A primary goal of molecular physiology is to understand how conformational changes of proteins affect the function of cells, tissues, and organisms. Here, we describe an imaging method for measuring the conformational changes of the voltage sensors of endogenous ion channel proteins within live tissue, without genetic modification. We synthesized GxTX-594, a variant of the peptidyl tarantula toxin guangxitoxin-1E, conjugated to a fluorophore optimal for two-photon excitation imaging through light-scattering tissue. We term this tool EVAP (Endogenous Voltage-sensor Activity Probe). GxTX-594 targets the voltage sensors of Kv2 proteins, which form potassium channels and plasma membrane–endoplasmic reticulum junctions. GxTX-594 dynamically labels Kv2 proteins on cell surfaces in response to voltage stimulation. To interpret dynamic changes in fluorescence intensity, we developed a statistical thermodynamic model that relates the conformational changes of Kv2 voltage sensors to degree of labeling. We used two-photon excitation imaging of rat brain slices to image Kv2 proteins in neurons. We found puncta of GxTX-594 on hippocampal CA1 neurons that responded to voltage stimulation and retain a voltage response roughly similar to heterologously expressed Kv2.1 protein. Our findings show that EVAP imaging methods enable the identification of conformational changes of endogenous Kv2 voltage sensors in tissue.
Introduction
To move the field of voltage-sensitive physiology forward, we need new tools that indicate when and where voltage-sensitive conformational changes in endogenous proteins occur. Many classes of transmembrane proteins have been found to be voltage sensitive (Bezanilla, 2008). One important class of voltage-sensitive proteins is the voltage-gated ion channels. Electrophysiological techniques have enabled remarkably precise studies of the voltage sensitivity of ionic conductances, primarily under reductionist experimental conditions where the channels have been removed from their native tissue. In addition to their canonical function as ion-conducting channels, voltage-gated ion channel proteins have nonconducting functions that are independent of their ion conducting functions (Tanabe et al., 1988; Kaczmarek, 2006). These nonconducting protein functions are largely inaccessible to study by electrophysiology and are more poorly understood. Novel approaches are needed to learn more about voltage sensing in intact tissues and to unlock the mysterious realm of nonconducting voltage-sensitive physiology. Here, we present a new approach to image the conformational changes of endogenous Kv2 voltage-sensitive proteins.
Kv2 proteins form voltage-gated K+ channels (Frech et al., 1989), bind endoplasmic reticulum proteins to form plasma membrane–endoplasmic reticulum junctions (Johnson et al., 2018; Kirmiz et al., 2018a), regulate a wide variety of physiological responses in tissues throughout the body (Bocksteins, 2016), and integrate their response to voltage with many other cellular processes, including phosphorylation (Murakoshi et al., 1997), SUMOylation (Plant et al., 2011), oxidation (MacDonald et al., 2003), membrane lipid composition (Ramu et al., 2006), and auxiliary subunits (Gordon et al., 2006; Peltola et al., 2011). Kv2 proteins are members of the voltage-gated cation channel superfamily. The voltage sensors of proteins in this superfamily comprise a bundle of four transmembrane helices termed S1–S4 (Long et al., 2005, 2007). The S4 helix contains positively charged arginine and lysine residues, gating charges, that respond to voltage changes by moving through the transmembrane electric field (Aggarwal and MacKinnon, 1996; Seoh et al., 1996; Islas and Sigworth, 1999). When voltage sensor domains encounter a transmembrane voltage that is more negative on the inside of the cell membrane, voltage sensors are biased toward resting conformations, or down states, in which gating charges are localized intracellularly. When voltage becomes more positive, gating charges translate toward the extracellular side of the membrane, and voltage sensors are progressively biased toward up states in a process of voltage activation (Armstrong and Bezanilla, 1973; Zagotta et al., 1994; Tao et al., 2010; Xu et al., 2019). Channel pore opening is distinct from, but coupled to, voltage sensor movement. In some voltage-gated ion channel proteins, voltage sensor movement is coupled to nonconducting protein functions (Tanabe et al., 1988; Kaczmarek, 2006). To study the functional outputs of voltage sensors, it is essential to measure voltage sensor activation itself. Conformational changes of voltage sensors have been detected with electrophysiological measurements of gating currents (Armstrong and Bezanilla, 1973; Schneider and Chandler, 1973; Bezanilla, 2018) or by optical measurements from fluorophores inserted near voltage sensors by genetic encoding (Lin and Schnitzer, 2016) or chemical modification (Zhang et al., 2015). However, the following experimental limitations prevent these existing techniques from measuring conformational changes of voltage sensors of most endogenous proteins: gating currents can only be measured when the proteins are expressed at high density in a voltage-clamped membrane; engineered proteins differ from endogenous channels; most chemical modification strategies result in off-target labeling; and conjugation of fluorophores into voltage sensors irreversibly alters structure and function. Here, we develop a different strategy to reveal conformational states of Kv2 proteins.
To image where in tissue the voltage sensors of Kv2 proteins adopt a specific resting conformation, we exploited the conformation-selective binding of the tarantula peptide guangxitoxin (GxTX)-1E, which can be conjugated to fluorophores to report Kv2 conformational changes (Tilley et al., 2014; Fletcher-Taylor et al., 2020; Stewart et al., 2021). Here, we synthesize GxTX-594, a Ser13Cys GxTX variant conjugated to Alexa Fluor 594, a fluorophore compatible with two-photon excitation imaging through light-scattering tissue. GxTX-594 dynamically binds Kv2 channels in living tissue. When GxTX-594 binds, it becomes immobilized and fluorescently labels Kv2 proteins at the cell surface. When Kv2 channels become voltage activated, GxTX-594 unbinds, resulting in unlabeling (see illustration). This labeling/unlabeling dynamic is similar to a recently reported point accumulation for imaging of nanoscale topology superresoluton imaging method (Legant et al., 2016), yet the method reported here is sensitive to changes in protein conformation. GxTX-594 labeling of Kv2 proteins equilibrates on the time scale of seconds, revealing the probability (averaged over time) that unbound voltage sensors are resting or active. Here, we develop a method to calculate the average conformational status of unlabeled Kv2 proteins from images of GxTX-594 fluorescence and deploy the GxTX-594 probe in brain slices to image voltage-sensitive fluorescence changes that reveal conformational changes of endogenous neuronal Kv2 proteins. We refer to this type of imaging tool as an Endogenous Voltage-sensor Activity Probe (EVAP). This EVAP approach provides an imaging technique to study conformational changes of endogenous voltage-sensitive Kv2 proteins in samples that have not (or cannot) be genetically modified.
Materials and methods
GxTX-594 synthesis
We used solid-phase peptide synthesis to generate a variant of GxTX, an amphiphilic 36-amino acid cystine knot peptide. We synthesized the same peptide used for GxTX-550, Ser13Cys GxTX, where a free thiolate side chain of cysteine 13 is predicted to extend into extracellular solution when the peptide is bound to a voltage sensor (Tilley et al., 2014). GxTX-1E folds by formation of three internal disulfides, and cysteine 13 was differentially protected during oxidative refolding to direct chemoselective conjugation. Following refolding and thiol deprotection, Alexa Fluor 594 C5 maleimide was condensed with the free thiol, and Ser13Cys (Alexa Fluor 594) GxTX-1E (called GxTX-594) was purified (Fig. S1).
The Ser13Cys GxTX peptide was synthesized as previously described (Tilley et al., 2014). Methionine 35 of GxTX was replaced by the oxidation-resistant noncanonical amino acid norleucine to avoid complications from methionine oxidation, and serine 13 was replaced with cysteine to create a spinster thiol. Ser13Cys GxTX was labeled with a Texas Red derivative (Alexa Fluor 594 C5 maleimide, cat. #10256; Thermo Fisher Scientific) to form GxTX-594. Ser13Cys GxTX lyophilisate was brought to 560 μM in 50% acetonitrile (ACN) + 1 mM Na2EDTA. 2.4 μl of 1M Tris (pH 6.8 with HCl), 4 μl of 10 mM Alexa Fluor 594 C5 maleimide in DMSO, and 17.9 μl of 560 μM Ser13Cys GxTX were added for a final solution of 100 mM Tris, 1.6 mM Alexa Fluor 594 C5 maleimide, and 0.4 mM GxTX in 24 μl of reaction solution. Reactants were combined in a 1.5-ml low-protein–binding polypropylene tube (LoBind, cat. #022431081; Eppendorf) and mixed at 1,000 rpm at 20°C for 4 h (Thermomixer 5355 R; Eppendorf). After incubation, the tube was centrifuged at 845 RCF for 10 min at room temperature. A purple pellet was observed after centrifugation. The supernatant was transferred to a fresh tube and centrifuged at 845 RCF for 10 min. After this second centrifugation, no visible pellet was seen. The supernatant was injected onto a reverse-phase HPLC C18 column (Biobasic 4.6-mm RP-C18 5 μm, cat. #2105-154630; Thermo Fisher Scientific) equilibrated in 20% ACN, 0.1% trifluoroacetic acid (TFA) at 1 ml/min, and eluted with a protocol holding in 20% ACN for 2 min, increasing to 30% ACN over 1 min, then increasing ACN at 0.31% per minute. HPLC effluent was monitored by fluorescence and an absorbance array detector. 1-ml fractions were pooled based on fluorescence (280-nm excitation, 350-nm emission) and absorbance (214 nm, 280 nm, and 594 nm). GxTX-594 peptide–fluorophore conjugate eluted at ∼35% ACN, and mass was confirmed by mass spectrometry using a Bruker ultrafleXtreme matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF; Fig. S1). Samples for identification from HPLC eluant were mixed 1:1 in an aqueous solution of 25% MeOH and 0.05% TFA saturated with α-cyano-4-hydrocinnamic acid, pipetted onto a ground-steel plate, dried under vacuum, and ionized with 60–80% laser power. Molecular species were detected using a reflector mode protocol and quantitated using Bruker Daltonics flexAnalysis 3.4 software. Lyophilizate containing GxTX-594 conjugation product was dissolved in cell external (CE) buffer (defined below) and stored at −80°C. GxTX-594 concentration was determined by 280-nm absorbance using a calculated molar attenuation coefficient of 18,900 M−1 cm−1.
