The Selectivity Filter of the Cation Channel TRPM4*

Transient receptor potential channel melastatin subfamily (TRPM) 4 and its close homologue, TRPM5, are the only two members of the large transient receptor potential superfamily of cation channels that are impermeable to Ca2+. In this study, we located the TRPM4 selectivity filter and investigated possible structural elements that render it Ca2+-impermeable. Based on homology with known cation channel pores, we identified an acidic stretch of six amino acids in the loop between transmembrane helices TM5 and TM6 (981EDMDVA986) as a potential selectivity filter. Substitution of this six-amino acid stretch with the selectivity filter of TRPV6 (TIIDGP) resulted in a functional channel that combined the gating hallmarks of TRPM4 (activation by Ca2+, voltage dependence) with TRPV6-like sensitivity to block by extracellular Ca2+ and Mg2+ as well as Ca2+ permeation. Neutralization of Glu981 resulted in a channel with normal permeability properties but a strongly reduced sensitivity to block by intracellular spermine. Neutralization of Asp982 yielded a functional channel that exhibited extremely fast desensitization (τ < 5 s), possibly indicating destabilization of the pore. Neutralization of Asp984 resulted in a non-functional channel with a dominant negative phenotype when coexpressed with wild type TRPM4. Combined neutralization of all three acidic residues resulted in a functional channel whose voltage dependence was shifted toward very positive potentials. Substitution of Gln977 by a glutamate, the corresponding residue in divalent cation-permeable TRPM channels, altered the monovalent cation permeability sequence and resulted in a pore with moderate Ca2+ permeability. Our findings delineate the selectivity filter of TRPM channels and provide the first insight into the molecular basis of monovalent cation selectivity.

Transient receptor potential channel melastatin subfamily (TRPM) 4 and its close homologue, TRPM5, are the only two members of the large transient receptor potential superfamily of cation channels that are impermeable to Ca 2؉ . In this study, we located the TRPM4 selectivity filter and investigated possible structural elements that render it Ca 2؉ -impermeable. Based on homology with known cation channel pores, we identified an acidic stretch of six amino acids in the loop between transmembrane helices TM5 and TM6 ( 981 EDMDVA 986 ) as a potential selectivity filter. Substitution of this six-amino acid stretch with the selectivity filter of TRPV6 (TIIDGP) resulted in a functional channel that combined the gating hallmarks of TRPM4 (activation by Ca 2؉ , voltage dependence) with TRPV6like sensitivity to block by extracellular Ca 2؉ and Mg 2؉ as well as Ca 2؉ permeation. Neutralization of Glu 981 resulted in a channel with normal permeability properties but a strongly reduced sensitivity to block by intracellular spermine. Neutralization of Asp 982 yielded a functional channel that exhibited extremely fast desensitization ( < 5 s), possibly indicating destabilization of the pore. Neutralization of Asp 984 resulted in a non-functional channel with a dominant negative phenotype when coexpressed with wild type TRPM4. Combined neutralization of all three acidic residues resulted in a functional channel whose voltage dependence was shifted toward very positive potentials. Substitution of Gln 977 by a glutamate, the corresponding residue in divalent cation-permeable TRPM channels, altered the monovalent cation permeability sequence and resulted in a pore with moderate Ca 2؉ permeability. Our findings delineate the selectivity filter of TRPM channels and provide the first insight into the molecular basis of monovalent cation selectivity. TRPM4 1 is a Ca 2ϩ -and voltage-dependent non-selective cation channel belonging to the melastatin subfamily of transient receptor potential (TRP) membrane proteins (1,2). It has been proposed to be the molecular correlate of Ca 2ϩ -activated nonselective cation channels in several excitable and non-excitable cell types, and it has been implicated in important physiological processes including T-cell activation, myogenic vasoconstriction, and cardiac function (3)(4)(5).
TRPM4 and its close homologue, TRPM5, exhibit two salient features that are unique within the TRP superfamily. First, they represent the only known TRP channels that are directly gated by increases in intracellular Ca 2ϩ ([Ca 2ϩ ] i ) (1, 2, 6 -11). The Ca 2ϩ sensitivity of TRPM4 activation is strongly modulated by several cellular factors, including protein kinase C phosphorylation, calmodulin binding, and ATP (11,12). Second, both channels are impermeable to Ca 2ϩ (1, 6 -8). This contrasts with all other functionally expressed TRPs, which form either Ca 2ϩ -permeable non-selective cation channels or even highly Ca 2ϩ -selective channels.
