Potassium Channel Modulation by a Toxin Domain in Matrix Metalloprotease 23*

Peptide toxins found in a wide array of venoms block K+ channels, causing profound physiological and pathological effects. Here we describe the first functional K+ channel-blocking toxin domain in a mammalian protein. MMP23 (matrix metalloprotease 23) contains a domain (MMP23TxD) that is evolutionarily related to peptide toxins from sea anemones. MMP23TxD shows close structural similarity to the sea anemone toxins BgK and ShK. Moreover, this domain blocks K+ channels in the nanomolar to low micromolar range (Kv1.6 > Kv1.3 > Kv1.1 = Kv3.2 > Kv1.4, in decreasing order of potency) while sparing other K+ channels (Kv1.2, Kv1.5, Kv1.7, and KCa3.1). Full-length MMP23 suppresses K+ channels by co-localizing with and trapping MMP23TxD-sensitive channels in the ER. Our results provide clues to the structure and function of the vast family of proteins that contain domains related to sea anemone toxins. Evolutionary pressure to maintain a channel-modulatory function may contribute to the conservation of this domain throughout the plant and animal kingdoms.

Mechanisms that fine tune the activity of potassium channels are crucial to a cell's ability to integrate and respond to a plethora of internal and external signals. Peptide toxins from venomous creatures have served as vital tools to define the molecular mechanisms underlying K ϩ channel function (1,2). It has been suggested that toxins evolved from endogenous genes that function in normal cellular pathways (3,4). Indeed, venomous creatures possess tox-ins with homology to several proteins, including acetylcholinesterases (5), phospholipases (6,7), nerve growth factor (8), endothelins (9), Lynx-1 (10,11), Kunitz-type serine protease inhibitors (12), and the ion channel regulatory (ICR) 5 domains of cysteinerich secretory proteins (CRISPs) (3,13,14). Mammalian proteins containing toxin-like domains (TxDs) that block K ϩ channels have not been characterized previously.
BgK, a 37-residue peptide toxin from the sea anemone Bunodosoma granulifera (15,16), and ShK, a 35-residue peptide toxin from the sea anemone Stichodactyla helianthus (17,18) are potent inhibitors of K ϩ channels. The Simple Modular Architecture Research Tool (SMART) (available on the World Wide Web) predicts the existence of a large superfamily of proteins that contain domains (referred to as ShKT domains in the SMART data base) resembling these two toxins (Fig. 1A). Many of these proteins (ϳ70 proteins) are metallopeptidases, whereas others are prolyl-4-hydroxylases, tyrosinases, peroxidases, oxidoreductases, or proteins containing epidermal growth factor-like domains, thrombospondin-type repeats, or trypsin-like serine protease domains (Fig. 1B). The only human protein containing a ShKT domain in the SMART data base is MMP23 (matrix metalloprotease 23). Matrix metalloproteases belong to the metzincin superfamily and play important roles in tissue remodeling, development, and the immune response (19).
MMP23 is expressed in many tissues and exists either as a type II transmembrane protein in ER/nuclear membranes or as a secreted form following cleavage of the RRRRY motif just N-terminal to the Zn 2ϩ -dependent metalloprotease domain (20 -23). The ShKT domain of MMP23 (MMP23 TxD ) lies between the metalloprotease domain and an immunoglobulinlike cell adhesion molecule (IgCAM) domain ( Fig. 2A). MMP23 has been implicated in prostate, brain, and breast cancer (24 -26). In humans, two related sequences, MMP23A (a pseudogene) and MMP23B, are co-located on chromosome 1p36 (20). We have investigated MMP23 to gain insight into the structure and physiological functions of ShKT toxin domains and describe the solution structure of the MMP23 TxD domain, its structural similarity to the sea anemone toxins BgK and ShK, and its functional role in blocking K ϩ channels.
