The Shaker-like family of voltage-gated K+ channels comprises four functionally independent gene subfamilies, Shaker (Kv1), Shab (Kv2), Shaw (Kv3), and Shal (Kv4), each of which regulates distinct aspects of neuronal excitability. Subfamily-specific assembly of tetrameric channels is mediated by the N-terminal T1 domain and segregates Kv1–4, allowing multiple channel types to function independently in the same cell. Typical Shaker-like Kv subunits can form functional channels as homotetramers, but a group of mammalian Kv2-related genes (Kv5.1, Kv6s, Kv8s, and Kv9s) encodes subunits that have a “silent” or “regulatory” phenotype characterized by T1 self-incompatibility. These channels are unable to form homotetramers, but instead heteromerize with Kv2.1 or Kv2.2 to diversify the functional properties of these delayed rectifiers. While T1 self-incompatibility predicts that these heterotetramers could contain up to two regulatory (R) subunits, experiments show a predominance of 3:1R stoichiometry in which heteromeric channels contain a single regulatory subunit. Substitution of the self-compatible Kv2.1 T1 domain into the regulatory subunit Kv6.4 does not alter the stoichiometry of Kv2.1:Kv6.4 heteromers. Here, to identify other channel structures that might be responsible for favoring the 3:1R stoichiometry, we compare the sequences of mammalian regulatory subunits to independently evolved regulatory subunits from cnidarians. The most widespread feature of regulatory subunits is the presence of atypical substitutions in the highly conserved consensus sequence of the intracellular S6 activation gate of the pore. We show that two amino acid substitutions in the S6 gate of the regulatory subunit Kv6.4 restrict the functional stoichiometry of Kv2.1:Kv6.4 to 3:1R by limiting the formation and function of 2:2R heteromers. We propose a two-step model for the evolution of the asymmetric 3:1R stoichiometry, which begins with evolution of self-incompatibility to establish the regulatory phenotype, followed by drift of the activation gate consensus sequence under relaxed selection to limit stoichiometry to 3:1R.
Shaker-like Kv channels regulate neuronal excitability, including many aspects of action potential repolarization and timing. The Shaker-like Kv gene family consists of four functionally independent subfamilies, which provide a diverse array of depolarization-gated K+ currents: Shaker (Kv1), Shab (Kv2), Shaw (Kv3) and Shal (Kv4; Wei et al., 1990; Covarrubias et al., 1991). Many of the delayed rectifier and transient A-type currents observed in neurons are encoded by Shaker-like Kv family genes, and some of their most notable roles are described below. Kv1 channels localize to the axon initial segment and juxtaparanodes of mammalian neurons, where they participate in axonal action potential repolarization (Wang et al., 1993; Dodson et al., 2002; Ogawa et al., 2008; Trimmer, 2015). They appear to underlie the classical delayed rectifier of the squid giant axon (Rosenthal et al., 1996). Kv2 channels encode the majority of somatodendritic delayed rectifiers (Tsunoda and Salkoff, 1995b; Murakoshi and Trimmer, 1999; Du et al., 2000; Malin and Nerbonne, 2002; Misonou et al., 2005), but they can also be found in ankyrin-free zones of the axon initial segment in mammalian neurons (King et al., 2014). Mammalian Kv3 channels underlie rapid high threshold delayed rectifiers that facilitate high spike rates in fast-firing neurons (Wang et al., 1998; Lau et al., 2000; Rudy and McBain, 2001; Lien and Jonas, 2003). Kv4 channels, in contrast, encode classical somatodendritic A-currents found in many mammalian and invertebrate neurons (Tsunoda and Salkoff, 1995a; Malin and Nerbonne, 2000, 2001; Carrasquillo and Nerbonne, 2014), although it should be noted that Kv1 subfamily channels can contribute a kinetically distinct component to somatodendritic A-type currents in at least some mammalian neurons (Malin and Nerbonne, 2001; Carrasquillo and Nerbonne, 2014).
Shaker-like Kv channels are tetrameric (MacKinnon, 1991; Long et al., 2005a), with each subunit containing a canonical voltage-gated cation channel core motif of six transmembrane domains (S1–S6). S1–S4 comprise the voltage sensor domain (VSD), while S5–S6 comprise the pore domain (PDs) with the K+ selectivity filter formed on the extracellular side by a highly conserved loop (Jiang et al., 2003; Long et al., 2005a,b). Each channel has four spatially independent VSDs, but a single pore formed by extensive intersubunit contact between the PDs. The unique and defining feature of Shaker-like Kvs relative to other voltage-gated K+ channels is the presence of a cytoplasmic N-terminal domain, T1, which promotes assembly of tetramers and forms another large intersubunit interface (Shen and Pfaffinger, 1995; Xu et al., 1995; Kreusch et al., 1998; Long et al., 2005a). Fig. 1, A–D, summarizes the structural layout of tetrameric Shaker-like Kv channels, including the two major intersubunit interfaces in T1 and the inner pore. T1-mediated tetramer assembly requires physical interaction between neighboring T1 domains and is subfamily-specific because T1s from distinct subfamilies are not compatible and do not interact (Shen and Pfaffinger, 1995; Xu et al., 1995). The T1 domain therefore plays a key role in maintaining functional segregation of the Kv1, Kv2, Kv3, and Kv4 subfamilies.
Kv1–4 subunits can typically assemble as functional homotetramers, and the isolated T1 domain is itself able to self-assemble into a tetrameric structure (Pfaffinger and DeRubeis, 1995; Kreusch et al., 1998). A notable exception to this rule of assembly is a group of ten “silent” or “regulatory” subunits discovered in mammals, Kv5.1, Kv6.1–6.4, Kv8.1–8.2, and Kv9.1–9.3, that are self-incompatible and thus not able to form homotetramers, but can form functional heteromeric channels with novel biophysical properties when coexpressed with Kv2.1 or Kv2.2 (Post et al., 1996; Patel et al., 1997; Salinas et al., 1997a,b; Kramer et al., 1998; Zhu et al., 1999; Ottschytsch et al., 2002). While the names of these channels imply that they belong to distinct gene subfamilies, phylogenetic analysis places them within the Kv2 subfamily (Jegla et al., 2012; Li et al., 2015b), thus preserving the principle of subfamily-specific assembly. In the absence of Kv2.1, Kv6.4 regulatory subunits do not assemble into tetramers and do not traffic to the plasma membrane (Ottschytsch et al., 2002, 2005). Mammalian regulatory subunit T1s also do not oligomerize in vitro, suggesting T1 self-incompatibility plays a significant role in generation of the self-incompatible regulatory subunit phenotype (Salinas et al., 1997b; Kramer et al., 1998; Ottschytsch et al., 2002).
Crystal structures of Shaker channels show that a T1 domain only physically interacts with T1s from neighboring subunits; there is no interaction between T1s of diagonally opposed subunits (Kreusch et al., 1998; Long et al., 2005a). T1 self-incompatibility of regulatory subunits therefore predicts that regulatory subunits could combine with Kv2.1 in two possible channel stoichiometries: (1) an asymmetric stoichiometry with three Kv2.1s and a single regulatory subunit (3:1R), or (2) a symmetric stoichiometry with two Kv2.1s and two diagonally opposed regulatory subunits (2:2R; Fig. 1 E). T1 self-incompatibility rules out 1:3R channels and 2:2R channels with adjacent regulatory subunits, but in theory would not be able to distinguish between a 3:1R or diagonal 2:2R stoichiometry. However, FRET-based determination of the Kv2.1:Kv9.3 stoichiometry suggests that heteromers form predominantly in the asymmetric 3:1R stoichiometry (Kerschensteiner et al., 2005). Measurements of gating charge components attributable to Kv6.4 and Kv2.1 support a 3:1R stoichiometry for Kv2.1:Kv6.4 heteromers (Bocksteins et al., 2017), and therefore suggest that the 3:1R stoichiometry might be shared among mammalian Kv2 family regulatory subunits.