The structural basis of TRPM channel permeation has not yet been studied. The region between the fifth and sixth transmembrane helices (TM5 and TM6), which is known to form the pore in the other tetrameric cation channels, shows only limited sequence homology to other (TRP) cation channels but is highly conserved among members of the TRPM subfamily. This region consists of a conserved hydrophobic region, a putative pore helix, followed by a hydrophilic region that contains a fully conserved aspartate residue (residue 984 in TRPM4), which may be part of the selectivity filter (13). There are, however, some clear sequence differences. In TRPM4 and TRPM5, the putative selectivity filter is highly acidic, with a cluster of three (TRPM4) or four (TRPM5) aspartates or glutamates. The other TRPM channels, which all exhibit some degree of Ca 2ϩ permeability, have only one or two acidic residues in this region. Moreover, the span between the putative pore helix and selectivity filter is one amino acid shorter in TRPM2 and TRPM8 than in the other TRPM channels. We hypothesized that the lack of Ca 2ϩ permeability of the TRPM4 and TRPM5 pores may be related to these structural differences.
In this study, we used a site-directed mutagenesis approach to investigate how changes to the putative selectivity filter affect the pore properties of TRPM4. Our data indicate that this region determines the TRPM4 permeability properties and its sensitivity to block by intracellular spermine.

EXPERIMENTAL PROCEDURES
Cell Culture-HEK293 human embryonic kidney cells were grown in Dulbecco's modified Eagle's medium containing 10% (v/v) human serum, 2 mM L-glutamine, 2 units/ml penicillin, and 2 mg/ml streptomycin at 37°C in a humidity-controlled incubator with 10% CO 2 .
Transient Expression of Human TRPM4 and Mutagenesis-We used the recombinant bicistronic expression plasmid pdiTRPM4b, which carries the entire protein-coding region for the human TRPM4b (Gen-Bank TM accession number AX443227) (2) and for the green fluorescent protein (GFP) coupled with an internal ribosomal entry site sequence. HEK293 cells were transiently transfected with the pdiTRPM4b vector using previously described methods, and successfully transfected cells were visually identified by their green fluorescence in the patch clamp set up (2). For all mutations and deletions we used the standard PCR overlap extension technique (14) with the human TrpM4 cDNA constructed in the pCAGGSM2/IresGFP. All constructs were verified by sequencing analysis. Cell Surface Biotinylation Assay-GFP-tagged TRPM4 and mutants were expressed in HEK293 cells, labeled with biotin on ice, and analyzed after precipitation with streptavidin-agarose exactly as previously described (11).
Solutions-This study was mostly carried out in cell free inside-out patches. The standard pipette solution for inside-out patch clamp measurements contained 150 mM NaCl, 5 mM CaCl 2 , 1 mM MgCl 2 , 10 mM glucose, and 10 mM HEPES, buffered at pH 7.4 with NaOH. In some experiments, 100 mM CaCl 2 in the pipette solution was used, buffered with Ca(OH) 2 to pH 7.4. Before patch excision, the extracellular bath solution was changed to an "internal solution" for inside-out patch clamp measurements, which contained 150 mM NaCl (or KCl, LiCl, or CsCl), 1 mM MgCl 2 , 10 mM HEPES, 5 mM EGTA ,and 5.297 mM CaCl 2 (resulting in 300 M free [Ca 2ϩ ] i ) calculated by the CaBuf program (ftp://ftp.cc.kuleuven.ac.be/pub/droogmans/cabuf.zip). Note that we added 5 mM EGTA to have the same amount of EGTA at the inner site of the membrane for comparison with experiments in which Ca 2ϩ was changed (2,10,11). The pH of all solutions was adjusted to 7.2 with NaOH (LiOH, KOH, or CsOH). In all inside-out studies, internal solutions were ATP-free. In whole-cell measurements for the TRPV6-TRPM4 pore swap, extracellular solution consisted of 150 mM NaCl, 5 mM KCl, 1.5 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES. This solution was made divalent-free by omitting CaCl 2 and MgCl 2 and adding 2 mM EDTA. For experiments with only Ca 2ϩ as charge carrier, 30 mM CaCl 2 was added together with 105 mM NMDG and 10 mM HEPES, adjusted with HCl to pH 7.2. Intracellular solution in these experiments consisted of 156 mM CsCl, 1 mM MgCl 2 , 5 mM EGTA with 10 mM HEPES, adjusted with CsOH to pH 7.2. All experiments were performed at room temperature (22-25°C).
Electrophysiology-Currents were monitored in the whole-cell or inside-out patch clamp configurations with an EPC-9 amplifier (HEKA Elektronik, Lambrecht, Germany). Patch electrodes had a DC resistance between 2 and 4 megaohms. An Ag-AgCl wire was used as a reference electrode. Currents were sampled at 2-4 kHz and filtered at 1 kHz. Either voltage ramps or steps were applied to the whole-cell and excised patches. For most experiments, the holding potential was 0 mV. Voltage ramps were applied from Ϫ100 to ϩ100 mV within 400 ms (for details, see the figure legends). In most of the experiments, step protocols consisted of a 500-ms step to Ϫ100 mV, followed by a 250-ms step to ϩ100 mV. The interval between the pulses was 2 s. Patches were excised in a bath of various Ca 2ϩ (intracellular) concentrations. The first data point during a time course experiment was obtained within 2 s after excision, which corresponds to the stimulation interval. All currents were corrected for capacitance and leakage currents.
