Intracellular Trafficking of the KV1.3 Potassium Channel Is Regulated by the Prodomain of a Matrix Metalloprotease*

Background: Noncanonical functions of matrix metalloproteases are poorly characterized. Results: The prodomain of MMP23 co-localizes with and traps KV1.3 channels intracellularly, thereby suppressing KV1.3 currents. Conclusion: A novel metalloprotease-independent channel-modulating function of the MMP23 prodomain has been identified. Significance: The topological similarity of the prodomain of MMP23 and KCNE proteins suggests a shared mechanism of channel modulation. Matrix metalloproteases (MMPs) are endopeptidases that regulate diverse biological processes. Synthesized as zymogens, MMPs become active after removal of their prodomains. Much is known about the metalloprotease activity of these enzymes, but noncanonical functions are poorly defined, and functions of the prodomains have been largely ignored. Here we report a novel metalloprotease-independent, channel-modulating function for the prodomain of MMP23 (MMP23-PD). Whole-cell patch clamping and confocal microscopy, coupled with deletion analysis, demonstrate that MMP23-PD suppresses the voltage-gated potassium channel KV1.3, but not the closely related KV1.2 channel, by trapping the channel intracellularly. Studies with KV1.2-1.3 chimeras suggest that MMP23-PD requires the presence of the KV1.3 region from the S5 trans-membrane segment to the C terminus to modulate KV1.3 channel function. NMR studies of MMP23-PD reveal a single, kinked trans-membrane α-helix, joined by a short linker to a juxtamembrane α-helix, which is associated with the surface of the membrane and protected from exchange with the solvent. The topological similarity of MMP23-PD to KCNE1, KCNE2, and KCNE4 proteins that trap KV1.3, KV1.4, KV3.3, and KV3.4 channels early in the secretory pathway suggests a shared mechanism of channel regulation. MMP23 and KV1.3 expression is enhanced and overlapping in colorectal cancers where the interaction of the two proteins could affect cell function.

Matrix metalloproteases (MMPs) 4 are a family of zinc-dependent endopeptidases that degrade extracellular matrix proteins, cleave cell surface receptors, release apoptotic ligands, and activate chemokines and cytokines (1). Their metalloprotease activities are involved in tissue remodeling, cell proliferation, cell migration, differentiation, angiogenesis, apoptosis, and the immune response (2,3). MMPs are synthesized as zymogens, from which the active enzymes are released after proteolytic cleavage and removal of their prodomain. Current knowledge about MMPs is derived almost exclusively from studies of the active enzymes and their metalloprotease activity. The only described functions of the prodomains also involve metalloprotease-dependent regulation (4,5). Noncanonical functions of MMPs are poorly defined.
MMP23, a 391-residue type-II trans-membrane protein, is synthesized as a zymogen, which is anchored to ER/nuclear membranes via a trans-membrane segment located within an N-terminal 80-residue prodomain (see Fig. 1A) (6 -10). The MMP23 prodomain lacks the cysteine switch (PRCGXPD motif) present in other MMPs, which maintains these enzymes in an inactive state (9). However, MMP23 contains an RRRR motif that can be cleaved by the serine protease furin (7), separating the membrane-tethered prodomain from the secreted active enzyme. The secreted protein contains three domains (see Fig. 1A) (2,3,10). At the N terminus is a Zn 2ϩ -dependent catalytic domain (Cat domain) like those in other metalloproteases. This is followed by a cysteine-rich region (TxD) with sequence and structural similarity to BgK and ShK, 35-residue peptide inhibitors of K V 1 channels from the sea anemones Bunodosoma granulifera and Stichodactyla helianthus, respectively. At the C-terminal end is an immunoglobulin-like cell adhesion molecule (IgCAM) domain with sequence similarity to the ROBO1, ROBO3, and ROBO4 proteins (supplemental Fig. S1).
We reported previously a novel channel-modulating, metalloprotease-independent, function for MMP23 (10). Full-length MMP23 zymogen suppresses voltage-gated K V 1.3 and K V 1.6 potassium channels by intracellular retention, but has no effect on related K V 1 channels (e.g. K V 1.2, K V 1.7) (10). We hypothesized that this trapping function resided in the K V 1.3-and K V 1.6-blocking TxD in MMP23.
In the present study, we used deletion analysis to test this hypothesis. To our surprise, MMP23 retained its trapping function even after deletion of the TxD. Here we report that the prodomain of MMP23 (MMP23-PD) co-localizes with and suppresses K V 1.3 channels, but not closely related K V 1.2 channels, by intracellular trapping. This trapping is likely to take place in the ER because MMP23 co-localizes with the ER protein Stim-1. Purified MMP23-PD in lipid micelles is shown by NMR to consist of a single bent trans-membrane ␣-helix joined by a short linker to a juxtamembrane ␣-helix, which is associated with the surface of the membrane. Experiments with K V 1 chimeras suggest that MMP23-PD requires the presence of the region of K V 1.3 from the S5 segment to the C terminus in order for it to co-localize and trap the channel intracellularly. MMP23 and K V 1.3 proteins are both present in normal human colon epithelium, and both are up-regulated in colonic cancers where channel modulation may occur.

