In contrast to most voltage-gated ion channels, hyperpolarization- and cAMP gated (HCN) ion channels open on hyperpolarization. Structure-function studies show that the voltage-sensor of HCN channels are unique but the mechanisms that determine gating polarity remain poorly understood. All-atom molecular dynamics simulations (~20 μs) of HCN1 channel under hyperpolarization reveals an initial downward movement of the S4 voltage-sensor but following the transfer of last gating charge, the S4 breaks into two sub-helices with the lower sub-helix becoming parallel to the membrane. Functional studies on bipolar channels show that the gating polarity strongly correlates with helical turn propensity of the substituents at the breakpoint. Remarkably, in a proto-HCN background, the replacement of breakpoint serine with a bulky hydrophobic amino acid is sufficient to completely flip the gating polarity from inward to outward-rectifying. Our studies reveal an unexpected mechanism of inward rectification involving a linker sub-helix emerging from HCN S4 during hyperpolarization.
The hyperpolarization-activated and cyclic nucleotide activated ion channels are found in the pacemaking cells of heart and brain where they play a singular role in regulating rhythmic electrical oscillations (Brown et al., 1979; Ludwig et al., 1998; Santoro et al., 1998). Unlike other members of the voltage-gated ion channel family, these channels open at membrane potentials below the threshold of action potential and depolarize the membrane by increasing the permeability to Na+ ions (Brown et al., 1979). Their slow gating kinetics and low permeability determine the frequency of action potential spikes (Larson et al., 2013; Sharpe et al., 2017). In addition to voltage, the gating properties of these channels are regulated by PIP2 and second messengers such as cAMP which mediates the ‘fight or flight’ response by increasing the activity of HCN channels in sinoatrial nodal cells of the heart (Brown et al., 1979; Pian et al., 2006).
From a structural standpoint, the HCN channels are remarkably similar to other voltage-gated ion channels. They are tetrameric and each subunit consists of a voltage-sensing domain (VSD) (S1-S4 helices) and pore-forming domain (S5-S6 helices) (Lee and MacKinnon, 2017). Like other members of CNBD family, the VSD and pore domain (PD) are arranged in a non-domain swapped arrangement (Lee and MacKinnon, 2017; Whicher and MacKinnon, 2016). The C-linker region connects the cyclic nucleotide binding domain (CNBD) in the carboxy terminal end to the S6 transmembrane segment. The resting state structure of the human HCN1 channel (hereafter referred to as HCN1) reveals several distinctive features. First, the S4 segment of HCN1 channel is at least two helical turns longer than the corresponding S4 in depolarization activated ion channels. Second, at the cytosolic amino terminal end, a 45 residue stretch forms a small 3-helix bundle which interacts with the voltage-sensor and CNBD domain. This domain is unique to this channel family and is called the HCN domain. Finally, the S4 helix of the HCN1 channel is in tight juxtaposition with the S5 helix, unlike in the related depolarization activated ether-á-go-go (EAG) channel (Lee and MacKinnon, 2017).
Recent studies show that the voltage-sensing domains of the HCN channels have an intrinsic ability to drive the channel in both hyperpolarizing and depolarizing directions (Cowgill et al., 2019). This ability to open the channel on hyperpolarization neither depends on the extra length of the S4 segment nor on the HCN domain suggesting that there are other non-obvious determinants of gating polarity in the voltage-sensing S4 of HCN channels. Cysteine accessibility studies to probe the nature of these conformational changes show that the S4 undergoes a large downward motion much like the voltage-sensor of the Shaker potassium channel (Bell et al., 2004; Männikkö et al., 2002; Vemana et al., 2004). One notable difference, however, is that the residues in the top part of the S4 helix showed larger displacements compared to those at the lower end. Recently, transition metal FRET studies also propose that the S4 undergoes a downward motion with a slight bend at the tail end of the S4 helix (Dai et al., 2019). Nevertheless, it remains unclear whether these subtle differences in the voltage-sensor motions between outward (depolarization-activated) and inward rectifying (hyperpolarization-activated) channels can account for the distinct gating phenotypes.
In order to probe the structural transitions through the gating cycle, we took advantage of the special purpose Anton 2 supercomputer (Shaw et al., 2014) to carry out two independent runs of tens of microseconds long molecular dynamics (MD) simulations. Our simulations showed that the downward movement of the voltage-sensors is accompanied by breaking of the S4 helix into two sub-helices and the lower sub-helix adopting an orientation parallel to the membrane plane and turning into a surrogate S4-S5 linker. To probe the importance of this helix breaking transition, we utilized bipolar chimeras which exhibit both hyperpolarization and depolarization-activated currents (Cowgill et al., 2019). Many of the mutations at this region in the wild type background result in a non-functional phenotype. The bipolar background enables us to robustly quantify the relative contributions of various molecular manipulations on hyperpolarizing vis-à-vis depolarizing currents. Our studies reveal that both the location of the breakpoint and nature of amino acid therein is critical for gating polarity in the channels of this superfamily.
HCN channels are notoriously slow-activating such that even the fastest family members require tens of milliseconds to open. Because this timescale is not accessible in even the most advanced simulations, we employed two strategies to accelerate channel kinetics. First, we applied strongly hyperpolarizing conditions (−550 mV) similar to the approach used for simulations of Kv1.2/2.1 deactivation (Jensen et al., 2012). Next, we sought to facilitate the movement of gating charges by further hydrating water crevices inside the voltage-sensing domain (VSD). Inspired by the work of Lacroix et al. (2013), we designed a mutant that was anticipated to increase hydrophilicity in the VSD. By comparing the sequences of HCN and EAG homologs, which activate and gate much faster, we identified two positions in S1 with large hydrophobic residues in HCN channels compared to smaller, less hydrophobic residues in the EAG family (Figure 1—figure supplement 1A,C). Based on this information, we ran preliminary simulations (~3 µs) on the wild-type and mutant (M153T/I160V) HCN1 under hyperpolarizing electric field. In agreement with our prediction, the mutant showed signs of faster activation than the wild type (WT) channel (Figure 1—figure supplement 1B), therefore we proceeded to run Anton simulations with the mutant channel. Functional testing shows that these mutations are well tolerated (Figure 1—figure supplement 1D).
Substantial conformational change was observed in 6 out of 8 voltage sensors from two independent ~20 µs simulations of the full length channel (Figure 1A,B). A large downward movement of the S4 helix drives two positively charged residues across the focused electric field at the charge transfer center (F186) (Figure 1A,C). This large vertical movement agrees with previous predictions based on cysteine accessibility (Bell et al., 2004; Vemana et al., 2004) and voltage clamp fluorometry (Dai et al., 2019). However, the simulations reveal an unexpected break in the S4 helix, causing the C-terminal half to adopt an orientation parallel to the membrane plane (Figure 1A–E, Video 1) when the last gating charge is transferred. Overlay of 6 independent voltage sensors confirms that they converge to the same activated (Down) structure (Figure 1B, Figure 1—figure supplement 2). This helix breaking model is in stark contrast to what has previously been observed in simulations of voltage-gated potassium channels and new resting state structures of voltage-gated sodium channels (Delemotte et al., 2011; Amaral et al., 2012; Jensen et al., 2012; Vargas et al., 2012; She et al., 2018; Xu et al., 2019; Wisedchaisri et al., 2019). In these channels, the consensus view is that S4 moves vertically and twists as a unit in what is typically referred to as the helical screw model (Figure 1E). Interestingly, the structure of the cyclic nucleotide-gated (CNG) channel TAX4 (Li et al., 2017), a close relative to HCN channels (Baker et al., 2015), shows a bent S4 helix reminiscent of what we observe for HCN1 (Figure 1D,E). Thus, the S4 segment in these two closely related clades may have an intrinsic propensity to bend.