We reasoned there must be additional factors beyond T1 self-incompatibility of regulatory subunits that bias formation of functional channels in the 3:1R stoichiometry. There is indeed evidence suggesting that T1 self-incompatibility is not the only factor guiding assembly of regulatory subunits. For instance, substitution of the Kv2.1 T1 into Kv6.4 is not sufficient to restore assembly of functional homomeric channels (Ottschytsch et al., 2005), and N terminus/C terminus interaction may play a role in blocking the ability of Kv6.4 to assemble with Kv3 channels (Bocksteins et al., 2014). To find channel regions that might be responsible for determining regulatory subunit stoichiometry, we used an evolutionary approach to look for regulatory subunit sequence signatures near the two major subunit interfaces in Shaker-like Kv family tetramers, the T1 contacts and the PD contacts. T1-containing Shaker-like Kvs can be traced to a common ancestor of all extant metazoans (Li et al., 2015b), and diversification of voltage-gated K+ channels (including the Shaker-like Kv, KCNQ, and EAG families) into the functional classes, or subfamilies, found in vertebrates occurred before the divergence of bilaterians and cnidarians (Jegla and Salkoff, 1994, 1997; Jegla et al., 1995, 2012; Sand et al., 2011; Martinson et al., 2014; Li et al., 2015b,c). Kv1–Kv4 channels can all be found in cnidarians, and the functional properties of all four subfamilies, including subfamily-specific assembly, are highly conserved between cnidarians and bilaterians. Most important for this study, the regulatory subunit phenotype evolved independently in cnidarians in the Kv1, Kv3, and Kv4 subfamilies (Jegla and Salkoff, 1997; Jegla et al., 2012; Li et al., 2015b), providing a rich dataset for identification of regulatory subunit sequence signatures.
We show here that alteration of the highly conserved sequence of the pore’s intracellular activation gate is a universal signature for evolution of the regulatory subunit phenotype. The activation gate forms part of the interface between subunits (Fig. 1 B), and extensive mutagenesis studies of Shaker subfamily channels have demonstrated that various substitutions in the gate can block channel maturation, interfere with gating, and alter conductance (Hackos et al., 2002; Kitaguchi et al., 2004; Pau et al., 2017). We tested the hypothesis that these gate substitutions could play a role in determining the stoichiometry of heteromeric channels containing regulatory subunits using a combination of TIRF microscopy and electrophysiology.
Materials and methods
Mouse Kv2.1 and Kv6.4 cDNA were isolated from whole brain mRNA using RT-PCR, cloned into pOX plasmid (Jegla and Salkoff, 1997) using HindIII/XbaI (Kv2.1) or EcoR1/Xba1 (Kv6.4), and the sequence was confirmed to code proteins identical to Refseq (Kv2.1, NP_032446; Kv6.4, NP_080010). Kv6.4-Kv2.1T1, Kv6.4-PIPIIV-Kv2.1CT, and Kv6.4-PIPIIV chimeras were generated from two PCR fragments using standard overlap PCR techniques and cloned into pOX using EcoR1/Xba1. For the Kv6.4-Kv2.1T1 chimera, the following primers were used: piece 1 (Kv2.1 T1), 5′-TATAGAATTCCACCATGCCGGCGGGCAT-3′ (sense) and 5′-AGCCTCCTCCCGGCGCAGCGTCTCAGCC-3′ (antisense), and piece 2 (Kv6.4 core to C terminus), 5′-AGGCTGAGACGCTGCGCCGGGAGGAGG-3′ (sense) and 5′-TCTCTCTAGAGTGGTGGGAGTTACAT-3′ (antisense). Kv6.4-PIPIIV-Kv2.1CT was generated with the following primers: piece 1 (Kv6.4 upstream of gate), 5′-TCTCGAATTCCACCATGCCCATGTCT-3′ (sense) and 5′-GTTATTGACGATAATTGGAATTGGGA-3′ (antisense), and piece 2 (Kv2.1 gate through C terminus), 5′-CTTATCATGGCTTTCCCAATTCCAATT-3′ (sense) and 5′-TCTCTCTAGATTCAGATACTCTGATCC-3′ (antisense). Kv6.4-PIPIIV was generated in the same manner except using a piece 2 containing the Kv6.4 C terminus generated with the following primers: 5′-TCATGCCTTTCCCAATTCCAATTATCG-3′ (sense) and 5′-TCTCTCTAGAGTGGTGGGAGTTACAT-3′ (antisense). Kv6.4-Kv2.1CT, Kv6.4-PITIIV-Kv2.1CT, Kv6.4-PIPIIF-Kv2.1CT, and Kv6.4-PITIIF-Kv2.1CT were generated by cloning PCR fragments into Kv6.4-PIPIIV-Kv2.1CT using an internal BamH1 site in S6 upstream of the gate and Xba1. Fragments were generated using the antisense primer 5′-TCTCTCTAGATTCAGATACTCTGATCC-3′ and the following sense primers: 5′-AGCGGGATCCTTATCATGGCTTTCCCAGCCACATCCATCTTCCACACCTTCTCCGAGTTCTAC-3′ (Kv6.4-Kv2.1CT); 5′-AGCGGGATCCTTATCATGGCTTTCCCAATTACAATTATCGTCAATAAC-3′ (Kv6.4-PITIIV-Kv2.1CT); 5′-AGCGGGATCCTTATCATGGCTTTCCCAATTCCAATTATCTTCAATAACTTCTC-3′ (Kv6.4-PIPIIF-Kv2.1CT); and 5′-AGCGGGATCCTTATCATGGCTTTCCCAATTACAATTATCTTCAATAACTTCTC-3′ (Kv6.4-PITIIF-Kv2.1CT). Pore cysteine mutants for Cd2+ block experiments were constructed using the QuikChange protocol (Wang and Malcolm, 1999) with the following primers: Kv2.1 I383C, (sense) 5′-GTTGGTTACGGAGACTGCTACCCTAAGACTCTCCTG-3′ and (antisense) 5′-CAGGAGAGTCTTAGGGTAGCAGTCTCCGTAACCAAC-3′; Kv6.4 M422C (sense) 5′-GTGGGCTATGGGGACTGCGTCCCTCGCAGCGTCCCG-3′ and (antisense) 5′-CGGGACGCTGCGAGGGACGCAGTCCCCATAGCCCAC-3′. Enhanced GFP (Zhang et al., 1996) tags were cloned onto the N terminus of Kv2.1- or Kv6.4-based constructs using NheI/HindIII or NheI/EcoRI, respectively. The flexible linkers between the GFP tag and channel N termini were QQQAST for Kv2.1 and QQQNST for Kv6.4. Oligonucleotide primers were obtained from IDT (Integrated DNA Technologies), and all constructs were confirmed by sequencing.
Complementary RNA (cRNA) synthesis and Xenopus oocyte preparation
For all constructs, capped cRNAs were made by run-off transcription from NotI-linearized templates using the T3 mMessage mMachine kit (Life Technologies). Prior to injection, cRNA was purified using lithium chloride precipitation and analyzed by gel electrophoresis for integrity. cRNAs were stored at −80°C and diluted to the desired concentration for injection in a 1:20 mix of the RNase inhibitor SUPERase-In (Invitrogen) and nuclease-free water. Optimal amounts of cRNA for injection were empirically determined for each batch of cRNA. Mature stage V/VI Xenopus laevis oocytes were injected with 50 nl of cRNA using a NanoJect II injector (Drummond Scientific). Defolliculated oocytes were isolated from X. laevis ovaries (Xenopus 1) using collagenase digestion (Type II Collagenase; Sigma Aldrich), as previously described (Clancy et al., 2009) and maintained at 18°C in an oocyte culture solution consisting of 98 mM NaCl, 2 mM KCl, 1.8 mM CaCl2, 1 mM MgCl2, 5 mM HEPES, 2.5 mM Na-pyruvate, 100 U/ml penicillin, and 100 µg/ml streptomycin, pH 7.2.
