Phosphoinositides modulate the voltage dependence of two-pore channel 3

Two-pore channels or two-pore Na⁺ channels (TPCs) have two homologous repeats of six-transmembrane helices (IS1–IS6 and IIS1–IIS6; Peiter et al., 2005; Calcraft et al., 2009). Each of the repeats is a functional unit containing a voltage sensor domain (VSD) and a pore domain, which are commonly observed in the superfamily of voltage-gated cation channels (Yu and Catterall, 2004). As four units form one functional channel in this superfamily, TPCs function as dimers (Guo et al., 2016; Kintzer and Stroud, 2016). In VSDs in general, basic amino acid residues that appear at every third position in helix S4 are essential for voltage sensing (Armstrong et al., 1974; Aggarwal and MacKinnon, 1996; Seoh et al., 1996). Although these basic residues are retained in both of the two VSDs in TPCs (VSD1 and VSD2), the number and the positions are different between the two VSDs and also between TPC subtypes. Animal TPCs were initially characterized as nicotinic acid adenine dinucleotide phosphate–dependent Ca²⁺ release channels (Brailoiu et al., 2009; Calcraft et al., 2009; Zong et al., 2009), but recent electrophysiological analyses have reported that they are activated by a type of phosphoinositide (PI), which is a minor phosphatidylinositol bisphosphate (PIP₂) species, PI(3,5)P₂, and are highly Na⁺-selective channels (Wang et al., 2012; Cang et al., 2013, 2014b). Therefore, the ligands and the permeating ion species of TPCs are still under debate.

The TPC family has three members, each of which has a different localization and biophysical properties. TPC1 and TPC2 are localized in endosomes and lysosomes (Xu and Ren, 2015), and are important for the physiological aspects related to these intracellular organelles, such as autophagy (Lin et al., 2015; García-Rúa et al., 2016), mammalian target of rapamycin (mTOR)–dependent nutrient sensing (Cang et al., 2013), and Ebola virus infection (Sakurai et al., 2015). While TPC1 requires both voltage and PI(3,5)P₂ stimuli for opening, TPC2 is simply a PI(3,5)P₂-activated channel. Recent cryo-EM structures of mouse TPC1 (MmTPC1), with or without PI(3,5)P₂, revealed that PI(3,5)P₂ is bound to the regions containing the IS4-S5 linker and IS6 in domain I (She et al., 2018). The critical importance of this region was also verified in the action of nicotinic acid adenine dinucleotide phosphate (Patel et al., 2017). Mutagenesis studies showed that the positive charges in IS4 do not affect voltage sensing, but those in IIS4 do. Taken together with the structural analyses, these findings suggest that the two domains are specialized for each stimulus, namely, domain I for PIs and domain II for voltage.

In contrast to TPC1 and TPC2, fewer studies of TPC3 are reported, in either biophysical or physiological aspects. TPC3 is localized in the plasma membrane and in cortical granules in starfish oocytes (Ramos et al., 2014). It is evolutionally conserved from echinoderms, such as sea urchins, to some mammals, but is lacking in primates and some rodents (Brailoiu et al., 2010; Cai and Patel, 2010). TPC3 has been believed not to be activated by any PIs, but to be simply a voltage-dependent Na⁺-selective channel (Cang et al., 2014a). Characteristically, when TPC3 is expressed in the Xenopus laevis oocyte expression system, a long depolarizing step pulse generates slowly developing currents that last >10 s, by a so-called “induction” mechanism (Cang et al., 2014a).

PIs, especially three types of PIP₂, serve as important cofactors for membrane proteins (Balla, 2013). The regulatory roles of PI(4,5)P₂ on the activity and trafficking of many ion channels and transporters are well known. PI(3,5)P₂ is, so far, known to activate only two families of ion channels that are localized in endosomal and lysosomal membranes, namely transient receptor potential channel mucolipin 1–3 (TRPML 1–3), TPC1, and TPC2 (Dong et al., 2010; Wang et al., 2012; Cang et al., 2014b). The third type of PIP₂, PI(3,4)P₂, is considered to be localized mainly in the plasma membrane and early endosomes (Posor et al., 2013). PI(3,4)P₂ is produced by PI3-kinase (PI3K) from PI(4)P, as is Phosphatidylinositol (3,4,5)-trisphosphate (PIP₃), which is also known to be an important signaling PI, as represented by Akt activation (Manning and Toker, 2017). Although PI(3,4)P₂ has received less attention, several emerging pieces of evidence have indicated the unique and critical importance of PI(3,4)P₂, such as in spatially localized PI3K signaling (Braccini et al., 2015), endocytosis (Posor et al., 2013; Boucrot et al., 2015), neuronal death (Sasaki et al., 2010), membrane ruffling (Hogan et al., 2004), and cell migration (Bae et al., 2010; Li et al., 2016). PI(3,4)P₂ has been known to bind some membrane-associated proteins (Hogan et al., 2004; Li and Marshall, 2015). However, despite the physiological importance of PI(3,4)P₂, there are few reports of transmembrane proteins, including ion channels and transporters, that respond specifically to PI(3,4)P₂.

We investigated the molecular mechanism of induction of TPC3 using the two-electrode voltage clamp technique in the Xenopus oocyte expression system. By analysis of the induction, we newly found that the induction is caused by modulation of voltage dependence by PI(3,4)P₂, which is produced by the endogenous system in Xenopus oocytes. The application of PI to inside-out patches revealed that TPC3 is sensitive to both PI(3,4)P₂ and PI(3,5)P₂, but not to PI(4,5)P₂. We also identified the key amino acid residues for the coupling of PI binding with voltage-dependent gating, which is not observed in the case of PI(3,5)P₂ binding to TPC1. Our findings revealed features of the PI dependence and PI-dependent modulatory mechanism of TPC3 that are unique among the TPC family, providing us with insight into subtype-specific differences.

All animal experiments were approved by the Animal Care Committee of the National Institutes of Natural Sciences (an umbrella institution of National Institute for Physiological Sciences, Tokyo, Japan).

All soluble analogues of PIs, such as PI(3,4)P₂-diC8, were purchased from Echelon Biosciences and dissolved in water as a 0.2-mM concentration of stock solution. PI stock solutions were directly used for PI injection experiments and diluted into the bath solution for inside-out patch experiments. Insulin (Wako) was dissolved in 10 mM HCl solution and kept as a 1-mM stock solution. Thapsigargin (Sigma-Aldrich) and wortmannin (Wako) were dissolved in DMSO at 1 mM and 10 mM for stock solution, respectively.