EXPERIMENTAL PROCEDURES
Cell Lines, Transfection, and Cell Culture-Enhanced GFPtagged human K V 1.2 and human K V 1.3 channels and their chimeric channels were transiently co-transfected with either DsRed monomer or one of the DsRed-tagged MMP23 constructs in COS-7 cells for 24 -48 h. COS-7 cells were maintained in Dulbecco's modified Eagle's medium containing 10% heat-inactivated FCS (Summit Biotechnology), 4 mM glutamine, 1 mM sodium pyruvate, and 500 g/ml G418 (Calbiochem) to 60 -80% confluence in culture and were transfected with 0.4 -1 g of DNA using Lipofectamine 2000 (Invitrogen) in Opti-MEM I medium as per the manufacturer's protocol. After 48 h, transfection efficiency was assessed by fluorescence microscopy (Olympus).
The human K V 1.2 channel was GFP-tagged on the C terminus in the pCMV6-AC-GFP vector (OriGene). The human K V 1.3 cDNA coding region was amplified from a pCDNA3.1 vector using primers containing SgfI (AsiSI) and MluI restriction sites. This K V 1.3 PCR fragment was then directionally cloned into the pCMV6-AC-GFP plasmid to create a GFPtagged K V 1.3 channel similar to the K V 1.2 channel.
We exploited the unique BstXI restriction site located after Phe-305 (K V 1.2 numbering) to construct K V 1.3-1.2-GFP and K V 1.2-1.3-GFP-tagged chimeric channels. This domain swap created a K V 1.2-1.3 chimera containing the K V 1.2 region from its N terminus up to its trans-membrane segment S4 (K V 1.2 Phe-305) and K V 1.3 trans-membrane segment S4-S5 linker up to the K V 1.3 C terminus (see Fig. 3A). The K V 1.3-1.2 chimera was constructed similarly utilizing the complementary domain exchange.
The rat MMP23 cDNA was cloned into the pDsRed vector, and subsequent deletion constructs of the MMP23 protein were created by PCR-mediated site-directed mutagenesis using this vector as the template. Point mutations were inserted to create AUU stop codons, and changes were confirmed by sequencing. The YFP-Stim-1 plasmid was a kind gift from Dr. Michael D. Cahalan (University of California Irvine).
Electrophysiology-The whole-cell patch clamp technique (11) was used to characterize the current amplitude of COS-7 cells transiently expressing K V channels as described previously (12,13), in the presence of different MMP23 constructs. Currents were recorded in normal Ringer solution with a Ca 2ϩ -free pipette solution containing (in mM): 145 KF, 10 HEPES, 10 EGTA, and 2 MgCl 2 , pH 7.2, 300 mosM. Transfected cells were visualized by epifluorescence microscopy of co-expressed green and red fluorescent proteins. Data acquisition and analysis were performed using pClamp software.
Immunohistochemistry-Human colon sections (6 m; formalin-fixed, paraffin-embedded) were obtained from the University of California Irvine (UCI) Experimental Tissue Resource. The slides were de-identified so that no private health information was discoverable. Sections were dewaxed with xylene (3 times, 5 min each) and rehydrated through graded alcohols (100 -30%). Before staining with the primary antibody, citrate buffer-based (Vector H3300) antigen retrieval was performed for 20 min to unmask antigens masked by paraffin embedding. Endogenous peroxidase activity was blocked by 1.5% hydrogen peroxide solution in PBS for 20 min. After blocking with 5% corresponding serum, 5% BSA, and 0.1% sodium azide in PBS, sections were incubated with primary antibody overnight at 4°C. We used a rabbit anti-K V 1.3 polyclonal antibody (a gift from Dr. Hans Gunther-Knaus, University of Innsbruck, 1:50 dilution) and rabbit anti-MMP23 polyclonal antibody (Millipore, 1:100 dilution) described in the Human Protein Atlas. As the isotype control, we used rabbit polyclonal IgG in place of primary antibodies. Bound primary Abs and the isotype control were detected with a goat antirabbit biotinylated secondary antibody followed by an HRPconjugated avidin complex using VECTASTAIN Elite ABC kit (Vector Laboratories). Peroxidase activity was visualized with 3,3Ј-diaminobenzidine using the diaminobenzidine substrate kit for peroxidase (Vector Laboratories). Sections were counterstained with hematoxylin (Fisher Scientific) and bluing solu-tion, dehydrated through graded alcohol and xylene washes, and mounted with Permount (Fisher Scientific). Two independent investigators scored the intensity of K V 1.3 and MMP23 expression on a scale of 1-3 in normal epithelium, colon adenomas, and colorectal cancers.
Cloning, Expression, and Purification of MMP23-PD-The cDNA for the prodomain of human MMP23 (residues 1-78) was amplified by PCR from the vector pCMV6-XL5 (OriGene Technologies Inc., Rockville MD) using the forward primer 5Ј-TAA ACC ATG GCA CTG GAA GTT CTG TTT CAG GGC CCG CCC TTC CCC ACG GTG GCC ACC ACC CCA CCG-3Ј and the reverse primer 5Ј-TTT TAC AAG CTT CTA TTA TCA GCG GCG TCT GCG GGG GGC CAG TGG-3Ј. The amplified product was cloned into the expression vector pET32a(ϩ) (Novagen Inc., La Jolla, CA) via the NcoI and Hin-dIII restriction sites. The final constructs contained cDNA for the N-terminal thioredoxin fusion protein followed by a hexahistidine tag, a HRV 3C protease cleavage site, and the MMP23-PD protein.
