and E.M. designed research; W.K., G.Z., T.P.A.H., I.K.W., W.Z., N.G., H.A.F., G.L.H.,K.T., W.P., X.H., N.d.Q.R., K.L., J.P., and B.B.F. performed research; T.P.A.H., I.K.W., A.D.S.,E.E.C., M.M.D., J.L.R., B.B.F., P.M.V., W.M.W., H.G., F.M.A., and E.M. contributed newreagents/analytic tools; W.K., G.Z., T.P.A.H., I.K.W., W.Z., N.G., H.A.F., G.L.H., P.M.V.,W.M.W., H.G., F.M.A., and E.M. analyzed data; and W.K., G.Z., W.M.W., H.G., F.M.A.,and E.M. wrote the paper.

Reviewers: D.L.B., The Scripps Research Institute; and R.Z., Columbia University.

Conflict of interest statement: F.M.A. and E.M. have financial interests in Genma Biosci-ences and Octagon Therapeutics, companies that are engaged in developing antimicro-bial compounds. E.M.’s and F.M.A.’s interests were reviewed and are managed by RhodeIsland Hospital (E.M.) and Massachusetts General Hospital and Partners HealthCare(F.M.A.) in accordance with their conflict of interest policies. The remaining authors de-clare no competing financial interests.

Published under the PNAS license.1To whom correspondence may be addressed. Email: ausubel@molbio.mgh.harvard.eduor emylonakis@lifespan.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1904700116/-/DCSupplemental.

Published online July 29, 2019.

www.pnas.org/cgi/doi/10.1073/pnas.1904700116 PNAS | August 13, 2019 | vol. 116 | no. 33 | 16529–16534

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

29,

202

0

persisters (6). They are also relatively nontoxic because they ex-hibit a significant amount of selectivity for Gram-positive bacterialmembranes compared with mammalian membranes (6). Likedaptomycin, the antimicrobial retinoids also appear to insert intoGram-positive bacterial membranes, causing rapid permeabilizationand cell death. In contrast to daptomycin, however, a subset ofthe membrane-active synthetic retinoids also kill nongrowingMRSA persister cells (6). In addition to the synthetic retinoids,we have described 2 additional membrane-active compounds withanti-MRSA activity: NH125, a histidine-kinase inhibitor (11), andnTZDpa, a nonthiazolidinedione peroxisome proliferator-activatedreceptor gamma partial agonist (12). Although NH125 and nTZDpahave excellent anti-MRSA persister activity, they both causesubstantial hemolysis of red blood cells at high concentrations(>32 μg/mL) (12, 13), which is a drawback for further devel-opment as anti-MRSA therapeutic agents.The membrane-active retinoids, NH125, and nTZDpa were

identified by screening ∼82,000 synthetic compounds by using ahigh-throughput Caenorhabditis elegans–MRSA infection screen-ing assay for compounds that block the ability of MRSA to kill thenematodes (6, 11, 12). We subsequently screened 185 “hit” com-pounds for the additional ability to permeabilize MRSA persistersby using a SYTOX Green membrane permeability assay (11).Another putative membrane-active antimicrobial identified in thisscreen was the clinically approved anthelminthic drug bithionol,which was chosen for additional studies because of its well-established pharmacokinetic, toxicity, and safety profiles (14). Inthis paper, we describe the therapeutic potential of bithionol as ananti-MRSA persister antibiotic agent and provide a putative modeof action that explains the ability of bithionol to selectively disruptbacterial compared with mammalian membrane lipid bilayers.Moreover, by studying a panel of bithionol and nTZDpa analogs,we found a strong correlation between the potency of anti-MRSApersister activity and the ability to increase membrane fluidity.

ResultsBithionol Shows Bactericidal Activity Against both Antibiotic-ResistantS. aureus and S. aureus Persister Cells. Bithionol (Fig. 1A) was pre-viously described as an antimicrobial agent with a minimal inhibitoryconcentration (MIC) of ∼8 to 15 μg/mL against Gram-positive bac-teria, including S. aureus (15). We confirmed that bithionol exhibitsantimicrobial activity against a variety of Gram-positive pathogens,including vancomycin-resistant S. aureus (VRSA) and vancomycin-and daptomycin-resistant enterococcal strains. In our hands, however,bithionol was more potent than previously reported, with MICs of0.5 to 2 μg/mL, comparable to daptomycin (SI Appendix, TableS1). We also confirmed that bithionol exhibits relatively weakantimicrobial activity against Gram-negative pathogens (MICs of16 to >64 μg/mL; SI Appendix, Table S1).In contrast to a previous study that described bithionol as

bacteriostatic (16), we found that bithionol eradicated ∼107 CFU/mLexponential-phase S. aureus strain MW2 within 3 h at 10 μg/mL(10× MIC; SI Appendix, Fig. S1A). The rate of killing wascomparable to daptomycin (at 10× MIC) and significantly fasterthan the killing kinetics of vancomycin (at 10× MIC; SI Appendix,Fig. S1A). Consistent with its bactericidal activity, 10 μg/mLbithionol caused a time-dependent decrease in optical density ofS. aureus cells comparable to the antiseptic detergent benzyldi-methylhexadecylammonium chloride (16-BAC; SI Appendix, Fig.S1B), indicating that bithionol has bacteriolytic activity. Indeed,transmission electron micrographs (TEMs) showed that 10× MICbithionol disrupts MRSA membranes, causing the intracellularformation of mesosome-like structures, abnormal cell divisions, andcell lysis (Fig. 1B), which we previously observed when S. aureuscells were treated with membrane-active antimicrobials (12).We also found that bithionol killed stationary-phase MRSA

MW2 planktonic and biofilm persisters in a dose-dependentmanner and completely eradicated them at 32 μg/mL (32× MIC)

within 2 h and 24 h, respectively (Fig. 1 C and D). In contrast,MW2 persisters displayed a high level of tolerance to 100× MICof several conventional antibiotics, including daptomycin andlinezolid, which were tested for activity against planktonic per-sisters. Similarly, bithionol killed vancomycin-resistant S. aureus(VRSA) strain VRS1 persisters, whereas linezolid and daptomycinexhibited no and nominal activity, respectively (Fig. 1E).

