State-stabilizing Interactions in Bacterial Mechanosensitive Channel Gating and Adaptation*

We outline several principles that we believe define the gating of two bacterial mechanosensitive channels, MscL and MscS. Serving as turgor regulators in bacteria and other walled cells, these molecules are tangible models for studying conformational transitions in membrane proteins driven directly by membrane tension. MscL, a compact pentamer, reversibly opens a gigantic 30-Å pore at near-lytic tensions. MscS, a heptameric complex, exhibits transient activation of a smaller pore at moderate tensions, thereby entering a tension-insensitive inactivated state. By comparing the structures and predicted transitions in these channels, we concluded that opening is commonly achieved through tilting and outward motion of the pore-lining helices, which is kinetically limited by hydration of the pore. The intricate adaptive behavior in MscS appears to depend on specific interhelical associations and the flexibility of the pore-lining helices. We discuss physical factors that may direct the transitions and stabilize main functional states in these channels.

We outline several principles that we believe define the gating of two bacterial mechanosensitive channels, MscL and MscS. Serving as turgor regulators in bacteria and other walled cells, these molecules are tangible models for studying conformational transitions in membrane proteins driven directly by membrane tension. MscL, a compact pentamer, reversibly opens a gigantic 30-Å pore at near-lytic tensions. MscS, a heptameric complex, exhibits transient activation of a smaller pore at moderate tensions, thereby entering a tension-insensitive inactivated state. By comparing the structures and predicted transitions in these channels, we concluded that opening is commonly achieved through tilting and outward motion of the porelining helices, which is kinetically limited by hydration of the pore. The intricate adaptive behavior in MscS appears to depend on specific interhelical associations and the flexibility of the pore-lining helices. We discuss physical factors that may direct the transitions and stabilize main functional states in these channels.
Osmotic forces are strong, which necessitated development of osmoregulation along with the first semipermeable membrane delineating the early cell. A simple estimation shows that a 1-m cell behaving as an ideal osmometer would sustain a downshock no stronger than 20 mM, after which membrane tension would exceed the lytic limit of 10 -12 dynes/cm. Thus, a cell without a reinforcing envelope or protective valves is very vulnerable. Free-living and enteric microorganisms cycling through the soil and experiencing drastic environmental changes developed robust mechanisms to maintain volume and integrity (1). The mechanosensitive channels MscS and MscL (mechanosensitive channels of small and large conductance, respectively) have been identified as primary osmolyte release valves limiting the turgor pressure under acute osmotic shock (2)(3)(4).
Without mscS and mscL genes, Escherichia coli survives a 300 mosM osmotic downshock (2), its resistance attributed to the peptidoglycan layer partially restraining swelling. However, expression of either MscS or MscL allows cells to withstand a 700 -800 mosM downshock through release of small osmolytes (2). Purification and reconstitution proved that MscL and MscS respond directly to tension in the lipid bilayer (5)(6)(7). Both channels reside in the inner (cytoplasmic) membrane (8), with MscL localized at the cell poles, bearing high curvature (9).
As primary components of the turgor regulation system, E. coli MscS and MscL became convenient models for studies of tension-driven conformational transitions in membrane proteins (10). The crystal structures of closed-state Mycobacterium tuberculosis MscL (11) and E. coli MscS in two distinct conformations (12,13) provided invaluable initial points to explore their gating mechanisms, in which computational methods play increasingly important roles.

Opening of MscL Creates a Large Stable Pore Accompanied by Protein Expansion in the Plane of the Membrane
It is accepted that many membrane-embedded mechanoreceptors undergo transitions by obeying lateral tension (␥) applied to the bilayer. This force biases the receptor distribution between a narrow resting state and wider open states, causing an in-plane area increase (⌬A). According to Boltzmann, the work done by tension (Ϫ␥⌬A) changes occupancy of both states: P o /P c ϭ exp(Ϫ(⌬E Ϫ ␥⌬A)/kT), where ⌬E is the intrinsic energy gap between the states (14). Electrophysiological experiments combined with patch imaging estimated the midpoint of the E. coli MscL activation curve (␥1 ⁄ 2 ) to be 9.5-12 dynes/cm, depending on the lipid environment (7,15). From the midpoint and slope of the activation curve, both ⌬A and ⌬E between the closed and open states were estimated as 20 nm 2 and ϳ125 kJ/mol, respectively (16). MscL forms an ϳ30-Å nonselective pore and stays open from milliseconds at the activation threshold to seconds at near-saturating tensions (15). The cytoplasmic bundle of the C-terminal segments was predicted to form a stable pre-filter in both open and closed conformations (17), but recent results suggested that it may occasionally disassemble, as frequently firing gain-of-function MscL mutants can pass polypeptides up to 6.5 kDa (18).