The peptide was prepurified to ϳ75% using preparative RP-HPLC (Kromasil-C18 10u) and lyophilized. Purification of 250 mg yielded 40 mg of semipure material, which was folded by dissolving the reduced and prepurified peptide in DMSO and then diluting to 0.25 mg/ml with 10% DMSO, 10% isopropyl alcohol, 2 M guanidine HCl and adjusting the pH to 7.6 with NH 4 OH. The solution, which remained clear during the pH adjustment, was then allowed to oxidize overnight. Liquid chromatography-mass spectrometry samples were taken at 2, 4, 6, and 24 h. Two peaks with the correct mass formed initially, but with longer reaction times, one peak became dominant. After 36 h, the folding was stopped by acidifying the solution with HCl to a pH of 4.0. The solution was diluted 50% with H 2 O and filtered before preparative HPLC purification over an reverse phase high performance liquid chromatography C18 column using a gradient of 15-45% B in 45 min (buffers A and B were 0.1% trifluoroacetic acid in H 2 O and acetonitrile, respectively). Mass spectral analysis determined the (M ϩ H) to be 4427, consistent with the formation of three intramolecular disulfide bonds. The final yield of 26 mg was at a purity of 95%.
NMR Spectroscopy-Synthetic MMP23 TxD (6 mg) was dissolved in 600 l of H 2 O containing 6% 2 H 2 O, and the pH was adjusted to 5.0. Two-dimensional homonuclear total correlation (TOCSY) spectra with a spin-lock time of 70 ms, nuclear Overhauser enhancement (NOESY) spectra with mixing times of 250, 150, and 50 ms, and double quantum-filtered correlation (DQF-COSY) spectra were acquired at 600 MHz on a Bruker DRX-600 spectrometer. Spectra were acquired at 20°C unless otherwise stated and referenced to dioxane (3.75 ppm). TOCSY and NOESY spectra were also collected at 5°C. The water resonance was suppressed using the WATERGATE pulse sequence (28,29). A series of one-dimensional spectra over the temperature range 5-25°C, at 5°C intervals, was collected. Amide exchange rates were monitored by dissolving freezedried material in 2 H 2 O at pH 5.2 and then recording a series of one-dimensional spectra, followed by 70-ms TOCSY, 50-ms NOESY, and an exclusive correlation (E-COSY) spectra, all at 5°C. In addition, 1 H-13 C HSQC spectra for the assignment of 13 C chemical shifts and a 1 H-15 N HSQC spectrum for the assignment of 15 N chemical shifts (28,30,31) were collected at 20°C on the Bruker DRX-600 and a Bruker Avance 500 spectrometer equipped with a TXI-cryoprobe, respectively. Diffusion measurements were performed at 5 and 20°C using a pulsed field gradient longitudinal eddy-current delay pulse sequence (31,32) as implemented (33). Spectra were processed using TOPSPIN (version 1.3, Bruker Biospin) and analyzed using XEASY (version 1.3.13) (34).
Structural Constraints-3 J HNHA coupling constants were measured from DQF-COSY spectra at 600 MHz and then converted to dihedral restraints as follows: 3 J HNH␣ Ͼ 8 Hz, ϭ Ϫ120 Ϯ 40°; 3 J HNH␣ Ͻ 6 Hz, ϭ Ϫ60 Ϯ 30°. 1 angles for some residues were determined based on analysis of a short mixing time (50 ms) NOESY spectrum. In addition, TALOS (35) was used to predict torsion angle ( and ) restraints based on chemical shifts. Predicted and angles of residues that gave good prediction scores (17 residues: 12-16, 22-25, and 27-34) were constrained (range Ϯ 40°) in structural calculations in XPLOR. Two 1 angles (Asp 5 and Phe 36 ) were constrained in final structure calculations. The number of final dihedral angle constraints is listed in supplemental Table S1, and details have been deposited along with distance constraints in BioMagResBank (36) as entry 15900. Because disulfide bonds for MMP23 TxD had not been mapped, these were not included as structural restraints in preliminary calculations. Subsequently, disulfide bond connectivities were determined based on Cys A H ␣ -Cys B H ␤ and Cys A H ␤ -Cys B H ␤ intercysteine NOEs observed in NOESY spectra (37) as well as Cys A C ␤ -Cys B C ␤ and Cys A S ␥ -Cys B S ␥ intercysteine distance calculations in preliminary structures (supplemental Table S4). Disulfide bonding was determined to be Cys 3 -Cys 37 , Cys 10 -Cys 30 , Cys 19 -Cys 34 , which is the same pattern as in ShK and BgK. These were added as restraints for final structure calculations. No hydrogen bond restraints were included.