Bithionol Interacts with and Disrupts the Bacterial Mimetic LipidBilayer. We performed all-atom molecular dynamics (MD) simu-lations of bithionol interacting with simulated bacterial mem-branes to elucidate a potential mechanism of action. The topologyand parameters of bithionol for the GROMOS54a7 force field(17) were generated by Automated Topology Builder (18). Weused the previously established model of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC)/1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (DOPG) at a 7:3 ratio (19) to simulate negativelycharged S. aureus membranes (SI Appendix, Materials and Meth-ods). The MDmodeling showed that bithionol is initially recruitedto the membrane surface by the binding of polar hydroxyl andchlorine groups to hydrophilic lipid heads (Fig. 2A and Movie S1).After several hundred nanoseconds of sustained attachment,bithionol penetrates into the membrane interior, maximizing in-teractions between nonpolar benzene rings and hydrophobic lipidtails (Movie S1). After penetration, bithionol embeds in the outerleaflet of the lipid bilayer (Fig. 2A and Movie S1). The insertion of

SCl

Cl

OHCl

Cl

OHBithionol Control

0 1 2 3 4

9876543

Time (h)

Log

(CFU

/mL)

100 μg/mL Van 100 μg/mL Gm 50 μg/mL Cipro 100 μg/mL Dap

200 μg/mL Lin Non-treated

0 1 2 3 4

9876543

Time (h)

Log

(CFU

/mL)

32 μg/mL BT16 μg/mL BT8 μg/mL BT

0 μg/mL BT

S. aureus MW2 persisters

0 1 2 3 4

9876543

Time (h)

Log

(CFU

/mL)

200 μg/mL Lin100 μg/mL Dap

Non-treated

0 1 2 3 4

9876543

Time (h)

Log

(CFU

/mL)

32 μg/mL BT 16 μg/mL BT8 μg/mL BT 0 μg/mL BT

S. aureus VRS1 PersistersNon

-trea

ted

100 μ

g/mL V

an

100 μ

g/mL G

m

50 μg

/mL C

ipro

876543

Log

(CFU

/mem

bran

e)

S. aureus MW2 biofilms

0 8 16 32

876543

BT (μg/mL)

Log

(CFU

/mem

bran

e)

A B

C

E

D

Fig. 1. Bithionol exhibits bactericidal activity against S. aureus persisters. (A)Chemical structures of bithionol. (B) TEM micrographs showing the formationof intracellular mesosome-like structures (red arrows), an abnormal cell di-vision (brown arrow), or cell lysis (blue arrow) in S. aureus MW2 treated with10× MIC (10 μg/mL) bithionol or 0.1% DMSO (control) for 2 h. (Scale bars,500 nm.) (C and D) Viability of MRSA MW2 stationary-phase (C) or biofilm (D)persister cells treated with 100× MIC of the conventional antibiotics vanco-mycin (Van), gentamicin (Gm), ciprofloxacin (Cipro), daptomycin (Dap), line-zolid (Lin), or the indicated concentrations of bithionol (BT) for 4 h (C) and24 h (D), respectively. (E) The viability of S. aureus VRS1 stationary-phasepersisters treated with 100×MIC (200 μg/mL) linezolid (Lin), 100×MIC (100 μg/mL)daptomycin (dap), or the indicated concentrations of bithionol (BT) as afunction of time. The data points on the x-axis are below the level of de-tection (2 × 102 CFU/mL, or 2 × 102 CFU per membrane). Individual data points(n = 3 biologically independent samples) are shown; error bars representmeans ± SD.

16530 | www.pnas.org/cgi/doi/10.1073/pnas.1904700116 Kim et al.

Dow

nloa

ded

by g

uest

on

June

29,

202

0

even a single bithionol molecule causes a substantial local increasein lipid bilayer disorder (SI Appendix, Fig. S2 A and C). These MDsimulations demonstrate that the polarity of branch groups andthe hydrophobicity of core rings play important roles in membraneattachment and penetration.We further investigated the effects of bithionol on lipid bila-

yers by using biomembrane-mimicking giant unilamellar vesicles(GUVs) consisting of a single lipid bilayer with a diameter of 10to 100 μm. GUVs were constructed by using DOPC/DOPG lipidsat a 7:3 ratio as in the MD simulations. Lipid aggregates formedon the surfaces of the GUVs exposed to 1 μg/mL bithionol, and,at 10 μg/mL, the GUVs burst (Fig. 2C and Movies S3–S5),

indicating that bithionol interacts with and disrupts a bacterialmimetic lipid bilayer.