Chinese hamster ovary (CHO) cell methods
CHO cell culture and transfection
The CHO-K1 cell line (American Type Culture Collection) and a subclone transfected with a tetracycline-inducible rat Kv2.1 construct (Kv2.1-CHO; Trapani and Korn, 2003) were cultured as described previously (Tilley et al., 2014). To induce Kv2.1 expression in Kv2.1-TREx-CHO cells, 1 μg/ml minocycline (cat. #ALX-380-109-M050; Enzo Life Sciences), prepared in 70% ethanol at 2 mg/ml, was added to the maintenance media to induce Kv2.1 expression. Minocycline was added 40–48 h before imaging and voltage-clamp fluorometry experiments. Minocycline was added 1–2 h before whole-cell ionic current recordings to limit K+ conductance such that voltage clamp could be maintained. Transfections were achieved with Lipofectamine 2000 (cat. #1668027; Life Technologies). 1.1 μl Lipofectamine was diluted, mixed, and incubated in 110 μl of Opti-MEM (product no. 31985062, lot no. 1917064; Gibco-BRL) in a 1:100 ratio for 5 min at room temperature. Concurrently, 1 μg of plasmid DNA and 110 μl of Opti-MEM were mixed in the same fashion. DNA and Lipofectamine 2000 mixtures were mixed and left at room temperature for 20 min. Then, the transfection cocktail mixture was added to 2 ml of culture media in a 35-mm cell culture dish of CHO cells at ∼40% confluency and allowed to settle at 37°C in 5% CO2 for 4–6 h before the media were replaced. Cells were given 40–48 h recovery following transfection before being used for experiments. Rat Kv2.1-GFP (Antonucci et al., 2001), rat Kv2.2-GFP (Kirmiz et al., 2018b), rat Kv1.5-GFP (Li et al., 2001), rat Kv4.2-GFP (Shibata et al., 2003), mouse BK-GFP, rat KvBeta2 (Shibata et al., 2003), and rat KCHiP2 (An et al., 2000) plasmids were all gifts from James Trimmer (University of California, Davis, Davis, CA). Identities of constructs were confirmed by sequencing from their cytomegalovirus promoter. To minimize any day-to-day variances, the cells in experiments shown in Fig. 4 or Fig. S2 were each plated for all transfections from a single-cell suspension, transfected in parallel, and imaged 2 d later using the same thawed aliquot of GxTX-594.
Confocal and Airy disk imaging
Confocal images were obtained with an inverted confocal system (LSM 880 410900-247-075; Zeiss) run by ZEN Black 2.1 software. A 63×/1.40 NA oil DIC objective (420782-9900-799; Zeiss) was used for most imaging experiments; a 63×/1.2 NA water DIC objective (441777-9970-000; Zeiss) was used for voltage clamp fluorometry experiments. GFP and YFP were excited with the 488-nm line from an argon laser (3.2 mW at installation) powered at 0.5% unless otherwise noted. GxTX-594 was excited with a 594-nm helium–neon laser (0.6 mW at installation) powered at 10% unless otherwise noted. Wheat germ agglutinin (WGA)-405 was excited with a 405-nm diode laser (3.5 mW at installation) powered at 1% unless otherwise noted. Temperature inside the microscope housing was 27–30°C.
In confocal imaging mode, fluorescence was collected with the microscope’s 32-detector gallium arsenide phosphide detector array arranged with a diffraction grating to measure 400–700-nm emissions in 9.6-nm bins. Emission bands were 495–550 nm for GFP and YFP, 605–700 nm for GxTX-594, and 420–480 nm for WGA-405. Point spread functions were calculated using ZEN Black software using emissions from 0.1-μm fluorescent microspheres prepared on a slide according to manufacturer’s instructions (cat. #T7279; Thermo Fisher Scientific). The point spread functions for confocal images with the 63×/1.40 NA oil DIC objective in the x–y direction were 228 nm (488-nm excitation) and 316 nm (594-nm excitation).
In Airy disk imaging mode, fluorescence was collected with the microscope’s 32-detector gallium arsenide phosphide detector array arranged in a concentric hexagonal pattern (Airyscan 410900-2058-580; Zeiss). After deconvolution, the point spread functions for the 63×/1.40 NA oil objective with 488-nm excitation was 124 nm in x–y and 216 nm in z and with 594-nm excitation, 168 nm in x–y and 212 nm in z. For the 63×/1.2 NA water objective, the point spread function with 488-nm excitation was 187 nm in x–y and 214 nm in z and with 594-nm excitation, 210 nm in x–y and 213 nm in z.
Unless stated otherwise, cells were plated in uncoated 35-mm dishes with a 7-mm inset no. 1.5 coverslip (cat. #P35G-1.5-20-C; MatTek). The CHO CE solution used for imaging and electrophysiology contained (in mM): 3.5 KCl, 155 NaCl, 10 HEPES, 1.5 CaCl2, and 1 MgCl2, adjusted to pH 7.4 with NaOH. Measured osmolality was 315 mOsm/liter. When noted, solution was supplemented with either 10 mM glucose (CEG) or 10 mM glucose and 1% BSA (CEGB).
For time lapse, GxTX-594 concentration–effect experiments, Kv2.1-CHO cells were plated onto 22 × 22-mm no. 1.5H coverglass (Deckglaser), and Kv2.1 expression was induced with minocycline 48 h before experiments such that all Kv2.1-CHO cells expressed Kv2.1. Prior to imaging, cell maintenance media were removed and replaced with CEGB, then the coverslip was mounted on an imaging chamber (cat. #RC-24E; Warner Instruments) with vacuum grease. We performed three 10-fold serial dilutions of 1,000 nM GxTX-594 to generate the range of concentrations used for this concentration–effect experiment and applied each concentration of GxTX-594 to Kv2.1-CHO cells for 15 min followed by 15 min of washout before the next concentration of GxTX-594 was applied. Solutions were added to the imaging chamber perfusion via a syringe at a flow rate of ∼1 ml per 10 s. Images were taken every 5 s. Laser power was set to 0.5% for the 488-nm laser and 1.5% for the 594-nm laser. For colocalization experiments with GFP-tagged proteins, cells were incubated in 100 nM GxTX-594 for 5 min and then washed with 1 ml CEGB three times before imaging.
Whole-cell voltage clamp for CHO cell imaging
Kv2.1-CHO cells plated in glass-bottom 35-mm dishes were incubated in minocycline for 48 h to induce Kv2.1 channel expression. Cells were washed with CEGB, placed in the microscope, and then incubated in 100 μl of 100 nM GxTX-594 for 5 min to label cells. Before patch clamp, the solution was diluted with 1 ml of CEG for a working concentration of 9 nM GxTX-594 during experiments. We determined the time required for GxTX-594 to reach a stable fluorescence after dilution from 100 nM to 9 nM by time-lapse imaging during dilution (Fig. 5 A). The rate of fluorescence change (kΔF) indicated that, on average, equilibration was 85% complete 9 min after dilution to 9 nM. After dilution to 9 nM, the mean fluorescence intensity decreased by 39.1 ± 8.4% and remained stable (Fig. 5 B). Cells with obvious GxTX-594 surface staining were voltage clamped in whole-cell mode with an EPC-10 patch-clamp amplifier (HEKA Elektronik) run by Patchmaster software (v2 × 90.2; HEKA Elektronik). The patch pipette contained the following potassium-deficient Cs+ internal pipette solution to reduce voltage error by limiting outward current: 70 mM CsCl, 50 mM CsF, 35 mM NaCl, 1 mM EGTA, and 10 mM HEPES, brought to pH 7.4 with CsOH. Osmolality was 310 mOsm/liter. The liquid junction potential was calculated to be 3.5 mV and was not corrected. Borosilicate glass pipettes (cat. #BF150-110-10HP; Sutter Instruments) were pulled with blunt tips to resistances <3.0 MΩ in these solutions. Cells were held at −80 mV (unless noted otherwise) and stepped to indicated voltages. The voltage step stimulus was maintained until any observed change in fluorescence was complete. Cells were stepped to −80 mV for at least 1 min and were visually inspected to determine sufficient fluorescence recovery before being stepped to another voltage. This recovery time did not always allow GxTX-594 labeling to equilibrate fully. For stimulus frequency dependence experiments, cells were given 2-ms steps to 40 mV at the stated frequencies (0.02, 5, 10, 20, 50, 100, 150, or 200 Hz). Images for voltage clamp fluorometry were taken in Airy disk imaging mode with the settings described above.
