Tension mediated mechanical activation and pocket delipidation lead to an analogous MscL state

The MscL channel gates in response to membrane tension changes to allow the exchange of molecules through its pore. Lipid removal from transmembrane pockets leads to a MscL response. However, it is unknown whether there is correlation between the tension mediated state and the state derived by pocket delipidation in the absence of tension. Transitions between MscL states may follow a similar pathway to cover the available conformational space but may not necessarily sample the same discrete intermediates. Here, we combined pulsed-EPR and HDX-MS measurements on MscL, coupled with molecular dynamics under membrane tension, to investigate the changes associated with the distinctively derived states. Whether it is tension or pocket delipidation, we find that MscL samples a similar expanded state, which is the final step of the delipidation pathway but only an intermediate stop of the tension mediated path. Our findings hint at synergistic modes of regulation in mechanosensitive channels.


91
Here we endeavored to investigate whether there is a structural analogy between the 92 physiologically relevant tension-activated state, and the one stabilized by modifications that result 93 in pocket lipid removal. To this end, we have combined untargeted (HDX-MS) and single-residue 94 (3pESEEM) methods to probe the architecture of MscL upon pocket delipidation and independently 95 generate a tension-activated MscL state by MD simulations. We find these two differently derived 96 states lead to an analogous state, suggesting a direct link between tension-and pocket lipid 97 removal-activation in MS channels. Further, our data suggest that upon tension-activation 98 additional structural changes can trigger substantial further opening of the channel pore, suggesting 99 that structural plasticity of these mechanosensitive ion channels enables them to respond 100 differently upon receipt of discrete stimuli.

105
The substitution of a tryptophan at position L89 in Mycobacterium tuberculosis MscL (TbMscL) has 106 been shown to restrict lipid access to channel pockets and destabilise the closed state, leading to 107 an expanded (sub-conducting) MscL state(8). In order to give a mechanistic insight into the 108 transitions occurring between the two states, we used HDX-MS to measure relative differences in 109 deuterium uptake between the WT (closed) and L89W (expanded by pocket delipidation) channel 110 proteins. We succeeded in obtaining 95% peptide coverage of the entire resolved MscL structure 111 where the gating occurs, i.e., residues 1-125, including the entire TM domain and the largest part 112 of the cytoplasmic helical bundle (83% overall coverage for the entire construct including the C-113 terminal 6xHis-tag). Differences in uptake, detected at the peptide level, between the two states 114 allowed us to identify protein regions that became deprotected from exchange in the sub-  Table S1). Residues 37-53 run from the top of TM1 and end after the first β-sheet 120 in the loop connecting TM1 to TM2. Residues 58-69 correspond to the middle of the periplasmic 121 loop up until TM2, and residues 97-111 cover the bottom of TM2 to the top of the C-terminal helical 122 bundle (Fig 1 and S1). This gave us the first glimpses into MscL's pocket lipid removal activation 123 and, which protein domains are mostly impacted by the L89W (pocket entrance) modification 124 across the entire MscL's length.

126
Solvent accessibility mapping at single residue resolution by 3pESEEM 127 Next, we sought to investigate the feasibility of 3pESEEM to study the structure of MscL and identify 128 sites that are potentially dynamic during gating and solvent accessible. To this end, we generated 129 twenty-six single cysteine mutants, accounting for 20% of the total TbMscL length, spanning all 130 protein domains. Specifically, L2 and F5 on the S1 helix, N13, L17, V21, I23, V31 and F34 on TM1, 131 L42, and F48 on the periplasmic loop, residues N70, V71, L72, L73, S74, F79, F84, A85, Y87, F88, 132 L89, R98, K99, and K100 on TM2, and E102 and V112 on the cytoplasmic loop and helical bundle 133 (CHB) respectively, were mutated to Cys in a Cys-free WT TbMscL background (Fig 2). We 134 4 subsequently expressed, purified and spin labelled each one of these twenty-six mutants (MTSSL 135 modification is denoted as R1 hereafter) and performed 3pESEEM solvent accessibility 136 measurements ( Fig S2). Solvent accessibility is based on the modulation depth of deuterium in the 137 time domain signal, which is proportional to its associated signal intensity in the frequency domain.