Two-electrode voltage clamp (TEVC) recordings for biophysical characterization of heteromers and expression ratio titrations used the following bath solution: 96 mM NaOH, 2 mM NaCl, 2 mM KCl, 1 mM CaCl2, 1 mM MgCl2, and 5 mM HEPES, pH 7.0, using methanesulfonic acid. For Cd2+ block experiments, the following base solution was used: 100 mM LiCl, 2 mM CaCl2, 1 mM MgCl2, and 5mM HEPES, pH 7.4. A LiCl-based solution has previously been found to optimize Cd2+ block for rat Kv2.1 I379C, the equivalent of mouse Kv2.1 I384C used here for Cd2+ block experiments (Krovetz et al., 1997). All chemicals were obtained from Sigma-Aldrich. Glass electrodes (1–3 MΩ) were pulled using a P-1000 Flaming/Brown Micropipette Puller (Sutter Instruments) and filled with 3 M KCl. Oocytes were clamped using a Dagan CA-1B amplifier and the pClamp 10 acquisition software for data collection and analysis (Molecular Devices). Data were sampled at 10 kHz and low-pass-filtered at 2 kHz using a four-pole Bessel filter. Voltage-activation (GV) curves were determined from isochronal tail currents recorded after test pulses. Tail currents were fit with a single Boltzmann: , where G is the conductance at voltage V, A1 is the initial value, A2 is the final value, V50 is the midpoint and dx is the slope factor. Reported V50 and slope factor values are the mean ± SEM of fits from individual oocytes, and data were normalized before averaging for display. Steady-state inactivation (SSI) data were determined from peak currents measured during a test pulse to +40 mV following prepulses to varying voltages and fit with the same equation. Non-inactivating pedestal current fractions were defined as A2/A1.
The procedure for data collection for the cRNA expression ratio titrations shown in Fig. 8 was developed in a series of pilot titration experiments to identify conditions that minimized inter-oocyte variation in current size, the optimal time window for a recording session, and the optimal amount of Kv2.1 cRNA to provide sufficient dynamic range for quantification while minimizing the chance of nonspecific current suppression. Batch-to-batch variation in current size is significant and required a single batch of oocytes to be used for each of the final cRNA titration experiments. We estimated that 15–20 µA of current was needed in the control Kv2.1 oocyte recordings in order to adequately resolve Kv6.4-dependent loss in current. We injected the minimum amount of Kv2.1 cRNA required to reach this level in ∼36 h of incubation at 18°C (0.47 ng/oocyte); oocytes were tested for expression level, and experiments were performed when optimal expression was reached. Because of the low amount of Kv2.1 cRNA needed, total cRNA injected never exceeded 50 ng/oocyte even at the most Kv6.4-biased expression ratios. This is below the 50–100 ng of cRNA we injected to maximize expression in oocytes for patch clamp analysis (Jegla et al., 2012; Li et al., 2015a). We and others have observed preservation of subfamily-specific dominant negative suppression for Shaker-like channels and hyperpolarization-activated cation channels with ≥50 ng total cRNA and ≥20 μA total current (Xue et al., 2002; Li et al., 2015b). It is therefore unlikely that current loss observed in the expression titration experiments here is due to nonspecific suppression related to channel expression level and trafficking or oocyte toxicity. Data for the RNA titration experiments in Fig. 8 were normalized to the average peak current in control Kv2.1 oocytes from the same recording session. Each RNA titration experiment lasted ∼4–6 h. To neutralize the effect of small changes in channel expression over this period of time, recordings from oocytes expressing distinct cRNA ratios were interleaved.
For the Cd2+-block experiments in Fig. 9, peak current before Cd2+ exposure served as an internal control for each individual oocyte. Current size after Cd2+ exposure was normalized to this control before data averaging. Because each oocyte has an internal control, the timing of recordings post-injection did not need to be tightly synchronized as for the titration experiment described above. Data from two different batches of oocytes were pooled for each condition.
For on-cell patch recordings of single channels, recordings were made 1–3 d postinjection, and cRNAs were diluted ≥10-fold more than for TEVC recordings. Kv2.1 and Kv6.4 were injected at ratios suitable to observe both homomeric and heteromeric currents as determined by TEVC (see Fig. 8 H). Vitelline membranes of oocytes were mechanically removed as previously described (Li et al., 2015a) and transferred to a recording dish filled with internal solution (138 mM KMES, 4 mM KCl, 10 mM HEPES, 5 mM EGTA, pH 7.2; MES). Patch pipettes (0.4–1 MΩ) were coated with Sticky Wax (Kerr Dental Laboratory Products), fire-polished, and filled with a solution containing 140 mM KMES, 2 mM KCl, 1 mM MgCl2, 0.2 mM CaCl2, and 10 mM HEPES, pH 7.2. Ground was isolated with a 1 M NaCl agarose bridge, junction potential was canceled before patch formation, and pipette capacitance was compensated. Data were collected using a Multiclamp 700A amplifier and the pClamp 9 acquisition package (Molecular Devices). Data were sampled at 20 kHz and filtered at 1.4 kHz.
Prior to imaging, oocyte vitelline envelopes were mechanically removed, and stripped oocytes were mounted on Number 0 Coverslips (VWR) in a custom-made chamber filled with oocyte culture solution. Care was taken to place the stripped oocyte onto the chamber such that the animal pole faces the microscope objective; this was done to limit autofluorescence during movie acquisition. The chamber was loaded onto a Nikon TE-2000 inverted microscope outfitted with a 60× numerical aperture 1.45 objective (Nikon). An 80-mW argon ion laser (Spectra Physics) was used for illumination, and a Cascade 512B electron-multiplying charge-coupled device (Roper Scientific) was used for detection. 240-s movies were acquired at 5 frames per second using MetaVue software (Molecular Devices). TIRF experiments were performed within 12–18 h after injection, and cRNAs were diluted 2–10-fold more than for electrophysiology to achieve optimal channel density. To ensure sufficient numbers of analyzable photobleaching spots, at least three movies were obtained for each egg, and photobleaching data were obtained for three to six different eggs from two to four different batches of oocytes for each GFP-tagged construct.
TIRF photobleaching data were analyzed in a semi-automated manner using a custom MATLAB (Mathworks) script. Briefly, fluorescence intensity over time traces for well-isolated spots were acquired from tagged image file format (TIFF) stacks, which summed counts in a 7-pixel diameter circle drawn around candidate channels. Fluorescence changes in a larger 15-pixel diameter circle near the chosen spot were used to estimate the background fluorescence in order to correct baseline changes. The number of photobleaching steps was then determined using a published step-finding algorithm (Chen et al., 2014). Only spots that were stationary and had stable baselines were included in the analysis. Analysis of all observed step dwell times indicated the time constant for a GFP bleaching step to occur was ∼68 s, so >97% of all bleaching events should be captured in 240-s movies.
Sequences were aligned for display using the CLUSTALW algorithm as implemented in MEGA7 (Kumar et al., 2016). Sequence logos displayed in Fig. 2 were generated from alignments using WebLogo (Crooks et al., 2004). Supplemental Table 1 contains sequences and accession numbers for all channels included in the sequence logos, and Supplemental Table 2 contains amino acid frequencies at each gate position for the sequence logos.
Structural models were based on published open and closed state computational models of Kv1.2 (Pathak et al., 2007) and included the entire conserved T1–S6 regions of Kv2.1 and Kv6.4. Sequence alignments of Kv2.1 and Kv6.4 to Kv1.2 for structural modeling were generated using MUSCLE, as implemented by Jalview (Waterhouse et al., 2009); a FASTA file of aligned sequences is provided (Supplemental text file). Models were built from the alignments using MODELLER (9v8; Sali and Overington, 1994; Martí-Renom et al., 2000; Eswar et al., 2007). For each target (Kv2.1, Kv6.4 and Kv6.4-PIPIIVN), we generated 30 initial models. MODELLER uses a variety of methods (e.g., Z-DOPE, molpdf, GA341) to assess the models. These scores are associated with the estimated accuracy of the model. For each target-based model set, we selected the top candidate model guided by the molecular PDF score (molpdf), which is the sum of all spatial restraints achieved when transferring structural regions from the template to the target. These candidates also had the lowest Z-DOPE scores of the 30 initial models generated for each template. Assembly and visualization of all the tetrameric models was done in VMD (Humphrey et al., 1996) using visual manipulations and custom-written Tcl scripts. We generated four tetramer arrangements: Kv2.1 homomer, Kv2.1:Kv6.4 (3:1R), Kv2.1:Kv6.4 (2:2R diagonal), and Kv2.1:Kv6.4-PIPIIVN (2:2R diagonal). For the first three models, local structural clashes in the models were removed through 1,000-step energy minimization using the Conjugate Energy algorithm, followed by relaxation using short (1 ps) Molecular Dynamics simulation in vacuo at 300°K, and a 100-step energy minimization. All the optimizations were done using NAMD (Phillips et al., 2005) with a CHARMM36 force field (MacKerell et al., 1998). For the Kv2.1:Kv6.4-PIPIIVN model, the gate of the final Kv2.1 homomer model was placed into the final Kv2.1:Kv6.4 2:2R model, and the model was optimized with a 100-step energy minimization, 1 ps thermal relaxation, and then another 100-step energy minimization again with symmetry restraints. The space accessible for the K+ ion center at the pore gate constriction in the open state models was visualized using “solvent” representation for the probe radius set to 1.8 Å (potassium ion in CHARMM36 force field [1.76375 Å] rounded to the nearest 0.1 A) and all the protein atoms having their Van der Waals radius derived from CHARMM36.