Data Analysis-The modeling of the pore structure was done by using the Swiss Model server (SIB Biozentrum Basel, swissmodel. expasy.org) (15)(16)(17). For modeling of the TRPM4 pore, we used the KscA pore as a template (18). All models were viewed by DS ViewerPro 5.0 software (Accelrys Inc., San Diego, CA). Electrophysiological data were analyzed using WinASCD software (ftp://ftp.cc.kuleuven.ac.be/pub/ droogmans/winascd.zip). Pooled data are given as the mean Ϯ S.E. of n cells. Origin 7.0 (OriginLab, Northampton, MA) was used to fit doseresponse curves and the kinetic model described below. Significance was tested with the unpaired Student's t test (significance was defined as p Ͻ 0.05).

RESULTS
Putative Structure of the TRPM4 Pore- Fig. 1A shows an alignment of the putative pore region of the TRPM subfamily and indicates the TRPM4 residues that were mutated in this study. Using the Swiss Model server, we modeled the structure of the loop between TM5 and TM6 based on the KcsA crystal structure (Fig. 1B). The resulting three-dimensional structure suggests that the selectivity filter is located between residues Glu 981 and Val 985 , with Glu 981 located at the inner entrance of the pore.
Permeation Profile of TRPM4 -Before evaluating the effects of the various mutations, we characterized the permeation profile of wild type (WT) TRPM4. The relative permeability of monovalent cations compared with Na ϩ was measured in inside-out patches by equimolar substitution of intracellular Na ϩ by the respective cation X ϩ and calculated according to Before calculation, the reversal potentials were corrected for liquid junction potentials (19) by where V LJ is the calculated liquid junction potential (Ϫ4.4 mV FIG. 1. Structure of the putative TRPM4 pore. A, alignment of the putative pore regions in the TRPM subfamily. A conserved hydrophobic region is present in all TRPMs and likely refers to the pore helix. Note the conserved negatively charged residues in all TRPMs. Highly conserved hydrophobic residues are shown in gray, the invariant aspartate is shown in red, and conserved residues with polar or charged groups are shown in orange (adapted from Ref. 13). B, modeling of the putative TRPM4 pore using the Swiss Model software (15)(16)(17) with the pore structure of KcsA as template (18).
for Cs ϩ , 0.4 mV for Na ϩ , 2.7 mV for Li ϩ , and Ϫ3.8 mV for K ϩ ) (see www.axon.com/axobits/AxoBits39.pdf and Ref. 20). In Eq. 1, ⌬V rev was Currents were activated by excision in 300 M [Ca 2ϩ ] i in symmetrical Na ϩ concentrations, and subsequently, Na ϩ was substituted by K ϩ , Cs ϩ , Li ϩ , or NMDG ϩ ( Fig. 2A). Reversal potentials were measured from voltage ramps (Fig. 2B). As summarized in Fig. 2C, we obtained a relative permeability sequence Na ϩ Ͼ K ϩ Ͼ Cs ϩ Ͼ Li ϩ . If the changes in conductance were analyzed (see also Fig. 2A), the sequence was K ϩ Ͼ Na ϩ Ͼ Cs ϩ Ͼ Li ϩ . The permeation sequence refers to an intermediate field-strength binding site (Eisenman sequence VII), which differs from the strong field-strength binding site (Eisenman sequence X) of TRPV5/TRPV6 (21,22) and the weak field-strength binding site (Eisenman sequence IV) of TRPV4 (23). In full agreement with previous studies (1), we found that TRPM4 is not measurably permeable for Ca 2ϩ (Fig. 2, E and F). Using Ca 2ϩ (100 mM) as the only charge carrier in the pipette solution, no inward currents could be measured using both step and ramp protocols. Fig. 2D shows a comparison with a pipette solution containing 140 mM Na ϩ . Even by applying huge driving forces for Ca 2ϩ entry (Ϫ160 mV), no inward Ca 2ϩ tail current could be detected (Fig. 2F). These data indicate that TRPM4 is virtually impermeable to Ca 2ϩ (P Ca /P Na Ͻ 0.001).