The movement of S4 in response to voltage results in the appearance of gating currents, which has been measured experimentally for the sea urchin HCN (spHCN) channel but not for other HCN channels presumably due to the slow nature of their gating (Bruening-Wright et al., 2007; Ryu and Yellen, 2012). The contribution of each residue to the net gating charge can be computed from our simulations (Figure 1F,G) (Roux, 2008; Treptow et al., 2009). In this approach, charge contribution of each residue is calculated based on the derivative of the local electrostatic potential with respect to applied voltage in the resting and activated states (the so-called coupling function, Figure 1G). These contributions will only differ between states if either the charged residues move relative to the local electric field or the local electric field rearranges around these residues. Summing the contribution of each residue shows the net gating charge per VSD is 1.85 eo (or 7.4 eo per channel), which is in good agreement with the eight charges per channel suggested previously through kinetic modeling of HCN channels activation (Hummert et al., 2018).
As expected, the per-residue contribution to this charge difference is dominated by the two positively charged residues on S4 (R267 and K270) which move relative to the charge transfer center, with lower contributions from the other charges of S4 (Figure 1F). Additionally, the negative countercharges on S1-S3 contribute to the net gating charge as was experimentally shown for the Shaker potassium channel (Seoh et al., 1996). Interestingly, the acidic residues of S1-S3 contribute to the gating charge despite little to no net displacement (Figure 1—figure supplement 2). This implies a redistribution of the local electric field inside the VSD upon activation, as previously suggested based on HCN1 cysteine accessibility studies (Bell et al., 2004). The change in the local electric field can be seen as a change in the coupling function between the activated and resting state (Figure 1G). The local electric field within the VSD is thought to be controlled primarily by hydration of the water accessible crevices (Islas and Sigworth, 2001); as expected, the hydration profile of the VSD is altered upon activation (Figure 1H). It is possible that the increase in hydration of the activated state comes from the mutations we introduced to accelerate gating kinetics. Alternatively, inward movement of the S4 positive charges may cause an increase in hydration of the countercharges on S1-S3 (Seoh et al., 1996). On the internal side, the increased hydration is likely related to the broadening of the internal crevice due to S4 bending. This increase in hydration of the VSD can be tested experimentally using substituted cysteine accessibility.
In substituted cysteine accessibility, cysteines are introduced in specific locations of the channel and their solvent accessibility can be monitored using thiol reactive methanethiosulfonate MTS compounds in a state-dependent manner. From the simulations, we identified positions within the VSD which undergo a predicted change in accessibility to a sphere of 2.9 Å radius (the approximate size of MTSET (2-(trimethylammounium)ethyl methanethiosulfonate). As expected, many of these positions lie on the S4 helix and have been probed previously (Bell et al., 2004; Vemana et al., 2004). The experimentally-attained accessibility of most of these residues agrees well with the accessibility pattern predicted from our model (Figure 2—figure supplement 1A). A key feature of this pattern is that there is a much larger change in accessibility for the internally accessible residues than external. This indicates that there is a larger conformational change in the internal crevice and lends support to the S4 bending movement observed in our simulations. However, we sought more direct experimental evidence to rule out a canonical vertical S4 movement.
In addition to the state-dependent accessibilities in S4, the simulation also predicts an increase in the accessibility of S1-S3 upon activation. These arise due to broadening of the internally facing crevice and would not be expected if S4 moved vertically as a unit (Figure 2A,B, see also Figure 2—figure supplement 1B,C). Surprisingly, in the activated state models extracted from the simulations, even the charge-transfer center (F186) becomes accessible in 3 out of the six activated states (Figure 2C). As the charge-transfer center is often considered the boundary between the internal and external crevices, it is an ideal candidate to test for the crevice broadening predicted in our broken helix model. Due to the high expression levels required for assessing internal accessibility with inside-out patches, we used spHCN rather than HCN1 (Männikkö et al., 2002). The WT spHCN channel shows no detectable functional change following MTSET application to either the closed (depolarized) or open (hyperpolarized) states as has been shown previously (Figure 2D and Männikkö et al., 2002). In the charge-transfer center mutant (F186C), internal application of MTSET at depolarized potentials leads to very slowly evolving increase in current amplitude (τmodification=7.5±0.5 M−1s−1). This confirms that the charge-transfer center has low solvent accessibility in the resting state. In contrast, application of MTSET during hyperpolarization results in a rapid increase in peak current amplitude that quickly reaches a steady state (τmodification=440±30 M−1s−1). This more than an order of magnitude increase in accessibility of the charge-transfer center upon channel activation is difficult to reconcile with the helical screw model and supports the helix-breaking model predicted by the simulations.
The unique voltage sensor movement we observe in MD simulations naturally raises the question about its role in determining the inverted gating phenotype of HCN channels. To explore this, we took advantage of two previously identified chimeras that only differ in the S3b-S4 voltage-sensing segment, yet show opposite gating polarities (Figure 3A and Cowgill et al., 2019). The HEHEH chimera containing the S3b-S4 voltage-sensing segment from HCN1 activates only upon hyperpolarization while the HEEEH containing the S3b-S4 voltage-sensing segment from the depolarization-activated EAG channel activates only on depolarization. If the helix bending movement underlies inverted gating polarity, we would predict that it is only the lower portion of S4 (lower paddle) that is required for hyperpolarization gating in these chimeras. Indeed, the chimera with only the lower paddle of HCN1 activates primarily upon hyperpolarization. At depolarized potentials, it conducts a leak current slightly larger than the HEHEH parent (Figure 3—figure supplement 1A–C). On the other hand, the upper paddle chimera activates only upon depolarization and its behavior closely resembles that of the parent chimera with the full EAG voltage sensor. Together, these results suggest that the source of inverted gating polarity can be localized to the lower portion of S4, precisely where we observe the helix break in our simulations.
A closer examination of the models from the simulations reveals that the S4 helix breaks in a highly consistent location centered at L271 (Figure 3B,C). Comparison of the consensus sequences for HCN and EAG channels show that this position is predominantly conserved between families. In fact, the overall sequence in this region is very similar between HCN and EAG with the exception of position 272. The consensus for HCN channels at this position is polar (such as serine for HCN1) whereas the consensus for EAG is hydrophobic. Furthermore, this residue is situated directly adjacent to the helix break, suggesting it may serve a vital role in channel gating. To test the role of S272 in channel gating, we introduced a series of substitutions (Figure 3—figure supplement 1D) in WT HCN1 channel. The lack of functional expression of mutants with hydrophobic residues at position 272 indicates that serine is critical for HCN1 channel gating but we must rule out the possibility that these channels are simply not trafficked to the membrane. As the constructs are C-terminally tagged with mCherry, we can use confocal microscopy to track surface expression. For WT HCN1, we observe mCherry surface fluorescence from oocytes displaying robust inward currents typical of HCN channels (Figure 3—figure supplement 2A–C). For the hydrophobic mutations, we observe surface fluorescence well above the background of uninjected oocytes, yet we still do not observe currents in these oocytes (Figure 3—figure supplement 2A–C). This confirms that the lack of currents detected in these mutants is due to non-functional channels and does not reflect a trafficking defect.
Functional experiments in the WT background are difficult to interpret because of confounding issues due to other structural elements in the channel. For instance, we have found that WT HCN channels rapidly inactivate upon depolarization due to the presence of tight interactions at the S4-S5 interface (Cowgill et al., 2019). To disentangle these effects and inspired by paddle chimera studies (Alabi et al., 2007), we utilized bipolar constructs such as HHHEH which can activate both upon depolarization or hyperpolarization (Figure 3D). We reasoned that, unlike wild type HCN channels, these channels have the gating machinery to activate in both directions and, therefore, allow us to test the relative contributions of these substitutions on outward and inward rectification. Strikingly, the S272L mutant is a predominantly outwardly rectifying channel in contrast to the parent HHHEH. We then attempted to rescue the inverted gating polarity phenotype by introducing serines throughout the surrounding region in the S272L background (Figure 3D and Figure 3—figure supplement 1E). Interestingly, only the L271S mutation was capable of restoring a primarily hyperpolarization-activated phenotype like the HHHEH parent (Figure 3D and Figure 3—figure supplement 1E)). This suggests that the location of the serine in the S4 helix is critical for the inverted channel gating phenotype.
