LettsScience

OK, time for part 2. If you missed part 1 of this two post series check it out here.

Just to recap, I am discussing the recent Science paper (April 13th issue) from the D.E. Shaw Research group, entitled “The Mechanism of Voltage Gating in Potassium Channels.” In this paper, Jensen et al. use their custom designed supercomputer Anton (named after Anton von Leeuwenhoek, the 18th century microscopist) to perform molecular dynamics (MD) simulations of voltage-gated K⁺ channels under the influence of different membrane potentials. As their starting point, they used the structure of the K_v1.2-2.1 paddle chimera solved by Long et al. in 2007 to 2.4 Å resolution (PDB accession code 2R9R) (Long et al., 2007). This is the structure is of the channel in the activated (open) conformation, which occurs during depolarization of the cell membrane (positive membrane potentials). Jensen et al. want to see if they can use Anton to calculate the deactivated (closed) state of the channel through MD simulations.

I work on a voltage-gated protein (H_v) and I have spent some time thinking about and studying the mechanism of how these voltage sensors operate to open and close the pore of the ion conduction pathway. So I have to say that watching the movies of these proteins in action generated from the MD simulations is pretty cool! However, just because something is cool doesn’t mean it is correct, so in this post I will try to overcome the coolness factor and be critical and really look at what we might be able to learn from these simulations.

Jensen et al.‘s paper is filled with complex diagrams tracking the movements of specific residues of interest within the voltage-sensor domain (VSD) and pore domain structure. All of these figures are generated from the 15 simulations discussed in the paper, representatives of which are shown in the supplementary movies on-line (Jensen et al., 2012 supplementary). I recommend checking out the paper and having a look at the figures since they highlight some of the most important conformational changes. Here I will mostly be using the movies themselves to discuss their results.

—

Activation – Opening the Channel

We already know the structure of the activated K_v channel (Long et al., 2007), so what Jensen et al. were trying to do here was to see whether applying a depolarizing voltage to their calculated deactivated (closed) structure, could they get the channel to activate (open) in their MD simulations.

First off this is a good point to mention that, in simulations 1-4 of the paper Jensen et al. performed some controls on the experimentally determined structure of the activated state of the Kv2.1-1.2 paddle chimera channel. Basically, after letting the channel relax in the membrane for ~2 μs at 0 mV, they ramped the voltage over 1 μs to +750 mV (simulations 1 and 2) or +375 mV (simulations 3 and 4). They did these controls to ensure that the crystal structure of the paddle chimera was indeed in the fully activated state. In other words, they wanted to examine if the channel would adopt an alternate conformation when they strongly depolarize the membrane. What they found was that the experimentally determined channel structure does not change, and they observed a steady outward conduction of K⁺ ions through the pore. From this they conclude that the crystal structure is indeed the fully activated open state of the channel. When they performed these simulations, they used either the standard CHARMM27 parameters (simulation 1) or the DER correction (simulations 2-4; I briefly discussed the DER corrections in the previous post) in both the absence (simulations 1-3) or presence (simulation 4) of the T1 domain (Jensen et al., 2012 supplementary). I would be curious to see what would happen if they used the DER2 correction in these simulations, which would further decrease the partial charge of the arginine side chain. It seems to me that to see any significant movement of gating charges in their simulations they needed to use the DER2 correction (simulations 9, 13 and 14), especially within the short duration of these control simulations (80 μs, 33 μs, 14 μs and 15 μs for simutations 1-4 respectively vs. 256 μs, 101 μs and 114 μs for simulations 9, 13 and 14).

Now, on to the activation simulations. In order to see the activation of the voltage sensors and outward movement of the gating charges, Jensen et al. did two simulations (13 and 14) using their calculated deactivated (closed) states from simulation 9 (in the presence of the T1 domain) as their starting model. In simulation 13, the starting model was taken from simulation 9 at 210 μs, i.e. the channel has all but one of the VSDs in the fully deactivated conformation (gating charges inward of Phe233). In simulation 14, the starting model was taken from simulation 9 at 254 μs, i.e. the channel has all of the S4 helices in the down conformation. The simulations were performed by linearly reversing the voltage from the simulation 9 value (-500 mV) to the depolarizing value (+375 for simulation 13 and +500 for simulation 14) over 1 μs.

