Interaction of the Pacemaker Channel HCN1 with Filamin A*

Pacemaker channels are encoded by the HCN gene family and are responsible for a variety of cellular functions including control of spontaneous activity in cardiac myocytes and control of excitability in different types of neurons. Some of these functions require specific membrane localization. Although several voltage-gated channels are known to interact with intracellular proteins exerting auxiliary functions, no cytoplasmic proteins have been found so far to modulate HCN channels. Through the use of a yeast two-hybrid technique, here we showed that filamin A interacts with HCN1, an HCN isoform widely expressed in the brain, but not with HCN2 or HCN4. Filamin A is a cytoplasmic scaffold protein with actin-binding domains whose main function is to link transmembrane proteins to the actin cytoskeleton. Using several HCN1 C-terminal constructs, we identified a filamin A-interacting region of 22 amino acids located downstream from the cyclic nucleotide-binding domain; this region is not conserved in HCN2, HCN3, or HCN4. We also verified by immunoprecipitation from bovine brain that the filamin A-HCN1 interaction is functional in vivo. In filamin A-expressing cells (filamin+), HCN1 (but not HCN4) channels were expressed in hot spots, whereas they were evenly distributed on the membrane of cells lacking filamin A (filamin–) indicating that interaction with filamin A affects membrane localization. Also, in filamin– cells the gating kinetics of HCN1 were strongly accelerated relative to filamin+ cells. The interaction with filamin A may contribute to localizing HCN1 channels to specific neuronal areas and to modulating channel activity.

The hyperpolarization-activated pacemaker current I f /I h has an established role in underlying generation and neurotransmitter-mediated modulation of cardiac pacemaker activity (1) and control of excitability and other functions, including integration of synaptic activity and modulation of synaptic strength, in several different types of neurons (2,3). The molecular correlates of f/h channels are the hyperpolarizationactivated cyclic nucleotide-gated channels, of which four isoforms (HCN1-4) are known. When expressed alone in heterologous systems, HCN channels elicit currents whose properties are qualitatively similar to those of native I f /I h currents, although they differ quantitatively in kinetics and cAMP sensitivity (4,5).
Recently, several studies have addressed the problem of the distribution of HCN isoforms in different cell types. HCN1, the first HCN isoform to be cloned (6), is extensively expressed in the brain with a selective distribution in specific brain areas. It is expressed for example in the layer 5 neocortical neurons, in the CA1, and to a lower degree, CA3 hippocampal regions, in the molecular cell layer of the cerebellum, and in the superior colliculus (3,(7)(8)(9). Interestingly, in some of the tissues of expression such as the neocortex, the retina, the hippocampus, and the taste receptors, the HCN1 isoform was found to be present in specific cell types and/or in specific subcellular compartments (e.g. the inner segment of rods and the cortical dendrites of CA1 hippocampal neurons) both at the RNA and protein levels (10 -14).
The correct functioning of various ion channels depends upon the interaction with auxiliary subunits (15) and with scaffolding proteins, which co-localize channels and regulatory components in subcellular compartments for improved efficiency of channel modulation (16 -18). The aim of this work was to investigate whether HCN1 channels interact with partner proteins in the brain and to check whether existing associations with any such proteins have a role in the specific cellular localization of the channels and modulation of channel function. We used the C terminus of mouse HCN1 (mHCN1), because this contains specific sequences for protein-protein interaction (including PDZ and SH3 1 domains), to screen a mouse brain cDNA library by yeast two-hybrid assay. We found that HCN1 associates with several proteins and specifically with filamin A, a high molecular mass cytoskeletal protein known to anchor different transmembrane proteins to the actin cytoskeleton and to act as a scaffold for various signaling proteins.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid System-We used the Matchmaker two-hybrid system 3 to screen a library of mouse brain cDNA inserted in pACT2 (Clontech), with a portion of the C terminus of mHCN1 in pGBKT7 as a bait (see below). The surviving clones were identified by PCR; the primers for amplification were mapped in pACT2, forward, 5Ј-CTAT-TCGATGATGAAGATACCCCACCAAACCC-3Ј) and reverse, 5Ј-GT-GAACTTGCGGGGTTTTTCAGTATCTACGA-3Ј. The proteins thus amplified included: filamin A (retrieved 8 times); Ral A (a protein of the Rho family of GTPases, (19)) (retrieved twice); GDI (GDP dissociation inhibitor, (20)) (retrieved once). Filamin A was selected for further analysis.