Although we were unable to observe the effect of the S272L mutation on HCN1 experimentally, we can assess its influence on channel gating in silico using free energy perturbation (FEP) (Rodinger and Pomès, 2005; Zhang et al., 2019). The free energy difference between the activation of HCN1 and its S272L mutant ∆∆G=∆Gr→a(mut)−∆Gr→a(HCN1) were calculated using FEP (Zwanzig, 1954) (Figure 4). Considering VSD-only simulations, ∆∆G was estimated via alchemical paths of a thermodynamics cycle as: ∆∆G=∆Ga(HCN1→mut)−∆Gr(HCN1→mut). In each path S272 was reversibly transformed into a leucine through linear interpolation of the Hamiltonians of the HCN1 and mutant systems, and the free energies of these transformations were then computed using the Bennet Acceptance Ratio method (BAR) (Bennett, 1976). The estimated free energy difference of 0.7 ± 0.4 kcal/mol equates to ~2.8 kcal/mol per channel suggesting that the S272L mutation shifts the equilibrium toward the resting state, thus disfavoring the bending motion. The observed difference in free energy is significant because the experimentally determined coupling energy between voltage sensor and pore for spHCN is only 1.3 kT (~0.8 kcal/mol) (Ryu and Yellen, 2012)
We introduced a series of substitutions at S272 for different bipolar backgrounds to better understand the physicochemical underpinnings of gating polarity (Figure 5A and Figure 5—figure supplement 1A–C). Both constructs tested showed a clear trend where hydrophobic substitutions favor depolarization activation while hydrophilic substitutions favor hyperpolarization activation. To quantify the relative strengths of the hyperpolarization and depolarization activation pathways, we used the ratio of the open probability at negative compared to positive membrane potentials (PO-150 mV/ PO50 mV). This parameter, which we refer to as the gating polarity index, is inversely correlated with residue hydrophobicity with a tight fit (Figure 5B). Previous work on model helices showed that the helical propensity is directly correlated to hydrophobicity in membrane environments (Li and Deber, 1994). Therefore, the negative trend we observed between gating polarity index and hydrophobicity may be caused by the helix-stabilizing nature of hydrophobic residues in transmembrane segments. Indeed, the strong, negative correlation is upheld when comparing the gating polarity index to an experimentally-derived scale for transmembrane helical propensity (Liu and Deber, 1998). Finally, the gating polarity index shows positive correlation to the independently-derived turn propensity (Monné et al., 1999), which has previously been used as a metric for helix breaking proclivity. Together, these correlations offer strong support for our helix breaking model and reveal that residue polarity at S272 is central to the gating mechanism of HCN channels.
Given the strong effect of the S272 mutants on gating phenotype, we wondered whether introduction of a serine at the equivalent position in EAG channels would be sufficient to confer hyperpolarization activation. The G362S mutant of hEAG1 shows depolarization-dependent opening with no signs of hyperpolarization activation (Figure 5—figure supplement 1D). This is not entirely surprising as both HCN and EAG channels have presumably evolved multiple structural elements to favor opening on either hyperpolarization or depolarization. Therefore, we hypothesized that a chimera with minimal components derived from HCN1 (EEHEH) would be especially sensitive to mutations of S272. We refer to this channel as a ‘proto-HCN’ chimera because it shares some properties of both HCN and EAG channels, which have evolved from a common ancestor. Remarkably, the leucine, valine, and isoleucine mutations in this background eliminate all detectable hyperpolarization activation in the EEHEH background, transforming an HCN-like parent into an EAG-like channel with a single point mutation (Figure 5C). Thus, divergence of hyperpolarization-activated channels from depolarization-activated channels could have occurred through a single point mutation in the S4 segment.
The activation mechanism of HCN channels (Bell et al., 2004; Männikkö et al., 2002; Vemana et al., 2004) involves a movement of S4 analogous to that observed in depolarization-activated channels (Larsson et al., 1996; Ahern and Horn, 2005), but the mechanisms which ultimately lead to inward rectification remain poorly understood (James and Zagotta, 2018). Here, we have resolved the activation pathway of the HCN1 voltage-sensing domain at the atomistic level using multi-microsecond long molecular dynamics simulations starting with the full-length HCN1 resting state structure. We observe that in response to a hyperpolarizing electric field, the S4 positive charges slide down by approximately two helical turns. Coupled to this motion is a ~ 80° bend in S4, which leads the bottom half of this helix to orient itself almost parallel to the membrane plane. Our down state model of the S4 voltage-sensor is consistent with existing S4 cysteine accessibility (Bell et al., 2004; Vemana et al., 2004) and transition metal FRET data (Dai et al., 2019). However, the predicted state-dependent accessibility of the residues in the gating scaffold is incompatible with the classical helical screw motion observed in voltage-gated potassium and sodium channels (Figure 2—figure supplement 1C), see also Pathak et al. (2007). Our studies show that the charge transfer center of HCN channels exhibit a distinct pattern of accessibility consistent with the molecular simulations. Furthermore, we find that this helix breaking transition is necessary for inward rectification and our studies show that the gating polarity strongly correlates with the energetics of transmembrane helix formation. Strikingly, in a proto-HCN background, the nature of the residue in the S4 breakpoint is the central determinant of gating polarity and the direction of rectification can be switched by a single amino acid substitution.
The mechanism of voltage-sensor activation has been a focus of many studies (Larsson et al., 1996, Chanda et al., 2005; Posson et al., 2005; Long et al., 2005; Swartz, 2008; Delemotte et al., 2011; Amaral et al., 2012; Henrion et al., 2012; Jensen et al., 2012; Kintzer et al., 2018; Li et al., 2015; Bezanilla, 2018; Wisedchaisri et al., 2019) over the past two decades, and the emerging consensus in the field is that VSD activation involves a helical screw motion across the membrane relative to the surrounding S1-S3 bundle which acts as a gating scaffold. In all these channels, the measured gating charge originates from the physical displacement of S4 in a relatively static electrostatic environment which is also conserved across different voltage sensor domains (Souza et al., 2014). During the gating process, step-wise rearrangement of salt bridges between S4 basic residues and acidic countercharges on S1-S3 (Tao et al., 2010; Pless et al., 2011) facilitates charge transfer from one side to the other. The extent of the S4 motion and the number of rearrangements (termed ‘clicks’) it undergoes depends, however, on the channel and the magnitude of the electrical force exerted on the S4 positive residues. In part due to lack of structural information, most of the mechanistic studies were focused on outward rectifying ion channels and very little was known about the mechanism of inward rectification. Our simulations based on the high-resolution structure paints an entirely different picture and seems to indicate that at least another type of activation mechanism exists, in which the downward motion of S4 is accompanied by a break in the center of the helix and a change in the electrostatic environment. Whether this activation mechanism is a hallmark found only in HCN channels or is also present in related families remains to be elucidated. Of particular interest now is to understand the activation mechanism in non-domain swapped ion channels activated by depolarization such as hERG and EAG (Trudeau et al., 1995) and examine the role of cytosolic domains in modulating voltage-dependence of activation (Xia et al., 2002; Miranda et al., 2018)
Until recently, all voltage-gated ion channels were also thought to be domain-swapped, with the voltage-sensing domain of one subunit sitting adjacent to the pore domain of its neighboring subunit. Allosteric coupling between the VSD and the pore domain were thought to involve direct contacts between the S4-S5 linker with the C-terminal end of S6 located at the intracellular gating interface (Lu et al., 2002; Soler-Llavina et al., 2006; Long et al., 2005; Muroi et al., 2010; Haddad and Blunck, 2011). The other gating interface involves contacts between the transmembrane residues in the S4 and those in the neighboring S5 segment (Hou et al., 2017; Fernández-Mariño et al., 2018; Carvalho-de-Souza and Bezanilla, 2019). Recently, several Kv channels including HCN1, EAG and hERG were shown to adopt a non-swapped architecture wherein the VSD of a given subunit is placed next to the pore domain of the same subunit, in an assembly that obviates the need for a S4-S5 linker helix (Lee and MacKinnon, 2017; Wang and MacKinnon, 2017; Whicher and MacKinnon, 2016). Our results provide clues towards elucidating the coupling mechanism between the VSD and the pore domain in HCN channels. While we did not observe pore opening in our simulations, we noticed a consistent motion of S5 in all channel subunits (Figure 6A,B). The S5 N-terminus was displaced away from its original position (Figure 6A), and towards one closer to the putative open conformation of TAX4 and hERG channels (Figure 6B). Such hinge-like conformational switches are often observed in simulations of channel pores and are thought to be a common mechanism of channel gating (Forrest and Sansom, 2000). An analysis of contacts maintained throughout the simulations showed that the VSD and S5 communicate via two interaction networks (Figure 6C). When the lower S4 sub-helix torques towards the membrane, the S4/S5 interactions at the lower gating interface causes the lower part of the S5 to tilt while the top of the S5 remains anchored in place due to interactions with the relatively static S1 segment at the upper gating interface (Figure 6D). The conservation of contacts explains how spHCN functions as a split channel in which the covalent link between S4 and S5 is severed (Flynn and Zagotta, 2018). We speculate that this splaying of S5 from the central pore in an iris-like motion provides S6 with the space needed for pore opening.