Bithionol Induces Rapid Membrane Permeabilization and an Increasein Membrane Fluidity.By using a SYTOXGreen permeability assay,we found that, in contrast to daptomycin, bithionol induced dose-dependent membrane permeability in both exponential-phaseMRSA MW2 cells and stationary-phase MRSA MW2 persisters(Fig. 2D and SI Appendix, Fig. S1C). The fluorescence intensitypeaked at 4 μg/mL and then decreased at higher concentrations ofbithionol (Fig. 2D), most likely as a consequence of nucleic acidsreleased form lysed cells (20), consistent with the observation that10 μg/mL bithionol lyses MRSA persisters (Fig. 1B). The dose-dependent effects of bithionol on membrane permeability corre-lated with its dose-dependent killing kinetics (Figs. 1C and 2D andSI Appendix, Fig. S1C), suggesting that the bactericidal activity ofbithionol results from its membrane-disrupting activity.Insertion of particular membrane disrupting compounds into

lipid bilayers is known to cause dramatic changes in membranefluidity that disrupts the normal liquid-crystalline phase of themembrane (21–23). This results in passive permeabilization, lossof membrane protein functions, leakage of cellular components,and bacterial death (24). We therefore tested whether bithionolalters MRSAMW2 membrane fluidity by utilizing the membranefluidity-sensitive dye Laurdan (25), which exhibits a fluorescenceemission wavelength shift dependent on the amount of adjacentwater molecules (25). Because water molecule penetration intolipid bilayers is in turn determined by lipid packing density andlipid bilayer fluidity, bacterial membrane fluidity can be quanti-fied by using the Laurdan generalized polarization (GP) metric:GP = (I440 − I490)/(I440 + I490), ranging from −1 (most fluid anddisordered) to +1 (most rigid and ordered) (25). As shown inFig. 2E, bithionol at concentrations greater than 8 μg/mL in-duced a significant decrease in Laurdan GP in S. aureus MW2,similar to 50 mM benzyl alcohol (BA), a known membrane flu-idizer. These data are consistent with our observation thatbithionol concentrations greater than 16 μg/mL exhibited sig-nificant killing activity against MRSA MW2 persisters (Fig. 1C).

Bithionol Does Not Penetrate Mammalian Membranes.We used MDsimulations to determine whether bithionol specifically penetratesbacterial compared with mammalian membranes. Mammalianmembranes are comprised of phospholipids, sphingolipids, andcholesterol (26), but can be modeled as a simplified bilayercomposed of the zwitterionic lipid 1,2-palmitoyl-oleoyl-sn-glycero-3-phosphocholine (POPC) mixed with cholesterol ranging from 20to 50 mol% (27, 28). At a POPC/cholesterol ratio of 7:3, MDsimulations using the GROMOS54a7 force field showed thatbithionol fails to penetrate the simulated mammalian bilayer (Fig.2 A and B, SI Appendix, Fig. S2B, and Movie S2). Moreover, theenergy barrier and transfer energy for bithionol increases withincreasing percentages of cholesterol (SI Appendix, Fig. S3A).These results are consistent with our observation that the presenceof cholesterol results in a more ordered alignment of the mem-brane lipids (SI Appendix, Fig. S3D), as well as with configura-tional and thermodynamic analyses (29, 30), which demonstratethat cholesterol condenses the hydrophobic region of the mem-brane (SI Appendix, Fig. S3B) and decreases membrane fluidity(SI Appendix, Fig. S3C). An independent MD simulation using theCHARMM force field (31) also showed that bithionol preferen-tially penetrates bacterial-mimetic compared with mammalian-mimetic lipid bilayers (SI Appendix, Fig. S4).Consistent with the MD simulations, bithionol did not cause

observable effects on GUVs formed with 7POPC/3cholesterol(Fig. 2C and Movies S6–S8). Moreover, bithionol exhibited rel-atively little hemolytic activity against human erythrocytes, withan HC50 >64 μg/mL (SI Appendix, Fig. S5A), and did not induceSYTOX Green membrane permeabilization of HKC-8 human

0 ns168 ns

290 ns500 ns

7POPC/3cholesterol0 ns

117 ns314 ns

7DOPC/3DOPG

500 ns

0 20 40 600

200400600800

1000

Time (min)

Fluo

resc

ence

inte

nsity S. aureus MW2

32 μg/mL 16 μg/mL 8 μg/mL 4 μg/mL2 μg/mL 1 μg/mL 0 μg/mL

0 20 40 600

200400600800

1000

Time (min)Fl

uore

scen

ce in

tens

ity HKC-8

64 μg/mL32 μg/mL16 μg/mL0 μg/mL

1 2 3 4-15

0

15

30

(kBT)

Pot

entia

l of m

ean

forc

e

(nm)Distance from bilayer center

7DOPC/3DOPG 7POPC/3chol

1 μg/ml 10 μg/ml 0.1%DMSOBithionol

7DO

PC/3

DO

PC

1 μg/ml 10 μg/mlBithionol

7PO

PC/3

chol

este

rol

DMSO0.1%

B.A.