In simulation 14, the voltage was reduced from +500 mV to +375 mV after 80 μs. This reduction in the driving force may have been related to the fact that Jensen et al. had to transiently impose α-helical restraints on the S4 helix for 2 μs (from 42 to 44 μs in the simulation) due to “helix deterioration” (Jensen et al., 2012 supplementary, Table S1 legend). When the goal of the simulation is to try and see the conformational changes of the protein when exposed to a depolarizing membrane potential, it makes me a little uncomfortable when the structure is constrained to an expectation of α-helicity. In addition, as I mentioned in part 1, the experimentally determined structure of the Kv1.2-2.1 paddle chimera shows that the top half of the S4 helix is a 3₁₀-helix not an α-helix (Long et al., 2007).

Movie S7 – Deactivation and activation from simulations 9 and 14 in the presence of the T1 domain

Movie S7 is a combination of simulations 9 and 14 in which both the deactivating and activating transitions can be seen explicitly (and in super slow motion). The movie shows a single VSD with helices S1-S3a colored in green and helices S3b and S4 in red. The color of the bar at the bottom of the movie indicates either hyperpolarizing (blue) or depolarizing (red) membrane potential (switches at ~254 μs). The transparent structure is an overlay of the Kv1.2-2.1 paddle chimera structure for comparison (PDB accession code 2R9R) (Long et al., 2007). The deactivating gating charge transitions occur at ~80 μs and ~244 μs. The activating transitions at ~255 μs, ~264 μs and ~330 μs return the structure nearly to that of the starting model (transparent structure).

Again, I have to say that watching these proteins in action is pretty cool. However, with these activation simulations there is one significant issue: activation of the VSD from the calculated deactivated structure did not result in opening of the pore. This is because not all of the VSDs activated and the pore stayed closed. It is well established that all four VSDs must adopt the activated conformation for the pore to open (Bezanilla, 2000). In simulation 13, one VSD remains with two gating-charge S4 arginines “down” (below Phe233). In simulation 14, two VSDs have one gating-charge carrying arginine down. For some reason Jensen et al. did not run the simulations long enough to allow full activation of the channel. Whereas the deactivation simulation 9 was allowed to run for 256 μs, with the final gating charge transitions happening at ~244 μs, the activating simulations 13 and 14 were only run for 101 μs and 114 μs respectively.

In order to see the channel open, Jensen et al. used a shortcut: they started the simulation from a structure generated in simulation 8 at 45 μs (simulation 10) or 48 μs (simulations 11 and 12), in which the pore was either partially (simulation 10) or fully closed (simulations 11 and 12). I believe there is a typo in the paper when referring to the model used for simulations 10-12. In the main text they say, “…starting from a computationally determined T1^– configuration in which all VSDs, save one [my empahsis], were ‘up’…” this conflicts directly with what they write in the supplementary information, “Activation simulations 10-12 were started from closed-pore snapshots of simulation 8 … in which all [my empahsis] VSD gating-charge residues were still ‘up’…” It is difficult to confirm which of these statements is correct since none of the figures in the paper that show gating-charge movement include any data from simulation 8. I would also be interested to know the degree of dissociation of the VSDs from the pore at the two simulation 8 time points used for these models. Without this information it is very difficult to interpret the results of these activation simulations. However, the mechanism of pore opening that Jensen et al. propose is not unexpected. In essence, it is the opposite of what they observed in their previous paper, in which they did simulations on the isolated pore domain (Jensen et al., 2010). In those pore-only simulations Jensen et al. were not able to open their calculated closed channel structure. In contrast, the presence of the activated VSDs holds the S4-S5 linker in such a conformation that the S6 gate of the pore domain opens (simulations 10-12). As was previously seen in the crystal structure, the open conformation is stabilized by specific interactions between S6 and the S4-S5 linker (Long et al., 2007). Therefore, cool as it may be to watch the pore pop open Jensen et al. do not provide new insight into this mechanism.

What I would like to see is a full activation simulation in which the dissociated VSD re-associate with the pore, activate and open the channel. Anything less than this just leaves questions about whether or not, after the dissociation of the VSDs from the pore domain seen in the deactivation simulations, it is still possible for the VSDs to open the pore.