Structure Calculations-Intensities of NOE cross-peaks were measured in XEASY and calibrated using the CALIBA macro of the program CYANA (version 1.0.6) (38). NOEs providing no restraint or representing fixed distances were removed. The constraint list resulting from the CALIBA macro of CYANA was used in XPLOR-NIH to calculate a family of 200 structures using the simulated annealing script (39). The 55 lowest energy structures were then subjected to energy minimization in water; during this process, a box of water with a periodic boundary of 18.856 Å was built around the peptide, and the ensemble was energy-minimized based on NOE and dihedral restraints and the geometry of the bonds, angles, and impropers. From this set of structures, final families of 20 lowest energy structures were chosen for analysis using PRO-CHECK-NMR (40) and MOLMOL (41). In all cases, the final structures had no experimental distance violations of Ͼ0.2 Å or dihedral angle violations of Ͼ5°. The structures have been deposited in the Protein Data Bank (42) with code 2K72. Structural figures were prepared using the programs MOLMOL (41) and PyMOL (DeLano Scientific, San Carlos, CA).
Electrophysiology-All experiments were conducted in the whole-cell configuration of the patch clamp technique, as described previously (1,43,44). Data acquisition and analysis was performed using pClamp software.
For flow cytometric studies to determine surface Kv1.3 channels using ShK-F6CA, the MMP23 construct from pEGFP-C1 was subcloned into pdsRED-C1 monomer (Clontech) at 5Ј HindIII and 3Ј BamHI restriction sites. We then co-transfected COS7 cells with human Kv1.3 and pDsRED-C1 (Clontech) or pDsRED-MMP23 for 30 h. Cells were trypsinized and incubated with 10 nM ShK-F6CA (44) in phosphate-buffered saline plus 2% goat serum for 30 min and were then washed three times with phosphatebuffered saline plus 2% goat serum. The intensity of ShK-F6CA staining (a measure of Kv1.3 cell surface expression) was determined by flow cytometric analysis (FACSCalibur flow cytometer and BD CellQuest Pro software, BD Biosciences). The D value, a measure of the difference in mean fluorescence intensities (MFI) of stained and unstained cells, was calculated as follows.

RESULTS
Phylogenetic Relatedness of ShKT Domain-containing Proteins-The MMP23 ShKT domains (henceforth referred to as MMP23 TxD ) from humans to hydra exhibit remarkable sequence conservation with no gaps or insertions in the domain ( Fig. 2A). We compared the MMP23 TxD sequence with that of sea anemone toxins as well as representative members of the ShKT domain family from worms, cnidarians, and plants (Fig. 2, B and C). We included in the sequence alignment a second human protein, microfibril associated protein MFAP2, with an ShKT domain that is not mentioned in the SMART data base (45). Because the ICR domains of CRISPs share structural similarity with ShKT domains (46 -50), we also included snake and human ICR domain sequences.
When compared with sea anemone toxins, MMP23 TxD s appear most similar to BgK with identical or equivalent substitutions at 14 of 36 positions (Fig. 2B). Asp 5 is conserved in all members of the ShKT domain family but is absent in the ICR domains of CRISPs (Fig. 2C). In ShK, the carboxylate of this aspartate (Asp 5 in ShK) forms a salt bridge with the ⑀-ammonium group of Lys 30 , and this salt bridge is necessary for proper folding of the peptide (18,51,52). Lys 32 and Arg 32 at the equivalent position in MMP23 TxD s from sea anemone, hydra, rat, mouse, and puffer fish ( Fig. 2A) could form a salt bridge with the aspartate. Other MMP23 TxD s contain Ser 32 ( Fig. 2A), and in these domains, Asp 5 may make hydrogen bonding interactions with the side chain hydroxyl or the peptide backbone. In the multiple sequence alignment, most proteins, with the exception of the three snake CRISPs, contain a serine or threonine at position 33 (Fig. 2C).