138
We used two independent analysis methods to determine solvent accessibility, both yielding very 139 similar results (Fig S3). Unlike PELDOR which requires high spin labelling efficiency to obtain high 140 sensitivity data acquisition(8, 13), 3pESEEM can be performed with no major losses in sensitivity 141 even if the sample of interest shows lower spin labelling efficiencies for individual sites. This allowed 142 us to obtain good quality spectra and quantify the solvent (D2O) accessibility for all twenty-six sites 143 (Fig 2, S2 and S4). Residues in the S1 helix showed low accessibility, which is consistent with 144 previous reports suggesting that this helix is buried within the bilayer(32). TM1 pore-lining residues 145 presented low to intermediate accessibility, as would be expected for a closed channel pore(55).

146
I23, V31 and F34 on TM1 are intermediately solvent-exposed (or relatively buried) and positioned 147 on the same side of TM1 (Fig 2 and S4). This could be due either to the presence of native lipids, 148 detergent used for membrane protein extraction and/or hinderance by the presence of TM2 on the 149 TM1 interface. L42 and V48 are intermediately to highly exposed, as they form part of the 150 periplasmic loop connecting TM1 and TM2. N70, V71, L72, L73, and S74 present the largest 151 disparity across our whole data set. L72 is the most buried residue we measured (~ 3% compared 152 to the most solvent-exposed cytoplasmic-facing labelled residue K100R1) while N70 (~90%) and 153 V71 (~80%) are two of the most exposed (~90%), comparable to D2O (solvent) accessibilities 154 measured for exposed cytoplasmic sites (Fig 2, S2 and S4). The remaining residues present 155 intermediate accessibilities to these two extremes, as expected for subsequent residues, which 156 form a helical turn with different space orientation. Residues N70 and V71 are solvent-exposed in 157 the closed MscL state, while residues L72 and L73 are buried. The latter pair of residues is known 158 to become exposed during opening, due to an anti-clockwise TM2 rotation (8,19,56 residues are solvent-exposed in the closed state. R98, K99 and K100 form a positively charged 164 cluster for attracting negatively charged lipids(57) and our data suggest that the non-ionic DDM 165 binds either differently or at a different site. Substitution of these residues with the neutral Gln had 166 no effect on MscL's conformation, suggesting that specific lipid headgroup binding on this particular 167 site does not influence MscL gating(8). E102 (cytoplasmic loop) and V112, which is located in the 168 upper portion of CHB and is expected to move apart upon MscL opening (59), presented high 169 solvent accessibilities. Combined, this has enabled us to validate 3pESEEM as a suitable tool to 170 map the structure of MscL.

172
Identification of channel gating regions with single residue resolution 173 HDX-MS resolution is usually limited to the peptide level and, therefore, it is often unable to resolve 174 changes that occur at different residues within the same peptide. Here we have already shown that 175 the solvent accessibility of individual residues of MscL can be probed using 3pESEEM. This  PELDOR's distance measurement range (< 19 Å) (Fig 1 and 3B). However, 3pESEEM is highly 185 sensitive to the local spin label environment, does not need a reference state as it is the case for 186 HDX-MS, and is not subjected to distance restriction between sites like for PELDOR. Differences 187 in the vapour-lock region may not have been detected by HDX-MS as structural changes may have 188 5 resulted in complex changes to hydrogen bonding networks, whereby some residues were 189 deprotected from exchange and others became protected, resulting in only minor/no changes in 190 deuterium uptake at the peptide level (Fig 1 and S1). To resolve this, we labelled the V21 and N13 191 sites crucial for the channel's vapour lock, and six other sites within the regions identified by HDX-

192
MS, which undergo conformational changes when the L89W modification is present (Fig 3, S5 and 193 S6). Overall, the motivation behind this was to confirm the peptide-level HDX-MS data and 194 complement this with single residue resolution 3pESEEM data for crucial sites where changes may 195 have not been detected, due to limitations associated with peptide-level analysis in HDX-MS.