Online supplemental information
Supplemental Table 1 contains sequences and accession numbers for all channels included in the sequence logos. Supplemental Table 2 contains amino acid frequencies at each gate position. The supplemental text file is a FASTA file of the Kv2.1 and Kv6.4 alignment used for structural modeling.
The highly conserved S6 activation gate sequence degenerates in regulatory subunits
To determine whether there are sequence signatures that accompany the regulatory subunit phenotype (i.e., inability to form functional homotetramers), we compared the amino acid sequences of eight evolutionarily distinct clusters of regulatory subunits comprising 43 proteins from mouse and sea anemone (Nematostella vectensis) that arose independently as determined by phylogenetic analyses (Jegla et al., 2012; Li et al., 2015b). We restricted the analysis to these two species because the phenotype for each regulatory subunit cluster has been experimentally verified. In mammals, the 10 Kv2 subfamily regulatory subunits can be split into two groups with separate evolutionary origins: (1) Kv5.1, which first appeared in chordates, and (2) Kv6–9, which later arose independently in vertebrates (Li et al., 2015b). In N. vectensis, six additional evolutionarily independent clusters of regulatory subunits can be found in the Kv1 (four clusters), Kv3 (one cluster), and Kv4 (one cluster) subfamilies (Jegla and Salkoff, 1997; Jegla et al., 2012; Li et al., 2015b). We found no consistent regulatory subunit sequence alterations in the subfamily-specific T1 assembly domain, but 42 of 43 regulatory subunits contained unusual substitutions in the C terminal of S6 (PVPVIV in Drosophila Shaker), a region that is typically among the most highly conserved in Shaker-like Kv channels (Fig. 2, A–C). This sequence includes the proline gating hinge and forms extensive intersubunit contacts in the structure of the mammalian Shaker channel Kv1.2 (Long et al., 2005a). The bold residues, V474 and V478 in Drosophila Shaker, face inward toward the conduction pathway, forming a hydrophobic seal in the closed state (Long et al., 2005a). Access to V474 in Shaker is gated by pore opening, while access to V478 is possible in both open and closed channels, suggesting that V478 forms the cytoplasmic boundary of the intracellular activation gate (Liu et al., 1997; del Camino and Yellen, 2001; Hackos et al., 2002; Webster et al., 2004; del Camino et al., 2005). To facilitate comparison of gate sequences between channels, in this paper we will number residues by their position within the six-amino acid (PVPVIV) gate sequence; in this nomenclature, V474 is V2 and V478 is V6. Each independent cluster of regulatory subunits has a distinct pattern of substitutions in this activation gate sequence (Fig. 2, D–K), with several types of substitutions occurring multiple times. These include loss or displacement of one of the gating hinge prolines, insertion of polar residues that disrupt gate hydrophobicity, or insertion of large aromatics that could potentially cause steric conflicts. In the mammalian Kv6–9 cluster, P3 is lost (nine of nine proteins) and replaced with a polar residue (S/T, eight of nine proteins), and an aromatic is inserted in place of V6 (eight of nine proteins; N in Kv8.1; Fig. 2 D). Site-directed mutagenesis of Shaker has found that hydrophilic substitutions at P3 can destabilize the closed state (Sukhareva et al., 2003), and aromatic substitution of V6 (V6W) results in nonconducting channels (Hackos et al., 2002).
Since mutations within the Shaker activation gate have been shown to disrupt channel function and expression (Hackos et al., 2002; Kitaguchi et al., 2004), we hypothesized that the unusual substitutions observed in regulatory subunit gate sequences could play a role in restricting the functional stoichiometry of regulatory subunit-containing heteromers. To test this hypothesis, we chose mouse Kv2.1/Kv6.4 heteromers as a model system because analysis of gating current components supports a 3:1R stoichiometry (Bocksteins et al., 2017), and heteromeric channels can be functionally distinguished from Kv2.1 homomers based on gating properties (Fig. 3 and Table 1). In oocytes coexpressing Kv2.1 and Kv6.4, currents arising from Kv2.1:Kv6.4 heteromers can be differentiated from currents arising from Kv2.1 homomers based on closed state inactivation, which is present only in the heteromers (Fig. 3, B–D). Heteromer contributions to currents elicited by a depolarizing test pulse to +40 mV can be selectively eliminated with a prepulse to −40 mV (Fig. 3, C and D), allowing for functional estimates of heteromer:homomer ratios in oocytes expressing both channels.
TIRF microscopy reveals multiple Kv2.1:Kv6.4 heteromer stoichiometries
We used a TIRF microscopy single-molecule bleaching assay to determine the stoichiometry of Kv2.1:Kv6.4 heteromers. TIRF single-molecule bleaching assays have been successfully used to determine the subunit stoichiometry of many different types of ion channels and other membrane-associated proteins (Ulbrich and Isacoff, 2007; Nakajo et al., 2010; Coste et al., 2012; Upadhyay et al., 2016). We chose this approach because it allows identification and quantification of minor heteromer species that would likely be missed by macroscopic analyses such as intersubunit FRET and gating currents that have previously been used to look at Kv2.1:Kv9.3 and Kv2.1:Kv6.4 stoichiometry (Kerschensteiner et al., 2005; Bocksteins et al., 2017). In this technique, individual membrane surface–expressed channels containing fluorescently tagged subunits (visible as fluorescent dots) are bleached with a laser, and the number of bleaching steps is counted to infer fluorescent subunit numbers (see Methods and Ulbrich and Isacoff, 2007).
The N terminus of Shaker-like Kv channels is highly variable even among closely related channels, so we reasoned that N-terminal GFP fusions could be used for TIRF experiments with minimal effect on channel function. When we expressed GFP-Kv2.1 by itself in Xenopus oocytes to exclusively form Kv2.1 homotetramers, or Kv2.1 with GFP-Kv6.4 to introduce heteromer formation, we observed fluorescent spots that bleached in discrete steps (Fig. 4, A and B). For GFP-Kv2.1 homomers, fluorescent spots bleached in one, two, three, and four steps (Fig. 4 A). From the distribution of channels bleaching in one to four steps, we estimated the probability for detection of an individual GFP fluorophore to be ∼69% (Fig. 4 C). Coexpressing GFP-Kv2.1 with Kv6.4 significantly shifted the distribution toward fewer bleaching steps (P < 0.05, Fisher’s exact test), consistent with a reduction in the average number of Kv2.1 channels/subunit through heteromer formation (Fig. 4 D). Note that while the percentage of channels in this Kv2.1:Kv6.4 coexpression experiment bleaching in four steps is significantly reduced, their presence suggests some Kv2.1 homomers are still formed. Residual homomer formation makes it difficult to calculate the stoichiometry of the heteromers with GFP-Kv2.1. Therefore, we coexpressed WT Kv2.1 with GFP-Kv6.4 in a 1:50 ratio to maximize formation of heteromers and to selectively detect heteromers to calculate their stoichiometry. With the GFP tag on Kv6.4 instead of Kv2.1, spots predominantly bleached in one step (consistent with a 3:1R stoichiometry), but to our surprise, there were also a significant number of spots bleaching in two steps (Fig. 4, B and E). Using the GFP bleaching probability determined from the GFP-Kv2.1 experiment (0.69), 2:2R heteromers were estimated to represent ∼30% of all heteromers in this experiment (Fig. 4 F). The absence of spots bleaching in three to four steps is consistent with the idea that T1 incompatibility would block assembly of heteromeric channels with more than two Kv6.4 subunits. Coexpressing Kv2.1 with GFP-Kv6.4 in a 15:1 ratio decreases 2:2R significantly to 13% (P < 0.05, χ2 test), demonstrating that the proportion of 2:2R heteromers formed depends on the Kv2.1:GFP-Kv6.4 ratio and is therefore not an artifact of the bleaching step counting procedure. These results provide evidence that Kv2.1:Kv6.4 2:2R heteromers can form but also suggest that their formation is highly inefficient compared with 3:1R heteromers. If the two stoichiometries formed with equal probability, we would expect 2:2R heteromers to predominate at the 1:50 RNA ratio of Kv2.1:GFP-Kv6.4.