While this manuscript was in press, the structure of the HCN channel with voltage-sensor trapped in the down state (Lee and MacKinnon, 2019) appeared online. The down state structure shows that the S4 helix breaks and the bottom half of the S4 forms an interfacial helix. Their structure is consistent with our findings described here and provides compelling evidence that the HCN S4 adopts a unique conformation in the down state.
The structure of the HCN1 apo state (pdb code 5U6O) was used as a starting structure for both Anton runs. Missing residues were built using Modeller (Sali and Blundell, 1993). Additionally, the structure was refined before the second run due to instabilities of several protein domains observed during the first run (see the structure refinement section below). The system for molecular dynamics simulations was prepared using the CHARMM-GUI server (Jo et al., 2008). Briefly, the structure of the HCN1 apo state was embedded in a 1-palmytoyl-2-oleoyl-phosphatidylcholine (POPC) bilayer and initially solvated in a 150 mM KCl solution. The ions were later removed during the first Anton run, and before the second Anton run, after we noticed that they promoted unpacking of the voltage sensor domains. The CHARMM36 force field (Mackerell et al., 2004) was used for the protein, lipids and ions, and the TIP3P model for water. The M153T/I160V mutant was built using the mutator plugin of VMD (Humphrey et al., 1996).
The initial steps of the WT and M153T/I160V mutant equilibration were performed on the PDC supercomputer Beskow using Gromacs 2018.1 (Abraham et al., 2015). During these steps (~100 ns in total) the restraints applied to the protein were gradually released first from the sidechains and then from the backbone. The simulations were performed in the NPT ensemble; Nose-Hoover thermostat (Nosé, 1984) and Parrinello-Rahman barostat (Parrinello and Rahman, 1981) were used to keep the temperature and the pressure constant at 300 K and 1 bar; the timestep was set to 2 fs.
The ~3 μs runs of the WT and the M153T/I160V mutant systems under an electric field were performed on Beskow using Gromacs 2018.1 (Abraham et al., 2015). The transmembrane potential was set to 750 mV using the external electric field method. Other simulations parameters were kept identical.
The M153T/I160V mutant system was then transferred to Anton (Shaw et al., 2014) and further equilibrated for ~1 μs. Additional restraints were applied to the hydrogen bonds between the backbone groups of the S4 C-terminus prior to the second long run. These restraints were required to stabilize the secondary structure of the S4 C-terminus that was otherwise partially lost in one of the subunits during the first run. They were completely released after the 600 ns simulation. The simulations were performed in conditions of constant temperature (300 K, Nose-Hoover thermostat [Nosé, 1984]) and pressure (1 bar, MTK barostat [Zhang et al., 2017]). The multigrator approach was used to coherently update the thermostat every 24th step, barostat every 480th step and Newtonian particle motion (Lippert et al., 2013). The timestep was kept to 2 fs.
After the equilibration, we gradually increased the transmembrane potential to −750 mV, and to −550 mV in the first and second runs, respectively. In the first run, we later decreased it to −550 mV after ~1.5 µs. The overall length of the trajectory under an electric field was ~18.8 and 22.8 μs for the first and second runs, respectively. The simulations under transmembrane potential were performed in the NVT ensemble. Other parameters were duplicated from the equilibration.
During the first Anton run we observed that the HCN domain, the selectivity filter and the A’ helix of the C-linker deviated from their initial positions significantly. This instability may result from a relatively low local resolution structure (of the HCN domain specifically), absence of stabilizing factors (cAMP or phosphatidylinositol 4,5-bisphosphate PIP2 lipids, for instance), dynamic properties of these domains, or other factors. While we cannot rule out all of these possibilities, we were able to test whether alternative modeling of the aforementioned domains in the regions with ambiguous electronic density improved their stability in molecular dynamics simulations during the second Anton run. To do so, we refined the structure using Phenix real-space refinement (Adams et al., 2010) and the cryo-EM map (EMD-8512); 4-fold symmetry was imposed using non-crystallographic symmetry restraints. Overall five macro-cycles were performed with global minimization, local rotamer fitting and simulated annealing. The resulting model was then locally refined using Coot (Emsley and Cowtan, 2004). Briefly, several conformations for the domains regions where the electronic density is ambiguous were generated and refined using real space refinement; torsion, planar peptide and trans peptide restraints were applied, and the refinement weight was set to 40. The model with the best density fit and the smallest number of Ramachandran outliers was further used for the final refinement in Phenix. Simulated annealing was not considered and the remaining settings were identical to that in the first Phenix refinement. The resulting structure was used as a starting point for the second Anton run and is deposited as a part of Supplementary Material.
During the second Anton run the HCN domain and the A’ helix of the C-linker were still unstable indicating that their instability potentially finds its origin elsewhere rather than in the modeling of the 3-d structure into the cryo-EM density. These domains partially lost their secondary structure shortly after the restraints on the backbone have been released, similar to what happened in the first Anton run. The selectivity filter, on the other hand, remained stable for a much longer time in the second Anton run (~10 μs compared to only few ns in the first Anton run). After ~10 μs, however, Y361 rotated toward the channel axis and blocked the conductive pore. Similar collapse of the selectivity filter has been observed in other potassium channels at low ionic concentration or in the absence of ionic current (Cuello et al., 2010; Zhou et al., 2001) and has previously been proposed as a mechanism of voltage-dependent gating in K2P channels (Schewe et al., 2016). We hypothesize that while the additional refinement did increase the stability of the selectivity filter other factors such as low ionic concentration or the absence of ionic current resulted in its collapse.
The initial hidden Markov models of the HCN and EAG transmembrane regions were built based on HCN and EAG sequences extracted from the Swiss-Prot database. They were then submitted to HMMER (Eddy, 1995) and iteratively updated with newly identified homologous sequences. Overall 176 and 266 sequences homologous to HCN and EAG channels were found. The resulting multiple sequence alignments were finally used to build sequence logos (Crooks et al., 2004).
The per-residue and total gating charge was computed using the 'coupling function' method (Roux, 2008; Treptow et al., 2009). In this method, the total gating charge is decomposed into individual contributions from residues as , and each contribution is further expressed in terms of the so-called coupling function :
and in Equation (1) represent dimensionless coupling of the charge to the transmembrane potential in the activated (a) and resting (r) states, and the summation runs over all charges of the th residue. The coupling function is then approximated as the rate of change of the local electrostatic potential with respect to :
where and are two different transmembrane potentials.