196
3pESEEM measurements at position L42, located in the periplasmic loop, showed no significant 197 difference in solvent accessibility between the WT and the L89W states. For N13 which points 198 towards the cytoplasm and is solvent-exposed in the closed state, we observed similar exposure 199 in the expanded L89W MscL state (Fig 3 and S7). In L89W MscL, the solvent accessibility increased 200 for V21, which is consistent with channel pore hydration as its side chain sits within the TM domain 201 in the closed state ( Fig S7). An over ten-fold increase in solvent accessibility was observed for L72, 202 while a significant decrease was observed for K100, which sits at the C-terminal end of TM2 (   MscL's gating process. For these pore hydration events we monitored the annular lipids, which 253 make direct contact or reside in close proximity to the channel and simultaneously interrogated the 254 pore radius profile using HOLE(62) (Fig 5B, C and 6C). For WT, we first observed that following 255 constant tension application for ~200 ns, the pore becomes hydrated, and the channel undergoes  Table S2). Following bilayer stretching the total pocket surface area decreases and limits 260 access and availability to bilayer lipids throughout our simulations ( Fig 5B). Overall, the under-

270
We first tested whether L89W modification causes any distortion on TbMscL's secondary structure.

271
To this end, we calculated an overall RMSD of 1.8 Å between the L89W and WT TbMscL following  Table S2).

277
We observed that during the equilibration of our simulations, lipids intercalated into the pockets and 278 occupied them. Following tension application, lipids were pulled towards the membrane, but the 279 bulky tryptophan modification at the entrance disrupted the lipid exit path, despite tension 280 application sufficient to open the WT channel was applied (Fig 5). Interestingly, in the tension 281 activated state, F79 and L89 from adjacent subunits come very close to form two new smaller 282 pockets in the inner-and outer-leaflets of the now substantially thinner membrane (Fig 6A and B).

283
This spatial "refinement" may facilitate the next stage in MscL's activation, which follows this sub-284 conducting state, and requires additional and highly localised pressure for full pore opening of 3 nS  (Table 1). We found that the lipid interaction (Lennard-Jones) 296 7 with residue 89 is twice as strong in the modified compared to WT channel (Table 1). This suggests 297 that lipids are "stuck" by the modification in the pockets, reducing their ability to exchange with the 298 bulk bilayer during tension application (Fig 6A and B). Despite the substantial global structural 299 rearrangements, we observed in L89W MscL under tension, when lipids remain tightly associated 300 with the pockets, its pore cannot hydrate (Fig 5 and 6C).

302
Lipid order and accessibility during tension application

303
To assess the lipid contact profile with the MscL TM domain we calculated the relative difference 304 of lipid contacts between the with and no tension TbMscL states (Fig 6 and S9). We found that the 305 relative lipid contacts within bilayer facing residues on the TM domain were reduced over the course 306 of the simulation. In contrast, the lipid contacts of hydrophobic residues that were not lipid exposed 307 in the closed state increased substantially (Fig 6 and S9). This is consistent with TM2 rotation 308 occurring during MscL opening(8) and with pockets decreasing their surface contact area and 309 becoming less accessible to lipids (Fig 4B). To understand lipid rearrangement occurring upon 310 tension application we calculated the lipid order before and after bilayer tension application and 311 found that all lipids (particularly the end of the lipid chains) orient substantially more horizontally in 312 the presence of tension (Fig S10A and C). However, the pocket lipids specifically orient marginally 313 more horizontal compared to the bulk bilayer lipids, in the presence of tension ( Fig S10B). These 314 findings suggest that although tension application results in a more horizontal, in respect to the 315 membrane plane, lipid reorientation, additional horizontal tilting other than that caused by 316 membrane stretching is not required to favor lipid pocket preference or specificity (8, 10).

