Release of content through mechano-sensitivegates in pressurized liposomesMartti Louhivuoria,b,1, H. Jelger Risseladaa, Erik van der Giessenb, and Siewert J. Marrinka

aGroningen Biomolecular Sciences and Biotechnology Institute, Department of Biophysical Chemistry, University of Groningen, Nijenborgh 4, 9747AGGroningen, Netherlands; and bZernike Institute for Advanced Materials, Department of Applied Physics, University of Groningen, Nijenborgh 4,9747AG Groningen, Netherlands

Edited by Alan R. Fersht, MRC Centre for Protein Engineering, Cambridge, United Kingdom, and approved September 21, 2010 (received for review February2, 2010)

Mechano-sensitive channels are ubiquitous membrane proteinsthat activate in response to increasing tension in the lipid mem-brane. They facilitate a sudden, nonselective release of solutesand water that safeguards the integrity of the cell in hypo- orhyper-osmotic shock conditions. We have simulated the rapid re-lease of content from a pressurized liposome through a particularmechano-sensitive protein channel, MscL, embedded in the liposo-mal membrane. We show that a single channel is able to relax theliposome, stressed to the point of bursting, in a matter of micro-seconds. We map the full activation–deactivation cycle of MscLin near-atomic detail and are able to quantify the rapid decreasein liposomal stress as a result of channel activation. This providesa computational tool that opens the way to contribute to therational design of functional nano-containers.

coarse-grained ∣ molecular dynamics simulation ∣ MARTINI force field ∣protein channel ∣ drug delivery vehicle

Mechano-sensitive channels react to a sudden increase inmembrane tension by forming transmembrane pores that

counteract the pressure gradient buildup by balancing the osmo-tic conditions on either side of the cell membrane.(1–4) Theyhave a crucial role in diverse biological functions from sensoryfeedback to the prevention of cell death by membrane rupture(5, 6). The mechano-sensitive channel of large conductance(MscL) is a pentameric, prevalently helical membrane proteinweighing approximately 95 kDa. The crystal structure of an inac-tive MscL from Mycobacterium tuberculosis (Tb-MscL) has beenresolved (4, 7) by X-ray diffraction. Each of the five subunitsconsist of a transmembrane and a cytoplasmic domain connectedby a flexible linker. The two transmembrane helices TM1 andTM2 and the N-terminal helix S1 are arranged in a criss-crossmanner with TM1 and TM2 passing through the membraneand S1 parallel to the membrane surface (Fig. 1). The transmem-brane complex forms a ring-like structure with a hydrophobiclock region located slightly toward the cytoplasmic side from thecenter of the membrane. The cytoplasmic domains form a helicalbundle at the mouth of the membrane channel.

When activated, e.g., by increased membrane tension, MscLforms a nonselective transmembrane channel capable of quicklytransporting large amounts of solvent and solutes. The maximumflux through MscL (∼3 nS) is so high that only a handful of thechannels could (in theory) dehydrate an entire living cell withinseconds (4). In practise, MscLs are usually active no longer than afew hundreds of milliseconds with characteristic, rapidly flicker-ing activation–deactivation cycles plainly visible in single-channeltraces (8–10). The existence of multiple subconducting stateshas been proposed (8), and they have been hypothesised (11)to be the source of the rapid flickering. By introducing ingeniouspoint-mutations (12) at the channel walls the activation anddeactivation of MscL can be controlled by ambient pH and/orlight (13). This makes MscL a functional nano-valve with engi-neerable properties for a rapid, targeted drug release from a

suitable nano-container (e.g., a stable liposome) acting as a drugdelivery vehicle (14).

Here we study the spontaneous gating of an MscL channelembedded in a model liposome using molecular dynamics(MD) simulations. All system components, i.e., the lipid, protein,and solvent molecules, are represented at near-atomic detailusing the coarse-grained MARTINI force field (15, 16). We startfrom a system obtained by immersing the crystal structure ofTb-MscL in its closed state (4) in a small dioleoyl-phosphatidyl-choline (DOPC) lipid vesicle measuring approximately 16 nm indiameter. Subsequently, a hypo-osmotic shock condition was mi-micked by gradually increasing the internal water content, andhence interior pressure, of the vesicle over a 0.5 μs time window.