MscL Structure and Character of Transition
The M. tuberculosis MscL crystal structure (11) revealed a pentamer of two-TM 2 hairpin-like subunits. The complex looks like a tightly packed nonconductive bundle of 10 helices in the membrane. Short cytoplasmic N termini and bundled C-terminal domains reside in the cytoplasm. Soon after publishing the M. tuberculosis MscL structure, a homologous model for the E. coli MscL was proposed (19), which is in good agreement with the recently revised crystallographic model of M. tuberculosis MscL (Protein Data Bank code 2OAR) (10).
The original hypothesis in the crystallographic study (11) suggested that the tilted TM1 helices that form a tight hydrophobic constriction swing away from the pore axis and arrange themselves in a barrel-stave fashion with the TM2s. However, today's consensus is that tilting in an iris-like manner, not straightening up, leads to a wide pore and large in-plane expansion (20,21). The iris-like model (supplemental Fig. 1A) was supported by the engineered disulfide cross-links trapping MscL open (22). This model, constrained by experimental 3.2nanosiemen conductance and ϳ20-nm 2 in-plane barrel expansion, predicts an ϳ30°tilting and 12-Å outward movement of TM1s. The barrel-stave model predicts much smaller expansion and conductance. Independently, site-directed spin labeling and EPR experiments demonstrated that the pore is lined primarily with TM1s in closed and open states, whereas TM2s always face lipids (21), strongly supporting tilting. The two tilted models are not fully compatible because the EPR-based model implies a large rotation of the TM1 helix, contradicting the observed cysteine cross-links (22) and methanethiosulfonate accessibility data (23).

Why Tilting?
The iris-like model is consistent with the thermodynamic view that the system achieves deeper free energy minima by allowing tension to do more work over a larger area. But this does not explain why tilting would be preferred for the initial helix motion under tension unless we specify where exactly the tension is applied. Computational three-dimensional decomposition of the pressure tensor across the bilayer (24,25) predicted that the lateral pressure profile is non-uniform with deep minima (areas of tension) at the polar/apolar boundaries (supplemental Fig. 1C). There is a net positive pressure in the center of the bilayer (hydrocarbon) and in the most peripheral headgroup regions, which balances the tension at rest. The applied membrane stretch preferentially increases tension at the boundaries (25). Tilting is predicted to be a result of the outward radial forces applied to the ends of the pre-tilted helices. Tilting with a tangential component has an advantage over straightening, as it preserves extensive interhelical contacts in the course of expansion.

How Can the High Energy of MscL Activation Be Combined with a Stable Open State?
The tightly packed resting state of MscL is stable relative to the open state by 50 kT and requires an almost lytic tension to open. One of the contributions to the large energy gap may be distortion of the lipids by the flattened open channel (14,26,27). This thickness mismatch may exert a closing force. The second factor is high hydrophobicity of the gate formed by the rings of Val 23 and Leu 19 in E. coli MscL, which is dehydrated according to molecular dynamics (28). The vapor bubble in the gate makes MscL leak-proof, whereas surface tension of watervapor interfaces above and below the constriction additionally tightens the gate. Hydrophilic mutations in the gate (29 -31) pre-expand the channel and drastically reduce the transition barrier and opening energy, emphasizing the stabilizing role of pore dewetting.
Once opened, however, wild-type MscL stays open for tens of milliseconds, implying a deep open-state well separated from the closed state by a barrier. Kinetic analysis placed the ratelimiting barrier about two-thirds toward the open state on the in-plane expansion coordinate (15,31). Because the channel complex expands beyond the transition state, the closing rate diminishes with tension (15).
Another factor that may greatly stabilize the open state is the considerably less hydrophobic lining of the open pore. This points to the role of a highly conserved motif of periodic glycines (xxG 22 xxxG 26 xxxG 30 ) in TM1. In single-pass proteins capable of dimerization, such motifs permit tight knob-intohole packing of the helices (32). In MscS, this motif serves the same purpose, allowing the helices to form a tight gate (33). These glycines are buried in the resting state, making the constriction hydrophobic (supplemental Fig. 1, A and C). Upon expansion, the glycines expose their polar backbone. Replacing Gly 22 with alanine makes MscL considerably harder to open (30), which illustrates its role as a conditionally exposed polar group, lowering the critical radius of stable wetting (34).
In the iris-like model of MscL transition (22), paired TM1 and TM2 helices (11) act as rigid rods. If the helices were allowed to buckle, they would let the ends spread without pore opening. Because the pore-lining TM1 and lipid-facing TM2 never separate, tension will always be transmitted to the gate, and the channel remains sensitive to membrane tension at any time. As will be discussed next, the propensity of the porelining helices to buckle near the gate and the ability to separate from the lipid-facing helices appear to define the adaptive behavior of MscS.