The complementary DNA (cDNA) of TPC3 from Xenopus tropicalis (XtTPC3; XP_002940387) in pUC57-Simple was synthesized by Genscript with one base replacement (T1395 to C) from original cDNA (GenBank accession no. XM_002940341) for abolishing the endogenous NdeI site (provided by Y. Fujiyoshi, Nagoya University, Nagoya, Japan). The XtTPC3 cDNA fragment was digested by cleavage at the NdeI and XhoI sites, which were attached at the 5′ and 3′ ends of XtTPC3 cDNA in pUC57-Simple, extracted, and blunted by the Klenow fragment. The fragment was then inserted to the SmaI site of the pGEMHE vector. A TAG sequence on the 3′ side of the SmaI site of the pGEMHE vector was used as a stop codon. The Ciona intestinalis voltage-sensitive phosphatase (Ci-VSP) cDNA was provided by Y. Okamura (Osaka University, Osaka, Japan). The cDNAs of fluorescent reporters sensitive to PI(4,5)P₂ (F-PLC) and PI(3,4)P₂ (F-TAPP) were provided by E.Y. Isacoff (University of California, Berkeley, Berkeley, CA). Site-directed mutagenesis was accomplished by a PCR-based method using PrimeSTAR Max DNA Polymerase and the In-Fusion HD Cloning Kit (Takara Bio), following the manufacturer’s method. The plasmid sequences were confirmed by DNA sequencing. The cRNAs were synthesized with a mMESSAGE mMACHINE T7 kit (Ambion; Life Technologies) from the linearized cDNA.

Preparation and injection for Xenopus oocytes were performed as described previously (Kume et al., 2018). X. laevis for oocytes were purchased from Hamamatsu Seibutsu Kyouzai (Japan). The oocytes were collected from frogs anesthetized by 0.15% tricaine. The isolated oocytes were treated with 2 mg/ml collagenase (Sigma-Aldrich) for 6.5 h and stocked overnight or longer at 17°C in frog Ringer’s solution containing (in mM) 88 NaCl, 1 KCl, 2.4 NaHCO₃, 0.3 Ca(NO₃)₂, 0.41 CaCl₂, 0.82 MgSO₄, and 15 HEPES, pH 7.6, with 0.1% penicillin–streptomycin (Sigma-Aldrich). Then, the oocytes were injected with 50 nl of cRNA solution and incubated at 17°C in frog Ringer’s solution. The amount of injected cRNAs of XtTPC3 was 25 ng. In the case of coexpression experiments with Ci-VSP, 20 ng of XtTPC3 cRNA and 5 ng of Ci-VSP cRNA were mixed and injected. For fluorescent recording, 20 ng of XtTPC3 cRNA and 5 ng of Ci-VSP cRNA were mixed with 25 ng of F-PLC cRNA or 12.5 ng of F-TAPP. In the recording without Ci-VSP, 20 ng of XtTPC3 was mixed with 25 ng of F-PLC or 6.25 ng of F-TAPP. Currents were measured 2–6 d after injection, depending on the required current amplitude.

Currents were recorded under two-electrode voltage clamps using an OC-725C amplifier (Warner Instruments) and pClamp10.7 software (Molecular Devices). Data from the amplifier were digitized at 0.1 to 10 kHz, depending on recording protocols, through Digidata1440 (Molecular Devices), and digitally filtered at 100 Hz. The resistance of microelectrodes was adjusted to be 0.2–0.5 MΩ when filled with a solution of 3 M K-acetate and 10 mM KCl. The standard recording bath solution was ND-96, which contains (in mM) 96 NaCl, 2 KCl, 1.8 CaCl₂, 1 MgCl₂, and 5 HEPES, pH 7.4. Mes, pH 4.6, was used for low pH solutions. Unless especially noted, oocytes were held at −60 mV.

The membrane potential of oocytes expressing a large amount of XtTPC3 often increased above 0 mV in the ND-96 solution, thought to be caused by the autonomous induction and sustained opening of XtTPC3. Once XtTPC3 is activated, it slightly depolarizes the membrane potential toward the Na⁺ reversal potential and thus causes further opening of XtTPC3, like the firing of an action potential. The noninactivating property of XtTPC3 facilitates and sustains the depolarized condition, and finally induces cell death of the oocytes. This problem was frequently encountered when Ci-VSP was coexpressed, as it potentiates XtTPC3. To avoid this fatal effect, the oocytes were often incubated in a low or medium Na⁺ concentration solution instead of ND-96, where 96 mM NaCl was replaced with 10 mM NaCl and 86 mM NMDG-HCl (low) or with 18.6 mM NaCl and 77.4 mM NMDG-HCl (medium). The half-activation voltage (V_1/2) value of XtTPC3 WT was confirmed to be apparently not affected by this solution replacement.

To compare the conductance–voltage (G–V) relationship before and after induction, as in Fig. 1, C and D, a series of step pulses were given before and after the 80-mV pulse for 20 s. On recording of the current traces before induction, the −60 mV holding potential was sustained between step pulses to maintain the state before induction, generally for 1 min, and 15 or 30 s for the mutants, which show less induction. A 100-mV pulse for 50 ms before each step pulse, followed by returning to −60 mV, was given to confirm that the condition before induction is maintained. Otherwise, the currents would become gradually larger with the progress of step pulses. The step pulses were given from 10 mV to 200 mV for 50 ms, followed by 60-mV pulses for tail currents. On recording of the current traces after induction, the 100-mV depolarizing pulses for 500 ms, followed by returning to −60 mV, were given before each step pulse to maintain the induced condition with minimal interpulse interval. The step pulses were given from −20 mV to +170 mV for 50 ms, followed by 60-mV pulses for tail currents.

For the wortmannin or insulin pretreatments, the oocytes were incubated in the ND-96–based low Na⁺ solution containing 30 µM wortmannin or 10 µM insulin for >1 h.