During time-lapse imaging of voltage-clamped cells, we noticed that GxTX-594 fluorescence in the center of the glass-adhered surface responded more slowly to voltage changes than the periphery. At −80 mV, at the center of the glass-adhered cell surface, relabeling was incomplete after 500 s; however, at the cell periphery, relabeling neared completion within 200 s (Fig. S4 A). To quantify this observation, concentric circular regions of interest (ROIs) were drawn, with the smallest ROI in the center of the glass-adhered surface. The average fluorescence intensities from each of these ROIs were compared with each other and an ROI at the cell periphery (Fig. S4, B and C). We quantified kΔF by fitting the average fluorescence intensities from each ROI with a monoexponential function (Eq. 1 in Image analysis). In response to voltage change, the kΔF was consistently slower at the center of the glass-adhered surface and faster toward the outer periphery (Fig. S4, D and E). While kΔF was consistently slower at the center of the glass-adhered surface, we observed that kΔF was more pronounced during labeling at −80 mV than unlabeling at 40 mV. When cells were held at a membrane potential of −80 mV, kΔF at the periphery was ∼10-fold faster than the kΔF at the center of the cell. In comparison, when the membrane potential was held at 40 mV, kΔF at the periphery was approximately threefold faster than the kΔF at the center of the cell. The gradual change in fluorescence intensity over many seconds after a voltage step is inconsistent with a fast electrochromic effect leading to fluorescence change as the change in fluorescence intensity does not occur instantaneously when the membrane potential is stepped. Additionally, the slowing of kΔF in subcellular regions farther from the periphery of cells is inconsistent with a slower electrochromic effect as whole-cell voltage clamp should render the membrane surface isopotential.
K+ channel–GFP ionic current recordings
Whole-cell voltage clamp was used to measure currents from CHO cells expressing Kv2.1-GFP, Kv2.2-GFP, Kv1.5-GFP, Kv4.2-GFP, BK-GFP, or GFP that was transfected as described above. Cells were plated on glass-bottom dishes. Cells were held at −80 mV, then 100-ms voltage steps were delivered ranging from −80 mV to +80 mV in +5-mV increments. Pulses were repeated every 2 s. The external (bath) solution contained CE solution. The internal (pipette) solution contained (in mM) 35 KOH, 70 KCl, 50 KF, 50 HEPES, and 5 EGTA, adjusted to pH 7.2 with KOH. Liquid junction potential was calculated to be 7.8 mV and was not corrected. Borosilicate glass pipettes (cat. #BF150-110-10HP; Sutter Instruments) were pulled into pipettes with resistance <3 MΩ for patch clamp recording. Recordings were at room temperature (22–24°C). Voltage clamp was achieved with an Axon Axopatch 200B Amplifier (Molecular Devices) run by Patchmaster software (v2 × 90.2; HEKA Elektronik). Holding potential was −80 mV capacitance and ohmic leak were subtracted using a P/5 protocol. Recordings were low-pass filtered at 10 kHz and digitized at 100 kHz. Voltage clamp data were plotted with Igor Pro 7 (WaveMetrics). As the experiments plotted in Fig. S2 A were merely to confirm functional expression of ion channels at the cell surface, series resistance compensation was not used, and substantial cell voltage errors are predicted during these experiments.
Kv2.1 ionic current recordings
Prior to patching, Kv2.1-CHO cells were washed in divalent-free PBS and then harvested in Versene (cat. #15040066; Gibco-BRL). Cells were scraped and transferred to a polypropylene tube, pelleted, and washed three times at 1,000 g for 2 min and then resuspended in the same external solution as used in the recording chamber bath. Cells were rotated in a polypropylene tube at room temperature (22–24°C) until use. Cells were then pipetted into a 50-μl recording chamber (RC-24N; Warner Instruments) prefilled with external solution and allowed to settle for ≥5 min. After adhering to the bottom of the glass recording chamber, cells were thoroughly rinsed with external solution using a gravity-driven perfusion system. Cells showing uniform intracellular GFP expression of intermediate intensity were selected for patching.
Voltage clamp was achieved with a patch clamp amplifier (Axon Axopatch 200B; Molecular Devices) run by Patchmaster software. Borosilicate glass pipettes (BF150-110-7.5HP; Sutter Instruments) were pulled with blunt tips, coated with silicone elastomer (Sylgard 184; Dow Corning), heat cured, and tip fire-polished to resistances <4 MΩ. Capacitance and ohmic leak were subtracted using a P/5 protocol. Recordings were low-pass filtered at 10 kHz using the amplifier’s built-in Bessel function and digitized at 100 kHz.
For whole-cell ionic current measurements in Kv2.1-CHO cells, the external patching solution contained (in mM) 3.5 KCl, 155 NaCl, 10 HEPES, 1.5 CaCl2, and 1 MgCl2, adjusted to pH 7.4 with NaOH. The internal (pipette) solution contained (in mM) 70 KCl, 5 EGTA, 50 HEPES, 50 KF, and 35 KOH, adjusted to pH 7.4 with KOH. The osmolality was 315 mOsm/liter for the external solution and 310 mOsm/liter for the internal solution measured by a vapor pressure osmometer. Following establishment of the whole-cell seal, ionic K+ current recordings were taken in the presence of a vehicle, which consisted of 100 nM tetrodotoxin, 10 mM glucose, and 0.1% BSA prepared in external solution. Cells were held at −100 mV with channel activation steps ranging from −80 mV to +120 mV in increments of +5 mV (100 ms) before being returned to 0 mV (100 ms) to record tail currents. The intersweep interval was 2 s. To determine the bioactivity of GxTX-594, Kv2.1 ionic currents were recorded once more, 5 min following the wash-in of bath solution also containing 100 nM GxTX-594. Wash-ins were performed while holding at −100 mV; 100 μl was washed through the chamber and removed distally through vacuum tubing to maintain constant bath fluid level.
Ionic current analysis
The average current in the 100 ms before voltage step was used to zero subtract the recording. Outward current taken as the mean value between 90 and 100 ms of the channel activation step was used to calculate and correct for series resistance-induced voltage error. Tail current values were derived from the mean value between 0.2 and 1.2 ms of the 0-mV tail current step. Tail current was normalized by the mean activation step current from 50 to 80 mV and plotted against the estimated membrane potential, which had been corrected for voltage error and the calculated liquid junction potential of 8.5 mV. These tail GV plots were fit with a fourth-power Boltzmann function (Sack et al., 2004), and the fit parameters were used for statistical analysis.
Brain slice methods
Hippocampal slice culture preparation and transfection
All experimental procedures were approved by the University of California, Davis, institutional animal care and use committee and were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Animals were maintained under standard light–dark cycles and had ad libitum access to food and water. Organotypic hippocampal slice cultures were prepared from postnatal day 5–7 rats, as previously described (Stoppini et al., 1991) and detailed in a video protocol (Opitz-Araya and Barria, 2011). DIV15–30 neurons were transfected 2–6 d before imaging via biolistic gene transfer (160 psi, Helios gene gun; Bio-Rad) as described in a detailed video protocol (Woods and Zito, 2008). 10 μg of plasmid was coated to 6–8 mg of 1.6-μm gold beads.
Two-photon excitation slice imaging
Image stacks (512 × 512 pixels, 1-mm Z-steps, 0.035 μm/pixel) were acquired using a custom two-photon excitation microscope (LUMPLFLN 60XW/IR2 objective, 60×/1.0 NA; Olympus) with two pulsed Ti:sapphire lasers (Mai Tai; Spectra Physics) tuned to 810 nm (for GxTX-594 imaging) and 930 nm (for GFP imaging) and controlled with ScanImage software (Pologruto et al., 2003). After identifying a neuron expressing Kv2.1-GFP, perfusion was stopped, and GxTX-594 was added to the static bath solution to a final concentration of 100 nM. After 5-min incubation, perfusion was restarted, leading to washout of GxTX-594 from the slice bath. Red and green photons (565dcxr, BG-22 glass, HQ607/45; Chroma Technology) emitted from the sample were collected with two sets of photomultiplier tubes (R3896; Hamamatsu).
Whole-cell voltage clamp for brain slice imaging
Organotypic hippocampal slice cultures (6–7 DIV, not transfected) were transferred to an imaging chamber with recirculating artificial cerebrospinal fluid (ACSF) maintained at 30°C. To hold the slice to the bottom of the chamber, a horseshoe-shaped piece of gold wire was used to weight the membrane holding the slice. ACSF contained (in mM) 127 NaCl, 25 NaHCO3, 25 D-glucose, 2.5 KCl, 1.25 NaH2PO4, 1 MgCl2, 2 CaCl2, and 200 nM tetrodotoxin, pH 7.3, and aerated with 95% O2/5% CO2 (∼310 mOsm). 4 ml of 100 nM GxTX-594 in ACSF was used in the circulating bath to allow the toxin to reach the slice and reach the desired concentration of 100 nM throughout the circulating bath. Images were acquired beginning 3 min after GxTX-594 was added. Apparent CA1 neurons with GxTX-594 labeling in a Kv2-like pattern were selected for whole-cell patch clamp. Voltage clamp was achieved using an Axopatch 200B Amplifier (Molecular Devices) controlled with custom software written in MATLAB (MathWorks, Inc.). Patch pipettes (5–7 MΩ) were filled with intracellular solution containing (in mM) 135 Cs-methanesulfonate, 10 Na2-phosphocreatine, 3 Na-L-ascorbate, 4 NaCl, 10 HEPES, 4 MgCl2, 4 Na2ATP, and 0.4 NaGTP, pH 7.2. Neurons were clamped at −70 mV. Input resistance and holding current were monitored throughout the experiment. Cells were excluded if the pipette series resistance was >25 MΩ or if the holding current exceeded −100 pA. To activate Kv2 channels, a 50-s depolarizing step from −70 mV to 0 mV was given.