The expression construct containing the MMP23-PD gene was transformed into Escherichia coli BL21(DE3) cells (Novagen/Merck). Transformants were grown in M9 minimal medium containing 100 g/l ampicillin with [ 15 N]H 4 Cl and D-[ 13 C 6 ]glucose as the sole sources of nitrogen and carbon, respectively. Cells were grown at 37°C to an A 600 of 0.5-1.0 and then overnight at 28°C following the addition of 1 mM isopropyl-1-thio-␤-D-galactopyranoside to induce expression of the thioredoxin-MMP23 prodomain fusion protein (Trx-MMP23-PD). Cells were harvested by centrifugation, and the pellet was resuspended in lysis buffer composed of BugBuster protein extraction reagent (Merck) supplemented with 10 mM CHAPS (Sigma-Aldrich), 1 mM PMSF, EDTA-free Complete protease inhibitor mixture (Roche Applied Science), 1.0 mg/ml lysozyme, and 5 g/ml DNase I. The cell suspension was probesonicated (XL-2000 ultrasonic liquid processor; Misonix, Farmingdale, NY; power level of 8.5, 30-s pulses separated by 30 s) for 15 min on ice. The lysate was centrifuged at 4°C and 35,000 ϫ g for 20 min to remove insoluble cell debris. The supernatant containing the Trx-MMP23-PD fusion protein was loaded at room temperature onto a 5-ml HiTrap (Amersham Biosciences) chelating column charged with nickel. The column was extensively washed with buffer A (20 mM Tris buffer, pH 8.0, containing 150 mM NaCl and 10 mM CHAPS) containing 5 mM imidazole followed by buffer A containing 50 mM imidazole. The protein was eluted with buffer A containing 350 mM imidazole. Cleavage of the hexahistidine-tagged fusion protein was carried out overnight at 4°C at a protein concentration of 1.0 mg/ml using 1.0 unit of HRV 3C protease (Novagen) per mg of protein. The Trx tag was removed by reversephase HPLC on a Luna C8 column equilibrated in solvent A (99.9% water and 0.01% TFA). Proteins were eluted with an acetonitrile gradient in 0.01% TFA (0 -40% over 10 min followed by 40 -80% over 30 min). Eluted fractions containing MMP23-PD were lyophilized, dissolved in 20 mM sodium citrate, pH 5.0, containing 20 mM TCEP, and 10 mM CHAPS, and then purified on the same column using an acetonitrile gradient in 0.01% TFA of 0 -50% over 5 min followed by 50 -70% over 40 min. Fractions containing MMP23-PD were lyophilized and stored at Ϫ20°C. The concentration of the protein was determined from the A 280 using an extinction coefficient of 0.962 mg/ml protein per A 280 unit (14)in a 1-cm path length cell.
Circular Dichroism Spectroscopy-CD experiments were performed on a model 410SF CD spectropolarimeter (Aviv Biomedical). Far UV CD spectra were recorded in a 0.01-cm path length quartz cuvette over the wavelength range 190 -260 nm, with a resolution of 0.5 nm, a bandwidth of 1 nm, and a response time of 1 s. Spectra were averaged over eight separate scans and corrected for background. The sample contained 14 M MMP23-PD in 20 mM sodium citrate buffer, pH 5.0, containing 100 mM deuterated n-dodecylphosphocholine (d 38 -DPC; Sigma-Aldrich).
NMR Spectroscopy-Lyophilized uniformly 15 N-and 13 C-15 N-labeled MMP23-PD was dissolved in a buffer containing 20 mM sodium citrate (pH 5.0), 100 mM deuterated DPC, 20 mM TCEP, 0.02% (w/v) sodium azide (NaN 3 ), 90% H 2 O, and 10% 2 H 2 O, at a final protein concentration of 0.7 mM. Twodimensional 1 H-15 N heteronuclear single quantum coherence (HSQC) spectra of MMP23-PD in DPC micelles were recorded on a Varian Inova 600-MHz spectrometer equipped with a cryogenically cooled triple-resonance probe at various temperatures to determine the optimal conditions for subsequent experiments. 256 and 1024 complex points were acquired in the t 1 ( 15 N dimension) and t 2 ( 1 H dimension) time domains, respectively. The data were zero-filled to 512 ϫ 2048 and apodized using a sine-bell squared window function prior to Fourier transformation using NMRPipe (version 3.0).
NMR data were also acquired on Bruker Avance spectrometers operating at 1 H frequencies of 600 and 800 MHz and equipped with cryogenically cooled triple-resonance probes. Backbone resonance assignments were obtained at 45°C using the following three-dimensional experiments, employing conventional or nonuniform sampling in the 15 N and 13 C dimensions: HN(CA)CO, HNCO, HN(CO)CA, HNCA, CBCA-(CO)NH, and HNCACB (15). Nonuniform sampling datasets were processed using the multidimensional decomposition algorithm (16). Uniformly sampled datasets were processed using TOPSPIN (version 3.0) from BrukerBioSpin or NMRPipe (17). Data were analyzed using NMRView (18), and initial automated assignments were obtained with the program PINE (19). 15 N relaxation experiments were performed at 30°C on a Varian 600-MHz spectrometer. R 1 values were obtained from a series of 1 H-15 N correlation spectra with 0-, 10-, 50-, 100-, 300-, 500-, 800-, 1000-, 1500-, 2000-, and 2500-ms relaxation delays. R 2 values were acquired with 10-, 30-, 50-, 70-, 90-, 110-, 130-, 150-, 170-, 190-, and 210-ms delays. The relaxation rates were calculated in NMRView by least square fitting of peak intensities versus relaxation delay times to an equation for single-order exponential decay. The reported errors were standard deviations derived from the fit of the data. Steady-state 1 H-15 N NOE values were determined from the ratio of peak intensities for spectra collected with and without 3-s proton presaturation. The experimental uncertainty was estimated from the signalto-noise ratio for each spectrum.