In the PI injection experiments shown in Fig. 8, the water containing PIs was manually injected by positive pressure from a glass pipette. The injection of 0.2 mM PI(4,5)P₂–diC8 (Echelon Biosciences), the soluble derivative of PI(4,5)P₂, into the oocytes evoked the endogenous Ca²⁺-activated Cl⁻ channel (CaCC) currents, which are thought to be caused by Ca²⁺ release through the soluble PI(4,5)P₂–activated inositol trisphosphate (IP₃) receptor. To abolish contamination by the Cl⁻ currents, the oocytes expressing XtTPC3 for PI(4,5)P₂–diC8 injection were pretreated with Ca²⁺-free ND-96 (in mM: 96 NaCl, 2 KCl, 5 MgCl₂, 5 HEPES, pH 7.4, and 1 EGTA) containing 1 µM thapsigargin for >1.5 h, and then the current recording and PI injection were performed in the Ca²⁺-free ND-96 solution. The time constant of activation at 100 mV was quantified as a parameter to reflect the PI-dependent change of voltage dependence. Furthermore, to verify no residual contamination of CaCC currents, which have similar kinetics to XtTPC3 currents at 100 mV, the current amplitude at 50 mV was also monitored. The appearance of inward currents at 50 mV was a good parameter of potentiation of XtTPC3, because the Cl⁻ currents would appear as outward currents at this membrane voltage. Also, XtTPC3 showed almost no currents because of its low open probability in the condition before induction. In addition, voltage ramps following the 50- and 100-mV test pulses were also monitored to verify that the reversal potential was not significantly altered by the leak caused by PI-induced oocyte swelling nor by contamination with CaCC currents. For the injection of PIs with lower activity for the IP₃ receptor than that of PI(4,5)P₂, the oocytes were pretreated with frog Ringer’s solution containing 10 µM BAPTA-AM (Funakoshi) for ≥1.5 h to chelate the leaked Ca²⁺. The recording and PI injection were performed in the Ca²⁺-free ND-96 solution.

The aforementioned pulse stimulus from −30 mV holding potential was applied every 100 s, and the injection of 0.2 mM PIs was performed 100 s after the start of the recording. After 700 s, 80-mV depolarizing pulses were applied twice for 10 s, separated by a brief return to the holding potential as an induction stimulus. The current amplitude evoked by a 50-mV pulse at each step was normalized to that obtained after the induction stimulus. The time course of activation at 100 mV was fitted with a single exponential function. In some batches of oocytes, and especially in oocytes kept for several days, the kinetics were often fast, even before PI injection. This might have been caused by an increased PI(3,4)P₂ concentration. To detect the effects of PI injection clearly, oocytes kept for <2 d after injection and showing a time constant for activation >70 ms at 100 mV were chosen for PI injection experiments.

The oocytes were kept for >4 d in ND-96 solution with the medium Na⁺ concentration. Immediately before the experiments, the oocytes were incubated in a hypertonic medium (in mM: 192 NMDG-HCl, 4 KCl, 3.6 CaCl₂, 2 MgCl₂, and 10 HEPES-NaOH, pH 7.4) for 10–15 min. The vitelline membrane was manually removed using forceps, and then the oocytes were transferred to, and again kept in, the ND-96 solution with medium Na⁺ concentration. The glass pipettes were filled with ND-96 solution, and the resistance was 1–3 MΩ. Inside-out patches were excised from oocytes that had been transferred into the bath solution (in mM: 92 KCl, 10 NaCl, 0.5 MgCl₂, 1 EGTA, 0.5 K₂ATP, and 5 HEPES, pH 7.4). Currents were recorded using an AXOPATCH 200B amplifier (Axon Instruments) and pClamp10.8 software (Molecular Devices). Data from the amplifier were digitized through Digidata1440 (Molecular Devices) at 4 kHz. PIs were applied using a Perfusion Fast-Step System (SF-77B; Warner Instrument).

A homology model of XtTPC3 was generated based on the cryo-EM structure of MmTPC1 bound with PI(3,5)P₂ (PDB accession no. 6C9A) using SWISS-MODEL (Schwede et al., 2003). The resultant XtTPC3 model, where the PI molecules were not included, was aligned with the original MmTPC1 structure using PyMol software (Delano Scientific). PI(3,5)P₂ molecules in the MmTPC1 structure were manually modified to PI(3,4)P₂ molecules and shown with the aligned XtTPC3 structure.

Fluorescent detection of PI(3,4)P₂ and PI(4,5)P₂ levels, using the FRET-based probes F-TAPP and F-PLC, and data analysis were conducted as previously reported (Grimm and Isacoff, 2016). Fluorescence was detected using a fluorescence microscope through a 20× 1.0 numerical aperture (NA) objective (Olympus XLUMPLFLN 20X), with a xenon lamp (Hamamatsu Photonics) as the light source. Light intensity was diminished to 25% through a neutral density filter. The cyan fluorescent protein (CFP) fluorescence was excited by light passed through a band-pass excitation filter (426–450 nm). The emitted light was passed through a filter (460-nm longpass) and divided by a dichroic mirror (505 nm). The intensities of shorter wavelength light, passed through a 430–490-nm filter, were detected as CFP signal, and those of longer wavelength, passed through a 510–55- nm filter, were detected as YFP signal. The two fluorescence signals were obtained by Photomultipliers (Hamamatsu Photonics). Current and fluorescent signals were simultaneously digitized and recorded through a Digidata 1322A and pClamp10.8 software (Molecular Devices). The YFP and CFP signals were sampled at 0.2 kHz and low pass filtrated for the analysis. A slow bleaching was adjusted by assuming that the decrease until the first voltage pulse was linear. The FRET values were obtained by dividing the YFP signal by the CFP signal. The baseline signal was normalized as 1.0, and the difference from the baseline was calculated as ΔFRET.

Electrophysiological data were analyzed using Clampfit 10.7 (Molecular Devices) and Igor Pro (WaveMetrics). Tail current amplitude elicited by +60 mV after step pulses was used to obtain G–V relationships. The G–V relationships were calculated by fitting to a two-state Boltzmann equation,

where G indicates the conductance, and G_max and G_min are the maximum and minimum conductance, respectively. V is the membrane voltage, and V_1/2 is the half-activation voltage. z, F, R, and T are the effective charge, Faraday's constant, Gas constant and temperature, respectively. The normalized G/G_max was plotted as a function of membrane voltage.

To analyze the effects of Ci-VSP coexpression, as shown in Fig. 4, time constants for activation were obtained using Clampfit by fitting each trace with a single exponential function. In the coexpression experiments with Ci-VSP, the current traces at 100 mV showed an unusual development, and the time constants were apparently not fitted well by a single exponential function, possibly reflecting the mixed effects of Ci-VSP and XtTPC3 at 100 mV, such as Ci-VSP–dependent PIP₂ generation and the subsequent degradation, and the endogenous PIP₂ production capability of the Xenopus oocytes. The SEM of each fitting was at most 2% of the value of the time constants, which is far smaller than the change caused by the voltage difference in the presence of Ci-VSP. Therefore, despite of the inaccuracy of fitting, we used the time constants obtained by single exponential fitting as a parameter to indicate the change of the activation kinetics. The range of 200 ms after 100-mV depolarizing step pulses was used for fitting.