A phylogenetic tree based on the multiple sequence alignment of representative proteins from the ShKT domain family and generated with the PHYLIP program (available at the GeneBee Web site) places the MMP23 TxD s, the sea anemone toxins and the ICR-CRISP domains in distinct but related clades (Fig. 3). MMP23 TxD s also show phylogenetic relatedness to ShKT domains in MFAP2 (45); Caenorhabditis elegans proteins (astacin metalloprotease NAS14, tyrosinase Tyr3, ligandgated channel lgc22, and Mab7 (56)) hydra and jellyfish astacin metalloproteases (HMP2 and PMP1 (57,58)); and plant oxidoreductases (2OG-Fe(II)) and prolyl-4-hydroxylases (Fig. 3). TxDs in MMP23 and ICR domains of CRISPs are each encoded by a single exon (supplemental Fig. S1), raising the possibility that an ancient exon gave rise to these domains. Sea anemones may have co-opted and modified this exon to generate potent K ϩ channelblocking toxins.
Synthesis of MMP23 TxD -We synthesized the 37-residue MMP23 TxD on Ramage TM amide resin with an automated Fmoc/tert-butyl protocol. Following cleavage and deprotection, 36 h was allowed for folding and oxidative formation of three disulfide bonds under conditions similar to those used for ShK. Folding proceeded smoothly to a major product that was homogeneous by analytical RP-HPLC (Fig. 4). Electrospray ionization mass spectral analysis yielded an (M ϩ H) of 4427.33, consistent with the theoretical value following formation of three disulfide bonds (Fig. 4).
Solution Structure of MMP23 TxD -Details of the solution structure of MMP23 TxD are provided in the supplemental material (supplemental Figs. S2-S6). A summary of experimental constraints and structural statistics for MMP23 TxD is given in supplemental Tables S1-S4. The angular order parameters for ⌽ and angles in the final ensemble of 20 structures were both Ͼ0.8 for residues 4 -36. The mean pairwise r.m.s. devia-tion over the backbone heavy atoms of residues 4 -36 in this family of structures was 0.75 Å. Three short ␣-helices encompassing residues 10 -14, 23-29, and 31-34 characterize the closest to average structure of MMP23 TxD . Hydrogen bonds between Ala 14 NH and Cys 10 Table 1. MMP23 TxD shares greater structural similarity with BgK (r.m.s. deviation 2.28 Å) than ShK (2.77 Å). In fact, the structural similarity between MMP23 TxD and BgK is greater than that of BgK and ShK (r.m.s. deviation 2.78 Å). MMP23 TxD , BgK and ShK have a turn involving the fifth cysteine residue (Cys 30 in MMP23 TxD ); this is followed by a short ␣-helix (residues 31-34) in MMP23 TxD and BgK but not in ShK. The main differences between MMP23 TxD and BgK are in the length of the first two helices, with the first helix in MMP23 TxD being shorter and the second longer than in BgK.
MMP23 TxD Exhibits Greater Structural Similarity to BgK Than the CRISP-ICR Domains-We compared the structure of MMP23 TxD with that of the CRISP-ICR domains (13, 46 -50) based on a backbone alignment of matching residues in a multiple alignment (Fig. 7 and Table 1). The ICR domains of snake (stecrisp, natrin, triflin, and pseudechetoxin) and human (CRISP-2/Tpx-1) CRISPs showed considerable structural sim- ilarity (Fig. 7A). Among the snake proteins, the pairwise r.m.s. deviation values over the backbone heavy atoms ranged from 0.52 to 0.98 Å, and between the snake proteins and human CRISP-2/Tpx-1, the r.m.s. deviation was 0.98 -1.32 Å (Table  1). Comparing the ICR domain structure of stecrisp as a representative of the CRISPs with the closest to average structure of MMP23 TxD (Fig. 7B), the main differences were in the first and third loops (corresponding to residues 4 -9 and 24 -28 of MMP23 TxD ), possibly because MMP23 TxD has two fewer residues in the first loop and two more in the third loop (Fig. 2, B and C). Compared with BgK, the backbone of MMP23 TxD aligned relatively well throughout, with the exception of the N-terminal region (Fig. 7C). The pairwise r.m.s. deviation values between the five ICR domains and MMP23 TxD ranged from 2.98 to 3.07 Å. These ICR domains of CRISPs also showed less structural similarity to the sea anemone toxins BgK (r.m.s. deviations 2.97-3.09 Å) and ShK (r.m.s. deviations 3.53-3.79 Å) ( Fig. 7 and Table 1). These structural results, together with the phylogenetic data demonstrate that MMP23 TxD s and sea anemone toxins share a closer relationship with each other than with ICR domains of CRISPs.