Swapping the Selectivity Filter between TRPV6 and TRPM4 -To examine whether the proposed location of the TRPM4 selectivity filter (see above) is correct, we constructed a chimeric channel in which the selectivity filter of TRPV6 (TI-IDGP), a highly Ca 2ϩ -permeable channel (24), substituted the putative selectivity filter of TRPM4 ( 981 EDMDVA 986 ). Wholecell currents through WT TRPM4 are not influenced by changing the extracellular concentration of divalent cations. If Ca 2ϩ is the only charge carrier, no inward current is measured up to ϳϪ80 mV (Fig. 3A). In step protocols from positive potentials to Ϫ100 mV (the usually applied protocol to measure TRPV5/6 currents, Ref. 25), a fast deactivating current can be seen in WT TRPM4, but no apparent inward current is seen with Ca 2ϩ as the sole charge carrier (Fig. 3B). In the selectivity filter swap mutant, replacing Na ϩ with Ca 2ϩ induces a shift of the reversal potentials toward positive potentials, and clear inward current can be measured (Fig. 3C). In step protocols, currents carried by monovalent cations deactivate very rapidly. However, a small inward current or 17 Ϯ 4 pA/picofarad (n ϭ 5, 50 ms after step to Ϫ100 mV) can also be measured when Ca 2ϩ is the only charge carrier (Fig. 3D, bath, 30 mM CaCl 2 , 105 mM NMDGchloride) indicating that this chimera is Ca 2ϩ -permeable. In addition, switching to the divalent-containing solution always induced a remarkable inhibition of the outward current, which was not observed for WT TRPM4 (Fig. 3, A and B for WT and C and D for the selectivity filter swap mutant).
In inside-out patches with the pipette solution containing only monovalent cations, activation of the chimeric mutant only occurred after patch excision in Ca 2ϩ -containing solutions. Current density increased when [Ca 2ϩ ] i was elevated (data not shown), indicating that Ca 2ϩ sensitivity is preserved.
As for WT TRPM4, activation of the chimera was followed by desensitization (Fig. 3, E and F). The current kinetics at negative and positive potentials are similar to those of WT TRPM4 (Fig. 3E). Moreover, the currents through the chimeric channel displayed voltage-dependent activation and deactivation kinetics similar to those of WT TRPM4.
However, when 5 mM Ca 2ϩ and 1 mM Mg 2ϩ were included in the pipette, currents through the chimeric channel were dramatically reduced (Fig. 3, F and H). Block by extracellular Ca 2ϩ of the outward current at ϩ100 mV occurred with an IC 50 of 431 M (Fig. 3H). These data indicate that, as is the case for TRPV6, divalent cations block the permeation of monovalent Note the shift in reversal potentials. C, permeation ratios as determined by Eqs. 1-3. Data points are from at least five measurements. D, comparison of currents in inside-out patches with 5 mM Ca 2ϩ plus 140 mM Na ϩ (black line) or 100 mM Ca 2ϩ (gray line) in the pipette. Currents for 100 mM Ca 2ϩ were scaled. Note the absence of the inward current at negative potentials and the nearly unchanged kinetics of the current activation at ϩ100 mV (stepped from a holding potential of 0 mV to Ϫ100 and ϩ100 mV). E, comparison of currents from voltage ramps with 5 mM Ca 2ϩ /140 mM Na ϩ (black line) and the complete substitution of cations by 100 mM Ca 2ϩ (gray line). No inward current was present. Outward currents are nearly unchanged (scaled for 100 mM Ca 2ϩ ). F, lack of Ca 2ϩ permeation under rigorous changes of the driving force for Ca 2ϩ . A test step to ϩ180 mV was applied, followed by back steps from ϩ120 to Ϫ160 mV (in 40 mV decrements). Again, no tail current is inwardly directed, indicating the complete absence of Ca 2ϩ permeation.
cations through the chimeric pore, however, with a lower affinity (26). From this, we conclude that the transplanted fragment of TRPV6 reconstituted a divalent cation-sensitive selectivity filter in the TRPM4 backbone. Other hallmarks of TRPM4 such as desensitization, voltage-dependent kinetics, and Ca 2ϩ sensitivity remain basically the same, indicating that these properties have other structural determinants.
Additional support for the correct identification of the selectivity filter was deduced from the following experiment. We noticed that a short stretch of amino acids C-terminal of TM5, 943 AYGVA 947 , displays significant homology to the conserved selectivity filter of K ϩ channels. To examine whether this region influences the permeability of TRPM4, we constructed two mutants (VA946/947DM and AVA943/946/947GDM), reproducing the KcsA-like pore sequence, GYGDM). Although both mutants were functional, currents in 300 M Ca 2ϩ were extremely small (at ϩ100 mV: 59 Ϯ 21 pA, n ϭ 6 for VA946/947DM and 58 Ϯ 31 pA, n ϭ 4 for AVA943/946/947GDM). We found no indication that this stretch could eventually be a part of the pore itself.
Point Mutations in the Putative Selectivity Filter-As shown in Fig. 1, the putative TRPM4 selectivity filter contains three negatively charged residues. We neutralized these residues to alanines, either individually or in all possible combinations, and studied the pore properties of these mutants.