Constructs-To identify the shortest filamin A-interacting site of mHCN1, we made several constructs of mHCN1 C-terminal portions in pGKT7, as outlined. 1) aa 657-910, we cut pEGFPC1-mHCN1 with AccI (blunt) and BamHI, and the resulting fragment was inserted into pG-BKT7 EcoRI (blunt) and BamHI; 2) aa 657-858 and aa 859 -910, the inserts coding for these two stretches were obtained by cutting the pGBKT7-mHCN1 C terminus (construct 1) with EcoRI, AflIII, and with AflIII and BamHI, respectively; 3) aa 657-791, we inserted a stop codon (TGA) in construct 1 (aa 790) using three PCR cycles. The primers used were F1 and R3 (first cycle), F3 and R2 (second cycle), and F1 and R2 (third cycle) (see below sequence of primers; 4) aa 657-715, construct 1 was cut with EcoRI and PvuII, and the fragment was reinserted into pGBKT7 (EcoRI and SmaI); 5) aa 657-693, construct 1 was cut with PstI, and the vector was ligated; 6) aa 694 -715, two partially overlapping oligonucleotides covering this region and carrying two restriction sites were annealed in vitro, blunted with a Klenow polymerase, and digested with NcoI/BamHI; the final product was cloned into the pG-BKT7 plasmid; 7) aa 657-910 with the mutation PSLP 3 ASAA (aa 814 -817), we used three PCR cycles with the primers F1 and R1 (first cycle), F2 and R2 (second cycle), and F1 and R2 (third cycle). 8) aa 657-715 with the mutation PP 3 AA (aa 695-696), the C terminus of HCN1 was amplified by PCR from aa 694 to a 715 using a mutagenic forward primer, and the product was then cloned into the PstI restriction site of aa 657-715.
In Vitro Transcription/Translation-We performed transcription/ translation of the mHCN1 constructs generated by the above protocols and of filamin A by TNT Quick-Coupled Transcription/Translation System (Promega). All mHCN1 constructs were inserted in pGBKT7 and filamin A was inserted in pGADT7 (Clontech). [ 35 S]Methionine radiolabeled protein was obtained by an in vitro system based on mammalian reticulocyte lysate (TNT Quick Master Mix, Promega) to which plasmidic cDNA and [ 35 S]methionine (Amersham Biosciences) were added. Each sample (final volume of 50 l) was incubated for 60 -90 min at 30°C.
GFP Fluorescence-Cells transfected as described above were treated with cycloheximide (Sigma) 50 g/ml and kept for 4 h at 37°C in a 5% CO 2 incubator to block protein synthesis (22). The cells were then washed with phosphate-buffered saline at 4°C three times and fixed with paraformaldehyde (4%) for 5 min at 4°C; this was followed by a treatment with a quenching solution (0.1 M glycine/phosphatebuffered saline) for 30 min at 4°C. Cells were washed again with a 20 mM phosphate buffer and harvested on glass coverslips with 4,6-diamidino-2-phenylindole mounting medium. After fixation, cells were observed by confocal microscopy (TCS Laica) with sections at ϳ4-m intervals.
Immunoprecipitation and Membrane Preparation-We homogenized 10 g of adult bovine brain in 30 ml of a solution containing 10 mM Tris-HCl, sucrose 320 mM, 0.5% v/v protease inhibitor (Sigma), pH 7.4, with a tissue grinder and a loose fitting Dounce homogenizer (10 strokes). After a first centrifugation (2000 rpm for 10 min at 4°C), the pellet (P1) was discarded, and the supernatant was centrifuged at 20,000 rpm for 20 min at 4°C). The newly formed pellet (P2) was dissolved into 15 ml of a solution containing 10 mM Tris-HCl, 0.5% v/v protease inhibitor, 1% v/v Triton X-100 (Sigma), pH 7.4, and stirred for 20 min on ice. After a further centrifugation (20,000 rpm for 20 min at 4°C), the insoluble pellet (P3) was eliminated, and the supernatant used for the immunoprecipitation protocol.