The S6 gate of Kv6.4 limits formation of 2:2R heteromers
We next examined the role of the Kv6.4 T1 domain and the Kv6.4 S6 activation gate in restricting the formation of 2:2R heteromers. In keeping with our prediction that T1 self-incompatibility should not be sufficient to block formation of 2:2R heteromers (Fig. 1), coexpressing Kv2.1 with a Kv6.4 chimera containing the self-compatible T1 of Kv2.1 (GFP-Kv6.4-Kv2.1T1; Fig. 5 A) did not significantly increase the formation of 2:2R heteromers as measured by TIRF (Fig. 5 B). However, when Kv2.1 was coexpressed with a Kv6.4 chimera containing the Kv2.1 S6 activation gate (PIPIIV) and C terminus (Kv6.4-PIPIIV-Kv2.1CT; Fig. 5 A), 2:2R heteromer formation was significantly increased to ∼60% (Fig. 5 B; P < 0.01, χ2 test). This represents an approximately fourfold increase in the efficiency of 2:2R formation relative to 3:1R formation, because it increases the 2:2R abundance from roughly 0.4 times to 1.5 times the 3:1R abundance. 2:2R formation with the GFP-Kv6.4-PIPIIV-Kv2.1CT chimera was concentration dependent and higher than 2:2R formation with WT Kv6.4 over a broad concentration range (Fig. 5 C). We included the Kv2.1 C terminus in this original gate chimera because the C terminus of Kv2.1 has previously been implicated in channel assembly (Mohapatra et al., 2008). A Kv6.4 chimera that retains the Kv6.4 activation gate (PATSIF) and includes only the Kv2.1 C terminus (GFP-Kv6.4-Kv2.1CT; Fig. 5 A) did not increase 2:2R formation (Fig. 5 D), specifically implicating the activation gate as the key region for increasing 2:2R formation within the Kv6.4-PIPIIV-Kv2.1CT chimera. Furthermore, the chimera Kv6.4-PIPIIV, which contains only the Kv2.1 activation gate, was also able to increase 2:2R formation (Fig. 5 D). Adding the Kv2.1 T1 to the Kv6.4-PIPIIV-Kv2.1CT (Kv6.4-Kv2.1T1-PIPIIV-Kv2.1CT) did not further increase 2:2R formation (Fig. 5 D) and did not form homotetramers as measured by TIRF (see Fig. 5 legend for step counts) or outward current recorded at +40 mV (143 ± 14 nA, n = 3). This is consistent with previous observations that the substitution of the Kv6.4 T1 and S6 with Kv2.1 sequences is not sufficient to confer homotetramer formation to Kv6.4 (Ottschytsch et al., 2005). The Kv6.4-Kv2.1T1-PIPIIV-Kv2.1CT chimera supports our hypothesis that the self-incompatible Kv6.4 T1 is not responsible for preferential assembly of Kv2.1:Kv6.4 3:1R heteromers over 2:2R heteromers, but also suggests that T1 is not entirely responsible for Kv6.4 subunit self-incompatibility.
We next investigated the role of the two most unusual residues of the Kv6.4 activation gate (T3 and F6; Fig. 2) in reducing 2:2R formation by introducing them back into the GFP-Kv6.4-PIPIIV-Kv2.1CT chimera. We chose this chimera instead of GFP-Kv6.4-PIPIIV because it displayed the highest 2:2R formation. When T3 and F6 were reintroduced into Kv6.4-PIPIIV-Kv2.1CT together or individually (Fig. 5 A), the proportion of 2:2R heteromers was statistically identical to the 2:2R proportion we observed for WT Kv6.4 (Fig. 5 D). These results as a whole show that the unusual T3 and F6 substitutions in the Kv6.4 S6 activation gate, and not T1 incompatibility, play a significant role in suppressing Kv2.1:Kv6.4 2:2R heteromer formation.
The Kv6.4 gate limits functional contributions of 2:2R heteromers
While our TIRF results suggest that 2:2R Kv2.1-Kv6.4 heteromers do form at a low rate, and that the efficiency of formation is increased by substitution of Kv2.1 gate residues into Kv6.4, they do not address the question of whether 2:2R heteromers incorporating either WT Kv6.4 or Kv6.4-PIPIIV-Kv2.1CT contribute significantly to macroscopic currents. To examine 2:2R current contributions, we first verified that currents generated by Kv2.1:Kv6.4-PIPIIV-Kv2.1CT heteromers, like currents from Kv2.1:Kv6.4 WT heteromers, can be distinguished from Kv2.1 homomer currents based on SSI (Fig. 6 and Table 1). In oocytes expressing a 1:10 ratio of Kv2.1:Kv6.4-PIPIIV-Kv2.1CT, the V50 of steady-state we measured was −82.32 ± 1.43 mV, compared with −62.56 ± 0.58 mV for oocytes expressing Kv2.1:Kv6.4 WT. Thus a −40 mV prepulse can also be used to distinguish the current fraction contributed by Kv2.1:Kv6.4-PIPIIV-Kv2.1CT heteromers from the Kv2.1 homomer current fraction. We do not think it is safe to attribute the increased hyperpolarized shift in SSI to the presence of functional 2:2R channels, because the altered S6 gate of Kv6.4-PIPIIV-Kv2.1CT could also directly impact the gating of 3:1R heteromers in unpredictable ways. We did not observe obvious changes in SSI properties based on the Kv2.1:Kv6.4-PIPIIV-Kv2.1CT and therefore did not examine SSI in detail for signs of multiple functionally distinct heteromeric species.
We instead measured total current size and calculated heteromer current size in oocytes injected with a constant amount of Kv2.1 cRNA titrated against increasing amounts of either Kv6.4 or Kv6.4-PIPIIV-Kv2.1CT cRNA to gain insights into the functionality of 2:2R heteromers. Assuming that tetrameric channels form as a dimer of dimers, as has been previously suggested (Tu and Deutsch, 1999), the availability of Kv2.1 homodimers required for 3:1R formation should drop as the expression of Kv6.4 increases and favors heterodimer formation. Therefore, the absolute number of 3:1R channels that can be formed should drop and become extremely small at highly Kv6.4-biased expression ratios. If 3:1R channels are the only functional heteromeric species and/or 2:2R formation is highly inefficient, as for Kv2.1:Kv6.4, then current size should approach 0 as the Kv6.4 expression ratio is increased. Fig. 7 A depicts predicted changes in the composition and relative abundance of surface-expressed channels in oocytes expressing Kv2.1 and Kv6.4 at various expression ratios incorporating the efficiency of 2:2R formation we determined in TIRF. In contrast, if both the 3:1R and 2:2R heteromers are functional, and 2:2R formation is more efficient as we observed for Kv2.1: Kv6.4-PIPIIV-Kv2.1CT using TIRF, the current size should drop off less with increasing regulatory subunit expression due to current flowing through the 2:2R heteromers. A prediction for the composition and relative abundance of surface-expressed channels for Kv2.1:Kv6.4-PIPIIV-Kv2.1CT at various expression ratios using TIRF data on 3:1R/2:2R ratios is shown in Fig. 7 B. Based on these predictions, we reasoned it would be possible to tell whether Kv2.1:Kv6.4-PIPIIV-Kv2.1CT 2:2R heteromers are functional if current is significantly increased relative to Kv2.1:Kv6.4 in oocytes injected with high concentrations of the Kv6.4 species relative to Kv2.1. It would be difficult to make conclusions regarding 2:2R functionality for Kv2.1:Kv6.4 from an expression ratio titration alone, because the absolute number of 2:2R channels is predicted to be very small, and thus current might approach 0 at high Kv6.4 concentrations regardless of 2:2R functionality (Fig. 7 A).