Scorpion and sea anemone peptide toxins interact with a binding site in the outer vestibule of K ϩ channels (1,2,52,53). Toxin affinity for the binding site in Kv1.3 can be decreased either by titrating His 404 at the entrance to the pore by lowering the external pH to 6.0 or by occupying the potassium-binding site in the ion selectivity filter by increasing the external K ϩ concentration (1,52,53). Both of these manipulations destabilize the toxin channel interaction via electrostatic repulsion of the pore-occluding lysine (Lys 27 in kaliotoxin and Lys 22 in ShK) (1,52,53). We applied both of these tests to MMP23 TxD . Increasing external K ϩ concentration or pH titration of His 404 in Kv1.3 both significantly decreased MMP23 TxD s affinity for the channel (data not shown). These results suggest that MMP23 TxD interacts with the external vestibule of Kv1.3.
Full-length MMP23 Suppresses K ϩ Channels-Because MMP23 TxD lies between the metalloprotease and the IgCAM domains in MMP23, it may not be optimally positioned in the full-length protein to block K ϩ channels. To test whether MMP23 TxD retained K ϩ channel-blocking activity in fulllength MMP23, N-terminal eGFP-tagged MMP23 (eGFP-MMP23) or eGFP was co-expressed in mammalian cells with MMP23 TxD -sensitive or -resistant channels. Confocal microscopy and patch clamp experiments were performed 30 h later. eGFP-MMP23 co-localized with MMP23 TxD -sensitive Kv1.3 channels (Fig. 9, A and B). In contrast, MMP23 did not co-localize with MMP23 TxD -resistant Kv1.2 channels (Fig. 9, A  and B). eGFP did not co-localize with either channel (Fig. 9, A  and B). These results suggest that MMP23 TxD is required for MMP23 channel co-localization, although other domains in MMP23 may also contribute.
Published cell fractionation and confocal studies demonstrate that full-length MMP23 is an intracellular protein that is expressed primarily in ER/Golgi membranes (21,22). We verified these results in our system by demonstrating that eGFP-MMP23, but not eGFP, co-localized with the ER membrane marker SERCA-2 (Fig. 9C). Because MMP23 is a type II transmembrane ER protein (21,22), MMP23 TxD will lie within the ER lumen, where it has the potential to bind to the outer vestibule of TxD-sensitive channels (Fig. 10A). Once compartmentalized with a MMP23 TxD -sensitive channel, the diffusion of MMP23 away from the channel is likely to be constrained by the ER membrane. Such an interaction might trap MMP23 TxDsensitive channels in the ER and thereby decrease surface channel expression. In contrast, MMP23 TxD -resistant channels should be unaffected.
We performed two types of experiments to test for ER trapping. First, we performed whole-cell patch clamp experiments on cells co-expressing MMP23 and either MMP23 TxD -sensitive or -resistant channels. MMP23 suppressed MMP23 TxD -sensitive Kv1.6 and Kv1.3 currents but had no effect on MMP23 TxDresistant Kv1.2 and Kv1.7 currents (Fig. 10B). These results indicate that fewer functional Kv1.3 channels are expressed in the surface cell membrane. Second, we used fluoresceinated ShK (ShK-F6CA) in flow cytometry experiments to measure cell surface protein expression of Kv1.3 in COS7 cells expressing Kv1.3 and either pDSRED-MMP23 or the control pDSRED-C1 monomer. ShK-F6CA (44) is a highly specific Kv1.3 inhibitor that blocks the channel (IC 50 value 48 pM) 56,000-fold more potently than MMP23 TxD . The intensity of ShK-F6CA staining reflects the number of Kv1.3 tetramers on the cell surface because the peptide binds to the channel tetramer (44). Cell surface Kv1.3 expression was significantly lower in pDSRED-MMP23-expressing cells compared with pDSRED-C1-expressing controls (Fig. 10C). Taken together, these results suggest that MMP23 reduces surface expression of MMP23 TxD -sensitive Kv1.6 and Kv1.3 channels via intracellular trapping.