Functional Characteristics of MscS
MscS, the smaller channel, activates at moderate tensions and tunes the turgor during the normal bacterial life cycle. When purified and reconstituted with soybean lipids, MscS activates with a midpoint of ϳ5.5 dynes/cm (6), slightly lower than previously reported for E. coli spheroplasts (35). The slope of the open probability on tension suggests ⌬A ϳ 13-18 nm 2 (36,37). Macroscopic estimations predict that the open pore is at least 16 Å in diameter (38).
A prominent functional trait of MscS is its ability to adapt to tension. It readily responds to abrupt pulses, but if the same tension is applied slowly (as a 30-s ramp), only about half of the population opens (36). In response to a prolonged step of subsaturating tension, MscS opens only transiently. This adaptive behavior results from two sequential processes, desensitization (gradual right-shift of the activation curve) and inactivation (36,38). In the inactivated state, MscS is tension-insensitive.

The Original Crystal Structure of MscS Is Nonconductive and Unstable When Simulated in the Lipid Bilayer
In the original (Protein Data Bank code 1MXM) and revised (code 2OAU) crystallographic models of wild-type MscS (10,12), the hydrophobic gate is predicted to be vapor-plugged, hence nonconductive (38,39). Steered ions passing through a dehydrated pore require a 200 -400-piconewton force or a transmembrane potential of ϳ1.5 V (38 -41). The structure exhibits an ϳ30°splay of the peripheral helices, which can result from delipidation (10), and the N terminus is disordered. Embedded in the lipid bilayer in molecular dynamics simulations, the unrestrained structure was unstable and collapsed asymmetrically (39,40). With deep crevices separating TM2 from TM3, both crystallographic models indicate no direct mechanical connection between the tension-receiving helices and the gate region. The pore-lining TM3s form a relatively narrow barrel (TM3a) and then break and separate (TM3b), forming the roof of the cytoplasmic cage (Fig. 1, upper left). Although the C-terminal ends of TM3s are "fixed" on the cage ϳ17 Å away from the pore axis, the characteristic kinks at Gly 113 bring Leu 105 and Leu 109 together to form a vapor-locked gate (12). As in MscL, hydrophilic substitutions in the gate result in a drastic decrease in activating tension and toxic gainof-function phenotypes (42), implicating pore dewetting (38) as setting the threshold for MscS.
In the recent crystal structure of the A106V mutant (Protein Data Bank code 2VV5), the TM3s form a wider (ϳ13 Å), apparently conductive pore (13). In patch-clamp experiments, this mutant frequently displays a half-amplitude subconductive state (33). Although the structure was interpreted as open, the periplasmic rim's narrow conformation suggests that the structure may not represent the fully open state. Macroscopic estimations predict that it conducts at about one-third of the experimental open MscS conductance. 3 Undoubtedly a big step forward, this new structure awaits computational exploration of its conductive properties and stability in the bilayer. Alterna-tive computational (37) and EPRbased (43) models of the open state will be discussed below.

Computational Exploration of the Conformational Space of MscS
As we are unable to maintain the membrane and tension in the crystallization chamber, it may be principally impossible to solve a true open state using crystallography. Replacement of lipids with detergent may lead to structurally distorted states. We resorted to computational techniques to model the functional cycle for MscS. We first used the newly developed extrapolated motion protocol (44) based on iterative cycles driving the protein along the self-permitted energy valleys without lipids and solvent. Candidate conformations satisfying experimental parameters of conductance and in-plane area were then refined in all-atom molecular dynamics simulations (37,44).