For the analysis of simultaneous recording of current and fluorescence, as shown in Fig. 7, the recordings of the initial 10 s after depolarizing step pulses were fitted to a double exponential function. Among current traces elicited by 30, 50, and 70 mV, those at 30 mV and 50 mV at increasing phase could not be fitted. This is possibly because the slower voltage-dependent activation becomes rate-limiting in the lower range of voltage and makes the time courses complicated. For this reason, only the traces of currents and fluorescence at 70 mV were used.

To analyze the effects of PI perfusion on XtTPC3 in the inside-out configuration, as shown in Fig. 9, the tail current elicited at −30 mV immediately after +80-mV pulses for 200 ms was used. The ratio of the current amplitude after and before PI perfusion was analyzed as fold increase of current amplitude.

Data are presented as mean ± SD. Student’s t test was used to evaluate the V_1/2 before and after induction, the effects of PI injection on the time course, and the effects of wortmannin and insulin pretreatment. One-way ANOVA, followed by Dunnett’s test, was used to estimate statistical significance in comparison of the increased current amplitude in multiple PI injection and mutational effects. One-way ANOVA, followed by Tukey’s test, was used to evaluate the PI perfusion–induced fold increase of current amplitude in multiple comparison between groups. P < 0.05 was considered statistically significant.

Fig. S1 shows the combined effects of Ci-VSP coexpression and insulin pretreatment on the activation kinetics of XtTPC3. Fig. S2 shows the effects of thapsigargin pretreatment on the oocytes for PI(4,5)P₂ injection. Fig. S3 shows a series of G–V relationships of the XtTPC3 channels with mutations to the putative PI binding site. Fig. S4 shows the representative current recordings of the R187Q mutant coexpressed with Ci-VSP. Fig. S5 shows the effects of PI(3,4)P₂ injection to the oocytes expressing the R187Q mutant. Fig. S6 shows the comparison of MmTPC1 structures with or without PI(3,5)P₂.

XtTPC3 was expressed in Xenopus oocytes. On application of two sequential 50-mV depolarizing pulses for 20 s, the first pulse generated slowly developing and noninactivating inward currents, but the second one showed clearly more rapid activation (Fig. 1 A). Then, after the oocyte was maintained at −60 mV holding potential for 10 min, another depolarizing pulse again generated the slowly increasing current. This induction by depolarization stimulus and recovery phenomenon suggests that some properties of XtTPC3 could be reversibly changed. As such a property, we investigated whether or not the voltage dependence was changed before and after induction. Comparison of G–V relationships shows that the voltage dependence was significantly shifted in the hyperpolarized direction after induction (Fig. 1, B–D; V_1/2 = 128 ± 8.4 mV before, 61.8 ± 13 mV after, n = 6). Considering this depolarization-induced shift of voltage dependence, induction is explained as a gradual increase in the open probability of noninactivating XtTPC3 by the sustained depolarizing pulse, which appears as a slowly developing increase in the current amplitude.

Previous studies showed that Xenopus oocytes have a TPC3-like endogenous Na⁺ conductance that is also increased by long depolarization (Baud et al., 1982; Cang et al., 2014a). These endogenous Na⁺ currents are by far smaller and have much slower induction kinetics than the exogenously expressed XtTPC3 currents (Fig. 2, A and B). Comparison of the tail currents using the protocols to obtain G–V relationships showed that the tail current amplitudes at 60 mV from the oocytes without cRNA injection are negligible, compared with those from XtTPC3, both before and after induction (Fig. 2, C and D). For these reasons, we believe that, despite the endogenous Na⁺ conductance, the results of our analyses reflect the exogenously expressed XtTPC3 activity.

We assumed that the shift of voltage dependence by long depolarization is caused by the PI dependence of XtTPC3. In a previous study, TPC3 was concluded to be insensitive to either PI(3,5)P₂ or PI(4,5)P₂ (Cang et al., 2014a), but it remains possible that other PIs are related to the induction. Consistent with this hypothesis, a pretreatment with wortmannin, a PI3K inhibitor that perturbs the PI production pathway, attenuated the depolarization-induced G–V shift (Fig. 3).

To confirm the involvement of PIs more directly, Ci-VSP (Murata et al., 2005) was coexpressed with XtTPC3. In response to depolarization, Ci-VSP dephosphorylates PIP₃ to PI(3,4)P₂ and PI(4,5)P₂ at lower membrane voltage, and even to PI(4)P at higher membrane voltage (Fig. 4 A; Grimm and Isacoff, 2016). During a 50-mV depolarizing pulse, XtTPC3 coexpressed with Ci-VSP showed a far faster current increase than that without Ci-VSP and was then followed by current decrease (Fig. 4 B). The protocol for obtaining the G–V relationship was difficult to be applied to the Ci-VSP coexpression experiment, because the prepulses with various membrane voltage change both the open probability of XtTPC3 and the Ci-VSP activity, making the G–V relationship ambiguous. Alternatively, we used a two-pulse protocol, where the activity of Ci-VSP was evoked by prepulses to various membrane voltages, and then the effects on the time constant of activation were monitored by test pulses to 100 mV (Fig. 4 C). The relationship between time constant of activation and prepulse membrane voltage showed that the activation was accelerated by prepulses at around 30 mV and then decelerated by those at further higher membrane voltage (Fig. 4, D–G). This double mode of action, as well as the rapid current increase and subsequent decrease under the continuous depolarizing condition (Fig. 4 B), are considered to reflect the two steps of voltage-dependent activity of Ci-VSP (Fig. 4 A). These results indicate that XtTPC3 is potentiated, presumably by PI(3,4)P₂ and/or PI(4,5)P₂ that is produced at a depolarized potential at around 30 mV and degraded at a further higher membrane voltage by Ci-VSP.