Image analysis
For colocalization analyses, the Pearson coefficient was calculated using the JACoP plugin (Bolte and Cordelières, 2006). Colocalization analyses were conducted within ROIs defining individual cells. A Pearson correlation coefficient value of 0 signifies uncorrelated pixels between two images, and a Pearson correlation coefficient value of 1 signifies complete correlation between pixels from two images. Correlation between pixels from two fluorescent recordings of an image suggests spatial colocalization of proteins. Plotting and curve fitting was performed with Igor Pro 7 or 8 (WaveMetrics), which performs nonlinear least squares fits using a Levenberg–Marquardt algorithm. Sample sizes of n ≥ 3 were selected to confirm reproducibility. Sample sizes of n ≥ 6 were selected to power nonparametric statistical comparisons to discern P < 0.01. The α for statistical significance in nonparametric statistical comparisons was adjusted for multiple comparisons using the Bonferroni method. The Bonferroni-corrected P value = where α = 0.01 and n represents the number of comparisons made in an experiment. Error values from individual curve fittings are SDs. All other errors, including error bars, indicate SEs. Arithmetic means are reported for intensity measurements and correlation coefficients. As the distributions underlying variability in results are unknown, nonparametric statistical comparisons were conducted with Mann–Whitney U tests, and two-tailed P values were reported individually if P > 0.0001. Parametric statistical tests, which include the Student’s t test, ANOVA, and Tukey’s post hoc test, were performed with paired data and on sample sizes of n ≤ 6 due to the weak statistical power of nonparametric tests when comparing small sample sizes.
EVAP model
Online supplemental material
Fig. S1 pertains to the synthesis of GxTX-594. It shows a model of GxTX-594, and provides HPLC chromatograms and MALDI-TOF mass spectrometry profiles of Ser13Cys GxTX and GxTX-594. Fig. S2 shows GxTX-594 selectively labeling Kv2 proteins at the cell surfaces. Fig. S3 shows that GxTX-594 labeling of surface membranes requires Kv2 proteins. Fig. S4 demonstrates that extracellular access can impact GxTX-594 labeling kinetics. Fig. S5 demonstrates that variation in bath temperature does not account for variability of GxTX-594 kinetics. Fig. S6 is an extended image gallery of GxTX-594 labeling CA1 hippocampal pyramidal neurons transfected with Kv2.1 GFP. Video 1 is a time-lapse image sequence of GxTX-594 fluorescence on a voltage-clamped CA1 hippocampal pyramidal neuron while it is depolarized from −70 to 0 mV. Data S1 is a spreadsheet that sets up calculations to generate EVAP model predictions.
Results
GxTX-594 retains bioactivity for Kv2.1 after chemoselective modification
To monitor activation of Kv2 proteins in tissue slices, we synthesized an EVAP compatible with two-photon imaging. We previously presented an EVAP that was a synthetic derivative of GxTX conjugated to a DyLight 550 fluorophore (GxTX-550; Tilley et al., 2014). DyLight 550 has poor two-photon excitation properties, and for this study, it was replaced with Alexa Fluor 594, a persulfonated Texas Red analogue with a large two-photon excitation cross section and ample spectral separation from GFP, making it well suited for multiplexed, two-photon excitation imaging experiments (Zito et al., 2004). We refer to this EVAP variant as GxTX-594.
We performed electrophysiological analyses to determine whether GxTX-594 retains the ability to allosterically modulate Kv2.1 (Fig. 1). GxTX is a partial inverse agonist of Kv2.1, which lowers channel open probability by stabilizing voltage sensors in a resting conformation. Consequently, more positive intracellular voltage is required to activate voltage sensors and achieve the same open probability as without GxTX (Tilley et al., 2019). Previously, we estimated that a Kv2.1 voltage sensor with GxTX bound is 5,400-fold more stable in its resting conformation and requires more positive intracellular voltage to become activated (Tilley et al., 2019). To characterize the efficacy of GxTX-594 in allosterically modulating Kv2.1 gating, we voltage clamped Kv2.1-CHO cells and measured K+ currents in GxTX-594. We analyzed the Kv2.1 conductance–voltage (GV) relation by fitting with a fourth power Boltzmann function. The voltage at which the conductance of the fitted function is 50% of maximum, Vmid, was +73 ± 13 mV for 100 nM GxTX-594 (Fig. 1 G). For comparison, the Vmid of 100 nM GxTX was +67 ± 6 mV (Tilley et al., 2019). This shift of the GV indicates that GxTX-594 retains an efficacy similar to GxTX.
GxTX-594 labels Kv2 proteins
To determine the concentration range where GxTX-594 effectively labels Kv2.1 protein, we performed a concentration–effect experiment. As the Kd of our previously presented EVAP, GxTX-550, was 30.0 ± 3.9 nM at a holding potential of −100 mV (Tilley et al., 2014), we presumed that 1,000 nM would be a sufficient upper bound for concentration–effect experiments with GxTX-594 (Fig. 2, A and B). Cell surface fluorescence intensity was analyzed immediately after washout of 1, 10, 100, and 1,000 nM GxTX-594 to eliminate fluorescence from toxin in solution. This fluorescence labeling concentration–effect relation was fit with a Langmuir binding isotherm, resulting in a Kd of 26.9 ± 8.3 nM (Fig. 2 C). Due to the incomplete equilibration of labeling at 1 and 10 nM, our measure is expected to overestimate the Kd. The near saturation of the fluorescence concentration–effect relation with 100 nM GxTX-594, and the rate of fluorescence equilibration (Fig. 2 D) suggested that incubation with 100 nM GxTX-594 for 5 min results in near-maximal labeling on the glass-adhered surface of Kv2.1-CHO cells.
We assessed whether GxTX-594 fluorescence on CHO-K1 cells is due to the presence of Kv2 proteins. CHO-K1 cells were transfected with either a rat Kv2.1-GFP or Kv2.2-GFP construct, each of which yielded delayed rectifier K+ currents (Fig. S2 A). 2 d after transfection, Kv2.1-GFP or Kv2.2-GFP fluorescence displayed distinct subcellular regions of high density at the glass-adhered surface (Fig. 3 A), which we refer to as Kv2 clusters, a term used to refer to similar high-density regions in neurons and other mammalian cell lines (Kirmiz et al., 2018b). Correlation coefficients indicated a high degree of subcellular colocalization of both Kv2 proteins with GxTX-594 (Fig. 3 B), quantitating the observation that GxTX-594 is not evenly distributed throughout the membrane but is localized to Kv2 clusters. We did not detect a significant difference between the ratios of GxTX-594 to Kv2.1-GFP or Kv2.2-GFP fluorescence (Fig. 3 C; P = 0.74 by Mann–Whitney U test), consistent with the lack of discrimination of GxTX between Kv2.1 and Kv2.2 (Herrington et al., 2006). GxTX accesses the membrane-embedded voltage sensors of Kv2 proteins by partitioning into the outer leaflet of the plasma membrane bilayer (Milescu et al., 2009; Gupta et al., 2015), and a GxTX derivative labeled with a fluorophore that brightens in less polar environments is detectable in the membrane of CHO cells without Kv2 proteins (Fletcher-Taylor et al., 2020). However, we failed to find any GxTX-594 labeling of CHO cells in the absence of Kv2 proteins (Fig. S3).
GxTX-594 selectively labels Kv2 proteins
An important consideration for determination of whether GxTX-594 could reveal conformational changes of endogenous Kv2 proteins is whether the EVAP is selective for Kv2 proteins. Electrophysiological studies have concluded that the native GxTX peptide is selective for Kv2 channels, with some off-target modulation of A-type Kv4 channels (Herrington et al., 2006; Liu and Bean, 2014; Speca et al., 2014). However, electrophysiological testing cannot determine whether ligands bind unless they also alter currents (Sack et al., 2013). Furthermore, structural differences between wild-type GxTX and the GxTX-594 variant could potentially alter selectivity among channel protein subtypes.
To test whether GxTX-594 binds other voltage-gated K+ channel subtypes, we quantified surface labeling and analyzed colocalization of GxTX-594 with a selection of GFP-tagged voltage-gated K+ channel subtypes (Fig. 4 A). The ratio of GxTX-594 fluorescence to each GFP-tagged K+ channel subtype was not distinguishable from zero for Kv4.2, Kv1.5, or BK channels (Fig. 4 B), indicating minimal, if any, binding. Furthermore, no colocalization was apparent between Kv4.2, Kv1.5, or BK channels and the residual GxTX-594 fluorescence (Fig. 4 C). An additional set of experiments conducted under different microscopy conditions and without the auxiliary subunits of Kv4.2 or Kv1.5 also gave no indication of GxTX-594 labeling (Fig. S2). While we cannot be certain that GxTX-594 does not bind any of the >80 known mammalian proteins containing voltage sensor domains, the lack of labeling of the related voltage-gated K+ channels indicates that GxTX-594 does not promiscuously label voltage sensors.
The relationship between GxTX-594 cell-surface fluorescence and Kv2.1 voltage activation
To understand the relationship between channel gating and GxTX-594 fluorescence, we determined how fluorescence intensity on cells expressing Kv2.1 responds to changes in membrane voltage. We found that GxTX-594 equilibrated more quickly on the sides of cells than on their glass-adhered surface, presumably due to restricted access to the extracellular space by the glass (Fig. S4). Due to this observation, we chose an imaging plane where fluorescence from GxTX-594 resembled an annulus (Fig. 6 A). This imaging plane varied between cells but was >1 μm above the glass. We developed a labeling protocol in which Kv2.1-CHO cells were incubated for 5 min in a bath solution (CEG) containing 100 nM GxTX-594, which was then diluted with extracellular solution to 9 nM (Fig. 5). Once GxTX-594 fluorescence intensity stabilized at the cell membrane (at least 9 min; Fig. 5), cells were voltage clamped in whole-cell mode. We measured the fluorescence response of GxTX-594 when the membrane voltage of Kv2.1-CHO cells was stepped from a holding potential of −80 mV to more positive voltages that ranged from −40 mV to +80 mV (Fig. 6 A). ROIs corresponding to the cell surface were manually identified and average fluorescence intensity quantified from time-lapse sequences. The voltage-dependent reduction in fluorescence equilibrates to a value above background. Even when cells were given a +80-mV depolarizing voltage stimulus, some GxTX-594 fluorescence remained (Fig. 6 B). Most of the residual fluorescence appeared to be localized to the cell surface membrane (Fig. 6 A) and varied between cells (Fig. 6 C). As such surface labeling was not present on CHO cells in the absence of Kv2 proteins (Fig. S3), we consider this residual fluorescence to originate from voltage-insensitive Kv2 proteins. Voltage-insensitive fluorescence could potentially be from Kv2.1–GxTX-594 complexes with immobilized voltage sensors or internalized Kv2.1–GxTX-594 complexes that remain just under the cell surface (Deutsch et al., 2012; Weigel et al., 2012; Fox et al., 2013a; Weigel et al., 2013).