320
According to the lipids move first model (2)  global structural rearrangements (Fig 5 and 7). Despite this, only the WT channel pore increased 335 in diameter enough for pore hydration. The pore of the L89W channel was significantly smaller in 336 volume and diameter (Fig 2,3,5,6C and S8). Other previously reported loss-of-function or gain-of-

344
(under membrane tension) (Fig 6A, B and Table 1). Lipids are loosely associated with the pockets 345 in the "hydrated" WT MscL, following tension activation, but these low pairwise force-contacts are 346 not sufficient to close the channel (Fig 5B, 6B and Table 1). We find that the pockets become 347 substantially smaller under tension and consequently less accessible to lipids (Fig 4B, 5B and 6).

354
In the modified L89W MscL, F79 and W89 form a "molecular" bridge resulting in a two-compartment 355 repartitioning of the smaller pockets (Fig 6A and B). The WT (under tension) expanded state is 356 substantially different from the closed state x-ray structure used (PDB 2OAR) (Fig 4, S8 and Video 357 S1). Importantly, TM2 movement (translation and expansion due to anticlockwise TM2 rotation) 358 under tension like that occurring upon pocket removal activation, consistent with the archaeal x-ray 359 structure of expanded MscL (56), and the expanded (and sub-conducting) state, revealed by 360 PELDOR (2) (Fig 7A).

361
Previously, cross-sectional expansions not associated with a conducting pore have been proposed 362 for the initial stages of MscL activation(41, 68), suggesting MscL could act as a membrane tension 363 dampener and undergo several transitions with the pore closed (69). Our data support the notion 364 that there are structural transitions that occur prior to channel opening and agree with these models.

365
We observed transitions which are part of the closed-closed expansion gating process, since the 366 overall RMSD between the L89W and WT structures following tension application was small, but 367 they were both significantly different from the closed state (PDB 2OAR) (Fig 4, 7B and S8). This is 368 substantiated by the fact that the channel is unable to reach the intermediate state under applied 369 tension after the lipids are trapped within the pockets via the L89W modification, by preventing pore 370 hydration and stabilizing a closed (non-hydrated pore) state (Fig 3, 5, 7).

371
Our data suggest that for MscL to reach the intermediate hydrated state, triggered either by tension 372 or molecules, the lipids will have to substantially loosen their contacts with the pockets (Fig 5 and 373 Table 1). Lipids could move under tension and adopt a more horizontal orientation in respect to the 374 membrane plane in the WT channel, in the absence of pocket modifications that disrupted lipid 375 exchange with the bulk bilayer ( Fig S10). The addition of molecules which could compete for the 376 pockets may support or disrupt lipids acting as negative allosteric modulators within the pocket 377 sites (Fig 7C). Indeed, following similar modifications on MscL, which restrict lipid access to the 378 pockets, or when molecules that specifically bind these pockets (and compete with lipids) were     (16, 18). This is in agreement with the 395 structural changes observed by HDX and ESEEM for the modified L89W TbMscL channel (or the 396 structurally equivalent M94 EcMscL) and the tension activated WT channel (Fig 1,3,4,7 and S8).

V21R1 exposure increases in the intermediate (L89W) MscL state, and this increase is consistent
398 with a channel pore hydration and opening. N13R1's side chain is already solvent exposed in the 399 closed state, as it is pointing into the cytoplasm (Fig S7), and thus unaffected in the expanded 400 intermediate state (Fig 3).

401
Transition to the intermediate state increased solvent accessibility on the upper TM1 and TM2, the 402 cytoplasmic loops and the top portion of the cytoplasmic helical bundle (Fig 1 and 2), consistent 9 with stretching of the cytoplasmic loops, of which shortening influences MscL's gating properties, 404 pore conductance and oligomeric assembly(16, 42, 60, 73).