DISCUSSION
K ϩ channels constitute the most abundant and diverse family of ion channels, and they regulate a myriad of functions in both excitable and non-excitable cells. Peptide toxins have helped to define the molecular mechanisms underlying K ϩ channel function and to determine the relationship between K ϩ currents in native tissues and specific genes. The cnidarian   (184 -189, 192-196, and 201-215), MMP23 TxD (4 -9, 10 -14, 19 -27, 30 -34, and 36), Natrin (184 -189, 192-196, and 201-215), pseudechetoxin (174 -179, 182-186, and 191-205), triflin (184 -189, 192-196, and 201-215), CRISP2/Tpx-1 (206 -211, 214 -218, and 223-237), BgK (3-8,11-15, 20 -28, 30 -34, and 36), ShK (4 -9, 12-25, 28 -32, and 34 toxins ShK and BgK are potent inhibitors of voltage-gated and calcium-activated K ϩ channels (15,16,(51)(52)(53)(54)(55). SMART has identified domains in a vast number of proteins that resemble ShK and BgK (see the SMART Web site). A majority of these proteins are metallopeptidases belonging to the astacin/adamalysin family in C. elegans (Fig. 1). The three-dimensional structures and biological activities of these putative channel-blocking domains have not been determined. Because MMP23 is the only protein in humans identified by the SMART data base to contain such a domain, we have investigated the structure and physiological function of this domain. MMP23 TxD s from diverse species exhibit a high degree of sequence conservation ( Fig. 2A). They are evolutionarily related to the sea anemone toxins and to TxDs in a number of worm, cnidarian, and plant proteins and the ICR domains of snake and human CRISPs (Fig. 2C). Interestingly, we have found a second human protein (MFAP2), not included in the SMART data base, that contains a ShKT domain phylogenetically related to MMP23 TxD . MFAP2 is a matrix protein, which, like MMP23, is located on human chromosome 1p36 (45). The evolutionary relatedness of MMP23 TxD s is supported by the similarity of their three-dimensional structures. MMP23 TxD and the sea anemone toxins share greater structural similarity with each other than with ICR domains; the structure of MMP23 TxD is most similar to that of BgK (Figs. 5-7 and Table 1).
MMP23 TxD s and ICR domains are each encoded by a single exon (supplemental Fig. S1), suggesting that these two related structural motifs possibly arose from an ancient exon. In venomous creatures, this ancient module may have been modified to give rise to potent ion channel blockers, whereas the incorporation of this exon into plant oxidoreductases and prolyl hydroxylases and into worm astacin-like metalloproteases and trypsin-like serine proteases produced enzymes with potential channel-modulatory activity. A lysine residue in sea anemone toxins (BgK 25 , ShK 22 ) protrudes into and plugs the K ϩ channel pore (15,16,(51)(52)(53)(54)(55). MMP23 TxD s in opossum, chicken, zebra finch, and stickleback fish contain lysine at the corresponding position and should block K ϩ channels. We have shown that rat MMP23 TxD containing an arginine at the equivalent position blocks Kv channels (Kv1.6 Ͼ Kv1.3 Ͼ Kv1.1 ϭ Kv3.2 Ͼ Kv1.4, in decreasing potency) in the nanomolar to low micromolar range but has no effect on Kv1.2, Kv1.5, Kv1.7, and KCa3.1 at 100 M concentration. The ICR domain of the snake protein natrin contains the critical pore-plugging lysine, and natrin has been reported to block Kv1.3 (46). The ICR domains of the four mammalian CRISPs lack the pore-occluding lysine and therefore may not block K ϩ channels; CRISP-2/Tpx-1 blocks ryanodine receptors (13).