The rotational correlation time ( c ) was determined using the following equation where, in the limit of slow molecular motion ( c Ͼ Ͼ 0.5 ns), the correlation time of a protein is related to the MMP23 Regulation of Potassium Channels MARCH 1, 2013 • VOLUME 288 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 6453 ratio of the longitudinal (t 1 ) and transverse (t 2 ) relaxation times and the nuclear frequency ( N ) where c is the effective rotational correlation coefficient and N is the 15 N resonance frequency. Equation 1 was derived from Equation 8 in Kay et al. (20) by considering only J(0) and J() spectral densities and neglecting higher frequency terms. The effective hydrodynamic radius a was calculated from Stokes law where is the viscosity of the solvent, is the Boltzmann constant, and T is the temperature.
The Clean SEA-HSQC experiment (21)  Molecular Modeling-All molecular dynamics calculations were performed using the GROMACS (version 4.5.5) package of programs (23) employing the Martini force field (24,25). The time step used was 25 fs. Peptide, lipid, and solvent (including ions) were coupled separately to a thermal bath at 320 K employing velocity rescaling (26), applied with a coupling time of 1.0 ps. The pressure was maintained at 1 bar by applying a semi-isotropic coupling to a Berendsen barostat (27) with a coupling constant of 5.0 ps and compressibility of 4.5 ϫ 10 Ϫ5 bar. A shifted nonbonded potential was used, with a distance cutoff of 12 Å and a shift distance of 9 Å, applying a neighbor list update frequency of 10 steps (250 fs). Coulombic interactions were screened using a relative dielectric constant of 15.
An initial model of the prodomain (PD) peptide was generated employing only the secondary structure restraints determined from the NMR experiments; residues Gly-15 to Ala-27, Leu-28 to Leu-31, Pro-32 to Leu-40, and Ala-47 to Leu-58 were restrained to adopt an ␣-helical conformation. Generation of this atomistic model used the MODELLER program (version 9.10 (28)). From this atomistic model, a coarse-grain (CG) model was created, and the trans-membrane helix aligned along the Z-coordinate axis. The PD was inserted into a CG representation of a preformed dioleoylphosphatidylcholine lipid bilayer, with water (and antifreeze particles, 10%) and sufficient Na ϩ and Cl Ϫ ions to provide a solvent ionic concentration of 0.15 M. The insertion employed a simple transmutation in which ghost atoms (those in which the Lennard-Jones interaction parameters were set to zero) were transformed into the desired CG representation; this adiabatic transformation was conducted over 1.25 ns with the PD restrained at its original position. Following PD insertion, the simulation was continued with no restraints for a further 500 ns. An atomistic model of the final geometry from the molecular dynamics simulation was created using the CG representation of the PD as a template in comparative modeling using the MODELLER program; secondary structure restraints (described above) were applied, along with distance restraints between C␣ carbons restrained to the separation between the equivalent backbone CG particles.

RESULTS
The Prodomain of MMP23 (MMP23-PD) Selectively Suppresses K V 1.3 Channels without Affecting Closely Related K V 1.2 Channels-Full-length rat MMP23 (MMP23-FL), tagged at the N terminus with DsRed-monomer (Fig. 1B), was co-expressed in COS-7 cells with C-terminally GFP-tagged K V 1.3 or K V 1.2 channels. Whole-cell patch clamp experiments performed 48 h later showed that MMP23-FL suppressed K V 1.3 currents but not K V 1.2 ( Fig. 1, C and D). Cells co-transfected with K V 1.3 and vector expressed 2967 Ϯ 434 channels/cell (mean Ϯ S.E.), whereas cells co-transfected with K V 1.3 MMP23-FL expressed 1256 Ϯ 216 channels/cell (p Ͻ 5 ϫ 10 Ϫ4 ). K V 1.3 channels in cells co-transfected with rat MMP23-FL were indistinguishable in terms of voltage dependence of activation or inactivation from cells expressing K V 1.3 alone (supplemental Fig. S2). This result suggests that MMP23-FL decreases the number of functional K V 1.3 channels on the cell surface but does not alter the properties of the channel. Because MMP23-FL has not been detected in the plasma membrane (6,7,10), it is possible that the K V 1.3 channels on the surface of MMP23-FL-expressing cells are not associated, directly or indirectly, with MMP23-FL and may represent channels that have escaped from an intracellular pool.
To define the region of rat MMP23-FL responsible for K V 1.3 suppression, we generated three C-terminal deletion constructs of MMP23, tagged at the N terminus with DsRed monomer, by progressively removing the immunoglobulin-like cell adhesion molecule domain, TxD, and Cat domain (Fig. 1B). All three MMP23 deletion constructs suppressed K V 1.3 but not K V 1.2 currents (Fig. 1, C and D), highlighting the specificity of this suppression. MMP23-PD, the shortest construct, caused equivalent suppression of K V 1.3 to that of MMP23-FL (Fig. 1, C and D; 1181 Ϯ 269 K V 1.3 channels/cell, p Ͻ 5 ϫ 10 Ϫ5 ). MMP23-PD did not change the properties of K V 1.3 channels (supplemental Fig. S2). These data suggest that MMP23-FL and MMP23-PD have the same ability to serve as functional modulators of K V 1.3 channels, they both deplete the number of functional channels on the cell surface, and MMP23-PD is likely the region of MMP23 responsible for K V 1.3 suppression.