In practice, from each Anton run under an electric field we extracted ten conformations: five for the resting state and five for the activated one. For every conformation nine two ns molecular dynamics simulations under different transmembrane potentials were performed. The local electrostatic potential was then computed for each obtained trajectory using the PME Electrostatics plugin of VMD (Humphrey et al., 1996). The resulting values were further combined to obtain the coupling function for each extracted conformation. Finally, the coupling functions were used to compute the per-residue and total gating charge through Equation (1).
Per-residue helicity was estimated based on the and torsions:
where and are corresponding torsions in a perfect α-helix.
Solvent accessible surface area (SASA) was calculated using Gromacs 2018.1 SASA (Abraham et al., 2015). Radius of the solvent probe was set to 2.9 Å according to the size of MTSET. Renderings presented in all Figures except 2 and S5 were prepared using Visual Molecular Dynamics VMD (Humphrey et al., 1996).
Free energy perturbation (Zwanzig, 1954) was performed on the voltage sensor domain only (residues 140-290) on Beskow using Gromacs 2018.1 (Abraham et al., 2015). Representative conformations of the resting and activated states were extracted from the trajectories under an electric field, embedded into the system with a POPC bilayer and water, and equilibrated for 100 ns. For the resting state, 3 conformations were extracted from the equilibration trajectories while for the activated state we considered 5 conformations spanning its structural diversity. For every conformation the S272L transformation was performed through linear interpolation of the Hamiltonians of the systems with HCN1 voltage sensor and its mutant; was used as a coupling parameter such as when the Hamiltonian corresponded to the HCN1 system, while when it was of the S272L system. Overall, 21 consecutive windows were considered; in each, the system was first equilibrated for 4 ns and then the production run was performed during 8-20 ns. The free energy difference and the corresponding error were calculated using the Bennet acceptance ratio method (BAR) (Bennett, 1976) (Supplementary file 1). The free energy difference between the activation of HCN1 and its S272L mutant and the corresponding error are calculated as the average and the standard error based on the 3 to 5 independent estimations.
The chimeric constructs were generated between mHCN1 and hEAG1 in pUNIV vector as described previously (Cowgill et al., 2019). Note that although the mouse isoform is used here, we maintain the human HCN1 numbering throughout the figures and text for consistency with the structure and simulations. Additionally, the charge transfer center mutation (F186C) refers to hHCN1 numbering for clarity, which is equivalent to position F260 in spHCN. Site directed mutation was done by Quick Change mutagenesis reaction using a standard protocol. The coding sequences of all constructs were verified using Sanger sequencing. Constructs were linearized with SbfI overnight (15 hr) and transcribed using the T7 mMessage Kit according to the protocol. For experiments requiring inside-out patches, the T7 Ultra kit was used for transcription. RNA was isolated by lithium chloride precipitation and washed with 70% ethanol according to the manufacturer’s protocol then resuspended to 1–2 mg/ml using RNase free water.
Oocytes were surgically removed from adult Xenopus laevis under anesthesia in accordance with the protocol approved by the Animal Care and Use Committee of Wisconsin- Madison at University of Wisconsin-Madison. Stage V-VI oocytes were prepared by treating with 1 mg/ml collagenase A (Roche) in a calcium- free ND96 solution for 45 mins to 1 hr until the removal of follicular membrane. The oocytes were stored in ND96 solution with calcium containing BSA (ND96 solution: 96 mM NaCl; 2.5 mM KCl; 1 mM MgCl2; 5 mM HEPES; 1.8 mM CaCl2; pH 7.40). Oocytes were maintained in ND96 solution before injection and then transferred to ND96 containing antibiotics (50 μg/ml gentamicin and ciprofloxacin, 100 μg/ml tetracycline, penicillin, and streptomycin) and BSA (0.5 mg/ml) after injections called the post injection solution. Oocytes were injected with 20–90 ng of mRNA using Nanoject II (Drummond Scientific). Injections were done in ND 96 without calcium chloride solution. After injection oocytes were kept in the post-injection solution as mentioned above. Two-electrode voltage-clamp (TEVC) recordings were obtained at room temperature with an OC-725C amplifier (Warner) at a sampling rate of 10 kHz. Thin-walled glass pipettes (World Precision Instruments) were used with tip resistances of 0.3 to 0.8 MΩ filled with 3 M KCl. The external solution contains 100 mM KOH, 5 mM NaOH, 20 mM HEPES and 2 mM EGTA. Solutions were adjusted to pH 7.4 using methanesulfonic acid. All recordings were obtained with no leak subtraction. Data in Relative PO vs. voltage curves were fitted to a sum of two Boltzmann curves in Origin 2017 (OriginLab) with the function I(V) = O1 + (A1 − O2)/(1 + exp (k1(V − V1))) + O2 + (A2 − O2)/(1 + exp(k2(V − V2))), where A1 and A2 represent the amplitudes, O1 and O2 represent the offsets, V1 and V2 represent V1/2, and k1 and k2 represent the slope factors for two independent components. As many curves do not reach saturation, curves are primarily provided for visual reference.
Ionic currents were measured from inside-out patches excised from Xenopus oocytes after 1–3 days of mRNA injection. Currents were recorded with an Axopatch 1D amplifier, low-pass filtered at 2 kHz and sampled at 10 kHz. Pipettes were made from borosilicate glass with resistance ranging between 0.7–1.0 MΩ when filled with pipette solution (in mM): 120 KCl, 10 HEPES, and 1.0 CaCl2, pH 7.2. The intracellular solution contained (in mM): 55 KCl, 5 KF, 10 K4O7P2, 0.1 Na3VO4, 10 K2EGTA, 10 HEPES, diC8 PIP2100 nM, pH 7.2. KF, 10 K4O7P2 and 0.1 Na3VO4 (phosphatase-inhibitors) and diC8 PI(4,5)P2 were used. The 1 mM MTSET was prepared by diluting from a 1 M stock prepared daily and stored on ice in internal solution prior to the experiment. Inside-out patches containing sp-HCN channels were exposed to MTSET via a rapid perfusion system in which a computer-controlled valve was used to switch between the internal solution without or with MTSET 1 mM. The dead time of the perfusion apparatus was determined as previously described (Oelstrom et al., 2014) and it was typically 150 ms. This dead time was considered to make the MTSET perfusion coincide with the pulses that open and closed the channels. The time constant of MTSET modification was calculated by fitting peak current at the end of the hyperpolarizing pulse versus the cumulative exposure time to a mono-exponential function. The accessibility (calculated as the apparent second-order rate constants) of MTSET to each cysteine mutant was determined by dividing the reciprocal of the product of the time constant by the MTSET concentration.
Surface fluorescence was measured on a Leica SP8 3X STED Super-resolution microscope using 576 nm excitation and collecting emission from 590 to 650 nm with a 10X, 1.4 NA objective. Detector gain, pinhole diameter, Z-position, integration time, and all other instrument settings were kept constant between all oocytes measured. Surface fluorescence was defined as the integrated intensity within a box of fixed dimension using ImageJ. Following confocal measurements, oocytes were maintained individually to enable correlation of current recordings to confocal measurements. Currents were recorded in two electrode voltage clamp as described above with the exception that ND96 solution was used externally.
Clampfit (Molecular Devices) was used to quantitate currents at steady state. Origin was used to fit data points to a sum of two Boltzmann curves. Kinetic Model Builder was used for all kinetic simulations. PyMOL was used for the structural analysis in Figure 5—figure supplement 1D and USCF chimera for the one in Figure 2. Throughout the paper, n is used to denote the number of oocytes tested in each experiment as indicated in each figure legend.