—

Discussion

This is a very interesting paper but there are clearly some pieces still to fill in before we fully understand the mechanism of voltage gating in potassium channels. If you were to ask me what were the most interesting and important questions still to be answered concerning the voltage gating mechanism before this paper came out,, I would have said:

1. What does the closed conformation of a voltage-gated channel look like?
2. What is the role of the 3₁₀-helix in S4?
3. How are the conformational changes in the VSDs transmitted to the S6 gate?

Jensen et al. provide a model for the closed conformation of the voltage-gated potassium channel. In fact, they provide a model for several different closed states. Ever since the description of the Cole-Moore effect, it was postulated that the voltage sensor can occupy several different closed conformations (Cole & Moore, 1960). As previously proposed in the study of the charge transfer center (Tao et al., 2010), these closed states are observed in Jensen et al.‘s simulations to consist of different S4 gating charge residues occupying the Phe233 binding site. This is a significant contribution, but this model will have to be validated by experiments. In addition, some of the simulations resulted in conformations of the VSD that are not shown in any of the movies or figures or even discussed in the paper. For example, hyperpolarizing simulations 6 and 7 each have one VSD with five gating charges “down” (below or at Phe233). In the figures and movies, they only show models with at most three gating charges down. I would really like to see what these other conformations of the VSD look like. Clearly, there is a need for a better data-sharing mechanism from these types of simulations, a sort of “MD PDB“.

No mention of the 3₁₀-helix in S4 was made in this paper. This is a significant oversight and, therefore, Jensen et al. fail to provide any new insight into the possible role of the 3₁₀-helix in the mechanism of voltage gating. By not mentioning the 3₁₀-helix, they are making the implicit argument that, throughout the entire gating process, S4 is an α-helix and that the 3₁₀-helix in the experimentally determined structure has no mechanistic importance.

I am also not entirely convinced by the arguments that Jensen et al. put forward about the role of the VSD in opening the pore. To paraphrase their argument, upon all four VSDs reaching the activated conformation, the S4-S5 linker is held in a conformation which is primed to bind the S6 helix when thermal noise causes S6 to stray from its closed conformation. This then allows water to refill the pore cavity and the channel to start conducting ions. This argument, which states that the closed conformation of the pore is intrinsically more stable, may be true and is supported by experimental data. My skepticism comes simply from the fact that they were unable or unwilling to simulate the entire VSD-activation-pore-opening process in a single simulation. The lack of this complete activation simulation leaves too many questions in my mind. For example, did something non-physiological happen to the structure during the deactivation simulations that resulted in a model that, when depolarized, was unable to properly activate the channel? My skeptical mind leads me again to question the large dissociation of the VSDs from the pore.

—

The control simulations that I would like to see

In addition to the more rigorous depolarization controls mentioned above with the DER2 correction and the full activation simulation, there are a few other simulations that I think would be of general interest to the community and would give a better understanding of the accuracy of the simulations performed thus far.

A simulation in which the gating charges are immobilized – It has been shown that it is possible to immobilize the gating charges in the activated state by blocking the pore and preventing the closure of the S6 gate (Perozo et al., 1992). It would be possible to do this in two ways: using a K⁺ channel with an intact N-terminal inactivating particle or inactivating peptides present or, perhaps more simply, by adding tetraethylammonium (TEA) (Zagotta et al., 1990; Perozo et al., 1992). The N-particle and TEA have been shown to bind in the pore of the K⁺ channel and block the flow of ions (Zhou et al., 2001). Their occupation of the pore prevents the closure of the S6 gate, which, in turn, has been shown to prevent the VSDs from transitioning from the activated state to the deactivated state, thereby immobilizing the gating charges (Perozo et al., 1992). I would like to see if blocking the channel in this fashion in the simulations would prevent this VSD transition. If, however, the VSDs simply detached from the pore domain and underwent the conformational transition, we would know that there was something weird going on.