MMP23-PD corresponds exactly to the naturally occurring prodomain that is liberated by furin cleavage in human and mouse MMP23 (supplemental Fig. S3) (6, 7). In an earlier study (7), protein sequencing of the secreted processed mouse MMP23 enzyme showed an N-terminal sequence of YTLT-PARL (residues 79 -86), which indicates that the prodomain remaining after furin cleavage ends with the C-terminal sequence RRRR (residues 75-78) and is the same Arg-78 residue at the C-terminal end of rat MMP23-PD (supplemental Fig. S3).
Although rat MMP23-FL contains the furin cleavage site (supplemental Fig. S3), it is not processed in mammalian cells for reasons that remain unclear (6). Because we used COS-7 cells for our experiments, rat MMP23-FL would not have been processed, suggesting that the full-length MMP23 protein has the potential to act as a K V 1.3 channel suppressor without being cleaved.
MMP23-PD Co-localizes Intracellularly with K V 1.3, but Not K V 1.2 Channels-We used confocal microscopy to determine whether K V 1.3 channel suppression was associated with intracellular trapping of the channel. First, we studied the localization of MMP23-FL and MMP23-PD. C-terminally tagged mouse, rat, and human MMP23 all localize to perinuclear regions and the ER in mammalian cells (6,7). N-terminally GFP-tagged rat MMP23-FL also co-localizes with the ER marker sarcoplasmic reticulum Ca 2ϩ -ATPase (SERCA) (10). Corroborating the published data, N-terminally DsRed-tagged MMP23-FL and MMP23-PD both co-localized with the ER marker YFP-Stim-1 in COS-7 cells (Fig. 2, A and B). These results together with the published data demonstrate that MMP23 and MMP23-PD, either N-terminally or C-terminally tagged, localize to ER/Golgi and perinuclear membranes.
We then examined whether DsRed-tagged MMP23-FL and MMP23-PD co-localized with GFP-tagged K V 1.3 expressed in COS-7 cells. Both MMP23 proteins exhibited strong co-localization with K V 1.3 but not with the vector control or with K V 1.2 ( Fig. 2, C-E). The MMP23-TxD and MMP23-Cat domain proteins also co-localized with K V 1.3 (supplemental Fig. S4). Quantification of the degree of co-localization (co-localization percentage) indicated that MMP23-FL and MMP23-PD co-localized significantly and to equivalent extents with K V 1.3 but not with K V 1.2 (Fig. 2C). This co-localization is likely to be in an intracellular compartment because fewer K V 1.3 channels/cell are detected on the cell surface by patch clamp (Fig. 1D), and MMP23-FL and MMP-23-PD are both present in ER/Golgi and perinuclear membranes (Fig. 2, A and B). In support of this, less K V 1.3 protein was detected on the cell surface in cells co-expressing MMP23-FL and MMP23-PD than cells co-expressing the DsRed vector control (supplemental Fig. S5).For each cell, we plotted cell surface K V 1.3 expression versus intracellular K V 1.3 co-localization with DsRed vector, DsRed-MMP23-FL, or DsRed-MMP23-PD. Intracellular co-localization was inversely related to K V 1.3 surface expression; as intracellular co-localization increased, cell surface K V 1.3 decreased. These results suggest that MMP23-FL and MMP23-PD deplete K V 1.3 channels on the cell surface by a mechanism consistent with trapping K V 1.3 in intracellular compartments, most likely the ER. This co-localization and trapping may be via a direct or an indirect interaction.  MARCH 1, 2013 • VOLUME 288 • NUMBER 9

JOURNAL OF BIOLOGICAL CHEMISTRY 6455
The K V 1.3 Region from the S5 Segment to the C Terminus Is Required for MMP23-PD-mediated Suppression and Colocalization-We constructed C-terminal GFP-tagged chimeras of MMP23-modulated K V 1.3 and MMP23-resistant K V 1.2 to identify the K V 1.3 region required for modulation by MMP23-PD.The K V 1.2-1.3 chimera contains the N terminus through to the end of the S4 segment of K V 1.2 and S5 through to the C-terminal end of K V 1.3. The K V 1.3-1.2 chimera contains the N terminus through to the end of the S4 segment of K V 1.3 and S5 through to the C-terminal end of K V 1.2 (Fig. 3A). Notably, K V 1.2 and K V 1.3 share identical sequences across the S4-S5 chimera ligation region. Both chimeras produced robust currents when expressed in COS-7 cells (Fig. 3, B and C) and behaved as expected. Cumulative inactivation, a unique property of the K V 1.3 pore domain (29), was exhibited by the K V 1.2-1.3 chimera (containing the K V 1.3 pore domain) but not by the K V 1.3-1.2 chimera (containing the K V 1.2 pore domain) (Fig. 3,  B and C). In contrast, the K V 1.3-K V 1.2 chimera but not the K V 1.2-1.3 chimera exhibited use-dependent activation, a unique property of the pore domain of K V 1.2 (Fig. 3, B and C). Next, we expressed MMP23-PD in COS-7 cells together with the K V 1.2-1.3 or K V 1.3-1.2 chimeras. MMP23-PD decreased the number of functional K V 1.2-1.3 channels on the cell surface from 2573 Ϯ 714 channels/cell in cells co-expressing the DsRed vector to 764 Ϯ 313 channels/cell in cells co-expressing DsRed-MMP23-PD ( Fig. 3D; p Ͻ 0.05). In contrast, MMP23-PD had no effect on the K V 1.3-1.2 chimera (Fig. 3E). Cells co-transfected with the K V 1.3-1.2 chimera and the DsRed vector expressed 1028 Ϯ 203 channels/cell, and cells co-transfected with the K V 1.3-1.2 chimera and DsRed-MMP23-PD expressed 1104 Ϯ 299 channels/cell (Fig. 3E). In confocal experiments, MMP23-FL and MMP23-PD both co-localized significantly with the K V 1.2-1.3 chimera, but not with the K V 1.3-1.2 chimera (Fig. 4, A-C). Together, these results demonstrate that the region of K V 1.3 extending from the S5 segment through to the C-terminal end is required for MMP23-PD-mediated depletion of functional channels on the cell surface and for intracellular co-localization with MMP23-PD.