A long depolarizing pulse is known to increase PI(4,5)P₂ and PI(3,4)P₂ concentration in a different voltage-dependent manner in the endogenous system of Xenopus oocytes (Zhang et al., 2010; Ratzan et al., 2011; Kurokawa et al., 2012; Grimm and Isacoff, 2016). The two-pulse protocol was used to evaluate the voltage dependence of induction, similar to the Ci-VSP coexpression experiment (Fig. 5). The relationship between time constant of activation and prepulse membrane voltage showed that the V_1/2 value of induction was 32.9 ± 2.7 mV (n = 6). The voltage dependence of PI(3,4)P₂ production by the endogenous mechanism, which is monitored by a PI(3,4)P₂-specific fluorescent sensor, appears over +10 mV (Kurokawa et al., 2012). On the other hand, the V_1/2 value of PI(4,5)P₂ production by a long depolarization, monitored by the PI(4,5)P₂ sensitivity of KCNQ2/3 channel, was calculated as −26.1 mV in a previous report (Zhang et al., 2010)_. Therefore, PI(3,4)P₂ is a more plausible cause for induction.

The contribution of PI(3,4)P₂ and/or PI(4,5)P₂ to XtTPC3 was further verified by several lines of experiments. First, the effects of the Ci-VSP F161W/R232K mutant, which is known to attenuate the dephosphorylation activity from PI(3,4)P₂ and PI(4,5)P₂ to PI(4)P (Grimm and Isacoff, 2016), were investigated. As expected, coexpression of XtTPC3 with Ci-VSP F161W/R232K attenuated the decreasing phase under continuous depolarization (Fig. 4 B) and abolished the deceleration of activation at highly depolarized voltage in the two-pulse protocol (Fig. 4, F and G). Second, oocytes expressing XtTPC3 were pretreated with insulin, which is known to increase PI(3,4)P₂ and PIP₃ concentration through endogenous PI3K activity (Liu et al., 1995; Grimm and Isacoff, 2016). To evaluate the potentiation of XtTPC3 by insulin, we used a 100-mV test pulse to monitor the time constant of activation, and a 50-mV pulse for inward currents because the inward currents at 50 mV are a good indication of Na⁺ influx in XtTPC3 (Fig. 6). The results showed that insulin pretreatment accelerates the activation kinetics at 100 mV and increases the current amplitude at 50 mV, which indicates the occurrence of potentiation of XtTPC3 even without induction stimulus. The insulin pretreatment of oocytes coexpressing XtTPC3 with Ci-VSP showed that time constants of activation at 30 mV are not further accelerated by insulin, which supports the notion that insulin potentiates XtTPC3 through PIP₃ and/or PI(3,4)P₂ production (Fig. S1).

Third, we compared the time course of the change in PI(3,4)P₂ and PI(4,5)P₂ concentration and of the induction progress, using PI(3,4)P₂- and PI(4,5)P₂-specific fluorescence sensor proteins (Grimm and Isacoff, 2016). XtTPC3 was coexpressed with Ci-VSP and F-TAPP or F-PLC, the FRET-based fluorescent reporters for PI(3,4)P₂ and PI(4,5)P₂, respectively, and the XtTPC3 current and FRET change (YFP/CFP) were simultaneously monitored. The normalized recording of FRET using F-TAPP showed a rapid increase and subsequent decrease (Fig. 7 A), which reflects the increase and decrease in PI(3,4)P₂ concentration as previously reported (Grimm and Isacoff, 2016). The pattern of FRET changes of F-TAPP is apparently consistent with that of XtTPC3 currents. A double exponential function was fitted only to the current traces at 70 mV, because it was not well fitted to the apparent multistep increasing phase at 30 mV and 50 mV (See also Materials and methods). Comparison of the time constants of increasing and decreasing components between current and fluorescence showed that they are similar in order of magnitude (Fig. 7 B). On the other hand, the time course of FRET change from F-PLC showed a continuous decrease over the voltage range tested, which is clearly different from the current traces (Fig. 7 C). In addition, even in the absence of Ci-VSP, we detected a FRET increase of F-TAPP, but not of F-PLC (Fig. 7, D–F). These results suggest that the increase in PI(3,4)P₂ concentration causes the induction, namely XtTPC3 potentiation.

Finally, a series of PIs was directly injected into the oocytes expressing XtTPC3, one at a time. The injection of a soluble derivative of PI(4,5)P₂ to oocytes evoked the endogenous Ca²⁺-activated Cl⁻ currents that are thought to be caused by Ca²⁺ release through the soluble PI(4,5)P₂-activated IP₃ receptor (Fig. S2; Barish, 1983). Therefore, oocytes for PI(4,5)P₂ injection were pretreated with thapsigargin to deplete Ca²⁺ in the endoplasmic reticulum, and those for other PIs were incubated with BAPTA-AM to buffer the weak Ca²⁺ leak that occurs at high concentrations of PI injection. In addition, we used the same two parameters as shown in Fig. 6 A (Fig. 8 A), which are especially beneficial to minimize the residual contamination by the Cl⁻ current, because it appears as outward current at 50 mV while XtTPC3 shows inward current. Therefore, both the acceleration of activation kinetics at 100 mV and the increase in inward currents at 50 mV are appropriate for evaluation of XtTPC3 potentiation. Among all types of PIs, PI(3,4)P₂ was the most effective to accelerate the activation at 100 mV and to increase the currents at 50 mV, and PIP₃ was the second most effective (Fig. 8). The other PIs showed no significant potentiation effect. Considering the effects of Ci-VSP WT and F161W/R232K that indicated the potentiation by PI(3,4)P₂ increase, PI(3,4)P₂, rather than PIP₃, is thought to be the most potent XtTPC3 modulator among PI species in Xenopus oocyte membranes.

We next performed PI perfusion of inside-out patches from Xenopus oocytes expressing XtTPC3. Consistent with the PI injection experiments, PI(3,4)P₂ increased the XtTPC3 current (Fig. 9 A). The fold increase of the current amplitude by PI(3,4)P₂ application was significantly larger than that by PI(4,5)P₂, which clearly has no effects on XtTPC3 (Fig. 9, A and C). PIP₃ also tended to potentiate XtTPC3, although its effect is not significant. Unexpectedly, we found that PI(3,5)P₂ also potentiates XtTPC3 to an extent comparable to PI(3,4)P₂, which is inconsistent with the PI injection experiment (Fig. 8). The successive application of PI(3,4)P₂ and PI(3,5)P₂ to an inside-out patch clearly showed the potentiation effects of both PI species to XtTPC3 (Fig. 9 B). Given the results of direct PI injection in Xenopus oocytes, PI(3,4)P₂ is considered to be an effective ligand of XtTPC3 expressed in the plasma membrane. The apparent discrepancy in the results of Fig. 8 and Fig. 9 might be due to the preferable subcellular localization of injected PI(3,5)P₂ to organelle membranes, rather than the plasma membrane (Marat and Haucke, 2016), or due to any mechanisms in XtTPC3 and/or cytoplasm that perturb PI(3,5)P₂ sensitivity in the plasma membrane in whole oocytes.