ResultsMscL Activates at the Limit of Membrane Elasticity. Fig. 2 shows theresponse of the vesicle to the increase of interior pressure. Twocases are shown, either with (Fig. 2A) or without (Fig. 2D) anMscL channel embedded in the liposomal membrane. In bothcases, the liposome initially undergoes an elastic swelling in aneffort to minimize the pressure gradient across the membrane(Fig. 2 B and E). The liposome grows until a threshold is reachedand then, in the absence of a channel, it simply bursts, ventingexcess solvent out (Fig. 2C). Intriguingly, once isotonic conditionsare restored, i.e., inside and outside pressures are equalized,the liposome may heal the rupture by closing itself once again(Fig. S1). In the presence of MscL, however, the uncontrolledrupture of the liposome is completely prevented. Instead, dissi-pation of the internal pressure occurs by the release of excess sol-vent through an activated membrane channel (Fig. 2F). A close-up of the system at the point of MscL activation is shown in Fig. 3.

From the simulations one can estimate the internal pressureof the vesicle at the point of gating to be ð138� 1Þ bar. Eventhough this pressure seems rather high at first glance, one shouldrealize that it is a microscopic pressure instead of a macroscopicpressure. The smaller a liposome is, the higher pressure differenceis needed to induce the same amount of tension in the membrane.For the current liposome, this considerable pressure differencecreates a lateral tension of ð67� 1Þ mN∕m (see Materials andMethods). This exceeds the experimentally estimated (8) midpointactivation tension of 12 mN∕m. Channel activation, however,involves one or even multiple kinetic barriers and, consequently,depends on the time scale at which the tension is applied (loadingrate). The time scale of the simulations (μs) is short comparedto experiments (ms), and hence the loading rate used is also

Author contributions: M.L., H.J.R., E.v.d.G., and S.J.M. designed research; M.L. performedresearch; M.L. analyzed data; and M.L., E.v.d.G., and S.J.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: m.j.louhivuori@rug.nl.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1001316107/-/DCSupplemental.

19856–19860 ∣ PNAS ∣ November 16, 2010 ∣ vol. 107 ∣ no. 46 www.pnas.org/cgi/doi/10.1073/pnas.1001316107

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

relatively high. At a similar submicrosecond loading rate(140 kN∕m·s), we observe the vesicular membrane of a purelipid vesicle to rupture at an estimated membrane tension ofð69� 1Þ mN∕m, similar to the tension required for channel acti-vation. Although the onset of both channel gating and membranerupture are time scale dependent, the response of the bilayerappears faster; simulations performed with loading rates of660 kN∕m·s do not lead to channel activation but rather to rup-ture of the membrane as in the case of the liposome without anembedded channel (Fig. S2). It is clear that in our simulations,MscL activates only at the very limits of membrane elasticity,in agreement with experimental findings (8, 17) for Tb-MscL.

Opening Mechanism Is Asymmetric. The observed mechanism ofMscL opening follows roughly the proposed (9,11) iris-like modelwith the transmembrane helices reorienting more loosely and at amore pronounced angle from the membrane normal. Contrary tothe beautiful, symmetric opening of the iris-like model, the free,nonbiased MscL seems to open in a distinctly asymmetric manner(Fig S3). The transmembrane helices tilt simultaneously, butindependently, to accommodate the thinning of the membrane,while theN-terminal helices follow themovement of TM1, stayingparallel to the membrane surface, and do not show any indicationof either moving to block the channel or to line the rim of thechannel as has been suggested (11). The surface area of the openchannel (60 nm2, including both protein and channel) is more

than two times larger than that of the closed channel (26 nm2).The minimum radius of the open channel in the constricting re-gion, estimated from the minimal distances between residuesI14–V21 of opposing subunits, measures ð11.6� 0.8Þ Å. This issmaller than the 15–20 Å estimated from conductance measure-ments including increasingly large “blocker” molecules (18, 19)and from thermodynamic consideration of in-plane area changes(8). It is plausible that the permeation of larger molecules occursby forcing the channel to open even further than necessary for apure water solvent, which would explain the difference.