Non-tr

eated

1 μg/m

l

2 μg/m

l

4 μg/m

l

8 μg/m

l

16 μg

/ml

32 μg

/ml

0.15

0.20

0.25

0.30

0.35

0.40

Laur

dan

GP

** **

***

***

Incr

ease

in fl

uidi

ty

A B

C D

E

Fig. 2. Bithionol selectively disrupts bacterial lipid bilayers. (A) Representa-tive configurations of MD simulations of bithionol from left to right: onset,membrane attachment, membrane penetration, and equilibrium interactingwith 7DOPC/3DOPG or 7POPC/3cholesterol lipid bilayers. Bithionol and sodiumions are depicted as large spheres, phospholipids are represented as chains,and bonds in cholesterols are highlighted by thickened tubes. The atoms inbithionol, phospholipids, cholesterol, and sodium ions are colored as follows:hydrogen, white; oxygen, red; nitrogen, blue; sulfur, yellow; chlorine, green;carbon, cyan; phosphorus, orange; and sodium, purple. For clarity, watermolecules are not shown. (B) The free-energy profiles of bithionol penetratinginto the indicated lipid bilayers as a function of the center-of-mass (COM)distance to the bilayer. The dot-dashed blue and red lines mark the surface ofbacterial and mammalian membranes, respectively, averaged from the COMlocations of phosphate groups in the lipids of the outer leaflets. Error barsrepresent means ± SD from 3 independent simulations. (C) GUVs consisting ofDOPC/DOPG (7:3) or POPC/cholesterol (7:3) labeled with 0.005% Liss Rhod PEwere treated with the indicated concentrations of bithionol or 0.1% DMSO(control) and were monitored over time by using fluorescence microscopy.(Scale bars, 10 μm.) (D) Uptake of SYTOX Green (Ex = 485 nm, Em = 525 nm)by MRSA MW2 persister cells or human renal proximal cells (HKC-8) treatedwith the indicated concentrations of bithionol. Results are shown as means;n = 3 biologically independent samples. Error bars not shown for clarity. (E) S.aureus MW2 membrane fluidity treated with the indicated concentrations ofbithionol for 1 h was evaluated based on Laurdan generalized polarization(Laurdan GP). Laurdan GP = (I440 − I490)/(I440 + I490), where I440 and I490 are theemission intensities at 440 and 490 nm, respectively, when excited at 350 nm.The membrane fluidizer benzyl alcohol (50 mM) was used as a positive con-trol. Individual data points (n = 3 biologically independent samples) areshown; error bars represent means ± SD. Statistical differences betweencontrol and antibiotic treatment groups were analyzed by 1-way ANOVA andpost hoc Dunnett test (**P = 0.01 and ***P < 0.0001.).

Kim et al. PNAS | August 13, 2019 | vol. 116 | no. 33 | 16531

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

29,

202

0

renal proximal cells up to 64 μg/mL (Fig. 2D and SI Appendix,Fig. S5 B and C).

Structure–Activity Relationships. The MD simulations describedhere earlier predicted that 2 key elements for the membrane ac-tivity of bithionol are the initial binding to the membrane surfaceusing phenolic hydroxyl groups and membrane perturbation in-duced by chlorinated benzene (Fig. 2A and Movie S1). To furthertest this proposed mechanism and to obtain additional insightsinto the effects of functional groups on the potency of bithionol,we conducted structure–activity relationship (SAR) studies byusing a commercially available bithionol analog, Bitin-S, as well aswith 7 newly synthesized analogs (Table 1 and SI Appendix, Figs.S6 and S7). The effect of binding affinity on antimicrobial andmembrane activity was tested by using Bitin-S, a sulfoxide de-rivative of bithionol (32), and the bithionol methoxy analog BT-OMe (Table 1). The polar sulfinyl group of Bitin-S provides ad-ditional hydrophilic interactions with lipid heads. Bitin-S exhibiteddecreased antimicrobial activity (MIC 8 μg/mL) and reducedmembrane activity (Table 1 and SI Appendix, Fig. S6). Reducingpolarity by substituting methoxy groups for the 2 hydroxyl groups(BT-OMe; Table 1 and SI Appendix, Fig. S6) resulted in completenullification of both antimicrobial and membrane activity (Table 1and SI Appendix, Fig. S6), indicating that the phenolic hydroxylgroups are critical for antimicrobial activity.To test whether the size and polarization of the inserted mole-

cules can affect the extent of membrane perturbation (12), wesubstituted alternative halogens for chlorine. Replacement of chlo-rine with fluorine resulted in reduced membrane and antimicrobialactivities as seen with compounds BT-oF, BT-pF, and BT-opF, whereassubstitution with bromine showed similar membrane and antimi-crobial activities as bithionol as demonstrated by BT-oBr, BT-pBr,and BT-opBr (Table 1 and SI Appendix, Fig. S7). It is likely that thepolarity of the C-F bond in the fluoro derivatives may increase thehydrophilicity of the aryl rings, thus decreasing membrane perme-ability. The bromine derivatives showed increased energy barriervalues of 2 to 3 kBT (SI Appendix, Table S2), indicating that thelarger bromine atoms may cause more membrane perturbation than

the smaller chlorine and fluorine atoms, although the bromines maybe slightly disadvantageous for initial binding and penetration.

Bithionol in Combination with Gentamicin Shows Efficacy in a MouseDeep-Seated MRSA Infection Model.We previously showed that theaminoglycoside antibiotic gentamicin, which is used to treat se-vere chronic MRSA infections despite nephrotoxicity issues (33),exhibits significant synergism with synthetic retinoids (6). Althoughwe found that relatively low concentrations of bithionol (8 μg/mL)do not lead to a significant decrease in the viability of MRSA per-sisters (Fig. 1 C and D), 8 μg/mL bithionol combined with as little as2 μg/mL gentamicin completely eradicated MRSAMW2 stationary-phase persister cells (Fig. 3A). Similarly, 8 μg/mL bithionol + 16 μg/mLgentamicin eradicated biofilm persisters (Fig. 3B).We further tested the efficacy of bithionol in combination with

gentamicin in an MRSA mouse thigh infection model, whichmimics human deep-seated chronic infections (5). Consistentwith a previous study (5), vancomycin, gentamicin, or a combi-nation of the 2 did not significantly reduce MRSA CFUs in themouse thigh (Fig. 3C), suggesting that the infecting bacterialcells are persisters. Although 30 mg/kg bithionol alone showedno effect on the viability of MRSA persisters, 30 mg/kg bithionolcombined with 30 mg/kg gentamicin killed ∼90% of the MRSApersister cells (P < 0.001; Fig. 3C). We evaluated the hepatic andrenal toxicity of bithionol in the mice in the experiments de-scribed in Fig. 3C by measuring serum levels of alanine amino-transferase (ALT) and blood urea nitrogen (BUN) (SI Appendix,Fig. S8). Although the combination of vancomycin and genta-micin significantly increase BUN levels (P < 0.01), the combi-nation of bithionol and gentamicin increased neither.