Recent cryo-EM structures of MmTPC1 revealed the critical residues for PI(3,5)P₂ binding in domain I, which are located in the H1-IS0 loop in the N-terminal cytoplasmic region, IS3, the IS4-S5 linker, and IS6 (Fig. 10, A–C; She et al., 2018). Most of them are basic residues, such as arginines and lysines, that interact with negative charges of the phosphate groups of PI(3,5)P₂. Because many basic residues in these regions are also conserved in XtTPC3 (Fig. 10 C), we assumed that they form the PI(3,4)P₂ binding site and hence are important for induction. To gain insight into the structural basis of PI(3,4)P₂ recognition in XtTPC3, we built the MmTPC1 structure-based homology model, where PI(3,5)P₂ was simply replaced with a PI(3,4)P₂ molecule.

Mutations of the basic residues in the IS4-S5 linker and IS6 were found to affect the induction in a slightly different way from the case of MmTPC1 (Fig. 10, D–F; SI Appendix; and Fig. S3). In the IS4-S5 linker, mutations of both Arg187 and Arg191, whose corresponding arginines in MmTPC1 are essential to PI(3,5)P₂ binding, abolished the induction almost completely. It is notable that the mutations of Arg194 in the IS4-S5 linker and of Arg297 in IS6 also significantly attenuated the induction, although their corresponding residues are not conserved in TPC1 (Fig. 10 F and Fig. 11). Besides in the IS4-S5 linker and IS6, a basic residue in IS3 (Lys137 in XtTPC3 and Arg173 in MmTPC1) was also indispensable for both induction in XtTPC3 and PI(3,5)P₂ binding in MmTPC1. In contrast to domain I, neutralizing mutations of positively charged amino acid residues in domain II (IIS4-S5 linker and IIS6) did not show significant effects on induction (Fig. 10 F and Fig. S3).

Next, to prove that the residues important for induction actually contribute to PI(3,4)P₂ binding, the effects of Ci-VSP coexpression and PI(3,4)P₂ injection were investigated using R187Q, which was chosen as a representative mutant lacking induction. As expected, R187Q was less affected by Ci-VSP coexpression (Fig. 10 G and Fig. S4). Also, an injection of PI(3,4)P₂ into oocytes expressing R187Q showed no significant acceleration of the activation kinetics at 100 mV (Fig. S5). Furthermore, perfusion of either PI(3,4)P₂ or PI(3,5)P₂ failed to potentiate R187Q (Fig. 10, H and I). These results demonstrate that the basic residues responsible for induction form the PI(3,4)P₂ binding site in XtTPC3.

In the H1-IS0 loop region of MmTPC1, a neutral residue (Asn85) was shown to be important for recognizing the C4 hydroxyl group of PI(3,5)P₂ (Fig. 12, A and B; She et al., 2018). We evaluated the effects of alanine mutations of three residues in the H1-IS0 loop of XtTPC3 (Asn52, Arg53, and Asn54) on induction (Fig. 12, C and D). In contrast to MmTPC1, alanine substitution of Asn52, which corresponds to Asn85 in MmTPC1, did not show significant effects on induction, but alternatively, N54A did. The mutation of Arg53 revealed limited contribution of the basic residue at this position. The results indicate that Asn54, a neutral amino acid in a different position from TPC1, contributes to PI(3,4)P₂ recognition in XtTPC3.

As this region is considered to be responsible for the PI selectivity, we further evaluated the mutations of Arg53 and Asn54, including the replacement with those of TPC1. Various mutations of Asn54, including lysine, aspartate, glutamine, and threonine, failed to show an extent of induction comparable to WT (Fig. 12 D). These results indicate that the asparagine residue at this position is not replaceable for efficient PI(3,4)P₂-induced potentiation. In contrast, the lesser effects of both R53Q and R53A suggest a supplemental role for induction, while the reversed charge (R53D) was not permissive. R53D/N54K, in which the phosphate recognition site is changed to TPC1-type, completely abolished induction. Also, the PI injection experiments demonstrated that R53D/N54K lost the PI(3,4)P₂ sensitivity, similar to R187Q (Fig. 12, E and F). These results clearly indicate that the H1-IS0 loop is critical for PI(3,4)P₂ recognition, possibly in terms of PI selectivity.

Our preceding data suggest that PI(3,4)P₂ binding in XtTPC3 causes the potentiation of voltage dependence, which is hardly observed in the case of the binding of PI(3,5)P₂ to TPC1 (Cang et al., 2014b). Consequently, there should exist a mechanism unique in XtTPC3 for the coupling of the PI binding in domain I with the voltage-dependent gating that is governed by domain II. The XtTPC3 structural model indicates that Arg297 in IS6, which was identified to be important for induction (Fig. 10, D–F), apparently does not participate in the PI binding site, but has an electrostatic interaction with Glu665 in IIS6 (Fig. 13, A and B). In addition to Arg297, the hydroxyl group of Tyr293, which is replaced with phenylalanine in MmTPC1, also interacts with Glu665 with a hydrogen bond. These interactions are likely to contribute to the induction by coupling the PI-induced IS6 movement to the voltage-dependent IIS6 movement, rather than as the PI binding site itself.

To prove the electrostatic interaction between Arg297 and Glu665, the countercharges were introduced to Arg297 and Glu665 separately and together. R297E attenuated the induction, similar to R297Q (Fig. 13, C–E). Because E665R did not show detectable currents, the neutralizing mutant, E665Q, was alternatively evaluated (Fig. 13 C). Although the current amplitude of E665Q before induction was too small to be analyzed, the G–V relationship after induction showed less extent of induction than WT, similar to R297E (Fig. 13, C–E). In addition, the fatal effect of E665R was rescued by introduction of R297E, producing a charge-swapped double mutation (R297E/E665R), which has a V_1/2 value after induction comparable to that of WT.

In addition to the electrostatic interaction between Arg297 and Glu665, another contribution of Tyr293 was also evaluated. Unexpectedly, Y293A and even Y293F, which removes only the hydroxyl group, did not produce detectable current (Fig. 13 C). Based on this result, which indicates the critical importance of the polar moiety at the tip of a bulky side chain to form the hydrogen bond with Glu665, the mutations to glutamine and histidine in the 293 position were evaluated. As expected, both Y293Q and Y293H retained the current with the induction property. Although the current before induction of Y293H was too small to be analyzed (Fig. 13 B), the V_1/2 value after induction showed less change (Fig. 13 C). Y293Q also showed a similar V_1/2 after induction. These results indicate that the induction requires a bridging interaction of Tyr293, Arg297, and Glu665, which is likely to couple the PI-induced movement of IS6 with the voltage-dependent movement of IIS6.