Currents through E981A were large and showed a similar decay (desensitization) behavior as the wild type channel (Figs. 2A and 4D; see also Ref. 10). As measured from reversal potentials, E981A had a similar permeation profile as WT TRPM4 (Na ϩ ϳ K ϩ Ͼ Cs ϩ Ͼ Li ϩ ; Fig. 4A).
Because our modeling predicts that Glu 981 is located at the pore entrance from the inner site, we tested the sensitivity of this mutant to intracellular spermine. WT TRPM4 was completely blocked by 1 mM spermine (12), whereas the same concentration induced only 30% block of the E981A mutant (Fig. 4, F and G). This indicates that Glu 981 contributes to the blocking site for intracellular spermine.
Currents through D982A were also large immediately after excision in 300 M Ca 2ϩ , but they decayed so fast that an analysis of permeation or spermine block was impossible (Fig. 4E).
Currents through the double mutant E981A/D982A decayed slower in the presence of 300 M [Ca 2ϩ ] i , allowing further analysis. The selectivity sequence was not changed (data not shown), and block by 1 mM spermine was reduced to 27 Ϯ 9% (n ϭ 4), similar to the E981A mutation alone. Similar results were obtained for another double mutant, E981Q/D982N.
In contrast to the E981A and D982A mutants, no current could be measured for the D984A mutant or the E981A/D984A and D982A/D984A double mutants (Fig. 4H), suggesting that Asp 984 is crucial for the integrity of the TRPM4 pore. Strongly reduced currents could be measured in cells coexpressing WT TRPM4 and D984A in a 1:1 cDNA ratio. Desensitization properties were similar to the expression of WT alone (Fig. 5, A and  B). During the steady-state phase, cation substitutions were performed (Fig. 5, C and D). However, because it is difficult to measure reversal potentials, we have analyzed changes in conductance for Na ϩ , K ϩ , Cs ϩ , and Li ϩ (see the legend for Fig. 5C), which were not significantly different from the transfection with the WT alone (Fig. 5D). Average current amplitude immediately after application of 300 M Ca 2ϩ was 119 Ϯ 39 pA (n ϭ 8), which corresponds to 7% of the amplitude in cells expressing only the WT subunit (1559 Ϯ 197 pA; n ϭ 7; Fig.  5E). Assuming random assembly of WT and the mutant subunits, we would expect that ϳ6% of the tetrameric complexes would consist of four WT subunits and that all other complexes would contain at least one mutant subunit. Thus, the most straightforward explanation of the above data is that the D984A subunit is dominant negative and that the remaining  ). B, step protocol from ϩ70 mV holding to Ϫ100 mV. Note again the lack of inward current in the Ca 2ϩ solution and the fast deactivation of the inward current in the monovalent cation-containing, divalent-free solution. Note also the unchanged outward currents. C, same protocol as described in A, but with the chimera in which the TRPV6 selectivity filter replaced at the same position the putative selectivity filter of TRPM4 studied herein (see text; for the TRPV6 selectivity filter, see Ref. 24). After changing from the monovalent cation solution to the Ca 2ϩ solution, clear inward current was measured, and the I/V shifted to a more positive value (see inset). The outward current was dramatically reduced in the Ca 2ϩ solution. D, same protocol as described in B. A small inward current was measured in the Ca 2ϩ solution at Ϫ100 mV (see text). Note again the inhibition of the outward current. E, time course of currents at ϩ100 mV (triangles) and Ϫ100 mV (circles) in inside-out patches after excision for the chimera. Note the fast decay in divalent cation-free internal solutions. This decay and the current pattern are rather similar to WT TRPM4. The pipette solution was nominally Ca 2ϩ -and Mg 2ϩ -free and contained 2 mM HEDTA. F, the same experiment as described in E but with the presence of 5 mM Ca 2ϩ and 1 mM Mg 2ϩ in the patch pipette as in all other experiments with TRPM4. Currents were smaller than those in the absence of divalent cation and decayed very rapidly. G, same experiment as described in E showing the current trace immediately after excision as indicated by the black triangle in the time course. The inward current at Ϫ100 mV deactivated, and outward current at ϩ100 mV showed an activation phase similar to the kinetics of WT TRPM4 currents. H, the same experiment as described in currents in the coexpressing cells are conducted by channels consisting of four WT subunits. In line with this, the conductance sequence of these channels as shown above was identical to that of WT TRPM4.
One reason for the lack of function of the D984A mutation might be that this amino acid substitution interfered with the expression of the channel protein on the plasma membrane. To examine this, we performed a cell surface biotinylation assay.
As shown in Fig. 4I, D984A, E981A/D984A, and D982A/D984A are all expressed to similar levels as WT TRPM4, in terms of both total protein level and expression at the cell surface. Therefore, the loss of function of the D984A mutation is most likely due to a defect in permeation or gating, rather than a defect in the trafficking to the cell surface.