For immunoprecipitation, 400 g of soluble membrane proteins were dissolved in 0.5 ml of water and incubated with Ab anti-filamin (MAB 1680, Chemicon) overnight at 4°C on a rotary shaker. The following day we added 25 g of protein G-Sepharose beads (P-3296, Sigma), incubated for 2 h, and centrifuged at 13,000 rpm for 10 min at 4°C. The pellet was then washed three times with a buffer solution containing 50 mM Tris-HCl, 150 mM NaCl, 5 mM EDTA, Triton X-100 1%, 0.5% sodium deoxycholate, 0.1% SDS, pH 7.4. The proteins were separated by SDS-PAGE, transferred to nitrocellulose, and incubated with anti-HCN1 antibody (Alomone Labs) for blotting. A7 and M2 cells used for immunoprecipitation were grown in 10-cm Petri dishes, transfected as described above, washed in cold phosphate-buffered saline, and scraped from the bottom, and then extracts from three dishes were resuspended in 500 l of phosphate-buffered saline containing 0.4% v/v protease inhibitor (Sigma) and homogenized. Immunoprecipitation was carried out as described above for brain extracts.
For the immunoprecipitation of proteins obtained by the in vitro transcription/translation procedure described above, individual mHCN1 constructs were either used alone or mixed (as baits) with filamin A (as prey) for 1 h at room temperature. Protocols for immunoprecipitation with anti-HA (Santa Cruz) and anti-c-Myc antibodies (Sigma) were as described above. The gel was treated with a gel fixation solution (50% methanol, 10% glacial acetic acid, 40% water) and exposed to autoradiography to reveal the presence of [ 35 S]methioninelabeled proteins.
Electrophysiology-M2 and A7 cells were washed once in Tyrode solution, placed under a fluorescent microscope, and superfused at room temperature with Tyrode. Only fluorescent cells were chosen for recording. The Tyrode solution was composed as follows (in mM): NaCl, 140; KCl, 5.4; CaCl 2 , 1.8; MgCl 2 , 1; D-glucose, 5.5; Hepes-NaOH, 5, pH 7.4. During whole-cell recording, the perfusing solution was switched to a high K ϩ solution containing (mM): NaCl, 110; KCl, 30; CaCl 2 , 1.8; MgCl 2 , 0.5; Hepes-NaOH, 5; 1 BaCl 2 , 1; MnCl 2 , 2; NiCl 2 , 0.1; nifedipine, 0.02, pH 7.4. The intracellular-like solution contained (mM): KCl, 130; NaCl, 10; EGTA-KOH, 1; MgCl 2 , 0.5; ATP (sodium salt), 2; creatine phosphate, 5; GTP (sodium salt), 0.1; Hepes-KOH, 5, pH 7.2. From a holding potential of Ϫ20 mV, hyperpolarizing steps in the range of Ϫ25 to Ϫ125 mV followed by a step at Ϫ135 mV were applied to measure activation curves in standard two-step protocols. Tail currents at Ϫ135 mV were normalized to plot activation curves. Single activation curves were fitted to the Boltzmann equation where V is voltage, V1 ⁄2 is half-maximum activation potential, and s is the inverse slope factor, to yield V1 ⁄2 and s values, which were then averaged. Deactivation traces were recorded by fully activating the current at Ϫ135 mV and then stepping to the range of Ϫ55 to 45 mV for a time long enough to reach steady-state. Time constants were calculated by fitting activation/deactivation traces to a monoexponential function after an initial delay (5), and the values were averaged and plotted against voltage. Conductances were normalized to cell capacitance, measured by a specific 10-mV step protocol. Values are given as the mean Ϯ S.E. throughout. Statistical analysis was performed by a t test comparison with a significance level set to p ϭ 0.05.