By the addition of the present study, all the members of the TPC family are demonstrated to be PI-sensitive channels, although TPC3 is unique in its PI(3,4)P₂ sensitivity. Among the PI species that predominantly localize in the plasma-membrane (PIP₃, PI(4,5)P₂, and PI(3,4)P₂), XtTPC3 is more clearly potentiated by PI(3,4)P₂ and completely insensitive to PI(4,5)P₂ (Fig. 8 and Fig. 9). Some inwardly rectifying K⁺ channels, such as G protein–coupled inwardly rectifying potassium channel 1/2 (GIRK1/2), GIRK1/4, renal outer medullary potassium channel 1 (ROMK1), and ATP-sensitive potassium channel (K_ATP), are known to respond to PI(3,4)P₂ to a similar or lesser extent compared with PI(4,5)P₂ (Ho and Murrell-Lagnado, 1999; Rohács et al., 1999, 2003; Zeng et al., 2002). Epithelial Na⁺ channels are activated by both PI(3,4)P₂ and PIP₃, whose preference is unclear, and requires PI(4,5)P₂ for its basal activity (Tong et al., 2004; Pochynyuk et al., 2007). The activation of TRPML1 by PI(3,5)P₂ is competitively inhibited by PI(3,4)P₂, but also by PI(4,5)P₂ and PIP₃ (Zhang et al., 2012). Therefore, the PI(3,4)P₂ preference of XtTPC3 on the plasma membrane is a unique feature in the PI dependence of ion channels.

The PI perfusion experiment using inside-out patches, a simpler system lacking cytosolic components, clearly showed the PI(3,5)P₂ sensitivity (Fig. 9). The inconsistency of this result with the direct PI injection experiment (Fig. 8) indicates that some mechanisms in the whole oocyte prevent PI(3,5)P₂ action on XtTPC3. PI(3,5)P₂ is known to be selectively accumulated in endosomes and lysosomes, and its localization is strictly regulated by some PI kinases, phosphatases, and binding proteins (Marat and Haucke, 2016). Therefore, the PI(3,5)P₂ injected into Xenopus oocytes might be minimally localized in and/or translocated to the plasma membrane. Another possibility is the existence of any mechanism of XtTPC3 and/or cytoplasm to regulate its PI sensitivity. For example, the phosphorylation of TPC3, which could be affected by membrane excision, might regulate its PI sensitivity. Although the issue of the difference in the sensitivity to PI(3,5)P₂ in different conditions remains, the results imply that PI(3,5)P₂ could not behave as a ligand for XtTPC3 expressed in the plasma membrane in whole oocytes.

A previous work reported that TPC3 from D. rerio expressed in endolysosomal membranes is insensitive to PI(3,5)P₂ (Cang et al., 2014a). This is in remarkable contrast with our inside-out patch experiments (Fig. 9). We think it is due to low luminal pH in their experimental condition, since we newly found that low extracellular (i.e., luminal) pH (pH 4.6) significantly inhibited the “induced” (PI-bound) XtTPC3 (Fig. 14). Therefore, TPC3 in endosomes and lysosomes cannot be highly activated due to their low luminal pH (Xu and Ren, 2015), despite the enriched PI(3,5)P₂ there. These notions indicate that the actual physiological ligand and functioning location for TPC3 is PI(3,4)P₂ and the plasma membrane, rather than PI(3,5)P₂ and endosomes and lysosomes.

The residues important for PI recognition in XtTPC3 are quite similar to those for PI(3,5)P₂ in TPC1, although there are some differences, which could explain the PI(3,4)P₂ sensitivity of XtTPC3 (Fig. 10, B–F). In the MmTPC1 structure, three arginines in the IS4-S5 linker form salt bridges with the C3 phosphate of PI(3,5)P₂ and hence contribute to the exclusion of PIs lacking C3 phosphate, such as PI(4,5)P₂. XtTPC3 retains the corresponding arginines, two of which are important for induction (Arg187 and Arg 191), but the other one (Arg188) is apparently less critical (Fig. 10 F). Alternatively, Arg194, which is not conserved in TPC1, is critically important. The XtTPC3 homology model indicates that Arg187 and Arg191 interact with the C3 phosphate, similar to MmTPC1, and that Arg194 is located close to the carbonyl oxygens of acyl chains (Fig. 10 B). The model also shows that the side chain of Arg194 would collide sterically with the PI molecule. Therefore, to accommodate the acyl chains near Arg194, the PI molecule might be pushed and rotated from the original position, but the extensive salt bridges for C3 phosphate imply that the C3 position could be maintained. This repositioning might direct the C4 and C5 positions to the appropriate location to interact with residues in the H1-IS0 loop region, and therefore, the precise PI position has to be modified from our original model (Fig. 10 B and Fig. 12 A).

To recognize PI(3,4)P₂, Asn54 in the H1-IS0 region is thought to be critical (Fig. 12). The large effects on induction by a slight change of the side chain, such as elongation (glutamine) and a negative charge with the same size (aspartate), indicate the asparagine residue here is needed to exactly recognize PI(3,4)P₂. The intolerance of positively charged lysine and the mild effect of the substitution of polar threonine suggest that Asn54 does not interact with the negatively charged phosphate groups, but is likely to form a hydrogen bond with the polar part of PI(3,4)P₂ (Fig. 12, C and D). In the XtTPC3 model, Asn54 is located nearest to the C5 hydroxyl group of the PI(3,4)P₂ molecule, although they are located a little too distant from each other to form a hydrogen bond. This positional relationship is similar to the model for PI(3,5)P₂ discrimination in MmTPC1, where the amino group of Asn85 interacts with the C4 hydroxyl group but would sterically hinder binding in the presence of a C4 phosphate group (She et al., 2018). The repositioning, caused by the interaction with Arg194 as mentioned in the previous paragraph, may move the C5 hydroxyl group of PI(3,4)P₂ closer to Asn54, to form a hydrogen bond. PI(3,5)P₂ might somehow bind to XtTPC3 by the contribution of Asn52, whose corresponding residue in MmTPC1 is important for PI(3,5)P₂ binding. This model could be verified by further structural studies of TPC3.