A comparison of current sizes for experimental titrations of WT Kv6.4 and Kv6.4-PIPIIV-Kv2.1CT cRNA against a constant amount of Kv2.1 cRNA is shown in Fig. 8, A and B. During electrophysiological recordings, oocytes were held at −100 mV for 500 ms, followed by a 4-s hyperpolarizing prepulse to −120 mV to relieve closed state inactivation, and then pulsed to +40 mV for 500 ms to measure current size. With Kv6.4, total current elicited by a +40 mV depolarization dropped rapidly as Kv6.4 cRNA levels were increased and approached 0 at a Kv6.4:Kv2.1 100:1 ratio (Fig. 8 C). Current levels also decreased with increasing Kv6.4-PIPIIV-Kv2.1CT titration but were significantly higher across almost the entire titration, with 43 ± 3.7% of the control current level remaining at a 1:50 ratio of Kv2.1:Kv6.4-PIPIIV-Kv2.1CT, the highest expression level we tested (Fig. 8 C). cRNA ratios differed slightly between the two titrations, so we statistically compared means from the closest cRNA ratios (Fig. 8 C, brackets). The current size in the Kv2.1:Kv6.4-PIPIIV-Kv2.1CT titration compared with the Kv2.1:Kv6.4 titration increased from 1.06 ± 0.22-fold (mean ± SD) at low expression ratios to 4.22 ± 0.60-fold (mean ± SD) at high ratios (Fig. 8 D), suggesting that 2:2R heteromers that should become more abundant at high ratios are significantly contributing to currents in the Kv2.1:Kv6.4-PIPIIV-Kv2.1CT titration.
We do not think that differences in single-channel conductance Kv2.1:Kv6.4 and Kv2.1:Kv6.4-PIPIIV-Kv2.1CT 3:1R heteromers could explain the differences in the titrations. We compared single-channel conductance for Kv2.1 homomers and Kv2.1:Kv6.4 heteromers expressed in Xenopus oocytes using on-cell patches. We observed channels with a conductance of 8.3 ± 0.7 pS in oocytes expressing only Kv2.1 (Fig. 8, E and G). In oocytes expressing Kv2.1 and Kv6.4, we observed channels with conductance statistically identical to these Kv2.1 homomers and one additional group of channels with an approximately one-third smaller conductance, which presumably represent Kv2.1:Kv6.4 heteromers (Fig. 8, F and G). This small single-channel conductance drop would predict some dropoff in current at points in the titration where 3:1Rs are predicted to be abundant, but it is not sufficient to explain the near complete loss of current at high Kv6.4 cRNA concentrations. The latter current loss likely corresponds to the expected reduction in 3:1R formation at high Kv6.4 concentrations that is not compensated by current flowing through 2:2R heteromers because of their low abundance and/or reduced functionality. If the difference we observed in whole cell current size in the Kv2.1:Kv6.4 and Kv2.1:Kv6.4-PIPIIV-Kv2.1CT titrations at highly Kv6.4-biased expression ratios simply reflected a difference in single-channel conductance between 3:1R heteromers, then we would expect that the fold difference in current size across the two titration curves to remain constant, unlike the approximately fourfold change we observed across the titration (Fig. 8 D). We did not pursue single-channel analysis for Kv2.1:Kv6.4-PIPIIV-Kv2.1CT because we reasoned that the introduced gate changes might lessen differences in single-channel conductance, making it difficult to detect multiple heteromer species and unsafe to assign observed conductances to specific stoichiometries. While significant current remains at high concentrations of Kv6.4-PIPIIV-Kv2.1CT cRNA, the current level does diminish >50%. This dropoff could be due to remaining inefficiencies in 2:2R formation (we observed a maximum of ∼60% 2:2R heteromers in TIRF) and/or a modest reduction in the single-channel conductance of the 2:2R heteromers relative to 3:1R heteromers and Kv2.1 homomers.
The titration results might also be explained if Kv6.4 WT was translated far more efficiently than Kv6.4-PIPIIV-Kv2.1CT, offsetting the current size reduction in the Kv2.1:Kv6.4-PIPIIV-Kv2.1CT titration. However, we would expect such an offset would also offset the cRNA expression ratio required to achieve a given percentage of heteromeric current. We therefore estimated the contribution of heteromeric current to the total current measured at each point in the titrations using a prepulse to −40 mV to selectively inactivate heteromers. Because SSI of Kv2.1:Kv6.4 heteromers is incomplete, it leaves a pedestal current of a predicted size (Fig. 3 D, Fig. 6 D, and Table 1). We therefore calculated the size of the heteromeric current as the size of the current removed by the −40 mV prepulse plus the predicted size of the corresponding pedestal. Thus, the heteromeric current amplitude was calculated as 1.38 times and 1.44 times the size of the current inactivated by the −40 prepulse for Kv6.4 WT and Kv6.4-PIPIIV-Kv2.1CT, respectively. Plots of total current versus calculated heteromeric current (normalized to the homomeric Kv2.1 controls) are shown for the Kv6.4 WT and Kv6.4-PIPIIV-Kv2.1CT titrations in Fig. 8, H and I. Both titrations approach 100% heteromeric current at a 1:10 Kv2.1:Kv6.4 cRNA ratio, suggesting that there was no significant offset in protein subunit translation that could explain the differences between current size in the titrations. Therefore, we favor differences in of the efficiency of 2:2R formation, and possibly also in 2:2R conduction, as the most likely explanations for the differences between current size in the two expression titrations.
Reduced Cd2+ block of Kv2.1 I383C:Kv6.4-PIPIIV-Kv2.1CT heteromers supports 2:2R function
While the cRNA titration experiments shown in Fig. 8 suggest Kv2.1:Kv6.4-PIPIIV-Kv2.1CT 2:2R heteromers conduct and significantly contribute to macroscopic currents, the evidence is indirect. Furthermore, the data are not conclusive for determining whether Kv2.1:Kv6.4 2:2R heteromers are functional. To test functionality of the 2:2R heteromers in a more direct manner, we took advantage of the phenomenon of Cd2+ coordination in a Kv2.1 pore cysteine mutant. Rat Kv2.1 I379C is blocked by external cadmium, and block requires coordination by two I379C residues (Krovetz et al., 1997). Furthermore, rat Kv2.1 WT:Kv2.1 I379C heteromers are blocked by Cd2+ only if the two cysteines are present in two adjacent subunits; diagonally opposed cysteines do not support cadmium block (Krovetz et al., 1997). We hypothesized that if this mutant (I383C in mouse Kv2.1) was coexpressed with Kv6.4, 3:1R heteromers, which would have adjacent Kv2.1 I383C subunits, should be Cd2+-sensitive, while 2:2R heteromers, which would have only diagonally opposed Kv2.1 I383C subunits, would be Cd2+-resistant (Fig. 9 A). If 2:2R heteromers contribute to the heteromeric current, then we would expect less cadmium block as the Kv6.4 expression ratio increases. We therefore tested Cd2+-sensitivity of Kv2.1 I383C:Kv6.4 WT and Kv2.1 I383C:Kv6.4-PIPIIV-Kv2.1CT at multiple expression ratios to look for evidence of functional 2:2R heteromers.