RESULTS
cDNA Screening by Yeast Two-hybrid System-We screened over 4 million proteins using the final portion of the mHCN1 C terminus (aa 657-910) as a bait and found ϳ40 hypothetical interacting partners. Of these 40 proteins, some were retrieved more often than others. To check for true versus false positives, we performed a yeast mating assay on a high stringency medium, which led to the identification of three proteins that were confirmed to bind to the HCN1 bait, Ral A (Ras-like GTPase, a protein of the Rho family of GTPases) (19), GDI (GDP dissociation inhibitor) (20), and filamin A, retrieved 2, 1, and 8 times, respectively. We focused our attention on filamin A. The interaction of filamin A with HCN1 in yeast is shown in Fig. 1A.
Filamin A, or actin-binding protein, is a large 280-kDa cytoskeletal protein carrying an actin-binding site at the N terminus, 24 immunoglobulin (Ig)-like repeats, and two hinge regions linking repeats 15 and 16 and repeats 23 and 24 (23). Filamin A acts as a dimer and, because of its specific structure, behaves as a scaffold protein. For example it can assemble broad macrocomplexes by binding a variety of proteins such as membrane receptors (e.g. D2 and D3 dopamine receptor), phos-phatases (e.g. SHIP-2), ion channels (e.g. Kv4.2 potassium channel), monomeric G-proteins (e.g. Ral A), and structural proteins (e.g. ␤1-integrin and caveolin-1) (17,19,(23)(24)(25)(26)(27)(28). Immunoprecipitation from Bovine Brain-We proceeded to test whether filamin A and HCN1 also interact in vivo in a mammalian system, and specifically in the central nervous system where both proteins are highly expressed. Using specific anti-filamin A antibodies (see "Experimental Procedures"), we immunoprecipitated filamin A from 400 g of membrane proteins extracted from a bovine brain. The immunoprecipitated and total membrane proteins were loaded onto SDS-PAGE and transferred to nitrocellulose for Western blot analysis. In Fig. 2, the filamin A-immunoprecipitated proteins (lane 1) yielded a strong band near 116 kDa, corresponding to the expected molecular mass of the HCN1 isoform (12) as detected by isoform-specific antibodies. As a control, the same signal was confirmed in the total membrane proteins before immunoprecipitation (Fig. 2, lane 2). For further control, we verified the presence (Fig. 2, lane 3) and specificity (lane 4, primary antibody omitted) of the filamin A signal in the filamin A-immunoprecipitated proteins. Also, HCN1 was not detected by incubating the same extract with a preimmune serum (Fig. 2,  Protein-Protein Interaction Assay: Mating and in Vitro Transcription/Translation-To identify the minimum filamin A-interacting portion of mHCN1 and to investigate whether HCN isoforms other than HCN1 could bind filamin A, we performed a yeast mating assay (Fig. 3). This was achieved by transforming the (MAT␣) Y187 yeast strain with filamin A and the (MATa) AH109 strain with different C-terminal portions of either mHCN1 or with the C termini of mHCN2 or hHCN4 (see Table I and Fig. 3 legend). Yeasts were then plated in high stringency conditions allowing yeast growth only under condition of strong interaction between constructs.
In Fig. 3, A-K, a collection of yeast mating combinations of filamin A and HCN C-terminal sequences is shown. Fig. 3A represents the C-terminal portion of mHCN1 used as a control for screening (aa 657-910, where 910 is the last amino acid of the mHCN1 protein), which clearly interacts with filamin A. This portion contains a PXXP motif conserved among the HCN isoforms 1, 2, 3, and 4 ( 812 PSLP 815 in mHCN1), a sequence known to bind SH3 domains and to be involved in some cases in filamin A binding (6,17), although filamin A does not contain SH3 domains. We therefore first checked whether or not this putative consensus site could be responsible for the interaction between mHCN1 and filamin A and replaced by mutagenesis amino acids PSLP with ASAA, a substitution known to affect filamin A-protein interactions (17) (Fig. 3B). The yeasts were able to grow normally indicating that this sequence is not necessary for interaction.
We also noted that the C-terminal part of mHCN1 includes a consensus sequence, conserved among all HCN isoforms (not shown), for PDZ domains (aa 907-910, XSXL), which are known to be involved in the organization of supramolecular complexes and protein scaffolding (29). To check for a potential interacting site with filamin A in this region, we generated a small fragment from the C-terminal part of the protein (aa 859 -910) containing the PDZ consensus site. As shown in Fig.  3, C and D, this fragment did not interact with filamin A, whereas its complementary sequence (aa 657-858) did indicating that the PDZ consensus site is not responsible for this interaction.