To compare voltage response properties between cells, we used a normalization procedure to analyze only the voltage-sensitive fraction of the fluorescence from each cell, which we defined as the fluorescence that changed between −80 mV and +80 mV. The initial fluorescence at a holding potential of −80 mV was normalized to 100% F/Finit, and residual fluorescence after a +80-mV step was normalized to 0% F/Finit (Fig. 6 D). To characterize the voltage dependence of the Kv2.1–GxTX-594 interaction, fluorescence–voltage (FV) responses were fit with a Boltzmann distribution (Eq. 2). This fit had a half maximal voltage midpoint (V1/2) of −27 mV and a steepness (z) of 1.4 e0 (Fig. 6 D, bottom panel, black line). This is similar to voltage sensor movement in Kv2.1-CHO cells without any GxTX present: V1/2 = −26 mV, z = 1.6 e0 (Tilley et al., 2019). These results suggest that at 9 nM GxTX-594, the FV appears to be a good surrogate for the gating current–voltage (QV) response of unlabeled Kv2 channels.
To determine the voltage dependence of the kinetics of GxTX-594 labeling and unlabeling, we compared kΔF at varying step potentials. We quantified kΔF by fitting the average fluorescence from voltage-clamped cells with a monoexponential function (Eq. 1). In response to voltage steps from a holding potential of −80 mV to more positive potentials, kΔF increased progressively as step potential was increased above −40 mV and appeared to begin to saturate at higher voltages (Fig. 6 E). Upon return to −80 mV, kΔF was similar to −40 mV. While the kΔF did not clearly display saturation at positive voltages that would justify fitting with a Boltzmann function, a model of GxTX-594 dynamics, which we develop later in this study, indicated that Boltzmann fitting could yield physical insight (Fig. 6 E, bottom panel, black line). We noted that the degree of variability in kΔF measurements became greater at more positive potentials (Fig. 6 E, top and bottom panels). At −80 mV, there was a twofold range in kΔF values and a ninefold range at +80 mV. The relatively low variation in kΔF at −80 mV suggests that despite variance in fluorescence intensity after rebinding (Fig. S4 C), kΔF from fits of the upward relaxation at −80 mV are relatively consistent. The average kΔF equilibration at 10 nM GxTX-594 in concentration–effect experiments was comparable to Kv2.1-CHO cells incubated in 9 nM GxTX-594 and voltage clamped at −80 mV (0.0011 s−1 and 0.0014 s−1, respectively; Fig. S4 E and Fig. 2 D). This suggests that the Kv2 voltage sensors in the unclamped cells for concentration–effect experiments are in the same early resting conformation as voltage-clamped cells at −80 mV. Additionally, we determined that only a small fraction of the up to ninefold variability in kΔF at positive voltages could be attributed to temperature fluctuations (Fig. S5). Possible reasons for cell-to-cell variability at more positive voltages are discussed further in Limitations.
The relation between voltage sensor activation and GxTX-594 dynamics can be recapitulated by rate theory modeling
To enable translation of the intensity of fluorescence from GxTX-594 on a cell surface into a measure of Kv2 conformational change, we developed an EVAP model, a series of equations derived from rate theory that relate cell labeling to voltage sensor activation. The framework of the EVAP model is generalizable to fluorescent molecular probes that report conformational changes by a change in binding affinity. In the EVAP model, the proportion of labeled versus unlabeled Kv2 in a membrane is determined by the proportion of voltage sensors in resting versus activated conformations. The model assumes that the innate voltage sensitivity of the Kv2 subunit is solely responsible for controlling voltage dependence. EVAP labeling is voltage dependent because the binding and unbinding rates are different for resting and activated conformations of voltage sensors. Voltage activation of Kv2 channels involves many conformational changes (Scholle et al., 2004; Jara-Oseguera et al., 2011; Tilley et al., 2019). However, models that presume independent activation of a voltage sensor in each of the four Kv2.1 subunits accurately predict many aspects of voltage activation and voltage sensor toxin binding (Lee et al., 2003; Tilley et al., 2019). For simplicity, we model Kv2 proteins as having only resting and activated conformations that are independent in each voltage sensor and developed a rate theory model consisting of four interconnected states (Scheme 1).
When voltage sensors change from resting to activated conformations, the binding rate of the GxTX-594 EVAP decreases, and the unbinding rate increases. When the membrane voltage is held constant for sufficient time, the proportions of labeled and unlabeled proteins reach an equilibrium. EVAP labeling requires seconds to equilibrate (Fig. 6), whereas Kv2 channel gating equilibrates in milliseconds (Tilley et al., 2019), three orders of magnitude more quickly. These distinct time scales of equilibration suggest an approximation to model the reversible EVAP labeling response: voltage sensor conformations achieve equilibrium quickly such that only their distribution at equilibrium is expected to greatly impact the kinetics of labeling and unlabeling, allowing Scheme 1 to collapse into Scheme 2, which depicts the structure of the EVAP model used for calculations.
We constrained the EVAP model with measurements of GxTX-594 binding kinetics and GxTX impacts on Kv2.1 gating (see EVAP model and Data S1). We tested the viability of model predictions by comparison with GxTX-594 labeling measurements. The EVAP model predicts that the FV for GxTX-594 labeling will conform to a Boltzmann distribution. The FV prediction in 9 nM GxTX-594 had a V1/2 and z that differ by only −4 mV and 0.06 e0, respectively, from the Boltzmann fit of experimental data (Fig. 6 D, bottom panel, black and green lines). The EVAP model also predicts that the kΔF-V will conform to a Boltzmann distribution. The kΔF-V prediction differs by only 3 mV and 0.05 e0 from the Boltzmann fit of experimental data (Fig. 6 E, bottom panel, blue line). However, this fit was poorly constrained as we failed to obtain sufficient data at voltages above +80 mV where kΔF-V is predicted to saturate. In our attempts, the durations required at more positive voltages irreversibly increased membrane leak. The V1/2 and z values from the GxTX-594 FV and kΔF-V were not used as constraints of the EVAP model, and the similarity between the predictions and empirical findings seemed remarkable enough to warrant further exploration of the EVAP model predictions.
We used the EVAP model to investigate general principles of the relation between voltage sensor activation and reversible labeling. The EVAP model predicts that as GxTX-594 concentration decreases, the change in labeling (ΔF/ΔFmax) approaches the probability that unlabeled voltage sensors are resting (Fig. 7 A). This prediction explains the similarity between the FV in 9 nM GxTX-594 (V1/2 = −27 ± 3 mV, z = 1.4 ± 0.1 e0; Fig. 6 D) and the QV of Kv2.1 (V1/2 = −26 ± 1 mV, z = 1.6 ± 0.1 e0; Tilley et al., 2019). As the concentration of EVAP is increased, the FV shifts to more positive voltages such that the fractional change in fluorescence intensity is always less than the fraction of unlabeled voltage sensors that are active. As EVAP concentration increases and approaches the activated state Kd (1,790 nM), voltage sensor activation becomes less effective at dissociating the EVAP due to binding to activated voltage sensors (Fig. 7 B). The model predicts that at any concentration, this simple interpretation will be valid: A decrease in EVAP surface fluorescence indicates activation of unlabeled voltage sensors.
The EVAP model also yields a simple interpretation of labeling kinetics, it predicts that as GxTX-594 concentration decreases, the rate of fluorescence change kΔF approaches the probability that labeled channels are active (Fig. 7 C). This prediction explains the similarity between the kΔF-V in 9 nM GxTX-594 (V1/2 = 38 ± 15 mV, z = 1.4 ± 0.4 e0; Fig. 6 D) and the QV of Kv2.1 in saturating GxTX (V1/2 = 47 ± 1 mV, z = 1.6 ± 0.1 e0; Tilley et al., 2019). At low concentrations, the dependence of kΔF on the conformation of channels bound to GxTX-594 is due to the rate of unbinding dominating kΔF, with the rate of unbinding being solely determined by the conformation of channels bound to GxTX-594.
Repetitive action potential–like stimuli amplify the GxTX-594 response
Kv2 currents impact repetitive action potential firing (Du et al., 2000; Liu and Bean, 2014), making repetitive action potentials the voltage waveforms that are, arguably, most relevant to Kv2 function. However, action potentials occur on the millisecond time scale, orders of magnitude faster than the GxTX-594 response, which integrates Kv2 conformations occurring over many seconds. Kv2 channels have slow deactivation kinetics (Liu and Bean, 2014; Tilley et al., 2019), and high-frequency firing could prevent Kv2 proteins from fully deactivating before a next action potential is triggered, creating a kinetic trap that progressively accumulates activated voltage sensors. This behavior of Kv2 proteins indicates that high-frequency firing could evoke a more robust fluorescence signal than the EVAP model predicts, as the model assumes continuous equilibrium of voltage sensors and cannot kinetically trap activated conformations. To test this hypothesis, we crudely mimicked action potentials with trains of 2-ms voltage steps from −80 to +40 mV and observed the changes in fluorescence on GxTX-594–labeled Kv2.1-CHO cells (Fig. 8, A and B). To assess frequency response, step frequency was varied from 0.02 to 200 Hz. GxTX-594 unlabeling and kΔF increased with stimulus frequency (Fig. 8, C and D). We compared these fluorescence responses to those predicted by the EVAP model.