Full-length MMP23 has been reported previously to be expressed mainly in ER/Golgi membranes (21,22). We confirmed these results by demonstrating that MMP23 co-localizes with the ER membrane marker SERCA-2 in COS7 cells. Because MMP23 is a type II transmembrane protein, MMP23 TxD will lie within the ER lumen and could snare and trap MMP23 TxD -sensitive channels in the ER. Three lines of evidence support this idea. First, MMP23 compartmentalizes with MMP23 TxD -sensitive Kv1.3 channels but not with MMP23 TxD -resistant Kv1.2 channels. Second, MMP23 suppresses MMP23 TxD -sensitive channels while sparing MMP23 TxD -resistant ones. These results indicate that MMP23 TxD within the full-length MMP23 protein has sufficient accessibility to bind to and block K ϩ channels. Third, MMP23 decreases cell surface expression of MMP23 TxDsensitive Kv1.3 channels.
It has been suggested previously that the preponderance of Kv1 channel heterotetramers in many tissues is the result of Kv1 homotetramers being retained in the ER by an unknown protein containing a TxD that binds to a trafficking determinant in the outer vestibule of the channel (59 -61). Vacher et al. (61) provided support for this idea by showing that heterologously overexpressed, ER-luminal dendrotoxin competed with the TxD-containing ER protein for toxin-sensitive Kv1.1 homotetramers and allowed these channels to escape the ER and migrate to the plasma membrane. MMP23 fits the criteria for this ER protein because it contains a TxD that can trap sensitive Kv1 channels intracellularly. MMP23 is expressed in many tissues (lung, heart, uterus, placenta, ovary, testis seminiferous tubules, prostate, intestine, colon, pancreatic islets, cingulate cortex, adrenal cortex, osteoblasts, chondroblasts, cartilage, synovium, natural killer cells, dendritic cells, and tendons) (20, 23, 63-66) (see the BioGPS Web site) that overlap with the tissue expression of MMP23-sensitive K ϩ channels. Therefore, there is a reasonable likelihood that MMP23 will modulate K ϩ channels in vivo.
Our results for MMP23 help shed light on the vast family of proteins containing ShKT domains. Most of these proteins are found in C. elegans, many with multiple repeats of the domain in a single protein (Fig. 1). The ShKT domain in Mab7 is required for Mab7-mediated regulation of morphogenesis of sensory rays of the male C. elegans (56). The rays of Mab7-deficient males are virtually all malformed (56). Full-length Mab7, but not deletion constructs lacking the ShKT domain, rescues the abnormal phenotype (56). Because the C. elegans genome contains almost as many K ϩ channel genes (ϳ70) as humans, modulation of worm ion channels by the Mab7-ShKT domain could contribute to its role in morphogenesis. However, Mab7 lacks the critical pore-blocking positively charged residue, and a Mab7 hybrid construct containing the ShK toxin sequence in place of the TxD was not able to rescue the mutant ray phenotype (56). These findings suggest that the ShKT domain in Mab7 may not be involved in K ϩ channel modulation, although modu- lation of other types of ion channels cannot be excluded. Astacin-like metalloproteases in Hydra vulgaris (HMP2) and jellyfish (PMP1) possess ShKT domains that contain the critical pore-occluding lysine required for K ϩ channel block, and both of these proteins play critical roles in foot morphogenesis (57,58).
In summary, we have defined a novel channel-regulatory role for a metalloprotease and characterized the first functional K ϩ channel-blocking toxin domain in a mammalian protein. Our results provide insight into the structure and function of the ShKT-containing protein superfamily. It is tempting to speculate that the TxDs in each of these proteins FIGURE 9. MMP23 colocalizes with Kv1.3 in the ER. MMP23 co-localizes with MMP23 TxD -sensitive Kv1 channels and with ER marker SERCA-2. A, eGFP-MMP23 (green) co-localizes with Kv1.3 channels (red) but not with Kv1.2 (red); eGFP did not co-localize with either channel. B, quantification of co-localization between eGFP or eGFP-MMP23 with Kv1.3 or Kv1.2 (white bar, n ϭ 25 cells; black bar, n ϭ 30 cells) (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; ns, not significant). C, eGFP-MMP23 (green) co-localizes with ER membrane marker SERCA-2 (red). Areas of co-localization are shown in yellow in the overlay image. eGFP does not co-localize with SERCA-2. N, nucleus. regulate different types of ion channels and that the evolutionary pressure to maintain channel-modulatory activity underlies the conservation of this domain throughout the plant and animal kingdoms.