We then proceeded to shorten the test sequence of mHCN1 from the control aa 657-910 portion and prepared four truncated constructs to identify a shorter interacting region, as shown in Fig. 3, E-H (see also Table I). This approach revealed that the 22-aa-long sequence comprised between aa 694 and 715 (SPPIQSPLATRTFHYASPTASQ) is necessary for the interaction. Because proline-rich sequences may contribute to protein-protein interactions (for example with WW domains), we mutated the diproline tract 695-696 to AA in the sequence aa 657-715; the mutation did not disrupt the interaction (Fig.  3I), ruling against an involvement of this tract in the binding with filamin A.
The sequence aa 694 -715 is not conserved in any of the other HCN isoforms (HCN2, HCN3, HCN4). This was verified by performing multiple sequence alignments with ClustalW as well as a search of homologous sequences with BLAST, and neither method provided significant alignment scores (not shown). In agreement with this observation, the HCN2 and HCN4 regions homologous to the HCN1 aa 657-910 control FIG. 1. Yeast two-hybrid interaction of filamin A with mHCN1. A, yeast mating procedure used to verify the interaction between the mHCN1 C terminus (bait, aa 657-910) and filamin A (prey) yielded positive coupling (panel 1); positive (T large antigen ϩ p53) and negative (T large antigen ϩ laminin) controls are also shown in panels 2 and 3, respectively. B, the drawing represents the schematic structure of filamin A, which comprises 24 Ig-like repeats, the actin-binding domain (ABD), and two hinge regions. The segments found to interact with mHCN1 by the yeast two-hybrid screening protocol and the numbers of times they were retrieved are also shown.

FIG. 2. In vivo interaction of filamin A with HCN1.
Co-immunoprecipitation from bovine brain was performed to verify the in vivo interaction between the two proteins. Top, proteins precipitated with the anti-filamin A antibody were loaded in lane 1 and tested for the presence of HCN1. The anti-HCN1 antibody revealed a band near 116 kDa. As a positive control, total membrane proteins of bovine brain were tested in lane 2 with the same antibody and a similar band was detected. Bottom, lanes 3-7 were loaded with proteins precipitated with the anti-filamin A (Fil) antibody and incubated as follows. 3, anti filamin A antibody; 4, secondary anti-mouse antibody only (without primary antibody); 5, rabbit preimmune serum; 6, secondary anti-rabbit antibody only (without primary antibody); 7, anti-HCN1 antibody (as in lane 1). sequence (spanning from the C-terminal ends of the cyclicnucleotide binding domain (CNBD) downstream from the C termini of HCN2 or HCN4, see "Experimental Procedures") did not interact with filamin A (Fig. 3, J and K). The filamin A-interacting sequence was downstream from the CNBD in mHCN1 (Fig. 3L). As a way to confirm the results above, we also checked for protein-protein interactions using in vitro transcription and translation of the various constructs to which either a HA or c-myc tag was fused in-frame (see "Experimental Procedures" for details).
In Fig. 4A, filamin A translated in vitro with [ 35 S]methionine was immunoprecipitated using an anti-HA antibody, and its presence was revealed by autoradiography (left lane, as labeled). Similarly, the mHCN1 control sequence (aa 657-910) fused to c-myc was translated in vitro and immunoprecipitated using an anti-c-myc antibody, and its presence was revealed (Fig. 4A, center lane, as labeled). The two proteins were then mixed together, and the immunoprecipitation was performed with the anti-c-myc antibody (Fig. 4A, right lane). Autoradiography confirmed the presence of filamin A in the pellet. In Fig.  4, B-G, similar experiments were carried out using the same constructs as in Fig. 3, C-H (as indicated). Once again, the region between amino acids 694 and 715 was the shortest one yielding positive interaction. The data from the yeast mating assay and the in vitro transcription/translation assay are summarized in Table I.