When the stimulus frequency was <50 Hz, the EVAP model did a reasonable job of predicting fluorescence change. At and >50 Hz, fluorescence decreased by more than the EVAP model predicted, even without accounting for a voltage-insensitive fraction of Kv2.1 proteins (Fig. 8 C, bottom panel, green line). As discussed above, this divergence from the equilibrium-based EVAP model is expected at frequencies where voltage steps are shorter than Kv2 equilibration times. The time constant of activating gating current decay from Kv2.1-CHO cells was 1.3 ms at +40 mV (Tilley et al., 2019), which means that the majority of voltage sensors are effectively activated during the 2-ms +40 mV steps. In contrast, the time constant of deactivating gating current decay from Kv2.1-CHO cells was 22 ms at −80 mV (Tilley et al., 2019), which means that Kv2.1 is expected to become kinetically trapped in activated conformations when stimuli to +40 mV from −80 mV are ∼50 Hz or faster. Thus, the amplified EVAP response appears consistent with voltage sensors failing to deactivate before the next stimulus, leading to an accumulation of activated voltage sensors and a more dramatic fluorescence response than predicted by the EVAP model. Overall, these dynamics indicate that the magnitude of the change in GxTX-594 fluorescence intensity will be amplified during repetitive action potentials, a regimen of electrophysiological signaling where Kv2 currents are critical.
In contrast, the kinetics of the GxTX-594 response did not appear to deviate from EVAP model predictions at high frequencies (Fig. 8 D). As kΔF responds to the dynamics of Kv2 proteins bound by GxTX-594, it could be that the faster deactivation rate of voltage sensors bound by GxTX (Tilley et al., 2019) prevents the bound channels from being kinetically trapped.
GxTX-594 labels brain slices transfected with Kv2.1-GFP
To determine whether expression of Kv2 proteins embedded in tissue can be imaged with GxTX-594, we overexpressed Kv2.1-GFP in rat brain slices and examined CA1 pyramidal neurons of the hippocampus. We chose CA1 neurons for several reasons: They express Kv2 channels at a density typical of central neurons, the physiology of these neurons has been intensively studied, and their electrical properties are relatively homogeneous (Misonou et al., 2005). Organotypic hippocampal slice cultures prepared from postnatal day 5–7 rats were sparsely transfected with Kv2.1-GFP, resulting in a subset of neurons displaying green fluorescence. When imaged 2–4 d after transfection, GFP fluorescence was observed in the plasma membrane surrounding neuronal cell bodies and proximal dendrites (Fig. S6, A and B). Six days or more after transfection, Kv2.1-GFP fluorescence organized into clusters on the surface of the cell soma and proximal processes (Fig. 9 A and Fig. S6 C), a pattern consistent with a prior report of endogenous Kv2.1 in CA1 neurons (Misonou et al., 2005). After identifying a neuron expressing Kv2.1-GFP, solution flow into the imaging chamber was stopped, and GxTX-594 was added to the static bath solution to a final concentration of 100 nM. After 5 min of incubation, solution flow was restarted, leading to washout of excess GxTX-594 from the imaging chamber. After washout, GxTX-594 fluorescence remained colocalized with Kv2.1-GFP (Fig. 9 and Fig. S6), indicating that GxTX-594 is able to permeate through dense neural tissue and bind to Kv2 proteins on neuronal surfaces. Pearson correlation coefficients confirmed the colocalization of GxTX-594 with Kv2.1-GFP in multiple slices (Fig. 9 C). In most images of Kv2.1-GFP–expressing neurons, GxTX-594 also labeled puncta on neighboring neurons that did not express Kv2.1-GFP but at intensities that were roughly an order of magnitude dimmer (Fig. 9 B, white arrow). The clustered GxTX-594 fluorescence patterns on the cell body and proximal processes of CA1 neurons were strikingly similar to reported patterns of anti-Kv2 immunofluorescence patterns and are consistent with GxTX-594 labeling of endogenous Kv2 proteins in CA1 neurons. While we cannot exclude the possibility that CA1 neurons have a subset of Kv2 proteins on their surface that is not labeled by GxTX-594, we saw no indication of Kv2.1-GFP on neuronal surfaces that are not labeled by GxTX-594.
While we observed GxTX-594 fluorescence that morphologically resembles endogenous Kv2 protein localizations, we also found that GxTX-594 occasionally labels structures not consistent with Kv2 proteins (Fig. S6, bottom panel, arrows). This non-Kv2 labeling was most prevalent at the surface of the hippocampal slices and progressively decreased as the imaging plane was moved deeper into the tissue (data not shown). Our interpretation of this phenomenon is that GxTX-594 can accumulate in the dead tissue and debris that is present at the surface of a hippocampal section after it is cut. For this reason, we analyzed only GxTX-594 fluorescence with subcellular localizations consistent with Kv2 channels.
GxTX-594 labeling in brain slices responds to neuronal depolarization
To test whether reversible GxTX-594 labeling of neurons in brain slices is consistent with binding to endogenous Kv2 voltage sensors, we determined whether GxTX-594 labeling responds to voltage changes in tissue. First, we looked for Kv2-like patterns on CA1 pyramidal neurons in untransfected brain slices bathed in 100 nM GxTX-594. With two-photon excitation, optical sections are thinner than the neuronal cell bodies, and GxTX-594 fluorescence appeared as puncta circumscribed by dark intracellular spaces (Fig. 10 A). This was similar to the patterns of fluorescence in Kv2.1-GFP–transfected slices (Fig. 9 B) and consistent with the punctate expression pattern of Kv2.1 in CA1 pyramidal neurons seen in fixed brain slices (Misonou et al., 2005).
We tested whether the punctate fluorescence was voltage sensitive using voltage clamp. To ensure voltage clamp of the neuronal cell body, slices were bathed in tetrodotoxin to block Na+ channels, and Cs+ was included in the patch pipette solution to block K+ currents. In each experiment, whole-cell configuration was achieved with a GxTX-594–labeled neuron, and holding potential was set to −70 mV. At this point, time-lapse imaging of a two-photon excitation optical section was initiated. Depolarization to 0 mV resulted in loss of fluorescence from a subset of puncta at the perimeter of the voltage-clamped neuron (Fig. 10 B, red arrows; and Video 1). Other fluorescent puncta appeared unaltered by the 0-mV step (Fig. 10 B, white arrows). These puncta could represent Kv2 proteins on a neighboring cell, off-target labeling by GxTX-594, or voltage-insensitive Kv2 proteins, possibly due to near-surface internalization. To assess whether the fluorescence changes were due to the voltage change, we compared fluorescence of the apparent cell membrane region with regions more distal from the voltage-clamped cell body. An ROI 30-pixels (1.2-μm) wide containing the apparent membrane of the cell body was compared with other regions within each image (Fig. 10 C). The region containing the membrane of the cell body lost fluorescence during the 0-mV step (Fig. 10 D, ROI 1), while neither of the regions more distal (ROI 2, ROI 3) or the intracellular region (ROI 0) showed a similar change. In three hippocampal slices from separate rats, we found a significant decrease in fluorescence in ROI 1 compared with ROI 2 and ROI 3 when neurons were given a voltage step to 0 mV (Fig. 10 E; ANOVA P < 0.001; Tukey’s post hoc test P < 0.001). The kinetics of fluorescence response of the voltage-clamped membrane region were similar between slices (Fig. 10, F and K).
To determine whether the fluorescence response to depolarization was driven by the Kv2-like puncta on the cell membrane, the fluorescence along a path containing the apparent cell membrane was selected by drawing a path connecting fluorescent puncta surrounding the dark cell body region and averaging fluorescence in a region 10-pixels (0.4-μm) wide centered on this path (Fig. 10 G, yellow line). The fluorescence intensity along the path of the ROI revealed distinct peaks corresponding to puncta (Fig. 10 H, red trace). After stepping the neuron to 0 mV, the intensity of fluorescence of a subset of puncta lessened (Fig. 10 H, black trace, peaks 1, 3, 4, 5, and 8). While the data are noisy, it is clear that even more reduction is observed in individual fluorescence puncta in brain slice (Fig. 10 I, blue line). This suggests that the voltage sensors in these functionally selected puncta are extensively activated. The EVAP model predicts only a 55% reduction of fluorescence from CHO cells stepped to 0 mV in 100 nM GxTX. The greater response of the endogenous puncta could be due to voltage sensors activating at more negative voltages in neurons than CHO cells. However, the GxTX-594 concentration within tissue may be more dilute than in the bath solution, and consequently, the sensitive fraction calculated from the EVAP model is a lower bound. Other puncta maintained or increased their brightness after depolarization (Fig. 10 H, black trace, peaks 2, 6, and 7), but it was unclear whether these voltage-insensitive puncta correspond to Kv2 proteins on the surface of the same voltage-clamped neuron. When the kinetics of fluorescence intensity decay of individual voltage-sensitive puncta were fit with Eq. 1, kΔF values were similar between puncta (Fig. 10 I), consistent with these spatially separated puncta all being on the surface of the voltage-clamped neuron. The kΔF changes measured in these puncta were similar to those predicted by the Fig. 7 model (Fig. 10 I, blue line), although the variability of these measurements was substantial. To address whether the fluorescence change at the cell membrane was driven by decreases in regions of punctate fluorescence, the punctate and nonpunctate fluorescence intensity changes were analyzed separately. The regions with fluorescence intensities above average for the path (Fig. 10 H, dotted line) were binned as one group. The above-average group, by definition, contained all punctate fluorescence. When comparing the fluorescence before and during the 0-mV step, the regions that were initially of below-average fluorescence maintained the same intensity (103 ± 8%); regions that were initially of above-average fluorescence decreased in intensity (70 ± 8%; Fig. 10 J). This suggests that the detectable unlabeling along the cell membrane originated in the Kv2-like puncta, not the regions of lower fluorescence intensity between them.