Flow of Water Through the Channel Is Bidirectional. Our detailedmodel allows us to calculate the flow of water through the chan-nel. The open channel conducts on average ð4.4� 1.4Þ watermolecules per ns in the outward direction. This is in disagreementwith an earlier geometrical estimate (4) of 0.22 ns−1, when ΔP ¼0.1 bar, or >40 ns−1 when applied to the current system withΔP ¼ 138 bar. It should be noted that their analytical estimateis based on a macroscopic theory of solvent flow through a cylind-rical pipe, which is likely to break down for a nano-sized proteinchannel. Interestingly, we observe a bidirectional, competitive

Fig. 1. Mechano-sensitive channel of large conductance. X-ray crystal structure (white licorice) overlaid with a coarse-grained protein model (backbone shownin red; side chains in yellow). (A) Single subunit showing transmembrane helices TM1 and TM2, N-terminal helix S1, and the C-terminal helix, (B) complete MscLwith all five subunits, (C) periplasmic view inside the channel, and (D) cytoplasmic view at the C-terminal helix bundle.

Fig. 2. Response of a vesicle to increased interior pressure with or without amechano-sensitive channel (white) embedded in the liposomal membrane(orange/gray). (A–C) An osmotic shock causes a liposome to swell and ulti-mately to pop when the liposome reaches its elastic limits. (D–F) In contrast,a liposome equipped with a mechano-sensitive channel, such as MscL, cansurvive the same osmotic shock through a controlled release of excess solventand solutes (blue) via the membrane channel.

Fig. 3. A MscL in a liposome, right at the point of channel activation. Excesssolvent (blue) is released from the pressurized liposome interior through theprotein that acts as a nano-valve. For clarity, external water is not shown, theprotein is shown as a backbone trace on a transparent surface representation(white) with transmembrane helices depicted as cylinders (red), and some ofthe lipids are cut away to reveal the inside of the liposome.

Louhivuori et al. PNAS ∣ November 16, 2010 ∣ vol. 107 ∣ no. 46 ∣ 19857

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

flow in the simulation. The average unidirectional flows are foundto be ð6.0� 1.3Þ ns−1 outward and ð1.7� 0.3Þ ns−1 inward, withpeaks as high as 40 ns−1 outward and 28 ns−1 inward. It is remark-able that, despite the large interior pressure, both a substantialoutward and inward solvent flux occurs. By comparing the aver-age force pushing water molecules along the pressure gradient(22 pN) to the average force due to Brownian motion (121 pN),or the equivalent energies of 0.5 kJ∕mol and 2.6 kJ∕mol, respec-tively, it is clear that thermal noise still dominates the motionof water molecules.

Liposomal Stress Is Relaxed on a Submillisecond Time Scale. Afterchannel opening the simulation has been extended over a periodof 40 μs to monitor the relaxation of the system. During this per-iod the liposome undergoes a sustained venting of excess internalsolvent that eventually relaxes the built-up liposomal stress. Fig. 4quantifies the relaxation of liposomal stress, showing the tempor-al evolution of the liposome radius, the pressure differencebetween the inside and the outside of the liposome, the surfacetension in the membrane, radius of the MscL channel, and detailsabout the amount of solvent exchange. The radius of the lipo-some decreases from 9.7 nm to 8.4 nm and the inside volumefrom 2;195 nm3 to 1;071 nm3 approaching steadily the radius

(8.0 nm) and volume (843 nm3) of the initial, relaxed liposome.The amount of excess internal solvent (compared to the initialrelaxed liposome) is reduced by 82%, and the level of mixingof internal and external solvent reaches 66%. During this timethe membrane thickness (estimated by the average distancebetween the phosphate moieties in the lipid head groups) in-creases from 3.30 nm to 4.12 nm. For comparison, an equili-brated, tensionless DOPC membrane at 310 K has a thicknessof ð4.54� 0.02Þ nm.