K V 1.3 and MMP23 Share Overlapping Expression in the Human Colon-The
Human Protein Atlas reports strong expression of MMP23 in human colonic epithelium, and an earlier study described K V 1.3 expression in human colonic epithelium, especially in colorectal cancers (30,31). We therefore performed immunohistochemistry studies to determine whether the two proteins exhibited a similar expression pattern in colonic epithelium. In normal colon, we observed equivalent intensity of staining of MMP23 and K V 1.3 in colonic epithelium (Fig. 5, A and E). In tissues from patients with colon cancer, we scored for staining intensity in normallooking epithelium, in adenomas, and in malignant epithelium. Expression of both K V 1.3 and MMP23 was lower in the normallooking epithelium and in adenomas when compared with malignant epithelium (Fig. 5, B and E). Higher magnification images were consistent with K V 1.3 and MMP23 being mainly intracellular, both in primary colorectal tumors and in metastatic tumors in the lymph node (Fig. 5C) and liver (not shown). The overlapping intracellular pattern of MMP23 and K V 1.3, particularly in colorectal cancers, suggests that the two proteins may interact, directly or indirectly, in colonic epithelium.
Expression and Purification of MMP23-PD-Purified recombinant MMP23-PD eluted as two separate peaks on a reversephase HPLC column, corresponding to monomeric and dimeric forms of the protein (Fig. 6, B and C). SDS-PAGE analysis under reducing and nonreducing conditions confirmed that the dimer resulted from intermolecular disulfide formation via the single Cys residue in MMP23-PD. Monomeric 15 Nlabeled MMP23-PD ran as a single band on SDS-PAGE, and electrospray mass spectrometry analysis gave a mass of 8161.06 Ϯ 0.75 Da, consistent with the theoretical value of 8161.5 Da. MMP23-PD contains a stretch of predominately hydrophobic amino acid residues (Trp-20 -Leu-40) identified as a potential trans-membrane domain (TMD) (supplemental Fig. S6) that overlaps a region predicted to adopt a helical conformation (supplemental Fig. S7A). The N-terminal region of this helix ( 17 ERRW 20 ) contains residues that are often found at the membrane interface of TMDs (32). The prediction of secondary structure is variable in the region just C-terminal to the TMD, although there is a propensity for helical content in this region. In support of these predictions, the CD spectrum of MMP23-PD in DPC micelles exhibited two prominent minima at 208 and 222 nm, indicative of significant ␣-helical content (supplemental Fig. S7B).
NMR Assignments of MMP23-PD in DPC Micelles-1 H-15 N HSQC spectra of MMP23-PD were recorded at 30, 40, and 45°C (supplemental Fig. S8). Several backbone amide resonance peaks that appeared weak at 30°C were significantly more intense at higher temperatures. The 600-MHz two-dimensional 1 H-15 N HSQC spectrum acquired at 45°C for 0.7 mM 13 C-15 N-labeled MMP23-PD is shown in Fig. 6E. The spectrum revealed relatively narrow peak dispersion in the 1 H dimension, typical for a single ␣-helical trans-membrane protein. The spectrum contained the correct number of resonances for a protein of this size, and the majority were well resolved. Several weaker peaks were observed in the 1 H-15 N HSQC spectrum, suggesting a slow equilibrium of two conformations, presumably due to Pro cis/trans-isomerization. However, only the cross-peaks from the major conformer were analyzed here.
Backbone resonance assignments for MMP23-PD in DPC micelles at 45°C were determined using three-dimensional heteronuclear NMR experiments (Fig. 7). The majority (97%) of the 68 nonproline backbone 1 H and 15 N assignments for MMP23-PD were completed (excluding the two N-terminal residues arising from the vector). 13 C ␣ and 13 C ␤ backbone assignments were both 97% complete. Backbone amide resonances corresponding to Val-26 could not be assigned due to resonance overlap. The 13 C ␤ chemical shift for Cys-29 at 27.5 ppm clearly indicated that the thiol group for this residue existed in the reduced state and did not engage in any intermolecular disulfide bonding. Secondary 1 H ␣ , 1 H N , 13 C ␣ , and 13 C ␤ chemical shifts (Fig. 7, B-F) and predicted N-and C-helical capping regions (Fig. 7, G and H) indicated that MMP23-PD contains a relatively long helix (␣1) (residues Glu-17-Leu-40) incorporating the TMD, which is joined by a short linker to a juxtamembrane helical region (␣2) (residues Ala-47-Leu-58). These results are similar to predictions obtained using TALOSϩ (33) (data not shown) and Motif Identification from Chemical Shifts (MICS) (34) (Fig. 7I). Identification of helical capping motifs using the MICS program was used to aid in assigning the N-and C-terminal boundaries for each helix (Fig.  7, G and H). Three potential N-terminal helical capping motifs were identified at residues Glu-17, Ala-47, and Asp-54, whereas residues Leu-40 and Leu-58 were predicted as C-terminal helical capping motif. N-terminal helical capping motif residues Glu-17 and Ala-47 are located at the N termini of helices ␣1 and ␣2, respectively, whereas Asp-54 lies in the middle of ␣2. Secondary chemical shift data also indicate a possible disruption of helix ␣2 at residue 53, suggesting the possibility that helix ␣2 may in fact consist of two short ␣-helices. The data also suggest a disruption in the TMD ␣1 helix near residues Ala-27-Leu-28, roughly one helical turn N-terminal to Pro-32. Consistent with this observation, the resonances for Cys-29 and Ala-33 were shifted well up-field of other resonances, possibly due to distortion of the ␣1 helix within this region.