Kv2.1, Kv2.1:Kv6.4 WT (1:10), and Kv2.1:Kv6.4-PIPIIV-Kv2.1CT (1:10) were all blocked <15% by 500 µM Cd2+ (Fig. 9, B and C) compared with >75% block observed for Kv2.1 I383C (Fig. 9 D). While Cd2+ block of the WT heteromers was significantly greater than Kv2.1 homomers, the degree of block was small compared with Kv2.1 I383C and thus does not interfere with detection of enhanced block in channels containing adjacent Kv2.1 I383C subunits. 200 µM DTT increased Kv2.1 I383C current size by 17.9 ± 2.2% (n = 9, P < 0.05, t test), suggesting that partial basal oxidation may interfere with conduction. Therefore, we included 200 µM DTT in all experiments to remove basal oxidation as a potential confounding factor for comparisons of Cd2+-sensitivity. We then examined 500 µM Cd2+ block of Kv2.1 I383C:Kv6.4 WT heteromers at varying expression ratios (Fig. 9, D and E). Cd2+ significantly slowed the activation of heteromeric channels, but we did not explore the mechanism. We increased test pulse duration to allow currents to reach steady-state for current amplitude measurements. Kv2.1 I383C:Kv6.4 heteromers were blocked to a similar extent as Kv2.1 I383C homomers (∼75%), and the degree of block was insensitive to Kv2.1 I383C:Kv6.4 WT expression ratio. In contrast, Cd2+-block in Kv2.1 I383C:Kv6.4-PIPIIV-Kv2.1CT was sensitive to the expression ratio (Fig. 9, D and E). Block was similar to Kv2.1 I383C homomers and Kv2.1 I383C:Kv6.4 WT heteromers with lower concentrations of Kv6.4-PIPIIV-Kv2.1CT cRNA, but the current became significantly more resistant to Cd2+ at high Kv6.4-PIPIIV-Kv2.1CT cRNA concentrations (Fig. 9 E). Some Cd2+ block remained even at the most biased Kv6.4-PIPIIV-Kv2.1CT expression ratios, consistent with our TIRF experiments, which show that ∼40% of the channels at these ratios should still be Cd2+-sensitive 3:1R heteromers. Full block by 500 µM Cd2+ could be restored by insertion of an equivalent pore cysteine (M422C) into the Kv6.4-PIPIIV-Kv2.1CT chimera (Fig. 9 F), which ensures that all heteromer stoichiometries produced by coexpression with Kv2.1 I383C would have subunits with adjacent cysteines. The results of these Cd2+ block experiments provide a strong line of evidence that Kv2.1:Kv6.4-PIPIIV-Kv2.1CT 2:2R heteromers contribute significantly to macroscopic currents. In contrast, Kv2.1:Kv6.4 2:2R heteromers may not contribute significantly to whole cell currents, because at a 1:10 Kv2.1:Kv6.4 ratio (−log [Kv2.1 fraction] = 1), where virtually all the current is expected to flow through heteromers (Fig. 8 H), and ∼20% of the heteromers should be 2:2Rs (Fig. 5 C), we saw no increase in Cd2+ resistance. However, because the low 2:2R percentage provides only a small dynamic range for detection of Cd2+ resistance, it is not possible to differentiate whether Kv2.1:Kv6.4 2:2R heteromers are completely nonfunctional or have a significantly reduced single-channel conductance relative to Kv2.1:Kv6.4 3:1R heteromers.
Our results agree with previous work on Kv6.4 and Kv9.3 that suggest the predominant stoichiometry for mammalian Kv2 subfamily regulatory subunit-containing heteromers is 3:1R (Kerschensteiner et al., 2005; Bocksteins et al., 2017). However, our use of a single-molecule technique (TIRF photobleaching) here revealed that 2:2R heteromers are also formed. Because 2:2R heteromers represent less than a third of all heteromers even at expression ratios highly biased in favor of Kv6.4, 2:2R formation appears to be inefficient and would likely not have been detected in previous studies using macroscopic analysis techniques such as FRET (Kerschensteiner et al., 2005) or gating current components (Bocksteins et al., 2017). T1 self-incompatibility is a key feature of Kv2 subfamily regulatory subunits that helps prevent homophilic subunit interactions, and should rule out formation of 4R, 1:3R, and 2:2R tetramers with adjacent regulatory subunits. However, T1 self-incompatibility should not block formation of 2:2R heterotetramers with diagonally opposed regulatory subunits, the major 2:2R species that would be expected to assemble from Kv2.1:Kv6.4 heterodimers (Fig. 1 E). Our results agree with these expectations for the role of T1 self-incompatibility in assembly of Kv2.1:Kv6.4 heteromers: only 3:1R and 2:2R heteromers were detected in TIRF. However, the ratio of these two stoichiometries was unaffected when the self-incompatible Kv6.4 T1 was replaced with the Kv2.1 T1 (Fig. 5, A and D), suggesting that T1 does not play a major role in driving preferential 3:1R assembly for Kv2.1:Kv6.4 and may not be fully responsible for Kv6.4 subunit self-incompatibility.
While we observed formation of 2:2R Kv2.1:Kv6.4 heteromers, our results suggest that only the more prevalent 3:1R stoichiometry contributes significantly to macroscopic heteromeric currents (Figs. 8 and 9). The 2:2Rs are low abundance and may also have an attenuated single-channel conductance. The unusual activation gate of Kv6.4 and its P3T and V6F substitutions play a key role in limiting 2:2R formation and conduction (Figs. 5, 8, and 9). We did not experimentally examine the mechanism through which these mutations restrict stoichiometry, but we made draft homology models of the pore of Kv2.1, Kv2.1:Kv6.4 heteromers (3:1R and 2:2R) and Kv6.4-PIPIIV (2:2R) to try to gain insights. Open models and closed models were based on the Kv1.2 computational models presented by Pathak et al. (2007). The PD (S5–S6) includes numerous signpost residues universally conserved among Kv1.2, Kv2.1, and Kv6.4 that allow for precise structural alignment (Fig. 10 A). Views of the closed conduction pathway from an extracellular perspective, color-coded by hydrophobicity (Fig. 10, B–E), suggest a feature of the Kv6.4 activation gate that could potentially impact 2:2R heteromer formation. In the Kv2.1 homomer model, the hydrophobic inner gates likely form a vapor lock around the conduction pathway that excludes water from the intersubunit interfaces and allows for strong hydrophobic interactions between subunits (Fig. 10 B). A vapor lock securing a closed state of a channel is a phenomenon common among different unrelated channel families of pro- and eukaryotes, and can be also observed in model hydrophobic pores (Beckstein and Sansom, 2003; Roth et al., 2008; Anishkin et al., 2010). Surface tension of water meniscus above and below the “dry” gate is expected to provide a force contracting the pore walls together, stabilizing the pore assembly. In contrast, a hydrophilic cleft bisects the gate area in Kv2.1:Kv6.4 2:2R heteromers, breaking the vapor lock (Fig. 10 D). With the single Kv6.4 subunit in the 3:1R heteromer, hydrophobicity is locally reduced, but there is no trans-pore hydrophilic cleft (Fig. 10 C). When Kv6.4-PIPIIV is substituted for WT Kv6.4 in the 2:2R model, the hydrophilic cleft is eliminated, and the vapor lock is restored (Fig. 10 E). This suggests that the hydrophilic cleft in the 2:2R Kv2.1/Kv6.4 heteromers could potentially weaken tetramer stability and may play a role in favoring formation of 3:1R heteromers over 2:2R heteromers even at high Kv6.4 concentrations.
In addition, intracellular views of the pore at the activation gate show that the combination of T3 and F6 of the Kv6.4 gate significantly narrows the conduction pathway in the open state, especially in 2:2R heteromers (Figs. 10, F–M). The opening at the level of F6 in the 2:2R heteromer would require significant dehydration of a K+ ion for it to pass, suggesting that conduction for 2:2R heteromers could be energetically unfavorable compared with 3:1R heteromers. The conduction pathway diameter in the 2:2R Kv2.1:Kv6.4-PIPIIV heteromer model, in contrast, has a similar diameter to that of the Kv2.1 homomer model. However, the V6W mutation in Shaker and Kv1.2 does not completely block conduction but instead may energetically trap channels in the closed state, propagating structural changes to the selectivity filter that alter gating or conduction (Kitaguchi et al., 2004; Pau et al., 2017). Shaker V6W channels do conduct in a diagonal 2 WT:2 V6W configuration (albeit poorly), and conduction in homomeric V6W channels can be restored by a compensatory gate mutation P3Q that stabilizes the open state (Sukhareva et al., 2003; Kitaguchi et al., 2004). It should also be noted that because of the ability of individual pore mutations to propagate structural rearrangements broadly through the pore (Pau et al., 2017), the functional consequences of specific gate substitutions might vary significantly from channel to channel. The significance of the steric barrier observed in the Kv2.1:Kv6.4 2:2R model therefore requires experimental clarification. What the models do show is that the Kv6.4 gate will likely perturb pore structure, and that the degree of perturbation increases significantly when two Kv6.4 subunits participate in the gate formation (Fig. 10).