As more and more solvent is fluxed out of the liposome,the pressure difference across the membrane decreases fromð138� 1Þ bar to ð56� 1Þ bar toward the end of the simulation.At this point, the vesicular membrane is still under a considerabletension of ð23� 1Þ mN∕m, enough to keep the channel open.The radius of the channel fluctuates around 1.1 nm, a decreaseof approximately 10% with respect to the largest openingobserved during the first microseconds after gating. However, thechannel is still in an active state with a total water flow of ð0.6�0.2Þ ns−1 and unidirectional flows of ð1.2� 0.2Þ ns−1 outwardand ð0.6� 0.1Þ ns−1 inward. The pressure difference as well asthe surface tension decrease linearly after an initial, high-conduc-tive period of approximately 8 μs. (Note the crossover from high-conductive state to a low-conductive state around 8 μs in Fig. 4.)

Fig. 4. Relaxation of a stressed liposome after channel activation by MscL-mediated solvent flux. Radius of the liposome (R), the pressure difference (ΔP)between the inside and the outside of the liposome, the surface tension (γ) in the membrane, the net amount of internal solvent transported outside theliposome (∑ χ), and the normalized molar fraction (Φ) of internal solvent initially located outside the liposome are shown in the upper plot. The radius of thechannel (r) and the momentary flux events (χ) are shown in the lower plot with the white line showing the net flux over 80-ns intervals. Above, snapshots of theprotein and the surrounding lipids and water are shown (A) for the initial, closed channel; (B) for the activated, open channel; and (C) for the channel with apartially dissociated cytoplasmic helix bundle.

19858 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1001316107 Louhivuori et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

By extrapolating from this linear regime (Fig. S4), one can esti-mate that it takes ð86� 4Þ μs to reach a tensionless state andð93� 4Þ μs to eliminate the pressure difference. These estimatesare of course only valid if MscL stays open and continues toconduct at a steady pace. This, however, is questionable; inour simulation we already observe a partial dissociation of theC-terminal helix bundle, allowing the helices to intermittently en-ter the channel and to obstruct the free flow of solvent. By block-ing the flux through the channel, the helices may give MscL anopportunity to constrict once again without the need of dewettingthe channel first, pointing to a different role of the C-terminalhelix than hitherto assumed (10, 11, 20) (see SI Text for details).

In conclusion, it is quite remarkable that a single MscL iscapable of relaxing a 138-bar pressure difference in a matterof microseconds. This study demonstrates that the simulationof a biomimetic nano-container is feasible and provides a com-putational tool for the rational design of programmable drugdelivery vehicles.

Materials and MethodsSystem Setup. MD simulations of MscL in model liposomes were carried outusing amodified version (21) of GROMACS (22) with mean-field force approx-imation (MFFA) boundary potentials (software available upon request). TheMFFA boundary potential is an effective potential that mimics the bulk watersurrounding the liposome, with the computational advantage that most ofthe external water can be removedwithout any adverse effect on the proper-ties of the liposome (21). The MARTINI coarse-grained force field (15)was used in conjunction with its recently released extension (16) for proteinmodels. In the MARTINI model, an average of four atoms and associated hy-drogens are mapped to an effective interaction site. Likewise, a CG waterbead represents four water molecules. Previously the tension-induced gatingof MscL embedded in a lamellar membrane was simulated (23) using thesame MARTINI model.

The initial system was obtained by immersing the crystal structure ofTb-MscL in its closed state (4) in a spontaneously formed lipid vesicle andallowing the system to equilibrate. The approximately 16 nm diameter lipo-some contained 2,108 DOPC lipids and 5,444 water beads with an additional54,649 water beads forming an approximately 4-nm water layer around thevesicle, enclosed by the MFFA boundary potential. The temperature of thesystem was kept constant at 310 K and the external pressure at 1 bar. MscLactivation was triggered by gradually increasing the internal pressure of theliposome during 0.5 μs, followed by a 40-μs simulation of the active channel.For further details see SI Text.