The Killing of MRSA Persisters Is Correlated with Increased MembraneFluidity. The SAR studies reported here show that some com-pounds that permeabilize MRSA persister cell membranes toSYTOX Green, such as Bitin-S, BT-oF, or BT-pF, do not killthem (Table 1 and SI Appendix, Figs. S6 and S7). We reported asimilar result in a previous SAR study on nTZDpa (12), wherewe observed that nTZDpa-analogs 6 and 11 (Fig. 4 B and C)induced rapid SYTOX Green permeabilization, but did not af-fect the viability of MRSA persisters (12).Interestingly, the bithionol SAR series showed that only

compounds leading to a significant increase in membrane fluiditykilled MRSA persisters (Fig. 4A, Table 1, and SI Appendix, Figs. S6and S7). To test the hypothesis that there is a general correlationbetween an increase in membrane fluidity and antipersister potency,we measured the effect of nTZDpa and 11 nTZDpa derivatives(Fig. 4 B and C) (12) on MRSA membrane fluidity. Consistent withthe bithionol analogs, only nTZDpa analogs that exhibit anti-persister potency showed increased membrane fluidity at 32 μg/mL.Furthermore, we observed substantial correlation between theconcentration of compound required for persister killing and theamount of induced membrane fluidity as measured by LaurdanGP for all 21 tested compounds at 32 μg/mL (R2 0.64, P ofslope < 0.0001; SI Appendix, Fig. S9). Consistent with these data,a recent study reported that the antibiotic agent rhodomyrtonecauses increased membrane fluidity in Bacillus subtilis and killsits persister cells (23).The data in this section show that the killing of MRSA per-

sisters is achieved only when membrane damage by membrane-active agents is sufficiently severe to show increased membranefluidity, potentially making membrane fluidity a biophysical in-dicator to identify and measure the potency of antipersisterantimicrobial agents.

DiscussionMembrane-active agents have attractive properties as antimi-crobial agents, including fast killing rates, antipersister potency,synergism with other antibiotic agents, and a low probability of

Table 1. Structure–activity relationships for antibiotic activityand membrane activity

Cmpd R R1 R2 R3 MIC PKC MP MFIBithionol S OH Cl Cl 1 32 Yes YesBitin-S SO OH Cl Cl 8 >64 Yes NoBT-OMe S OMe Cl Cl >64 >64 No NoBT-oF S OH F Cl 2 >64 Yes NoBT-oBr S OH Br Cl 1 32 Yes YesBT-pF S OH Cl F 2 >64 Yes NoBT-pBr S OH Cl Br 1 32 Yes YesBT-opF S OH F F 8 >64 No NoBT-opBr S OH Br Br 1 32 Yes Yes

R1R

R1

R3R3

SOH

Cl

Cl

ClOH

Clbithionol

Zn dust (10 equiv)AcOH

OH

R3

SOH

R3100 ºC, 4 h

OH

R3

S ClO HO

R3

SHO

R3

OCl

AlCl3 (1 equiv)DCM

22 ºC, 8 h+(1 equiv)

R2 R2

R2 R2 R2 R2 R2

MIC, minimum inhibitory concentration (in micrograms per milliliter); PKC,persister killing concentration (in micrograms per milliliter) required to kill5 × 107 CFU/mL MRSA persister cells below the limit of detection; MP, memb-rane permeabilization (determined based on SYTOX Green fluorescence in-tensity); MFI, membrane fluidity increase (determined based on Laurdan GPmeasurement).

16532 | www.pnas.org/cgi/doi/10.1073/pnas.1904700116 Kim et al.

Dow

nloa

ded

by g

uest

on

June

29,

202

0

resistance selection (7). Unfortunately, most of these agents alsoindiscriminately disrupt mammalian membranes. However, evolu-tion has taken advantage of differences in bacterial and eukaryoticmembranes, as reflected in the production of cationic antimicro-bial peptides by animals and plants that specifically target bacterialcells. Gram-positive bacterial membrane lipid bilayers contain∼25% anionic phospholipids, whereas mammalian membranes arecomposed of zwitterionic (neutral) phospholipids and 20 to 50%cholesterol (27, 34). Cationic antimicrobial peptides bind prefer-entially to negatively charged bacterial membranes, and cholesterolin animal membranes creates a condensing effect that confersmembrane rigidity and prevents the penetration of antimicrobialpeptides (34).We found that bithionol penetration into negatively charged

bacterial mimetic lipid bilayers (7DOPC/3DOPG) is energeti-cally favorable, whereas bithionol penetration into cholesterol-rich mammalian mimetic lipid bilayers (7POPC/3cholesterol) isenergetically unfavorable (Fig. 2 A and B). Further, as the pro-portion of cholesterol increases from 0 to 30%, the penetrationof bithionol becomes increasingly unfavorable (SI Appendix, Fig.S3A), indicating that cholesterol plays a key role in bithionol’smembrane selectivity.MD modeling suggests that the 2 polarized phenolic hydroxyl

groups of bithionol play a major role in the presumed initial bindingto phospholipid headgroups via hydrogen bonding. SAR studiesshowed that a methoxy analog of bithionol nullified bioactivity,supporting this proposed mechanism of interaction (Table 1 and SIAppendix, Table S2 and Fig. S6). In addition to providing presumedpolar interactions with lipid headgroups, the bithionol chlorinemoieties also appear to play a key role in lipid bilayer perturbation.Briefly, after attachment, the binding affinity of bithionol is domi-nated by hydrophobic interactions between its aromatic rings andthe hydrophobic tails of the membrane lipids, which drives thepenetration of the chlorinated benzene into the outer leaflet of thelipid bilayer, causing lipid bilayer perturbation (SI Appendix, Fig. S2A and C). Accordingly, the replacement of chlorine with fluorineresulted in a decrease in antimicrobial potency (Table 1 and SI