PHENIX : a comprehensive Python-based system for macromolecular structure solutionActa Crystallographica Section D Biological Crystallography 66:213–221.https://doi.org/10.1107/S0907444909052925
Changes in local S4 environment provide a voltage-sensing mechanism for mammalian hyperpolarization-activated HCN channelsThe Journal of General Physiology 123:5–20.https://doi.org/10.1085/jgp.200308918
Efficient estimation of free energy differences from monte carlo dataJournal of Computational Physics 22:245–268.https://doi.org/10.1016/0021-9991(76)90078-4
Kinetic relationship between the voltage sensor and the activation gate in spHCN channelsThe Journal of General Physiology 130:71–81.https://doi.org/10.1085/jgp.200709769
The HCN channel voltage sensor undergoes a large downward motion during hyperpolarizationNature Structural & Molecular Biology 26:686–694.https://doi.org/10.1038/s41594-019-0259-1
Gating interaction maps reveal a noncanonical electromechanical coupling mode in the shaker K+ channelNature Structural & Molecular Biology 25:320–326.https://doi.org/10.1038/s41594-018-0047-3
Mode shift of the voltage sensors in shaker K+ channels is caused by energetic coupling to the pore domainThe Journal of General Physiology 137:455–472.https://doi.org/10.1085/jgp.201010573
Inactivation of KCNQ1 potassium channels reveals dynamic coupling between voltage sensing and pore openingNature Communications, 8, 10.1038/s41467-017-01911-8.
Electrostatics and the gating pore of shaker potassium channelsThe Journal of General Physiology 117:69–90.https://doi.org/10.1085/jgp.117.1.69
Structural insights into the mechanisms of CNBD channel functionThe Journal of General Physiology 150:225–244.https://doi.org/10.1085/jgp.201711898
CHARMM-GUI: a web-based graphical user interface for CHARMMJournal of Computational Chemistry 29:1859–1865.https://doi.org/10.1002/jcc.20945
A measure of helical propensity for amino acids in membrane environmentsNature Structural & Molecular Biology 1:368–373.https://doi.org/10.1038/nsb0894-558
Accurate and efficient integration for molecular dynamics simulations at constant temperature and pressureThe Journal of Chemical Physics 139:164106.https://doi.org/10.1063/1.4825247
Uncoupling hydrophobicity and helicity in transmembrane segments. Alpha-helical propensities of the amino acids in non-polar environmentsThe Journal of Biological Chemistry 273:23645–23648.https://doi.org/10.1074/jbc.273.37.23645
Coupling between voltage sensors and activation gate in voltage-gated K+ channelsThe Journal of General Physiology 120:663–676.https://doi.org/10.1085/jgp.20028696
Extending the treatment of backbone energetics in protein force fields: limitations of gas-phase quantum mechanics in reproducing protein conformational distributions in molecular dynamics simulationsJournal of Computational Chemistry 25:1400–1415.https://doi.org/10.1002/jcc.20065
Molecular determinants of coupling between the domain III voltage sensor and pore of a sodium channelNature Structural & Molecular Biology 17:230–237.https://doi.org/10.1038/nsmb.1749
A unified formulation of the constant temperature molecular dynamics methodsThe Journal of Chemical Physics 81:511–519.https://doi.org/10.1063/1.447334
Polymorphic transitions in single crystals: a new molecular dynamics methodJournal of Applied Physics 52:7182–7190.https://doi.org/10.1063/1.328693
Regulation of gating and rundown of HCN hyperpolarization-activated channels by exogenous and endogenous PIP2The Journal of General Physiology 128:593–604.https://doi.org/10.1085/jgp.200609648
Contributions of counter-charge in a potassium channel voltage-sensor domainNature Chemical Biology 7:617–623.https://doi.org/10.1038/nchembio.622
Enhancing the accuracy, the efficiency and the scope of free energy simulationsCurrent Opinion in Structural Biology 15:164–170.https://doi.org/10.1016/j.sbi.2005.03.001
Charge movement in gating-locked HCN channels reveals weak coupling of voltage sensors and gateThe Journal of General Physiology 140:469–479.https://doi.org/10.1085/jgp.201210850
Comparative protein modelling by satisfaction of spatial restraintsJournal of Molecular Biology 234:779–815.https://doi.org/10.1006/jmbi.1993.1626
Cyclic AMP reverses the effects of aging on pacemaker activity and if in sinoatrial node myocytesThe Journal of General Physiology 149:237–247.https://doi.org/10.1085/jgp.201611674
Anton 2: raising the bar for performance and programmability in a Special-Purpose molecular dynamics supercomputerProceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis. pp. 41–53.https://doi.org/10.1109/SC.2014.9
Initial response of the potassium channel voltage sensor to a transmembrane potentialJournal of the American Chemical Society 131:2107–2109.https://doi.org/10.1021/ja807330g
An emerging consensus on voltage-dependent gating from computational modeling and molecular dynamics simulationsThe Journal of General Physiology 140:587–594.https://doi.org/10.1085/jgp.201210873
A unified thermostat scheme for efficient configurational sampling for classical/quantum canonical ensembles via molecular dynamicsThe Journal of Chemical Physics 147:034109.https://doi.org/10.1063/1.4991621
Ranking reversible covalent drugs: from free energy perturbation to fragment dockingJournal of Chemical Information and Modeling 59:2093–2102.https://doi.org/10.1021/acs.jcim.8b00959
High‐temperature equation of state by a perturbation method. I. nonpolar gasesThe Journal of Chemical Physics 22:1420–1426.https://doi.org/10.1063/1.1740409
Richard W AldrichSenior and Reviewing Editor; The University of Texas at Austin, United States
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
Some voltage-gated ion channels are opened by hyperpolarizing (more negative) membrane potentials and others open with depolarization. This differential gating polarity of related voltage sensitive channels has been the subject of intense investigation both because of its physiological implications and because understanding the structural and biophysical mechanisms determining the different behaviors will reveal basic principles of voltage gating. This paper reports a combination of functional studies and molecular computations that proposes and supports a novel structural mechanism of voltage sensing and gating.https://doi.org/10.7554/eLife.53400.sa1
[Editors' note: we include below the reviews that the authors received from another journal, along with the authors’ responses.]
1) Authors should consider eliminating the term "paddle" throughout the manuscript. It is a specific term meaning helix-turn-helix entity with structural and functional implications. In the context of this paper the function or structure is not shown.
We appreciate the feedback but in this context the use of the “paddle” is appropriate. It is widely used to define the S3b-S4 helix and has been shown to be a portable module. In fact, this terminology has been used to describe the primary voltage-sensing region in many channels even when high-resolution structures of those channels were not available (Bosmans et. al. (2008) Nature 456, 202-208).
2) This reviewer has major concern regarding the length of the simulations. Longest simulations of ~350 µs have been done for Kv1.2/2.1 paddle chimera from David Shaw's Lab (Jensen et al., Science, 2012). In essence, the authors can improve the quality and length of simulations as per this reference. Reason being, MD simulations of 20 µs is too short for HCN. For HCN channels fastest family member requires milliseconds to open as measured in cells. I am concerned that simulations might not reflect the correct opening based on time constraints.
As mentioned in the manuscript we have used two strategies to speed up the activation of HCN1, which, as pointed out by the reviewer, is known to open on a millisecond timescale.
The first strategy was to apply a large electric field (-550 mV), similar to what has been done by Jensen et al. for Kv1.2/2.1. The ~350 µs simulations which the reviewer refers to were conducted to observe both deactivation and activation of the channel and were performed under a lower electric field (in an adaptive manner, varying between -375 to -500 mV, making it difficult to infer the average value used; see Jensen’s paper Figure 2A, B and Supplementary file 1).