Increasing Internal Pressure. The internal pressure of the liposome was gra-dually increased by adding more and more water into the liposome. Thiswas achieved by having an additional MFFA potential in the center of theliposome that acted as a water piston capable of creating a water-fillablecavity inside the vesicle. This cavity was then filled with water, and the process

was repeated. MscL activation tension was reached after 480 ns of “pump-ing” in six cycles (Table S1). To prevent the rapid loss of pressure throughbasal water flux across the membrane, the parameters of the lipid tails wereadjusted slightly without affecting other membrane properties (Fig. S5 andTable S2).

The pressure inside and outside the liposome was computed from theaverage 3D pressure field across the system, a method we have developed(24) recently. The surface tension in the membrane was estimated usingthe Laplace equation (25, 26)

ΔP ¼ 2γ

r; [1]

where ΔP is the pressure difference, γ the surface tension, and r the radiusof the liposome. A similar estimate of the tension was obtained from themechanical definition of tension (Fig. S6). See SI Text for details.

Limitations of the Model. To put our results into perspective, it is important toreflect on some of the limitations underlying our coarse-grained model. IntheMARTINI force field, the secondary structure of the protein is constrained.Possible unfolding/folding events of MscL during the gating are thereforenot considered. Tertiary structural changes, however, proceed in an unbiasedway. The iris-like gating mechanism observed in our simulations results fromthe specific protein–protein and lipid-mediated interactions. These interac-tions have been parameterized in MARTINI based on reproduction of ther-modynamic data for small building blocks. The four-to-one mapping schemeof MARTINI allows the keeping of chemical specificity, although the direc-tionality of, e.g., hydrogen bonds is lost. The specificity retained in the modelis good enough, for instance, to reproduce the dimeric structure of trans-membrane helices (27, 28) and allows us to discriminate between severalMscL mutants (Table S3 and Fig. S7) as described in detail in SI Text.

Another point warranting discussion is the time scale. Due to the coarsen-ing of the interaction potentials, the dynamics is faster than in correspondingall-atom models. In most applications as well as in the current manuscript, afactor of four has been applied to provide an approximate time scale. Thisfactor is based on the speedup of diffusion rates for water and lipids whenmodeled with MARTINI (15). However, this factor only accounts for the ne-glect of friction from the missing atomistic degrees of freedom. One shouldrealize that the dynamics of MscL structural reorganizations is largely deter-mined by the energy barriers of the transition states along the gating path-way, which are easily under- or overestimated by several kT, potentiallyleading to orders-of-magnitude shifts in the observed dynamics. Concerningdynamical aspects, the simulations should be considered as qualitative only.

ACKNOWLEDGMENTS. We gratefully acknowledge the financial support ofthe Marie Curie Research Training Network “SizeDepEn” (M.L., E.v.d.G.),the Academy of Finland (M.L.), and the Netherlands Organization forScientific Research (H.J.R., S.J.M.). The computational resources of CSC(Espoo, Finland) and HECToR (Edinburgh, Scotland) provided through DEISAand of Netherlands National Computing Facilities were used in this study.

1. Martinac B, Buechner M, Delcour AH, Adler J, Kung C (1987) Pressure sensitive ionchannel in Escherichia coli. Proc Natl Acad Sci USA 84:2297–2301.

2. Levina N, et al. (1999) Protection of Escherichia coli cells against extreme turgorby activation of MscS and MscL mechanosensitive channels: Identification of genesrequired for MscS activity. EMBO J 18:1730–1737.

3. Perozo E, Rees DC (2003) Structure and mechanism in prokaryotic mechanosensitivechannels. Curr Opin Struct Biol 13:432–442.

4. Steinbacher S, Bass R, Strop P, Rees DC (2007) Structures of the prokaryotic mechan-osensitive channels MscL and MscS. Curr Top Membr 58:1–24.

5. Nakamaru Y, Takahashi Y, Unemoto T, Nakamura T (1999) Mechanosensitive channelfunctions to alleviate the cell lysis ofmarine bacterium, Vibrio alginolyticus, by osmoticdownshock. FEBS Lett 444:170–172.

6. Kung C (2005) A possible unifying principle for mechanosensation. Nature436:647–654.

7. Chang G, Spencer RH, Lee AT, Barclay MT, Rees DC (1998) Structure of the MscLhomolog from mycobacterium tuberculosis: A gated mechanosensitive ion channel.Science 282:2220–2226.