Appendix, Fig. S7). Replacement of the chlorines with brominesdecreased initial binding to the lipid bilayer (SI Appendix, TableS2), but increased destabilization of the membrane compared withchlorine and fluorine after penetration. Collectively, the antimicrobialactivity of bithionol can be modulated by binding affinity to lipid headgroups, penetration depth, and molecule size.In contrast to bithionol and other membrane-active compounds

we have studied, daptomycin does not induce rapid SYTOXGreen membrane permeabilization or kill MRSA persisters(Fig. 1C and SI Appendix, Fig. S1C). Importantly, we discoveredthat bithionol, nTZDpa, and their analogs that exhibit anti-MRSA persister potency induce SYTOX Green membrane per-meabilization and an increase in membrane fluidity (Fig. 4, Table 1,and SI Appendix, Figs. S6 and S7). These data are consistent withprevious reports that show that insertion of compounds intomembrane bilayers can increase membrane disorder and fluidity,which subsequently causes passive membrane permeabilization (21,22). Interestingly, we found that the initiation of SYTOX Greenmembrane permeabilization occurs at a lower concentration ofthe membrane-active compounds than is required to increasemembrane fluidity (Fig. 2 D and E and SI Appendix, Figs. S6 andS7). Furthermore, some membrane-active agents, such as bitin-S,bithionol fluorine analogs, and nTZDpa-analogs 6 and 11, thatinduce SYTOX Green membrane permeabilization do not causea change in membrane fluidity and do not kill persisters (Fig. 4 Band C and SI Appendix, Figs. S6 and S7). Bacterial membranesconsist of lipid rafts organized into microdomains having differentlipid compositions (35). Depending on lipid compositions, onedomain may be less ordered and more fluid, whereas anotherdomain may be more ordered and rigid (35). Because ordered andrigid domains show more resistance to membrane active agents(35) up to a certain threshold concentration, only less ordereddomains would be affected by membrane-active compounds, thusmaking them SYTOX Green-permeable. However, this type of

0 1 2 3 4

9876543

Time (h)

Log

(CFU

/mL)

8 μg/mL BT + 2 μg/mL Gm 8 μg/mL BT + 4 μg/mL Gm8 μg/mL BT + 8 μg/mL Gm8 μg/mL BT + 16 μg/mL Gm8 μg/mL BT16 μg/mL GmNon-treated

Non-tr

eated

8 μg/m

L BT

8 μg/m

L BT +

2 μg/m

L Gm

8 μg/m

L BT +

4 μg/m

L Gm

8 μg/m

L BT +

8 μg/m

L Gm

8 μg/m

L BT +

16 μg

/mL G

m

16 μg

/mL G

m

Log

(CFU

/mem

bran

e) 876543

Contro

lVan Gm

Van+G

m BT

BT + Gm

10

9

8

7

6

Log

(CFU

/g)

***

A B C

Fig. 3. Bithionol shows synergism with gentamicin against MRSA persistersin vitro and in vivo. (A) Stationary-phase or (B) biofilm MRSA MW2 persisterswere treated with the indicated concentrations of bithionol (BT) combinedwith gentamicin (Gm). Colony forming units (CFUs) were measured by serialdilution and plating on TSA plates. The data points on the x-axis are belowthe level of detection (2 × 102 CFU/mL, or 2 × 102 CFU per membrane). In-dividual data points (n = 3 biologically independent samples) are shown;error bars represent means ± SD. (C) Ten infected mice per group (n = 10biologically independent animals) were treated with control (5% Killophor +5% ethanol, i.p.), vancomycin (30 mg/kg, i.p.), gentamicin (30 mg/kg, s.c.),bithionol (30 mg/kg, i.p.), or vancomycin (30 mg/kg, i.p.) or bithionol (30 mg/kg,i.p.) combined with gentamicin (30 mg/kg, s.c.) every 12 h for 3 d at 24 hpostinfection. At 12 h after the last treatment, mice were euthanized.Their thighs were excised and homogenized. CFUs from each mouse thighare plotted as individual points, and error bars represent the SD in eachexperimental group. Statistical differences between control and antibiotictreatment groups were analyzed by 1-way ANOVA and post hoc Tukey test(***P < 0.001).

B.A.

Non-tr

eated

Bithion

ol

Bitin-S

BT-OMe

BT-oF

BT-oBr

BT-pF

BT-pBr

BT-opF

BT-opB

r0.1

0.2

0.3

0.4

Laur

dan

GP

*** ****** ***

N OH

OClX

R1

R2

234

3

4

Cmpd R1 R2 XnTZDpa 4-Cl H S4 4-Cl 4-Cl S5 4-Cl 4-tBu S6 4-Cl H O10 3,4-Cl H O11 4-Cl 4-Cl O12 4-Cl 3,4-Cl O13 4-Cl 4-Br O14 4-Cl 4-I OS21 2,4-Cl H OS24 4-Cl 4-tBu OS26 4-Cl 4-CF3 O

******64

***16 ***

8

ns>64

**32

ns>64

***8 **

32 ***16 *

32

***16 ***

64

B.A.