Surprisingly, combined mutation of all three negative charges, E981A/D982A/D984A, resulted in a functional channel. As shown in Fig. 6, depolarizing steps activate large currents in WT TRPM4 (10,11). Current-voltage curves were plotted from the currents at the end of the step pulses (Fig. 6,  A and B). From tail current analysis (see Refs. 2 and 10), activation curves were measured and fitted by the Boltzmann equation (Fig. 6C). From these fits, the open probability at ϩ100 mV was calculated (Fig. 6G). Activation of the triple mutant was strongly shifted toward positive potentials, and currents were only detectable at potentials of Ͼϩ100 mV (Fig. 6, DϪF). To estimate the shift in voltage-dependent activation for this mutant, the open probability at ϩ100 mV (P open, ϩ100 mV ) was compared and showed a dramatic reduction relative to the WT channel (Fig. 6G). In addition to the dramatic shift, currents measured in inside-out patches immediately after excision were much smaller in the E981A/D982A/ D984A mutant compared with the WT TRPM4 (1.9 Ϯ 1.1 nA, n ϭ 4 for WT as compared with 0.4 Ϯ 0.2 nA, n ϭ 5, for the triple mutant). Due to the fast desensitization of the triple mutant (data not shown), a complete permeation study was not possible. However, when all cations in the pipette were replaced by Ca 2ϩ , again no inward current could be measured in this mutant, indicating that it remains Ca 2ϩ -impermeable (Fig. 6H). Similar to the WT channel (Fig. 2F), with just Ca 2ϩ in the pipette, switching the test pulse from ϩ100 to Ϫ160 mV did not elicit any inward tail current (n ϭ 4, data not shown). Biotinylation experiments showed that the triple mutant was efficiently expressed on the plasma membrane as WT TRPM4 and other mutants that contain the D984A substitution (Fig. 4I).
Mutations in the Linker between Putative Pore Helix and Selectivity Filter-TRPM6 and TRPM7, which have relatively high permeability for divalent cations, contain a glutamate following the hydrophobic region, whereas TRPM4 and TRPM5 contain a glutamine residue at the corresponding site (Gln 977 in TRPM4). To check whether this residue interferes with divalent cation permeation, we studied the Q977E mutant. This mutant produced large currents and showed a similar decay as WT TRPM4. The permeability for monovalent cations was changed to Eisenman type VI (K ϩ Ͼ Na ϩ ϳ Cs ϩ Ͼ Li ϩ , Fig.  7, A and C). With Ca 2ϩ as the only charge carrier in the pipette, small but significant inward currents could be measured (insets in Fig. 7, B and D). The permeability of Ca 2ϩ relative to Na ϩ was calculated from the absolute reversal potential measured with the respective Ca 2ϩ concentration of 100 mM in the extracellular solution, according to Ref. 27 where P X represents the permeability of the divalent cation, and [X] e is extracellular concentration. A relative Ca 2ϩ permeability (P Ca /P Na ) of 0.05 Ϯ 0.02 (n ϭ 6) was obtained. Another property of this mutant was the much more pronounced outward rectification. The ratio between the current at Ϫ100 and ϩ100 mV was 0.03 Ϯ 0.008 (n ϭ 6) for Q977E and 0.22 Ϯ 0.06 (n ϭ 12) for the WT TRPM4 (p Ͻ 0.001). This rectification was independent of the presence of divalent cations in the pipette solution.
FIG. 4. Mutations of the negatively charged residues in the selectivity filter. A, currents from voltage ramps obtained from the pore mutation E981A. Patches were held at 0 mV and excised in 300 M Ca 2ϩ . Voltage ramps were applied from Ϫ100 to ϩ100 mV. Currents between Ϫ20 and ϩ40 mV are shown to visualize changes in the reversal potentials due to various cation substitutions at the inner side of the membrane as indicated. The gray trace represents the K ϩ current. B, current-voltage curves measured from voltage ramps for E981A/D982A (same protocol as described in A). Permeation properties were similar to the wild type. C, from the same experiments as shown in A and B, permeation ratios were determined for the mutants E981A (f) and E981A/D982A (•). Data points are from at least five cells. D, desensitization of current through E981A in the presence of 300 M Ca 2ϩ . A steady-state component remains from which the spermine block can be reliably measured. Data are obtained from step protocols at Ϫ100 and ϩ100 mV (peak currents). E, currents through the D982A mutant showed a very fast decay. The same protocol as described in D was used. F, application of 1 mM spermine from the inner side of the membrane reduced the current, however, to a much smaller degree for E981A than for WT TRPM4 (see Ref. 12). G, dose-dependent block of TRPM4 by spermine for the WT channel and the E981A mutant (12). At ϩ100 mV, the block by 1 mM spermine and ϩ100 mV was reduced from 92 Ϯ 4% for WT (n ϭ 4) to 35 Ϯ 12% for E981A (n ϭ 5). H, pooled data for currents at ϩ100 mV obtained from the first step after excision in 300 M Ca 2ϩ for all mutations in the putative selectivity filter. Data points are from at least five cells. I, transiently transfected HEK293 cells expressing GFP-tagged WT TRPM4 and the indicated mutants were biotinylated on ice, washed, lysed, and precipitated with streptavidin-agarose. The amounts of TRPM4 and mutants in total cell lysates (top panel) and streptavidin-agarose precipitated (bottom panel) were determined by Western blotting using anti-GFP antibodies.