To determine whether the dynamics of the voltage-dependent responses of GxTX-594 fluorescence on neurons in brain slices were consistent with reversible labeling of Kv2.1, we performed experiments with Kv2.1-expressing CHO cells under similar conditions as brain slice: 100 nM GxTX-594, 30°C, Cs+-containing patch pipette solution. The rate of fluorescence changes in Kv2.1-CHO cells was similar to neurons in brain slices and consistent with the kΔF of 0.06 s−1 predicted by the Fig. 7 model (Fig. 10 K). However, the data underlying of kΔF measurements were noisy, limiting our ability to detect differences.
Discussion
The molecular targeting, conformation selectivity, and spatial precision of fluorescence from GxTX-594 enable identification of where in tissue the conformational status of Kv2 voltage sensors becomes altered. However, the utility of GxTX-594 as an EVAP is limited by several factors, including emission intensity, variability between experiments, and inhibition of Kv2 proteins. We discuss the potential utility and limitations of the EVAP mechanism underlying GxTX-594.
Unique capabilities of GxTX-594
The Kv2 EVAP presented here is the only imaging method we are aware of for measuring voltage-sensitive conformational changes of a specific, endogenous protein. As GxTX binding selectively stabilizes the fully resting conformation of Kv2.1 voltage sensors (Tilley et al., 2019), reversible GxTX-594 labeling is expected to bind with highest affinity specifically to the fully resting conformation of the Kv2 voltage sensor in which the first gating charge of the Kv2 S4 segment is in the gating charge transfer center (Tao et al., 2010). Images of GxTX-594 fluorescence reveal this conformation’s occurrence with subcellular spatial resolution. Importantly, the EVAP model we developed allows deconvolution of the behavior of unlabeled Kv2 proteins. This enables the subcellular locations where Kv2 voltage sensing occurs to be seen for the first time.
Electrophysiological approaches can detect the voltage-sensitive K+ conductance of Kv2 channels. However, the majority of Kv2 proteins on cell surface membranes do not function as channels and are nonconducting (Benndorf et al., 1994; Malin and Nerbonne, 2002; O’Connell et al., 2010), and Kv2 proteins dynamically regulate cellular physiology by nonconducting functions (Antonucci et al., 2001; Singer-Lahat et al., 2007; Feinshreiber et al., 2010; Dai et al., 2012; Fox et al., 2015; Johnson et al., 2018; Kirmiz et al., 2018a, 2018b, 2019; Vierra et al., 2019). The Kv2 EVAP reports on the conformation of Kv2 voltage sensors independently from ion conductance, enabling the study of voltage sensor involvement in Kv2’s nonconducting physiological functions. We present this EVAP as a prototype molecular probe for imaging voltage sensing by endogenous proteins in tissue with molecular specificity.
Here, during our initial testing of GxTX-594, we observed that the majority of Kv2 protein detected at discrete individual clusters was voltage sensitive. While it may not be surprising to find that voltage-gated ion channel proteins are voltage sensitive, the voltage sensors of surface-expressed proteins can be immobilized. For example, gating charge of the L-type Ca2+ channel Cav1.2 is immobilized until it is bound by an intracellular protein (Turner et al., 2020). Kv2.1 channel function is extensively regulated by neurons. In rat CA1 neurons, the clustered Kv2 channels are proposed to be nonconducting (Fox et al., 2013b). Our results show that clustered Kv2 proteins in rat CA1 neurons remain voltage sensitive. Interestingly, when Kv2.1 is expressed in CHO cells, a fraction of the GxTX-594 fluorescence is voltage insensitive (Fig. 6 C). This observation is consistent with voltage sensor immobilization of some surface-expressed Kv2.1 protein, although it could be due to intracellular Kv2.1–GxTX-594 proteins that appear to be at the cell surface.
We used images of GxTX-594 fluorescence to measure the coupling between endogenous Kv2 proteins and membrane potential at specific, subcellular anatomical locations. Similarly, GxTX-594 imaging should detect changes of voltage sensor status when Kv2 proteins become engaged or disengaged from nonconducting functions, such as formation of plasma membrane–endoplasmic reticulum junctions (Antonucci et al., 2001; Fox et al., 2015; Kirmiz et al., 2018a), regulation of exocytosis (Singer-Lahat et al., 2007; Feinshreiber et al., 2010), regulation of insulin secretion (Dai et al., 2012), interaction with kinases, phosphatases, and SUMOylases (Misonou et al., 2004; Park et al., 2006; Dai et al., 2009; Cerda and Trimmer, 2011; McCord and Aizenman, 2013), formation of specialized subcellular calcium signaling domains (Vierra et al., 2019), and interactions with astrocytic end feet (Du et al., 1998). GxTX-594 could potentially reveal conformational changes in organs throughout the body where Kv2 proteins are expressed, which include muscle, thymus, spleen, kidney, adrenal gland, pancreas, lung, and reproductive organs (Bocksteins, 2016).
Limitations of GxTX-594
There are important limitations to the GxTX-594 approach and of the underlying EVAP mechanism generally. We discuss several limitations that are worth considering in the design of any studies with GxTX-594.
GxTX-594 labeling is slower than channel gating
The kinetics of reversible GxTX-594 labeling are limited to measuring changes in Kv2 activity on a time scale of tens of seconds. While the temporal resolution of GxTX-594 is compatible with live imaging and electrophysiology experiments, labeling kinetics do not provide sufficient time resolution to distinguish fast electrical signaling events. The response time of GxTX-594 is far slower than the kinetics of Kv2 conformational change, limiting measurements to the probability, averaged over time, that voltage sensors are resting or active. It is worth noting that the probability of a conformation’s occurrence can be a valuable measure and is the ultimate quantitation of many biophysical studies of ion channels (e.g., open probability, steady-state conductance, and gating charge–voltage relation).
GxTX-594 dynamics are altered in confined extracellular spaces
The kinetics of GxTX-594 dynamics varied within different regions of the same CHO cell (Fig. S4). The location dependence of kΔF was more pronounced during GxTX-594 labeling at −80 mV than unlabeling at +40 mV, and such a difference can be explained by the distinct voltage-dependent affinities of Kv2.1 for GxTX-594. We suspect that the more extreme location dependence at −80 mV is due to a high density of Kv2.1 binding sites in the restricted extracellular space between the cell membrane and glass coverslip, such that GxTX-594 is depleted from solution by binding Kv2.1 before reaching the center of the cell. After unbinding at +40 mV, Kv2.1 proteins are in activated, low-affinity conformations, which are unlikely to rebind GxTX-594 and, thus, do not slow their diffusion across the cell surface. The space extracellular to Kv2.1 channel clusters of hippocampal and cortical interneurons is restricted by astrocytic end feet, which create an extracellular cleft only a few nanometers wide (Du et al., 1998). Such restricted spaces may slow the kinetics of labeling in the hippocampal slices (Fig. 10).
GxTX-594 dynamics are variable between CHO cells
The variability of GxTX-594 response rates and amplitudes between CHO cells limited the precision of results. Some of this variability is expected from technical imprecisions: Fits of kΔF where the final value of the relaxation was poorly determined by the data, small changes due to variations of room temperature, photobleaching, and other potential sources. However, we noticed that results were more consistent between stimuli of the same cell, and the variability was greatest between cells (Fig. 6, D and E; and Fig. 8, C and D). We suspect that cell-to-cell differences in Kv2.1 conformational equilibria are responsible for much of the variability in GxTX-594 response. The Kv2.1 conductance–voltage relation is regulated by many cellular pathways, including kinases, phosphatases, and SUMOylases (Misonou et al., 2004; Park et al., 2006; Dai et al., 2009; Cerda and Trimmer, 2011; McCord and Aizenman, 2013). Large cell-to-cell variation in Kv2.1 conductance–voltage and gating charge–voltage relations have been reported in CHO cells by our group and others (McCrossan et al., 2009; Tilley et al., 2014; Kang et al., 2019; Tilley et al., 2019). In this study, when we predicted the voltage sensor V1/2 of Kv2.1 from electrophysiology, we observed a 6.4-mV SD with a range of 19 mV, and this variance appeared to be exacerbated by GxTX-594 having a 9.7-mV SD and range of 36 mV (Fig. 1 E). As EVAP dynamics and the GV are both determined by voltage sensor activation, variability in the GxTX-594 response is expected. The hypothesis that cell-to-cell variation in fluorescence dynamics is due to the inherent variability of Kv2.1 voltage sensor activation could be more definitively tested by identifying whether a correlation exists between the V1/2 of the QV and fluorescence–voltage relationship from individual cells labeled with GxTX-594. While we have not attempted this, the structure of the variance in GxTX-594 fluorescence–voltage relationships is informative. The fluorescence–voltage relationships compiled from many cells become more variable near the midpoint of relevant voltage sensor movements. The response amplitude F/Finit (Fig. 6 D) is determined by unlabeled voltage sensor activation and appears most variable near the V1/2 of the unlabeled QV relation (−32 mV; Tilley et al., 2019). The response kinetics kΔF are determined by activation of voltage sensors, which have GxTX-594 bound and appear to become increasingly variable at voltages higher than −20 mV (Fig. 6 E, bottom panel). Despite this variability, the V1/2 and z from the Boltzmann fit of the kΔF–voltage relationship from many cells were remarkably close to the QV of the GxTX–Kv2.1 complex, with a V1/2 and z that differ by 3.3 mV and 0.07 e0, respectively (Tilley et al., 2019).