Non-tr

eated

nTZDpa 4 5 6 10 11 12 13 14 S21 S24 S26

0.1

0.2

0.3

0.4

Laur

dan

GP

A

B

C

Fig. 4. Relationship between antipersister activity and alteration in mem-brane fluidity. Membrane fluidity of (A) bithionol and its analogs or (B)nTZDpa and its analogs at 32 μg/mL was evaluated by Laurdan GP. Themembrane fluidizer benzyl alcohol (B.A.; 50 mM) was used as positive con-trol. (B) The blue-colored numbers above each bar indicate persister killingconcentration (PKC, in micrograms per milliliter) required to kill 5 × 107 CFU/mLMRSA persister below the limit of detection (2 × 102 CFU/mL). (A and B)Individual data points (n = 3 biologically independent experiments) areshown; error bars represent means ± SD. Statistical differences betweencontrol and antibiotic treatment groups were analyzed by 1-way ANOVAand post hoc Dunnett test; ns, no significance (P > 0.05), *P = 0.05, **P =0.01, and ***P < 0.001. Individual data points (n = 3 biologically in-dependent experiments) are shown; error bars represent means ± SD. (C)The structures of nTZDpa and its analogs.

Kim et al. PNAS | August 13, 2019 | vol. 116 | no. 33 | 16533

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

29,

202

0

localized membrane damage may not be sufficient to cause anoverall increase in membrane fluidity. Over the threshold concen-tration, most membrane domains would be disrupted, and, sub-sequently, overall membrane fluidity would increase. It is alsopossible that some compounds may attack only less ordered andmore fluid areas of membranes, which causes SYTOX Greenmembrane permeability, but not an overall increase in membranefluidity. In any case, the killing of MRSA persisters is apparentlyachieved only when the bacterial membrane is sufficiently dam-aged to show increased membrane fluidity as detected byLaurdan GP.In conclusion, we report that the clinically approved anthel-

mintic bithionol is an effective antimicrobial agent against bothmultidrug-resistant and -persistent Gram-positive pathogens.Bithionol kills Gram-positive (but not Gram-negative) bacterialcells by disrupting lipid bilayers while maintaining high selectivityfor bacterial compared with mammalian membranes, a conse-quence of the presence of cholesterol in mammalian membranes.Further, bithionol in combination with gentamicin effectivelyeradicates S. aureus persisters in vitro and significantly reducesbacterial burden in a mouse model of chronic deep-seatedMRSA infection. We also demonstrate that increased mem-brane fluidity is a biophysical indicator to identify potent anti-persister compounds. Our studies provide further understanding ofthe molecular mechanisms by which membrane-active smallmolecules selectively disrupt Gram-positive bacterial over

mammalian membranes and support the conclusion thatmembrane-active antimicrobial agents have promising potentialto be used for treating chronic infections caused by bacterialpersisters.

Materials and MethodsSI Appendix, Materials and Methods describes in detail the materials andprocedures used in this study, including bacterial strains and growth con-ditions, antimicrobial agents and chemicals, transmission electron micros-copy, a minimal inhibitory concentration (MIC) assay, a killing kinetics assay,a persister cell killing assay, a biofilm persister killing assay, a bacterialmembrane permeability assay, a mammalian membrane permeability assay,a membrane fluidity assay, a human blood hemolysis assay, a giant uni-lamellar vesicles (GUVs) assay, all-atom molecular dynamics (MD) simula-tions, a deep-seated mouse thigh infection model, and general proceduresfor the synthesis of bithionol analogs.

ACKNOWLEDGMENTS. This study was supported by National Institutes ofHealth Grants P01 AI083214 (to F.M.A. and E.M.), P20 GM121344 (toB.B.F.), and R35 GM119426 (to W.M.W.) and by National Science FoundationGrant CMMI-1562904 (to H.G.). We thank the Institute of Chemistry and CellBiology (ICCB)–Longwood at Harvard Medical School for providing thechemical libraries used in this study. We thank Dr. L. Rice for generouslyproviding the E. faecium strains. The simulations reported were performedon resources provided by the Extreme Science and Engineering DiscoveryEnvironment (XSEDE) through Grant MSS090046 and the Center for Compu-tation and Visualization (CCV) at Brown University. The NMR instrumentsused in this work were supported by National Science Foundation GrantCHE-1531620.

1. A. S. Lee et al., Methicillin-resistant Staphylococcus aureus. Nat. Rev. Dis. Primers 4,18033 (2018).

2. K. Lewis, Persister cells. Annu. Rev. Microbiol. 64, 357–372 (2010).3. B. P. Conlon et al., Persister formation in Staphylococcus aureus is associated with ATP

depletion. Nat. Microbiol. 1, 16051 (2016).4. K. R. Allison, M. P. Brynildsen, J. J. Collins, Metabolite-enabled eradication of bacterial

persisters by aminoglycosides. Nature 473, 216–220 (2011).5. B. P. Conlon et al., Activated ClpP kills persisters and eradicates a chronic biofilm in-

fection. Nature 503, 365–370 (2013).6. W. Kim et al., A new class of synthetic retinoid antibiotics effective against bacterial

persisters. Nature 556, 103–107 (2018).7. J. G. Hurdle, A. J. O’Neill, I. Chopra, R. E. Lee, Targeting bacterial membrane function:

An underexploited mechanism for treating persistent infections. Nat. Rev. Microbiol.9, 62–75 (2011).

8. W. R. Miller, A. S. Bayer, C. A. Arias, Mechanism of action and resistance to dapto-mycin in Staphylococcus aureus and Enterococci. Cold Spring Harb. Perspect. Med. 6,a026997 (2016).