In TRPM4/5, the linker between the end of the putative pore helix and the selectivity filter (Gly 976 ϪGln 980 in TRPM4) is one amino acid longer than that in the Ca 2ϩ -permeable TRPM2/8 channels. Surprisingly, deletion of Gln 980 in TRPM4 to mimic the corresponding regions of TRPM2/8 resulted in a functional channel exhibiting extremely fast current decay after patch excision (Fig. 8). Fig. 8A shows the normal time course of TRPM4 current after patch excision in 300 M Ca 2ϩ , as has been described in detail elsewhere (10). The first voltage steps after excision evoke large outward and inward currents (Fig.  8B). In contrast, currents through the ⌬Q980 mutant channel decayed already during the first step (Fig. 8, C and D). When all cations in the pipette were substituted by Ca 2ϩ , no inward tail currents could be observed, even upon application of large driving forces after current activation at ϩ180 mV, indicating that the mutant is not measurably Ca 2ϩ -permeable (Fig. 8E). DISCUSSION The Ca 2ϩ -activated non-selective cation channel TRPM4 represents a molecular candidate for a large number of functionally similar Ca 2ϩ -activated cation channels found in native cells types with, as a typical fingerprint, its lack of permeability for Ca 2ϩ . Together with TRPM5, this permeation feature is unique in the TRP superfamily. In this study, we localized the selectivity filter and identified amino acids that are crucial for cation permeation and spermine block in TRPM4.
The strongest evidence that the region between residues Glu 981 and Ala 986 forms the selectivity filter was obtained by replacing the putative selectivity filter of TRPM4 with the previously identified selectivity filter of TRPV6. This chimeric channel gives rise to large monovalent currents, but only in the absence of extracellular divalent cations. This is reminiscent of the potent block of monovalent TRPV5/6 currents by Ca 2ϩ and Mg 2ϩ (see Ref. 28 for a review). Importantly, the pore chimera showed no change in activation by Ca 2ϩ , desensitization, and voltage dependence, indicating that the molecular determinants of these gating parameters lie elsewhere in the TRPM4 protein (see also Ref. 11). Because of the fast current decay in the presence of extracellular Ca 2ϩ , a permeation analysis was difficult to perform. The inward rectification of the TRPV6 pore was not observed for the chimeric channel. However, this might indicate that rectification requires a larger part of the TRPV6 pore than only the selectivity filter. Consistent with this idea, FIG. 6. Effect of mutating all three negative charges in the putative selectivity filter. A, time course of currents activated by 300 M Ca 2ϩ in inside-out patches. From the holding potential of 0 mV, steps were applied from Ϫ100 to ϩ180 mV with a 40-mV increment and returned to Ϫ100 mV after each pulse. B, current-voltage curve measured at the end of the step pulses. C, open probabilities at different potentials of the WT channel were estimated from the tail currents. Currents were normalized to the extrapolated maximal current, I max , obtained from the fit with the Boltzmann equation (I ϭ I max /(1 ϩ exp(Ϫ(V Ϫ V1 ⁄2 )/s)). D, the same protocol as described in A, but for the E981A/D982A/D984A (EDD981XAAA) triple mutant. E, I/V curve as described in B, but for the triple mutant. F, open probability curve as described in C, but for the triple mutant. G, from the Boltzmann plots, the open probability at ϩ100 mV was calculated to compare the shift in voltage dependence for the triple mutant. Note that the open probability is strongly reduced for the E981A/D982A/D984A (EDD981XAAA) mutant. H, the same shift in voltage dependence can be deduced from experiments using voltage ramps (ramps from Ϫ150 to ϩ100 mV for WT and from Ϫ150 to ϩ200 mV for the triple mutant). Pipette solution contained 100 mM Ca 2ϩ and no other cation. The data also show that the triple mutant is Ca 2ϩ -impermeable.