405
The top of TM1 (residues 36 -41) partially unfolds within a single subunit of the MscL pentamer, 406 while the remaining four subunits bend without fully converting into a loop. This finding was 407 consistently observed across all three MD simulation repeats following the symmetric application 408 of tension along the membrane xy plane and suggests the presence of a subunit asymmetry during 409 MscL opening (41, 74, 75). In contrast to the WT channel, all five TM1 helices bend without a break 410 in helical symmetry or conversion into a loop in the L89W channel (Fig 4).

418
Subsequently, following initial expansion and conversion into loops, these loops could then provide 419 membrane handles for MscL in order to transit to its full opening state (Video S1).

420
Despite a significant membrane thinning (~1.2 nm), L89W's pore does not allow water molecules 421 to flow through (Fig 4 and 5). Membrane thickness and hydrophobic mismatch energetically

445
The structural similarities of these two differently derived states suggest that lipids could act as 446 molecular triggers on specific sites and mimic the naturally occurring tension activation in MS 447 channels, with implications for MS channel evolution and multimodality.

472
Protein Purification and Spin Labelling

473
The protocols followed in this study were similar to the ones previously described (8, 19, 35). In   3-pulse sequence used for the experiments is p/2-t-p/2-T-p/2-stimulated echo with a pulse length 501 tπ/2 = 16 ns and inter-pulse delay t that was adjusted to match either the blind spots of the proton 502 or deuterium. The delay T was incremented from 400ns in 12 ns steps. A four-step phase cycling 503 was used to eliminate the unwanted echoes. All the measurements were performed at the 504 maximum of the filed sweep spectrum of the nitroxide. For solvent accessibility determination, only 505 -3pESEEM data recorded with t that corresponds to the proton blind spot were used. The obtained 506 time-domain traces were background-corrected, apodized with a hamming window and zero-filled 507 prior to Fourier transformation. The solvent accessibility can be determined by different analysis 508 methods from the deuterium 3pESEEM. In the present study, we used a model developed by data was background corrected in a such way to obtain a deconvoluted and normalized nuclear 511 modulation function. We used two approaches to determine the solvent accessibility; the first one 512 by fitting the obtained nuclear modulation function for each mutant by a damped harmonic 513 oscillation function and the outcome from the fitting was used to estimate the solvent 514 accessibility(51) , the second approach by Fourier transformation of the nuclear modulation function 515 and in this case the water accessibility was determined directly form the intensities of the deuterium 516 peaks in the magnitude ESEEM spectra. The solvent accessibility parameters derived from both 517 methods are in a good agreement within the error bars as shown in Fig S3. Although the standard 518 deviations for the fitted parameters were less than 2% for all the mutants, we additionally accounted 519 for errors that might emanate from differences in relaxation behavior or in contribution from no-520 water protons between the different mutants and use an error of 5%. All ESEEM raw data for this

569
The set up and parameters of MD simulations without tension were as previously described(8). In 570 brief, CHARMM-GUI was used to insert the TbMscL structure (2OAR) into a pre-equilibrated patch 571 of POPC bilayer containing approximately 387 lipids and occupying an area of 120 × 120 Å2. The 572 protein and membrane bilayer were solvated with TIP3P water and 150 mM NaCl. The simulations 573 were performed in an NPT ensemble at 303.15 K and 1bar pressure on all xyz axes using 574 GROMACS_2016.4 with CHARMM36 force field. The particle mesh Ewald (PME) method was 575 applied to calculate electrostatic forces, and the van der Waals interactions were smoothly switched 576 off at 10-12 Å by the force-switch manner. The time step was set to 2 fs in conjunction with the 577 LINCS algorithm. After the standard minimization and equilibration steps using the GROMACS 578 input scripts generated by CHARMM-GUI, 100 ns dynamic simulation was calculated.

896
Errors are calculated conservatively at 5% to compensate for the fitting error and differences in 897 relaxation times of different spin labelled residues. There was no significant difference in solvent 898 accessibility of N13R1, L42R1, and N70R1 compared to their 89W double mutant counterpart.