9. B. Sharma, A. V. Brown, N. E. Matluck, L. T. Hu, K. Lewis, Borrelia burgdorferi, thecausative agent of lyme disease, forms drug-tolerant persister cells. Antimicrob.Agents Chemother. 59, 4616–4624 (2015).

10. G. S. Sahukhal, S. Pandey, M. O. Elasri, msaABCR operon is involved in persister cellformation in Staphylococcus aureus. BMC Microbiol. 17, 218 (2017).

11. W. Kim et al., Identification of an antimicrobial agent effective against methicillin-resistant Staphylococcus aureus persisters using a fluorescence-based screeningstrategy. PLoS One 10, e0127640 (2015).

12. W. Kim et al., Discovery and optimization of nTZDpa as an antibiotic effective againstbacterial persisters. ACS Infect. Dis. 4, 1540–1545 (2018).

13. W. Kim et al., NH125 kills methicillin-resistant Staphylococcus aureus persisters bylipid bilayer disruption. Future Med. Chem. 8, 257–269 (2016).

14. J. Keiser, J. Utzinger, Food-borne trematodiasis: Current chemotherapy and advanceswith artemisinins and synthetic trioxolanes. Trends Parasitol. 23, 555–562 (2007).

15. F. S. Barr, G. F. Collins, L. G. Wyatt, Potentiation of the antimicrobial activity of bi-thionol. J. Pharm. Sci. 54, 801–802 (1965).

16. A. L. Erlandson, Jr, C. A. Lawrence, Inactivating medium for hexachlorophene (G11) typesof compounds and some substituted phenolic disinfectants. Science 118, 274–276 (1953).

17. N. Schmid et al., Definition and testing of the GROMOS force-field versions 54A7 and54B7. Eur. Biophys. J. 40, 843–856 (2011).

18. A. K. Malde et al., An automated force field topology builder (ATB) and repository:Version 1.0. J. Chem. Theory Comput. 7, 4026–4037 (2011).

19. Y.-F. Chen, T.-L. Sun, Y. Sun, H. W. Huang, Interaction of daptomycin with lipid bi-layers: A lipid extracting effect. Biochemistry 53, 5384–5392 (2014).

20. B. L. Roth, M. Poot, S. T. Yue, P. J. Millard, Bacterial viability and antibiotic susceptibilitytesting with SYTOX green nucleic acid stain.Appl. Environ. Microbiol. 63, 2421–2431 (1997).

21. M. L. Cartron et al., Bactericidal activity of the human skin fatty acid cis-6-hexadecanoic acid on Staphylococcus aureus. Antimicrob. Agents Chemother. 58,3599–3609 (2014).

22. Y. Wu et al., Antibacterial activity and membrane-disruptive mechanism of 3-p-trans-coumaroyl-2-hydroxyquinic acid, a novel phenolic compound from pine needles ofCedrus deodara, against Staphylococcus aureus. Molecules 21, E1084 (2016).

23. D. Saeloh et al., The novel antibiotic rhodomyrtone traps membrane proteins invesicles with increased fluidity. PLoS Pathog. 14, e1006876 (2018).

24. R. M. Epand, C. Walker, R. F. Epand, N. A. Magarvey, Molecular mechanisms ofmembrane targeting antibiotics. Biochim. Biophys. Acta 1858, 980–987 (2016).

25. T. Parasassi, E. Gratton, Membrane lipid domains and dynamics as detected byLaurdan fluorescence. J. Fluoresc. 5, 59–69 (1995).

26. T. Harayama, H. Riezman, Understanding the diversity of membrane lipid composi-tion. Nat. Rev. Mol. Cell Biol. 19, 281–296 (2018).

27. M. Deleu, J.-M. Crowet, M. N. Nasir, L. Lins, Complementary biophysical tools to in-vestigate lipid specificity in the interaction between bioactive molecules and theplasma membrane: A review. Biochim. Biophys. Acta 1838, 3171–3190 (2014).

28. C. Hong, D. P. Tieleman, Y. Wang, Microsecond molecular dynamics simulations oflipid mixing. Langmuir 30, 11993–12001 (2014).

29. F. de Meyer, B. Smit, Effect of cholesterol on the structure of a phospholipid bilayer.Proc. Natl. Acad. Sci. U.S.A. 106, 3654–3658 (2009).

30. T. Róg, M. Pasenkiewicz-Gierula, I. Vattulainen, M. Karttunen, Ordering effects ofcholesterol and its analogues. Biochim. Biophys. Acta 1788, 97–121 (2009).

31. J. B. Klauda et al., Update of the CHARMM all-atom additive force field for lipids:Validation on six lipid types. J. Phys. Chem. B 114, 7830–7843 (2010).

32. T. Meshi, M. Yoshikawa, Y. Sato, Metabolic fate of bis(3,5-dichloro-2-hydroxyphenyl)-sulfoxide (bithionol sulfoxide). Biochem. Pharmacol. 19, 1351–1361 (1970).

33. C. Liu et al.; Infectious Diseases Society of America, Clinical practice guidelines by theinfectious diseases society of America for the treatment of methicillin-resistantStaphylococcus aureus infections in adults and children. Clin. Infect. Dis. 52, e18–e55 (2011).

34. J. R. Brender, A. J. McHenry, A. Ramamoorthy, Does cholesterol play a role in thebacterial selectivity of antimicrobial peptides? Front. Immunol. 3, 195 (2012).

35. M. Bramkamp, D. López, Exploring the existence of lipid rafts in bacteria. Microbiol.Mol. Biol. Rev. 79, 81–100 (2015).

16534 | www.pnas.org/cgi/doi/10.1073/pnas.1904700116 Kim et al.

Dow

nloa

ded

by g

uest

on

June

29,

202

0