The Resting State of MscS Is Predicted to Be More Compact than the Crystal Structure
Adding the N terminus and repacking the peripheral helices along the central TM3a barrel using the extrapolation protocol resulted in a compact nonconductive state (44). A similar type of barrel packing was proposed previously (45). Equilibration of the compact model in an explicit bilayer predicted stable interhelical contacts. The subsequent cross-linking experiments confirmed the proximity of TM2 to TM3 in the resting state (44). Restoring these contacts reconnected the lipid-facing helices with the gate region. These modifications did not change the crystal-like packing of TM3a segments, in which the conserved G 101 A 102 A 103 G 104 xA 106 xG 108 xA 110 motif allows for tight helical packing (33).
The compact resting-state model was stable in the bilayer; however, its membrane position was different from what was inferred in the original crystallographic paper and used in early simulations (39,40). Arg 46 , Arg 54 , and Arg 74 , initially proposed to face the membrane's hydrocarbon (12), now reside more favorably in the layer of the polar headgroups. The characteristic TM3 kink was relocated two helical turns down from Gly 113 to Gly 121 . TM3b segments moved 7 Å down and resolved their steric conflicts with the ends of TM1-TM2 loops that occur during repacking of TM1-TM2 along TM3s (44).
The more recent resting model of MscS based on EPR data (46) suggested a similar conformation of the TM3a barrel and overall compaction of the external helices; however, it did not restore the direct TM2-TM3 contacts. The EPR-based model has TM3 kinks 3 A. Anishkin, unpublished data. The splayed peripheral TM1 (gold) and TM2 (green) helices of the crystal structure are aligned with the TM3a segments (cyan) in simulations, and the missing N terminus (red) was modeled de novo (44). The reconstructed resting state (lower left) with the closed gate (Leu 105 and Leu 109 ; yellow van der Waals spheres) equilibrated in the explicit lipid bilayer, and TM3s bent at Gly 121 . Opening results in kink-free TM3s dilating the gate by 8 Å (lower right). Inactivation is associated with re-forming of the crystallographic kink at Gly 113 and uncoupling of the TM1-TM2 pairs from the gate-bearing TM3s.
at Gly 113 and the predicted N-terminal domains to allow for more extensive interactions with the bilayer. The constraints provided by the EPR methodology and the criteria by which they are used to build the model may account for the differences.

The Open-state Model of MscS Predicts That the TM3 Helices Straighten and Rotate, Acting as Collapsible Struts
The opening of MscS in our extrapolated simulations resulted in radial retraction and straightening of the TM3 helices, accompanied by a 10°tilting (37). The pore widened by 8 -10 Å and was formed by uniformly bent helices with no discernable kinks. This tendency to straighten was previously observed in conventional molecular dynamics simulations (47). Extrapolated pore expansion from the resting state resulted in a similar conformational set that satisfied both the experimental conductance and the in-plane expansion. The best open-state candidate had at least a 6-Å wider pore than the crystal structure of A106V MscS (13). Subsequent equilibration of the open-state model under moderate tension in the explicit bilayer produced a 53°axial rotation of TM3s, which reoriented the gate-keeping Leu 105 and Leu 109 to closely appose Leu 111 and Leu 115 on the neighboring helix. Although the conserved TM3 motif of alternating glycines and alanines allowed a tightly packed resting state, the helix rotation observed in simulations exposed the buried Gly 101 , Gly 104 , Gly 106 , and Gly 108 to the lumen, favoring a fully hydrated state. The model stabilized in this state without restraints. Explicit conductance simulations performed on this model matched the experimental conductance and selectivity of open MscS (37).
The general character of transition in our open-state model (tilting and retraction of the helices) was consistent with both the A106V partially open crystal structure and the independently proposed open-state model based on constraints from scanning cysteine mutagenesis, spin labeling, and EPR (43). The authors of the EPR model reported that MscS could be stabilized in a conductive state by adding large amounts of lysophosphatidylcholine to the reconstituted proteoliposomes. This treatment asymmetrically changes the lateral pressure profile in the membrane but may not result in the same overall protein conformation as uniform saturating membrane tension. The deduced model showed an 8-Å widening of the gate achieved primarily through tilting of TM3s compared with the EPR model of the resting state (46). Distinctive features of the EPR model are 160°counterclockwise axial rotation of the TM3a helices that buries the glycines and partially unwound TM3b helical segments. Similar experiments previously predicted large rotations for TM1 in MscL, generally inconsistent with other data, which could be a result of bending and "snorkeling" of the EPR probe. The distance from the ␣-carbon to the paramagnetic oxygen of the cysteine-attached MTSL ((1-oxyl-2,2,5,5-tetramethylpyrrolindin-3-methyl) methanethiosulfonate) probe is 11 Å. Because an every-residue scanning approach was taken in both cases (21,44), the conformations and perturbing effects of the probe in the confined spaces, such as the pore and crevices, require further clarification.