Because we were not able to reach 100% 2:2R formation with the Kv6.4-PIPIIV-Kv2.1CT or Kv6.4-Kv2.1T1-PIPIIV-Kv2.1CT chimeras even at highly biased expression ratios, regions of the channel other than T1 and the gate are also likely to play a role in suppression of the 2:2R stoichiometry. The inability of the Kv2.1 T1 to restore assembly of channels with more than two regulatory subunits in the Kv6.4-PIPIIV-Kv2.1CT background further suggests that other channel regions could also contribute to Kv6.4 self-incompatibility. A broader analysis of pore interface residues that differ between Kv2.1 and Kv6.4 might reveal additional factors limiting 2:2R formation. We focused on the gate here because of the consistent and striking evolutionary divergence of the gate sequence among regulatory subunits (Fig. 2). It is also possible that 2:2R formation could be limited over the expression ratio range we tested if Kv2.1 simply has higher affinity for itself than Kv6.4, independent of any structural perturbations. We did not test higher expression ratios here because the concentrations of cRNA required compromised oocyte health. However, if Kv2.1 homodimerization is preferred to Kv2.1:Kv6.4 heterodimerization, our data suggest that the T1 domain alone does not account for the affinity difference because we were unable to increase 2:2R formation by substituting the Kv2.1 T1 into Kv6.4 or Kv6.4-PIPIIV-Kv2.1CT (Fig. 5 D). Furthermore, differential affinity does not contradict the role we found for the Kv6.4 gate in limiting 2:2R formation.
Based on the results presented here, we propose a two-step model for evolution of the regulatory phenotype in Shaker-like Kv family channels that predicts degeneration of the highly conserved activation gate sequence and opens an evolutionary path to a 3:1R stoichiometry (Fig. 11). In the first step, the regulatory phenotype, which is characterized by an inability to form homotetramers, is established by a self-incompatibility mutation, most likely in T1 as depicted here (Fig. 11 B). Both 3:1R and 2:2R stoichiometries are predicted to be functional because T1 self-incompatibility mutations are unlikely to affect channel function beyond assembly. The self-incompatible mutated channel produced in step 1 should remain under positive selection to maintain functional channel numbers. The original self-incompatibility mutation could in theory occur outside T1, provided it does not disrupt heteromer function. While the precise nature of the self-incompatibility mutation(s) in mammalian Kv2 subfamily regulatory subunits is not clear, the T1 is a likely spot because their T1 domains are self-incompatible (Salinas et al., 1997b; Kramer et al., 1998; Ottschytsch et al., 2002). Furthermore, a single amino acid change in the T1 of Kv1.3 is sufficient to create T1 self-incompatibility while preserving the ability to form heteromers with WT channels (Robinson and Deutsch, 2005). Nevertheless, our results indicate that Kv6.4 self-incompatibility extends beyond T1. Regardless of the location of the original step 1 mutation, self-incompatibility limits functional channels to no more than two mutated subunits/tetramer (Fig. 1). Selective pressure on the gate sequence is therefore reduced, and the gate can now accumulate mutations that can be functionally accommodated in no more than two subunits of a tetramer. In step 2 of the model, this evolutionary drift of the gate sequence over time could eventually result in mutations that are tolerated only in a single subunit, thus fixing the functional stoichiometry at 3:1R (Fig. 11 B). While some subunits in step 2 may be lost to nonfunctional stoichiometries, as we observe for Kv2.1:Kv6.4, advantageous new biophysical properties could maintain positive selection. The physiological advantage of a heteromeric delayed rectifier with a low activation threshold, slower kinetics, and closed state inactivation (Fig. 3) in neurons expressing Kv6.4 remains to be determined. We believe it is unlikely that evolution of the regulatory phenotype would begin with a restrictive gate mutation in step 1, because gate incompatibility in the presence of intact assembly would tie most of the expressed subunits up in nonfunctional tetramers (Fig. 11 C). The gate mutation would therefore serve as a strong dominant negative and would likely face immediate negative selection. This may help explain why the gate is so highly conserved in self-compatible homomer-forming Shaker-like Kv subunits.
This evolutionary model predicts the gate consensus sequence degeneration we observe in all independent evolutions of the regulatory subunit phenotype within the Shaker-like Kv family. A 3:1R stoichiometry could be the most likely long-term outcome because it is logical that the gate would eventually acquire a mutation that can be tolerated in only a single subunit. While maintenance of two functional stoichiometries might initially be selectively neutral, fixation on a single functional stoichiometry could gain a selective advantage over time as the regulatory subunit accumulates mutations that would functionally differentiate 3:1R from 2:2R channels. Nevertheless, the model does not require that each evolutionary instance of the regulatory subunit phenotype will fix at 3:1R, or that the observed regulatory subunit group will have reached a point of fixation. We believe 2:2R fixation should be less common because (a) the regulatory phenotype is most likely first established with a loss-of-function mutation in assembly, and (b) it is unlikely that any gate mutations are better tolerated in a 2:2R configuration than a 3:1R configuration. Fixation on a 2:2R stoichiometry is possible within this model if this stoichiometry has unique, positively selected functional properties, but it might require secondary mutation(s) that favor heterodimer formation during assembly. Once a 2:2R stoichiometry is favored, gate mutations tolerated in only a single subunit would be selected against. A cnidarian Kv4 regulatory subunit jShalγ with only a single gate substitution (PVPIIQ) has been predicted to form functional channels in a 2:2R stoichiometry based on examination of the stoichiometry of N-type inactivation (Jegla and Salkoff, 1997), but in the absence of TIRF photobleaching analysis, it is unclear if 3:1R heteromers do or do not form.
Further functional characterization of independent evolutionary instances of regulatory subunits will be needed to test whether this model predictably describes the evolution of the regulatory subunit phenotype across the Shaker-like Kv family. It predicts that self-incompatibility should be a universal feature of regulatory subunits, and that an exclusive 3:1R functional stoichiometry should be common. It would also be interesting to see if there is correspondence between the predicted severity of gate mutations and regulatory subunit stoichiometry. Regardless, degeneration of the highly conserved activation gate sequence appears to have high predictive value for identifying evolution of the regulatory phenotype in Shaker-like Kv family. Another example of preferential assembly of a 3:1R stoichiometry in the voltage-gated cation channel superfamily is the heteromeric rod cyclic nucleotide-gated channels (Weitz et al., 2002; Zheng et al., 2002; Zhong et al., 2002). These channels lack the Shaker T1, but a cytoplasmic coiled-coil domain in the C terminus seems to play an analogous role; it is formed by three helices, one each contributed by the CNG1A subunits, while the regulatory CNGB1 is excluded and thus limited to a single subunit/channel. Our results suggest it would be interesting to look for coevolution of gate sequences and the regulatory phenotype across the breadth of the CNG channel family.
W.O. Hancock and K.J. Mickolajczyk were supported by National Institutes of Health grant R01GM076476. K.J. Mickolajczyk was also supported by National Institutes of Health grant 1F99CA223018-01. X. Li and W. Horton were supported by National Institutes of Health grant R01NS069842.
The authors declare no competing financial interests.
Author contributions: A. Pisupati, K.J. Mickolajczyk, W. Horton, and X. Li designed experiments and analyzed data. J. Chu-Luo and G. Busey constructed molecular reagents. A. Anishkin, S.V. Chintapalli, and D.B. van Rossum performed molecular modeling. A. Pisupati and T. Jegla conceived the study and wrote the manuscript with contributions from A. Anishkin, D.B. Van Rossum, K.J. Mickolajczyk, and W.O. Hancock.
Merritt C. Maduke served as editor.