Strong PI-dependent modulation of the voltage dependence is another special feature of TPC3. It is presumably achieved by two mechanisms, namely PI-induced IS6 movement, which is thought to be similar among TPCs, and a tight coupling of IS6 with IIS6. Based on the cryo-EM structures of MmTPC1 with or without PI(3,5)P₂, PI binding induces a conformational rearrangement in the H1-IS0 loop, the cytosolic part of VSD1, and the linker region between domain I/II, and finally, leads to IS6 movement to expand the activation gate (Fig. S6). In MmTPC1, although the PI binding increases macroscopic current amplitudes, it induces only a slight or no shift of voltage dependence (Cang et al., 2014b; She et al., 2018). This indicates that PI-induced IS6 movement and voltage-dependent IIS6 movement are relatively independent in TPC1. In the case of TPC3, the interaction between Tyr293 and Arg297 in IS6 and Glu665 in IIS6 is expected to strengthen the IS6–IIS6 coupling (Fig. 13 B). Because the disruption of this interaction, such as in R297Q, R297E, and Y293Q, showed no profound effects on the V_1/2 before induction (Fig. 10 F and Fig. 13 E), the interaction is likely to be formed in the PI(3,4)P₂-bound state (Fig. 15). The tight interaction between IS6 and IIS6 could transmit the conformational change of IS6 induced by PI(3,4)P₂ binding to IIS6 to facilitate its movement for gate opening (Fig. 15). As less energy is required to direct IIS6 toward its opening conformation thanks to PI(3,4)P₂-bound IS6, VSD2 needs less voltage stimulus to open the gate. This could be the cause of the negative shift of the G–V relationship after induction.

As Glu665 in IIS6 is conserved and the position for Arg297 in IS6 is replaced with asparagine in TPC1 orthologues (Fig. 11), the weak interaction using a hydrogen bond between glutamate and asparagine might be evolutionally conserved between IS6 and IIS6 in TPC1s. The necessity of, at least, the minimum interaction might reflect the indispensable linking between IS6 and IIS6 for channel gating, regardless of the presence or absence of PI-dependent modulation of the voltage dependence. In XtTPC3, a strong interaction is formed by a salt bridge and supported further by the hydroxyl group of Tyr293 (Fig. 13 B). Notably, Arg297 and Tyr 293 are also conserved among TPC3 orthologues (Fig. 11). These observations suggest that, in each TPC family, the extent of mutual dependence between PIs and voltage is optimized through the IS6–IIS6 interactions. The weak interaction between IS6 and IIS6 means that TPC1 opens only when both voltage and PI are sufficiently applied, while the tight interaction in TPC3 could integrate the two stimuli, where one stronger stimulus might compensate for the other weaker stimulus (Fig. 15). Interestingly, the tyrosine and the basic residues in IS6 are also conserved in the TPC2 family, as well as in TPC1 from A. thaliana (AtTPC1), which is gated by voltage and Ca²⁺ (Fig. 11; Guo et al., 2016; Kintzer and Stroud, 2016). Because TPC2 is a PI-dependent and non–voltage-dependent channel (Wang et al., 2012), PI binding needs to provide the energy sufficient to open the gate without any contribution from VSDs. Very recently solved cryo-EM structures of human TPC2 in the apo- and PI(3,5)P₂-bound states reveal that the distance between those residues decreases in the PI(3,5)P₂-bound state (She et al., 2019). The tight IS6–IIS6 interaction identified in this study might contribute to enabling movement of both IS6 and IIS6, so that gate opening in TPC2 occurs in response to PI binding alone. In AtTPC1, Ca²⁺ binding to the linker region between domains I and II, which is located just beneath IS6, is known to modulate the voltage dependence (Guo et al., 2016). Therefore, similarly to XtTPC3, Ca²⁺-induced IS6 movement might affect the voltage-dependent IIS6 movement through the IS6–IIS6 interaction in AtTPC1.

The PI(3,4)P₂-dependent modulation is thought to be evolutionally conserved among TPC3 orthologues, because the basic residues important for induction are almost completely conserved among most TPC3 orthologues (Fig. 11). TPC3 mRNA or protein is reported to exist in oocytes of sea urchin (Ruas et al., 2010; Patel et al., 2011) and X. laevis (Cang et al., 2014a), and known to play a role in fertilization and development in starfish oocytes (Ramos et al., 2014). Some eggs, including those of Xenopus, show a long-lasting depolarization, the so-called fertilization potential (Cross and Elinson, 1980; Grey et al., 1982), that is induced by sperm entry and causes polyspermy block. Interestingly, in X. laevis, Na⁺ conductance is reported to transiently appear during the fertilization potential in eggs (Peres and Mancinelli, 1985) or by applying a long depolarization in oocytes (Baud et al., 1982). In addition, the PI3K activity that produces PI(3,4)P₂ is known to increase before fertilization, during the maturation step, in Xenopus oocytes (Liu et al., 1995; Hehl et al., 2001). Taking these data into consideration, TPC3 is likely to be activated at fertilization, as both depolarization voltage and PI(3,4)P₂ stimuli are provided, and to contribute to the fertilization potential through its Na⁺ permeability. These data suggest a contribution of PI(3,4)P₂ to oocyte fertilization through TPC3.

All the data sets generated for this study are available from the corresponding author in response to reasonable requests.

We thank Dr. Y. Fujiyoshi (Nagoya University, Nagoya, Japan) for kindly providing us with the XtTPC3 cDNA, Dr Y. Okamura (Osaka University, Osaka, Japan) for the Ci-VSP cDNA, and Dr. E.Y. Isacoff (University of California, Berkeley, Berkeley, CA) for F-TAPP and F-PLC cDNAs. We also thank Ms. C. Naito for technical support, and all the members of the Kubo laboratory for valuable discussion. We thank Dr. A. Collins (Saba University School of Medicine, Dutch Caribbean) for editing and refining the manuscript.

This work was supported by the Grant-in-Aid for Young Scientists (18K15019 to T. Shimomura) and the Scientific Research (B) (17H04021 to Y. Kubo) from the Japan Society for the Promotion of Science.

The authors declare no competing financial interests.

Author contributions: T. Shimomura conceived the study with Y. Kubo and designed the experiments. T. Shimomura performed all experiments and data analyses. T. Shimomura wrote the initial draft, and T. Shimomura and Y. Kubo completed the manuscript. Both authors have approved the final version of the manuscript.

Merritt C. Maduke served as editor.