The Closed and Inactivated States of MscS Are Characterized by Two Distinct Positions of the TM3 Kink
Our open-state model with kink-free TM3s suggested that the pore-lining helices may act as collapsible "struts" holding the pore open. Because the C-terminal ends of the TM3 helices are separated, re-forming the narrow TM3a barrel can be achieved only if the helices break and allow Leu 105 and Leu 109 to come together (Fig. 1). The return to the compact state in repeated extrapolations revealed preferential bending of TM3 at Gly 121 , which, unlike Gly 113 , was found absolutely conserved (38), pointing to an alternative hinge.
Because MscS has at least two nonconductive states, resting and inactivated (2,36,48), one may ask whether each state is defined by a different breaking point in the compacted TM3 barrel. This conjecture was strongly supported by experiments in which the flexibility of TM3 was changed by altering the two main "hinges," Gly 113 and Gly 121 (48). Increased helical propensity at the crystallographic kink by the G113A substitution essentially removed inactivation, whereas increased flexibility by engineering two sequential glycines (Q112G) strongly favored inactivation. Helical "straightening" in the alternative hinge region by the G121A mutation impeded exit from the open state and opened an unnatural inactivation path directly from the resting state, bypassing the open state. The A120G mutation increased this hinge's flexibility, destabilizing the open state. When the A120G mutation was combined with G113A, the channel swiftly and reversibly desensitized, but did not inactivate, always remaining sensitive to tension. Finally, mutating both hinge glycines to alanines (G113A/G121A) drastically stabilized the open state; after release of tension, the double mutant remained open for several minutes (48). These data strongly suggest that the open state has kink-free TM3 helices acting as struts. The Gly 121 hinge thus appears to mediate the closed-to-open transitions as well as desensitization, whereas formation of the kink near Gly 113 can be attributed to inactivation. Therefore, the original crystal structure (11) should resemble the inactivated state.

How Does Inactivated MscS Become Completely Insensitive to Membrane Stress?
We presume that inactivation is a way to preserve or restore membrane integrity under tensions that may be above the threshold for MscS but below lytic tension. The inactivated channel somehow manages to disengage the gate from tensionreceiving domains and ignore tension. The splayed state of the TM1 and TM2 helices, a prominent feature of the crystal structure, may suggest the mechanics of decoupling. Although we cannot exclude tension transmission to the gate either through the flexible periplasmic TM2-TM3 linkers (13) or through the putative Asp 62 -Arg 128/131 salt bridges on the cytoplasmic side (39,49), our preliminary data suggest a different route.
Reconstruction of the TM2-TM3 association (44) revealed a number of conserved residues forming this buried hydrophobic contact. The Leu 111 and Leu 115 side chains on the gate-forming TM3 are opposed by Val 65 , Phe 68 , and Leu 69 on TM2. Small side chain hydrophilic substitutions such as L111S and F68S increase the activation tension, so the force is less effectively conveyed to the gate. These mutations also favor faster inacti-vation and open a path for silent inactivation directly from the resting state. This strongly suggests that the reconstructed TM2-TM3 interface is necessary for tension transmission from the peripheral helices to the gate (50). Thus, the detached state of the TM1-TM2 "paddles" in the crystal structures may resemble a tension-insensitive inactivated state. Because the splayed conformation of TM1-TM2 is predicted to distort the surrounding bilayer and may be unfavorable, we presume the paddles' positions in the native inactivated state to be different from those in the delipidated crystal structure. A greater understanding of the interplay between the gate and the cage domain could be obtained through clarification of the functional roles of the predicted Asp 62 -Arg 128/131 salt bridges (39,49).

Conclusions
MscL and MscS are convenient systems for studying opening pathways and physical factors stabilizing the main functional states in tension-sensitive channels. Computational exploration of the conformational spaces critically assists in predicting such states. Experimental parameters of pore conductance and in-plane protein expansion constrain the open-state models. Opening the hydrophobic gate in both channels is achieved through tilting and outward movement of the pore-lining helices. The dehydrated state of the pore constriction stabilizes closed conformations, and wetting of the pore appears to be the rate-limiting step for activation in both channels. Exposure of conserved glycines changes the pore lining from hydrophobic to hydrophilic. Although in non-inactivating MscL the gate is rigidly connected to the tension-receiving surfaces through the TM1-TM2 contact, in MscS, the presence of a separable shell of peripheral helices around the gate allows for inactivation. The flexibility of the pore-lining helices at two distinct hinge glycines, which may depend on the associations with the peripheral helices, defines the functional cycle of MscS. Our understanding of bacterial models will assist in recognizing the mechanisms and functional roles of similar channels in many cells for environmental adaptations, host-parasite interactions, and maintenance of internal compartments and organelles.