8. Sukharev SI, Sigurdson WJ, Kung C, Sachs F (1999) Energetic and spatial parametersfor gating of the bacterial large conductance mechanosensitive channel, MscL. J GenPhysiol 113:525–539.

9. Betanzos M, Chiang CS, Guy HR, Sukharev S (2002) A large iris-like expansion of amechanosensitive channel protein induced by membrane tension. Nat Struct Biol9:704–710.

10. Anishkin A, et al. (2003) On the conformation of the COOH-terminal domain of thelarge mechanosensitive channel MscL. J Gen Physiol 121:227–244.

11. Sukharev SI, Betanzos M, Chiang CS, Guy HR (2001) The gating mechanism of the largemechanosensitive channel MscL. Nature 409:720–724.

12. Koçer A, Walko M, Meijberg W, Feringa BL (2005) A light-actuated nanovalve derivedfrom a channel protein. Science 309:755–758.

13. Koçer A, Walko M, Feringa BL (2007) Synthesis and utilization of reversible andirreversible light-activated nanovalves derived from the channel protein mscl.Nat Protoc 2:1426–1437.

14. Koçer A (2007) A remote controlled valve in liposomes for triggered liposomal release.J Liposome Res 17:219–225.

15. Marrink SJ, Risselada HJ, Yefimov S, de Vries AH (2007) The MARTINI force field:Coarse grained model for biomolecular simulations. J Phys Chem B 111:7812–7824.

16. Monticelli L, et al. (2008) The MARTINI coarse-grained force field: Extension toproteins. J Chem Theory Comput 4:819–834.

17. Rawicz W, Smith BA, McIntosh TJ, Simon SA, Evans E (2008) Elasticity, strength,and water permeability of bilayers that contain raft microdomain-forming lipids.Biophys J 94:4725–4736.

18. van den Bogaart G, Krasnikov V, Poolman B (2007) Dual-color fluorescence-burstanalysis to probe protein efflux through the mechanosensitive channel MscL. BiophysJ 92:1233–1240.

19. Cruickshank CC, Minchin RF, Le Dain AC, Martinac B (1997) Estimation of the pore sizeof the large-conductance mechanosensitive ion channel of Escherichia coli. Biophys J73:1925–1931.

20. Ajouz B, Berrier C, Besnard M, Martinac B, Ghazi A (2000) Contributions of thedifferent extramembranous domains of the mechanosensitive ion channel MscL toits response to membrane tension. J Biol Chem 275:1015–1022.

Louhivuori et al. PNAS ∣ November 16, 2010 ∣ vol. 107 ∣ no. 46 ∣ 19859

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021

21. Risselada HJ, Mark AE, Marrink SJ (2008) Application of mean field boundarypotentials in simulations of vesicles. J Phys Chem B 112:7438–7447.

22. Hess B, Kutzner C, van der Spoel D, Lindahl E (2008) GROMACS 4: Algorithms for highlyefficient, load-balanced, and scalable molecular simulation. J Chem Theory Comput4:435–447.

23. Yefimov S, van der Giessen E, Onck PR, Marrink SJ (2008) Mechanosensitive membranechannels in action. Biophys J 94:2994–3002.

24. Ollila OHS, et al. (2009) 3D pressure field in lipid membranes and membrane-proteincomplexes. Phys Rev Lett 102:078101.

25. Rowlinson JS, Widom B (1982) Molecular Theory of Capillarity (Clarendon, Oxford).

26. Kralchevsky PA, Nagayama K (2001) Particles at Fluid Interfaces and Membranes

(Elsevier, Amsterdam).

27. Sengupta D, Rampioni A, Marrink SJ (2009) Simulations of the C-subunit of

ATP-synthase reveal helix rearrangements. Mol Membr Biol 26:422–434.

28. Sengupta D, Marrink SJ (2010) Lipid mediated interactions tune the association of

Glycophorin A helix and its disruptive mutants in membranes. Phys Chem Chem Phys

12:12987–12996.

19860 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1001316107 Louhivuori et al.

Dow

nloa

ded

by g

uest

on

Janu

ary

5, 2

021