Importantly, under a larger electric field (varying between -375 to -750 mV), the full deactivation of Kv1.2/2.1 (all four voltage sensors) was observed within ~ 65 µs (see Jensen’s paper Figure 1D). In our simulations, we used the voltage of -550 mV throughout the simulation #2 and starting from 1.5 µs in the simulation #1. We anticipate that the application of the large electric field greatly accelerated the activation of HCN1 bringing it from milliseconds to tens of microseconds timescale. In particular, extrapolation of the HCN1 opening time constant to strongly hyperpolarized voltages indeed suggests that it can be as low as ~ 20 µs (see Author response image 1). Using this approach, the time constant for Shaker return from the activated state to the resting state is predicted to be from 18.5 µs (at -500 mV) to 134 µs (at -375 mV), which is consistent with the timescale of observed movement by Jensen et al.,2012.
The second strategy we used was to perform the simulations of an HCN1 mutant with increased hydration of the voltage sensors, as suggested by Lacroix and Bezanilla, 2013, Neuron. These authors indeed showed that substitution of a hydrophobic residue by a hydrophilic one in voltage-gated potassium channels can speed up their activation up to 3 times. We therefore anticipate that this strategy further allowed us to shift the HCN1 activation time constant towards the timescales available to MD simulations performed on Anton. As controls, we have shown that the activation of the HCN1 mutant follows a similar path than that of the WT (Figure 1—figure supplement 1B of the original manuscript; in both channels the activation starts with the displacement of S4 charges along an electric field), and that the designed mutant is fully functional in electrophysiology experiments, which argues against the possibility that it has an altered activation mechanism. Finally, we have shown that our simulations are in excellent agreement with the already published experimental data (such as puzzling solvent accessibility of S4, the FRET experiments, etc.) and, importantly, provide novel predictions that were further tested in our study.
Finally, we would like to point out that only DE Shaw research can carry out hundreds of microseconds long simulations because no one has access to their special purpose machine outside of the one computer that was donated to PSC. This limits simulations to tens of microseconds at most for any individual research project.
3) Collapse of the filter is observed in the simulations. However, the electrical currents shown did not inactivate or diminished throughout the recordings. If the filter was collapsing the measured function of HCN should be altered (as it happened in KcsA and the authors mentioned the reference Cuello et al., 2010). There is discrepancy between simulations and electrophysiology experiments. How reliable are simulations to guide the experiments when the filter collapse in 10 µs and the currents did not inactivate?
The simulations here were performed for the closed structure of HCN1 (which remained closed throughout the MD runs), and therefore we cannot anticipate whether we would see the same collapse of the selectivity filter in the open state under conditions of ionic current. Based on this, we do not follow the reviewer’s argument and see no discrepancy between the simulations and the experimental data mentioned by the reviewer.
Additionally, the reviewer is probably aware that it is common practice to restrain the selectivity filter in molecular simulations to avoid collapse. The collapse is generally attributed to the inability of classical force fields to describe polarizability. This phenomenon is well known to be extremely specific to the issue of conduction of ions through the selectivity filter of potassium-selective and related cation-selective channels and has no bearing on all the other aspects of channel function, including voltage sensor domain activation.
4) Furthermore, authors state "Similar collapse of the selectivity filter has been observed in other potassium channels at low ionic concentration or in the absence of ionic current (Cuello et al., 2010; Zhou et al., 2001) and has previously been proposed as a mechanism of voltage-dependent gating in K2P channels (Schewe et al., 2016)."To the best of my knowledge in this study the KCl was 150 mM as mentioned in the Materials and methods. Then why we see the filter collapsed if low potassium was not tested?
There appears to be some misunderstanding here. We actually use low ionic strength condition during our simulations and it not clear to us why the reviewer believes that we used 150 mM KCl. Therefore, it is not surprising that the selectivity filter also tends to collapse in our simulations. The relevant section from the text is copied below:
"Briefly, the structure of the HCN1 apo state was embedded in a 1-palmytoyl-2-oleoyl-phosphatidylcholine (POPC) bilayer and initially solvated in a 150 mM KCl solution. The ions were later removed during the first Anton run, and before the second Anton run, after we noticed that they promoted unpacking of the voltage sensor domains. "
5) Figure 3—figure supplement 1D: This reviewer is concerned that the lack of currents for Leucine, Isoleucine, and Valine could be due to altered surface expression, as also mentioned by the authors in the explanation. There is a concussive way to show whether constructs express at the plasma membrane and if yes then indeed the mutants are non-functional.
We provide experimental data to show that the mutants are trafficked to the surface membrane. See Figure 3—figure supplement 2A and B.
1) Was the net gating charge of HCN measured previously or only suggested? In its current form it is not clear. If suggested then on what basis? "The reference net gating charge per VSD is 1.85 e (or 7.4 e per channel), which is in good agreement with the 8 charges per channel suggested previously (Hummert et al., 2018)."
The net gating charge of HCN2 channel was calculated previously by fitting electrophysiological behavior to a detailed kinetic model. The original language was unclear and we have clarified the sentence to reflect how the experimental charge determination was made.
2) There should be some discussion/explanation that why 6 out of 8 simulations were successful. "Substantial conformational change was observed in 6 out of 8 voltage sensors from two independent ~20 µs simulations of the full-length channel (Figure 1A, B)." Is it norm of this kind of experiments or technical challenge?
Molecular dynamics simulations predict the evolution of a system of particles along time. As such, they reproduce the stochasticity of motions that biomolecular system naturally undergo. The events observed in these simulations occur thus with a certain probability which increases with the length of simulations. With a longer MD run, we would thus be able observe the activation of all 8 voltage sensor domains. The stochasticity of MD simulations is also well documented in other computational works, such as the one by Jensen et al. mentioned in this reviewer's point 2. For instance, out of five simulations performed on the resting/closed Kv1.2/2.1 channel under depolarizing electric field, only two showed substantial conformational changes in the voltage sensor domains (see Jensen’s paper Supplementary file 1 simulations 10-14). Moreover, in those two simulations, the gating charges responded only in one (two) VSD for the first (and the second, respectively) simulation.
Is it possible that the hydrophilic 'breakpoint' residue is required not to break the S4 helix but to allow the lipid head group/glycerol backbone to move to that position? In that case, a hydrophobic amino acid substitution would interact with the acyl-chain, incorrectly positioning the lipid molecule and altering gating. The accessibility change would then be explained by movements of the lipids and not from breaking of the helix.
We are not sure that we understand the reviewer’s concern because it is not clear how the proposed mechanism can explain a complete switch in the gating polarity from inward rectifying to outward rectifying. Interaction with a single phospholipid may account for subtle changes in gating properties but it is difficult to imagine how it can lead to the observed phenotype.
Additionally, we are unaware of any experimental evidence to suggest that lipid interactions at S272 are involved in gating. Although HCN channels have been shown to be regulated by PIP2 levels in the membrane, this regulation has been primarily shown to involve interactions with the C-linker within the carboxy terminus of the channel.
Most importantly, if a lipid headgroup is involved, then the interactions would be electrostatic in nature and would depend on the charge of the residue. However, we observe no difference between polar residues and charged residues. Moreover, proline which lacks any charge but promotes helix disruption promotes inverted gating phenotype like the serine at that position. We should also point out that we observed strong correlation between turn propensity and gating polarity. To the best of our knowledge, such correlations have never been observed when considering interactions between lipid molecule and amino acid side-chains.
The cysteine accessibility studies are used to highlight that F186C is exposed in the HCN1 down state, which is inconsistent with helical screw model. It would be helpful to have a series of positions that show that the helical-breaking model is correct instead of primarily showing that the helical screw model is not correct. For example, the incompatibility with the helical screw model may arise from the misplacement of lipids in the proposed accessibility model (shown in the figure) instead of where they are actually located in the real membrane.