FIG. 5. Effect of coexpression of TRPM4 wild type and the pore mutant D984A. A, time course of currents activated by 300 M Ca 2ϩ and voltage ramps as shown in the inset. Currents were small and decayed but reached a steady state similar to the WT TRPM4 currents. B, current-voltage curves obtained from A showing desensitization without a change in the reversal potential. C, in the steady-state phase, cation substitutions were performed. For analysis, currents for different cations at ϩ100 mV were normalized to that for Na ϩ (see data in D). D, data obtained from experiments shown in C were pooled for coexpression experiments (E, WT ϩ D984A) and for experiments in which only WT TRPM4 (•) was expressed. Data are nearly identical for both series. E, currents were measured immediately after patch excision in 300 M Ca 2ϩ at ϩ100 mV. Note the dramatic current reduction in the coexpression experiments (all cells were from the same batch). rectification properties of TRPV6 were also changed by mutations outside the selectivity filter (21,29).
Mutating the three acidic residues in the putative TRPM4 selectivity filter (Glu 981 , Asp 982 , and Asp 984 ) had important functional consequences. Neutralization of Glu 981 did not change the permeation sequence, but it significantly reduced the sensitivity to block by spermine. Most likely, negative charges at the inner mouth of the channel pore provide a binding site for positively charged polyamines, inducing open channel block (30). Neutralization of Asp 982 resulted in a functional channel exhibiting extremely fast and complete current decay after activation. Because pore mutations are not expected to influence Ca 2ϩ binding to the activation site(s) (11), we hypothesize that the stability of the pore might be deterio-rated in this mutant such that the pore collapses rapidly after channel activation by Ca 2ϩ . A similar collapse may be caused by the D984A mutation, which apparently acts as a dominant negative subunit.
The putative pore loop is most strongly conserved among TRPM2, TRPM8, TRPM4, and TRPM5. These four channels share a conserved glutamine and proline in the span between the hydrophobic pore helix and the selectivity filter. A clear difference is that this span is one residue shorter in TRPM2 and TRPM8 than in TRPM4 and TRPM5, which contain an additional glutamine. Deletion of this glutamine causes a rapid closure of the pore after channel activation by Ca 2ϩ . We hypothesize that an intact linker is essential for the stability of the pore. FIG. 7. Effect of the Q977E mutation within the linker between the pore helix and the putative selectivity filter. A, current-voltage relationships for the Q977E mutant obtained by voltage ramps from Ϫ100 to ϩ150 mV for different ions as indicated (same protocol as described in Fig. 4A, K ϩ trace is in gray). Note the more pronounced rectification as compared with WT TRPM4. 5 mM Ca 2ϩ and 1 mM Mg 2ϩ are present in all solutions. The inset shows an enlarged part of the I/V curve demonstrating that the Na ϩ current reverses at 0 mV. B, all cations in the pipette solution were substituted by 100 mM Ca 2ϩ . The reversal potential shifted toward more negative voltages, and inward currents were obtained (see inset, ramps from Ϫ100 to ϩ200 mV). C, permeation was measured from the experiments shown in A and B. Points are from at least four cells. Ratios were calculated according to Eqs. 1-4. D, protocol to check for a possible Ca 2ϩ permeation by imposing large changes in the driving force for Ca 2ϩ . A test step to ϩ120 mV was applied, followed by back steps from ϩ80 mV to Ϫ160 mV (40-mV decrement). Small inward currents were observed at the negative potentials Ϫ80 and Ϫ120 mV. Pipette solution contained Ca 2ϩ as the only cation (100 mM CaCl 2 , see B and also see Fig.  2F). The inset shows the I/V relationship for the peak tail currents: note the clear inward currents.
FIG. 8. Effect of deletion of Gln 980 within the linker between the pore helix and the putative selectivity filter. A, time course of desensitization in a cell transfected with WT TRPM4 (insideout, 300 M Ca 2ϩ , steps from 0 to Ϫ100 and ϩ100mV). B, current measured immediately after patch excision in 300 M Ca 2ϩ (same experiment as described in A). C, deletion of Gln 980 caused extremely fast decaying current. D, first current obtained after patch excision from the experiment shown in C. Note that the decay already occurred during the voltage step. E, the top panel shows the step protocol used to rigorously test Ca 2ϩ permeability of the ⌬Q980 mutant. From a test potential of ϩ180 mV, back steps to Ϫ80 and Ϫ160 mV were applied. All cations in the patch pipette were substituted by Ca 2ϩ . No inward tail current was detected despite the strong driving force, indicating Ca 2ϩ impermeability of the deletion mutant.
The E977Q mutation, which introduces the corresponding residue of the divalent cation permeable TRPM6/7 channels, induced a small change in monovalent permeability, which changed from Eisenman sequence VII to VI. More importantly, this mutation rendered the TRPM4 pore permeable to Ca 2ϩ , albeit to a relatively low extent (P Ca /P Na Ͻ Ͻ 1).
In conclusion, we propose that the stretch between Glu 981 and Ala 986 forms all or part of the selectivity filter of TRPM4. Given the strong sequence conservation, we hypothesize that the variable permeability properties of the TRPM channels are due to the subtle amino acid variations in this region.