Employing only the secondary structure information obtained above, we constructed a model of MMP23-PD in a lipid membrane (Fig. 7J). The model illustrates how the TMD ␣1 helix can span the membrane with a slight kink associated with Pro-32, and the ␣2 helix remains associated with the lipid head groups and, along with the linker region, is consequently largely shielded from solvent.
Backbone Dynamics of MMP23-PD in DPC Micelles-Backbone 1 H-15 N steady-state NOEs and 15 N longitudinal (R 1 ) and transverse (R 2 ) rate constants for MMP23-PD are shown in Fig.  8. The average 1 H-15 N NOE, R 1 , and R 2 values for the fulllength protein and ␣-helical regions of MMP23-PD are given in supplemental Table S1. The observation of mostly positive NOE values for residues 14 -63 showed that the conformation of ␣1 (containing the TMD), ␣2, and the intervening linker was restrained, whereas the N and C termini (residues 1-13 and 64 -78, respectively) were disordered (Fig. 8A). Residues in helices ␣1 and ␣2 displayed large transverse relaxation rates (R 2 ) (Fig. 8C), giving rise to broader line widths and reduced peak intensities when compared with other regions of the protein (supplemental Fig. S9). Peak intensities over this region of MMP23-PD increased at higher temperatures, primarily due to a decrease in the correlation time for the MMP23-PD/micelle complex. R 1 , R 2 and R 1 /R 2 values indicated that the degree of rigidity within this region varied, with the backbone of ␣2 being slightly more mobile than ␣1 (Fig. 8, B and D). This was expected because the TMD of ␣1 is buried and dynamically constrained by lipid within the DPC micelles. However, R 2 val- ues indicated some degree of flexibility in the TM ␣1 helix around residues Ala-27 to Leu-28. Based on the R 1 /R 2 values for residues within the TMD (residues Glu-17-Leu-40), it was pos-sible to calculate an overall isotropic global correlation time ( c ) of 14 ns for MMP23-PD in DPC micelles at 30°C. This corresponds to the rotation of a globular particle with an effective hydrodynamic radius of 26.4 Å and an apparent volume of ϳ77.1 nm 3 (assuming that the MMP23-PD-DPC micelles are spherical), comparable with values measured for other peptides embedded in DPC micelles (35).
Paramagnetic Relaxation Enhancement and Solvent Exchange Studies of MMP23-PD in DPC Micelles-Protein localization within the DPC micelles was assessed by measuring the decrease in peak intensities in the 1 H-15 N HSQC spectrum of MMP23-PD in the presence of the water-soluble paramagnetic relaxation agent Gd(DTPA-BMA) (22). Comparison of 1 H-15 N HSQC peak intensities before and after adding the relaxation agent to MMP23-PD in 100 mM DPC (Fig. 8E) showed that residues in the MMP23-PD TMD were not accessible to Gd(DTPA-BMA), presumably because they were buried within the DPC micelles. Peak intensities for residues at the N-terminal end of the ␣1 helix (residues Glu-17-Arg-19) adjacent to the predicted TMD (residues Trp-20 -Leu-40) were significantly attenuated, suggesting that these residues protruded from the micelle and were exposed to the solvent. Backbone amides for residues in helix ␣2 and the linker region between ␣1 and ␣2 experienced only a moderate paramagnetic relaxation enhancement effect and were presumably located close to the surface of the DPC micelle with only limited access to Gd(DTPA-BMA).
The Clean SEA-HSQC experiment detects the exchange of amide protons with solvent and provides a convenient method for identifying solvent-exposed residues (21). Backbone amide protons for residues Arg-19 to Ser-61, which incorporate the TMD, helix ␣2, and the linker region, exhibited negligible solvent exchange, indicating that these residues were associated with the DPC micelles and protected from solvent exchange (Fig. 8F). In contrast, residues at the N terminus of helix ␣1 (Glu-17-Arg-18) adjacent to the TMD exhibited high levels of exchange, implying that this region of the helix extends from the DPC micelles and is exposed to the solvent.