A simulation in which the VSD is covalently linked to the pore – It has been shown that the contact made between S1 on the VSD and the pore helix is a functionally significant contact (Lee et al., 2009). To better understand the interaction of these two domains, it would be interesting to perform a simulation in which the VSD is covalently linked to the pore via engineered cysteine cross-linking across the S1-pore helix interface. When this experiment was performed in the voltage-gated K⁺ channel KvAP, the channel was still functional (Lee et al., 2009). It would be really interesting to see how this cross-linking would affect deactivation and activation of the VSD and opening and closing of the pore in the simulations.

—

Conclusions

With a title like “The Mechanism of Voltage Gating in Potassium Channels” you are going to raise a few eyebrows. Sometimes, in order to get your paper published in a top journal it helps to have a grandiose title. But is the paper actually good? Do the simulations presented give us any insight into how the VSDs work? Does this paper really deserve to be published in Science?

Although this review raises a lot of questions about the methods and conclusion in the paper, my answers to these three questions are: yes, yes and yes. Don’t get me wrong, the paper doesn’t fully live up to its title, but the accomplishments presented are far from insignificant. Science- the enterprise and the journal- should promote innovation, and what Jensen et al. have been able to do with Anton in this paper goes far beyond what anybody else has thus far been able to accomplish with MD. The people at D E Shaw Research are innovators. I certainly would not have been carrying this paper around with me for the last two weeks if I didn’t think it was interesting or important.

Is this paper the final word on the mechanism of voltage gating? No. Clearly there are still problems with the MD force fields, and simulations will always have to be experimentally verified. But, of course, the ultimate goal of MD simulations is to generate hypotheses for experimentalists to check. Hypothesis-generation is a powerful contribution to the scientific endeavor. I can’t wait to see what Anton2 and Anton3 with improved force fields are able to predict.

What do you think? Is this an important paper? What questions does it raise in your mind about the mechanism of voltage gating? Can you think of any voltage sensor simulations that you would like to see Anton do? Please write in your comments and ideas.

—

Works Cited in this Post

Bezanilla, F. (2000). The voltage sensor in voltage-dependent ion channels. Physiological Reviews, 80(2), 555–592.

COLE, K. S., & MOORE, J. W. (1960). Potassium ion current in the squid giant axon: dynamic characteristic. Biophysical Journal, 1, 1–14.

Jensen, M. Ø., Borhani, D. W., Lindorff-Larsen, K., Maragakis, P., Jogini, V., Eastwood, M. P., Dror, R. O., et al. (2010). Principles of conduction and hydrophobic gating in K+ channels. Proceedings of the National Academy of Sciences of the United States of America, 107(13), 5833–5838. doi:10.1073/pnas.0911691107

Jensen, M. Ø., Jogini, V., Borhani, D. W., Leffler, A. E., Dror, R. O., & Shaw, D. E. (2012). Mechanism of voltage gating in potassium channels. Science, 336(6078), 229–233. doi:10.1126/science.1216533

Lee, S.-Y., Banerjee, A., & Mackinnon, R. (2009). Two separate interfaces between the voltage sensor and pore are required for the function of voltage-dependent K(+) channels. PLoS biology, 7(3), e47. doi:10.1371/journal.pbio.1000047

Long, S. B., Tao, X., Campbell, E. B., & Mackinnon, R. (2007). Atomic structure of a voltage-dependent K+ channel in a lipid membrane-like environment. Nature, 450(7168), 376–382. doi:10.1038/nature06265

Perozo, E., Papazian, D. M., Stefani, E., & Bezanilla, F. (1992). Gating currents in Shaker K+ channels. Implications for activation and inactivation models. Biophysical Journal, 62(1), 160–8; discussion 169–71. doi:10.1016/S0006-3495(92)81802-7

Tao, X., Lee, A., Limapichat, W., Dougherty, D. A., & Mackinnon, R. (2010). A gating charge transfer center in voltage sensors. Science, 328(5974), 67–73. doi:10.1126/science.1185954

Zagotta, W. N., Hoshi, T., & Aldrich, R. W. (1990). Restoration of inactivation in mutants of Shaker potassium channels by a peptide derived from ShB. Science, 250(4980), 568–571.

Zhou, M., Morais-Cabral, J. H., Mann, S., & Mackinnon, R. (2001). Potassium channel receptor site for the inactivation gate and quaternary amine inhibitors. Nature, 411(6838), 657–661. doi:10.1038/35079500