Another possibility is that variability in surface membrane composition undergirds the variability of the GxTX-594 response. GxTX affinities are influenced by the dynamically changing complement of lipids in the plasma membrane lipid composition. Lipids bind to the voltage-sensing domain near the GxTX binding site (Milescu et al., 2009; Gupta et al., 2015). Sphingomyelinase D treatment, which alters the membrane composition by converting sphingomyelin to ceramide-1-phosphate, has been shown to enhance GxTX affinity for Kv2.1 fourfold (Milescu et al., 2009). Voltage activation of Kv2.1 is also affected by sphingomyelinase treatments. While variation in lipid composition is expected to cause variation in GxTX-594 dynamics, we do not know the degree to which the lipid composition varies between the Kv2.1-CHO cells in our studies.
Fluorescence intensity
Optical noise limits interpretation of EVAP imaging. While the GxTX-594 signal from CA1 neurons was sufficient to identify voltage sensing of endogenous Kv2 protein, fluorescence signal-to-noise issues limit interpretation with the EVAP model. This signal-to-noise ratio is influenced by the density of EVAP binding sites, the fluorescence intensity from each binding site, characteristics of the imaging system, and background fluorescence from unbound EVAP or other sources. As the concentration of fluorophore is lowered, background fluorescence from unbound EVAP will decrease, and the percentage of labeled binding sites will decrease. However, ΔF after voltage sensor activation remains similar, as does kΔF (Fig. 7). Due to this dynamic, the minimum concentration of EVAP that is detectable is expected to be determined by the background fluorescence and the density of EVAP binding sites. Here, we see labeling of Kv2.1-CHO cells with 1 nM GxTX-594 (Fig. 2), and previously, we found a robust fluorescence–voltage response in 1 nM GxTX-550 (Tilley et al., 2014). in each case, <10% of resting voltage sensors are bound by the EVAPs. In theory, fluorescence measurements from a single EVAP molecule immobilized by binding to a single-voltage sensor could be informative, as fluorescence from a single binding site is eliminated altogether after unbinding and diffusion away. If the EVAP imaging method does not have single-molecule resolution, the density of EVAP binding sites can limit the signal to noise.
CA1 hippocampal neurons express Kv2 proteins at a density typical of central neurons (Misonou et al., 2004; Vacher et al., 2008; Speca et al., 2014), and we expect that GxTX-594 imaging will have similar signal-to-noise characteristics in most brain regions. Improved signal to noise would be expected from such cells that express higher densities of Kv2 proteins, such as neurons of the subiculum, or the inner segment of photoreceptors (Maletic-Savatic et al., 1995; Gayet-Primo et al., 2018). Kv2 proteins are also expressed by many other cell types throughout the body (Bocksteins, 2016), where GxTX-594–labeling techniques may reveal Kv2 activity, if protein densities are sufficient.
GxTX-594 inhibits Kv2 proteins
GxTX-based probes inhibit the Kv2 proteins they label by stabilizing the resting conformation of Kv2 voltage sensors. The Kv2.1–GxTX-594 complex does not open to conduct K+ ions in the physiological voltage range (Fig. 1). Thus, GxTX-594 depletes the population of Kv2 proteins responding normally to physiological stimuli, which could alter Kv2 signaling. The concentration of an EVAP can be lowered such that only an inconsequential minority of proteins are bound, with the trade-off being dimmer fluorescence. With a related probe, we explored the impact of decreasing concentration on fluorescence response of a GxTX-based EVAP and saw substantial fluorescence responses to voltage while inhibiting only ∼10% of Kv2.1 current (Tilley et al., 2014). Here, we demonstrate that lower concentration and physiological stimuli are not always required for scientifically meaningful implementation of an EVAP. Even when GxTX-594 inhibits most Kv2 proteins, the behavior of unlabeled Kv2 proteins can be calculated using the EVAP model we have developed. Of course, the electrical feedback within cells will be altered by such protocols.
The EVAP model is oversimplified
Another limitation of the analysis developed here is that the model of Kv2 voltage sensor conformational change is an oversimplification. The gating dynamics of Kv2 channels are more complex than our model (Islas and Sigworth, 1999; Scholle et al., 2004; Jara-Oseguera et al., 2011; Tilley et al., 2019). Under some conditions, the assumption of voltage sensor independence will limit the model’s predictive power. With a related tarantula toxin, Hanatoxin, the concentration dependence for inhibition of Kv2.1 charge movement was consistent with independent binding to each voltage sensor inhibiting ∼25% of the total gating charge movement (Lee et al., 2003). This result suggests that the simplifying assumption of voltage sensor independence is reasonable. Additionally, the model of GxTX-594–reversible labeling developed here assumes that voltage sensors are in continuous equilibrium. These deviations from equilibrium likely explain the deviation of the model from the data in response to high-frequency voltage steps (Fig. 8, C and D). This model could be further tested with a Kv2 mutant, which shifts channel gating without interfering with GxTX-594 binding.
Conformation-selective probes reveal conformational changes of endogenous proteins
Measurements of dynamic reversible labeling by a conformation-selective probe such as GxTX-594 can enable deduction of how unlabeled proteins behave. This is perhaps counterintuitive because GxTX inhibits voltage sensor movement of the Kv2 protein it binds, and thus, only bound proteins generate optical signals. This approach is analogous to calcium imaging experiments, which have been spectacularly informative about physiological calcium signaling (Yang and Yuste, 2017), despite the fact that no optical signals originate from the physiologically relevant free Ca2+ but only from Ca2+ that is chelated by a dye. In all such experiments, fluorescence from Ca2+-bound dyes is deconvolved using the statistical thermodynamics of Ca2+ binding to calculate free Ca2+ (Adams, 2010). Similarly, GxTX-based probes dynamically bind to unlabeled Kv2 proteins, and the binding rate depends on the probability that unlabeled voltage sensors are in a resting conformation (Fig. 7). Thus, the conformations of unlabeled Kv2 proteins influence the dynamics of labeling with GxTX-based probes. Consequently, the dynamics of labeling reveal the conformations of unlabeled Kv2 proteins.
Deployment of GxTX-594 to report conformational changes of endogenous proteins demonstrates that conformation-selective ligands can be used to image occurrence of the conformations they bind to. The same principles of action apply to any conformation-selective labeling reagent, suggesting that probes for conformational changes of many different proteins could be developed. Probes could conceivably be developed from the many other voltage sensor toxins or other gating modifiers that act by a similar mechanism as GxTX yet target the voltage sensors of different ion channel proteins (McDonough et al., 1997; Sack et al., 2004; Catterall et al., 2007; Swartz, 2007; Schmalhofer et al., 2008; Peretz et al., 2010; McCormack et al., 2013; Ahuja et al., 2015; Dockendorff et al., 2018; Zhang et al., 2018). Conformation-selective binders have been engineered for a variety of other proteins, and methods to quantify conformational changes from their fluorescence are needed. For example, fluorescently labeled conformation-selective binders have revealed that endocytosed GPCRs continue to remain in a physiologically activated conformation (Irannejad et al., 2013; Tsvetanova et al., 2015; Eichel and von Zastrow, 2018). A means to determine the conformational equilibria of GPCRs from fluorescence images has not yet been developed. We suggest that the statistical thermodynamic framework developed here could provide a starting point for more quantitative interpretation of other conformation-selective molecular probes.
Acknowledgments
Christopher J. Lingle served as editor.
We thank Jim Trimmer (University of California, Davis) for numerous discussions and constructive critical reading of an early version of the manuscript. We thank Georgeann Sack (Afferent LLC) for critical reading, editing, and feedback.
This research was supported by National Institutes of Health grants R01NS096317 (to J.T. Sack and B.E. Cohen), U01NS090581 (to J.T. Sack), R21EY026449 (to J.T. Sack), R01NS062736 (to K. Zito), U01NS103571 (to K. Zito), T32GM007377 (to R.J. Sepela), University of California, Davis New Research Initiative award (to J.T. Sack) and American Heart Association grant 17POST33670698 (to P. Thapa). GxTX variants were synthesized at the Molecular Foundry, supported by the Director, Office of Science, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, U.S. Department of Energy under contract DE-AC02-05CH11231.
The authors declare no competing financial interests.
Author contributions: P. Thapa: Conceptualization, formal analysis, investigation, methodology, visualization, writing-original draft, writing-reviewing and editing. R. Stewart: Conceptualization, formal analysis, investigation, methodology, visualization, writing-original draft, writing-reviewing and editing. R.J. Sepela: Conceptualization, formal analysis, investigation, methodology, visualization, writing-original draft, writing-reviewing and editing. O. Vivas: Investigation, methodology, writing-reviewing and editing. L.K. Parajuli: Investigation, methodology, writing-reviewing and editing. M. Lillya: Conceptualization, formal analysis, investigation, methodology, visualization, writing-reviewing and editing. S. Fletcher-Taylor: Investigation, methodology, writing-reviewing and editing. B.E. Cohen: Conceptualization, funding acquisition, project administration, supervision, writing-reviewing and editing. K. Zito: Conceptualization, funding acquisition, investigation, methodology, project administration, supervision, writing-original draft, writing-reviewing and editing. J.T. Sack: Conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, supervision, visualization, writing-original draft, writing-reviewing and editing.
References
Author notes
P. Thapa and R. Stewart contributed equally to this paper.