We thank the reviewer for this comment, we agree that conclusions should generally not be reliant on the solvent accessibility of a single position. However, solvent accessibility experiments on the S4 helix of HCN1 had been already performed by two independent research groups in 2004. In Figure 2—figure supplement 1, we compare the predicted accessibility from our simulations to 11 positions tested in these studies. Our model correctly predicts the accessibility pattern of 9 of the 11 residues tested in those studies. It is worth noting that for the one of the two residues that are not consistent, we do not observe accessibility in the simulations for a position with very low experimentally determined accessibility. This may indicate that we were unable to sample timescales long enough to observe accessibility. More importantly, both experimental accessibility studies have previously shown that there are larger state-dependent accessibility changes for the internally facing S4 residues than for externally facing residues. This is consistent with our model showing that the S4 helix breaks in the down state and broadens the inside facing crevice. .
We agree that it is worth considering alternative explanations for our experimental results. However, we would like to point out that it is the steric occlusion within the voltage sensing domain that limits accessibility of F186 as well as the S4 helix. We have attempted to highlight this in in the surface representations of Figure 2A and in cartoon form in Figure 2—figure supplement 1. This occlusion is observed in the experimentally determined resting state. Therefore, the drastic changes in accessibility we (and others) show experimentally must be caused by substantial conformational change of the protein itself. Without knowing more details, it is hard to envision how rearrangement of the lipid around the protein could explain the changes in accessibility observed given the structure of the resting state of HCN1.
From a philosophical standpoint, we agree with the reviewer that we cannot rule out a more complex model involving additional parameters such as lipid molecule etc. Given the available data, our model involving helix breaking transition is the most parsimonious. Our experiments are designed to distinguish between existing models. As the helical screw model is the canonical model of voltage sensor movement, we elected to focus on distinguishing it from other proposed mechanisms. The only other model that has been proposed which would be consistent with our observed accessibility change to the charge transfer center is the so-called ‘transporter model’ in which S4 is relatively static while S1-S3 rearrange around it. Indeed, Seigelbaum and colleagues in their accessibility study favored this model over the canonical sliding helix type of model. Although the transporter model could be consistent with the accessibility observed by us and others, it would fail to explain why hyperpolarization activation is sensitive to mutations at S272. This position faces towards the pore and away from S1-S3 in the resting Up state. Given the strong agreement of our model with the accessibility patterns shown by us and others, as well as the structure-function data at the break point, we feel that we have strong experimental evidence supporting the S4 breaking model.
The sequence conservation suggests that other HCN channels will exhibit the same helix-breaking in S4. Are similar structural changes observed with these channels?
A similar solvent accessibility pattern of the S4 residues has been reported for spHCN, suggesting that this channel employs the same helix-breaking mechanism. In addition, in the original manuscript, we refer to the example of a CNG family member (TAX4, see Figure 1E) which is homologous to HCNs and whose structure was resolved in the open state. Remarkably, in this channel, the S4 helix is broken in the same position as we find in HCN1.
The FEP calculations of the serine (272) to leucine (272) transition support their model. Similar calculations upon mutating neighboring positions, such as those shown in Figure 3—figure supplement 1, should be made to further test their model and the appropriateness of the FEP calculations in this context.
Indeed, performing such additional simulations may give more confidence in the appropriateness of the FEP calculations. However, these calculations are computionally expensive and to do this throughly for various sites is beyond the scope of this manuscript.
Mutations are used to increase the rate of activation of the channel for the Anton simulations (M153T/L157V). What are the properties of the M153T/L157V mutant using electrophysiology?
We thank the reviewer for this comment. Both the simulations and the electrophysiology experiments (Figure 1—figure supplement 1) were performed for M153T/I160V. We apologize for the numbering error in the original manuscript and have changed the manuscript accordingly.
It's unclear whether Figure 1—figure supplement 1 (M153T/I160V) is an electrophysiology experiment or a computational calculation. This mutation isn't discussed in the main text, but M153T/L157V is. These mutations are near (in 3-dimensional space) to residue 272 (the break helix). Could these mutations be responsible for the observed helix break?
We thank the reviewer for this comment. Please see the response to the previous comment for the first part of the question. For the second part, we do not think that M153T and I160V are responsible for the S4 break. Though we were not able to run a control simulation of appropriate length without these mutations (see response to reviewer 1 for a discussion of the rationale for using such mutants), our experiments supporting the idea of the S4 break were performed on HCN1 and chimeras lacking these mutations. Again, there are also previously reported solvent accessibility and FRET experiments performed on WT spHCN channels that are compatible with our model.
In the text, the relationship between depolarization/hyperpolarization and inward/outward rectification needs to be stated for the general reader.
We agree this connection was not properly explained in the text and have clarified in revisions.
Jargon such as "minimal HCN background" should be explained in the text.
We now refer to these as “proto-HCN channels”. Because the HCN and EAG channels are related, we think of these as some type of ancestral channels which have some properties of both HCN and EAG channels.https://doi.org/10.7554/eLife.53400.sa2
- Baron Chanda
- John B Cowgill
- Jenna L Lin
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
This work was supported by grants from the Gustafsson Foundation and Science for Life Laboratory to LD; National Institutes of Health to BC (NS101723), JC (T32 HL-07936–18), JL (T32 GM008293); Anton 2 allocation to BC (PSC17025P and PSC18025P); Romnes faculty fellowship to BC; Science and Medicine Graduate Research Scholars (SciMed GRS) fellowship to JL. Preliminary simulations and free energy calculations were performed on resources provided by the Swedish National Infrastructure for Computing (SNIC) at the PDC Centre for High Performance Computing (PDC-HPC). Anton 2 computer time was provided by the Pittsburgh Supercomputing Center (PSC) through Grant R01GM116961 from the National Institutes of Health. The Anton 2 machine at PSC was generously made available by DE Shaw Research. The authors also thank N Nallappan and A Thangaraju for help generating chimeras and A Thangaraju, N Nallappan, and W Stevens-Sostre for performing frog surgeries. We thank Y Itoh for help with the structure refinement and RJ Howard for assistance in preparing Figure 2. We would like to thank GA Robertson for providing hEAG1, C Czajkowski for providing the pUNIV vector, B Santoro and SA Siegelbaum for providing mHCN1, and UB Kaupp for providing spHCN. We would like to thank S Chowdhury and other members of the labs for helpful discussions and input throughout the project.
- Richard W Aldrich, The University of Texas at Austin, United States
© 2019, Kasimova et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Downloads (link to download the article as PDF)
Download citations (links to download the citations from this article in formats compatible with various reference manager tools)
Open citations (links to open the citations from this article in various online reference manager services)
Protein-protein interaction specificity is often encoded at the primary sequence level. However, the contributions of individual residues to specificity are usually poorly understood and often obscured by mutational robustness, sequence degeneracy, and epistasis. Using bacterial toxin-antitoxin systems as a model, we screened a combinatorially complete library of antitoxin variants at three key positions against two toxins. This library enabled us to measure the effect of individual substitutions on specificity in hundreds of genetic backgrounds. These distributions allow inferences about the general nature of interface residues in promoting specificity. We find that positive and negative contributions to specificity are neither inherently coupled nor mutually exclusive. Further, a wild-type antitoxin appears optimized for specificity as no substitutions improve discrimination between cognate and non-cognate partners. By comparing crystal structures of paralogous complexes, we provide a rationale for our observations. Collectively, this work provides a generalizable approach to understanding the logic of molecular recognition.
Filoviruses such as Ebola and Marburg virus bud from the host membrane as enveloped virions. This process is achieved by the matrix protein VP40. When expressed alone, VP40 induces budding of filamentous virus-like particles, suggesting that localization to the plasma membrane, oligomerization into a matrix layer, and generation of membrane curvature are intrinsic properties of VP40. There has been no direct information on the structure of VP40 matrix layers within viruses or virus-like particles. We present structures of Ebola and Marburg VP40 matrix layers in intact virus-like particles, and within intact Marburg viruses. VP40 dimers assemble extended chains via C-terminal domain interactions. These chains stack to form 2D matrix lattices below the membrane surface. These lattices form a patchwork assembly across the membrane and suggesting that assembly may begin at multiple points. Our observations define the structure and arrangement of the matrix protein layer that mediates formation of filovirus particles.