DISCUSSION
In this study, we have identified and characterized a noncanonical, metalloprotease-independent, voltage-gated K ϩ channel-modulating role for the 81-residue single helical TMD-containing prodomain of MMP23, MMP23-PD. MMP23-FL and MMP23-PD caused an equivalent ϳ2-fold decrease in the number of functional K V 1.3 channels/cell, whereas not suppressing K V 1.2. The residual K V 1.3 currents in MMP23-FLand MMP23-PD-expressing cells exhibited the same biophysical properties as wild-type K V 1.3. Because MMP23 is not present in the surface membrane (6,7,10), it is likely that these channels are not associated with MMP23-FL or MMP23-PD. MMP23-FL and MMP23-PD both co-localized with the ER marker YFP-Stim-1, confirming earlier studies reporting that mouse, rat, and human MMP23 are ER proteins (6,7,10). Both MMP23-FL and MMP23-PD co-localized equivalently with K V 1.3, but not with K V 1.2 or the DsRed vector. Furthermore, K V 1.3 expression on the cell membrane was reduced as more of the channel co-localized intracellularly with MMP23-FL and MMP23-PD. K V 1.3 channel suppression may be mediated by a mechanism consistent with intracellular trapping of the channel in the ER, although we have no evidence as to whether this is through a direct or an indirect interaction. Studies with chime-ras of K V 1.2 and K V 1.3 show that the K V 1.3 region from S5 to the C terminus is required for MMP23-PD-dependent channel modulation.
Structural analysis by NMR reveals that MMP23-PD contains a 24-residue ␣-helical region encompassing residues Glu-17 to Leu-40 (␣1), incorporating the TMD (Fig. 7). Helix helical trans-membrane proteins, they usually occur several residues C-terminal to the bend (37), as observed here. We note also that the micelle environment may influence the nature of the kink, which might not be as pronounced in a bilayer membrane.
Binding of MMP23-PD to K V 1.3 channels would be likely to leave one face of the TMD lipid-exposed while the other face interacts with the channel. Comparison of MMP23-PDs from various species shows that conserved residues are localized to two separate regions on the surface of the TMD (supplemental Fig. S10); either or both of these may be required for MMP23-TM-mediated modulation of K V 1.3 trafficking. These residues may be optimally oriented to interact with K V 1.3 as a consequence of the likely kink in the TMD around residues 26 -28.
The ␣-helical TMD of MMP23-PD is linked to a shorter membrane-interacting ␣-helical region (Ala-47-Leu-58) (Fig.  7J). Intra-or extracellular helical segments adjacent to TMDs are not uncommon (39,40) and often play a functional role (41). KCNE1, another K ϩ channel-modulating protein, also contains an ␣-helical trans-membrane domain linked to two membraneinteracting juxtamembrane helical domains (40). Like MMP23-PD, KCNE1 and the related KCNE2 protein both suppress currents generated by homomeric K V 1.4, K V 3.3, and K V 3.4 channels by trapping them early in the secretory pathway (42)(43)(44). Furthermore, KCNE4, a related protein, suppresses currents through K V 1.3-containing homotetramers and heterotetramers by trapping the channels within the ER (45). KCNE1 has additional channel-modulating effects on K V 7.1 channels mediated by the C-terminal juxtamembrane helical domain, (46) which was not found with MMP23-PD.
We have shown expression of MMP23 and K V 1.3 in normal colonic epithelium and overlapping staining of both proteins in human colorectal cancers. It is tempting to speculate that MMP23-PD may regulate the surface expression of K V 1.3 channels in primary and metastatic colorectal cancer cells, and thereby regulate cellular function. Colon cancer cells may secrete processed active MMP23 enzyme containing the K V 1.3 channel-blocking toxin domain, which would block K V 1.3 channels (IC 50 ϭ 2.8 M) on infiltrating anti-tumor T cells and suppress them as a means of immune evasion. The tumor cell would protect itself from the channel-blocking toxin domain because its K V 1.3 channels are trapped intracellularly by MMP23-PD. In support of this idea, human malignant melanoma tumors with higher MMP23 expression contain significantly fewer tumor-infiltrating lymphocytes and are associated with a greater risk of recurrence among patients treated with immune biologics (47). We cannot exclude a role for K V 1.3 in the endoplasmic reticulum of tumors. MMP23 and K V 1.3 also exhibit overlapping tissue expression in many other tissues where they may interact: lung, heart, uterus, placenta, ovary, testis seminiferous tubules, prostate, intestine, pancreatic islets, cingulate cortex, adrenal cortex, osteoblasts, chondroblasts, cartilage, synovium, natural killer cells, dendritic cells, and tendons (6, 8, 10, 48 -55) (see BioGPS web site).
In conclusion, our studies reveal a novel noncanonical, channel-modulating role for a MMP. The prodomain of MMP23, an integral membrane domain, co-localizes and traps K V 1.3 channels intracellularly, resulting in suppression of the K V 1.3 current. MMP23 is present in many tissues that express K V 1.3 channels. In these tissues, trapping of K V 1.3 channels by MMP23-PD could deplete channels from the cell surface and thereby alter cellular function. The similarity in overall architecture and trapping function between MMP23-PD and the K V -trapping KCNE proteins suggests a common evolutionary mechanism of channel modulation and interaction by these trans-membrane proteins.  15 N relaxation rates (C), and ratio of R 1 to R 2 for MMP23-PD (D). E, ratio of peak intensities before and after the addition of the water-soluble paramagnetic spin label Gd(DTPA-BMA) (19). F, peak intensities for the Clean SEA-HSQC spectrum (Cleanex) (18) of MMP23-PD in DPC micelles, measuring the exchange of backbone amide protons with solvent. Sites for which no values were determined are denoted by a P for proline residues or an asterisk for unassigned residues. Secondary structure elements derived from CSI values are illustrated at the top and boxed by a dotted line in each figure.