To investigate the role of the interaction between filamin A and mHCN1, we performed fluorescence and electrophysiological experiments by transiently transfecting the mHCN1 isoform in-frame with GFP in a melanoma cell line lacking filamin A (filamin Ϫ or M2 cells) and compared the results with those obtained with the same cell line stably expressing the filamin A clone (filamin ϩ or A7 cells) (17,21).
We first verified by immunoprecipitation that filamin A is expressed in A7 but not in M2 cells (Fig. 5A, top) and that mHCN1 and filamin A, as expected, co-immunoprecipitate in A7 but not in M2 cells (Fig. 5A, left). The interaction was specific, because no signal was detected when untransfected cells were incubated with the anti-HCN1 antibody (Fig. 5A,  bottom right).
In Fig. 5, shown. There is a significant level of protein expression in both cell types. In Fig. 5, D and E, cells were treated with cycloheximide for 4 h to block protein synthesis and consequently reduce the fluorescence produced within endoplasmic reticulum and Golgi regions (see "Experimental Procedures" for details). Under these conditions, the mHCN1 fluorescence signal was concentrated in discrete spots on the plasma membrane of A7 cells (Fig. 5D), whereas in M2 cells it was evenly distributed on the membrane (Fig. 5E). The same protocol was applied after transfection of hHCN4-GFP in A7 and M2 cells; in this case, as apparent in Fig. 5, F and G, channels were expressed evenly on the membrane of both cell types.
These results indicate that filamin A contributes to a channel clustering process that takes place while HCN1 channels are inserted into the plasma membrane. At the same time, because the lack of filamin A does not appear to inhibit the membrane insertion of channels, the data suggest that this protein is not involved in HCN1 channel trafficking from the endoplasmic reticulum to the plasma membrane.
To identify the potential effects of the mHCN1-filamin A association on the biophysical properties of mHCN1 channels, we measured whole-cell mHCN1 currents by patch clamp in both A7 and M2 cells. In Fig. 6A, the typical currents elicited by hyperpolarization to the voltage range Ϫ25 to Ϫ125 mV (20-mV steps) are shown for A7 (left) and M2 cells (right).
On average, the normalized conductance was ϳ2 times smaller in A7 than in M2 cells (0.28 Ϯ 0.03 pS/pF, n ϭ 11, and 0.58 Ϯ 0.12 pS/pF, n ϭ 12, respectively, p Ͻ 0.05). The cell capacitance did not differ between the two groups (A7, 47 Ϯ 4.4 pF, n ϭ 11; M2, 46.8 Ϯ 4.7 pF, n ϭ 12). In Fig. 6B, the mean activation curves measured for the two types of cells are shown. In cells lacking filamin A (M2 cells, open circles), the activation curve was shifted to the positive direction relative to filamin A-expressing cells (A7, filled squares) by 7.6 mV. Fitting individual curves to the Boltzmann equation yielded half-activation voltages V1 ⁄2 ϭ Ϫ71.2 Ϯ 2.1 mV and Ϫ63.6 Ϯ 1.3 mV (significantly different) and inverse slope factors s ϭ 9.5 Ϯ 1 and 10.24 Ϯ 0.8 mV (not significantly different) for A7 (n ϭ 7) and M2 (n ϭ 8) cells, respectively. Time constants of activation and deactivation were measured and averaged (Fig. 6C). Both activation and deactivation time constants were much slower (1.4 -3-fold) in A7 cells than in M2 cells. Both the activation curve and the activation/deactivation time constant curves in A7 cells were similar to those measured previously for mHCN1 interaction with filamin A Data were obtained with either yeast two-hybrid or in vitro transcription/transduction (T/T) assay. The first nine rows refer to HCN1 constructs. The HCN1 constructs aa 657-910 mut and aa 657-715 mut carry the mutations 812 PSLP 3 ASAA 815 and 695 PP 3 AA 696 , respectively. Positive interaction in yeast two-hybrid protocols is indicated with ϩϩ or ϩϩϩ according to rate of growth; -indicates no growth. HCN2 and HCN4 constructs were not employed in in vitro T/T experiments. HCN

FIG. 4. Protein-protein interaction assay by in vitro transcription and translation.
A-G, the same HCN constructs as in Fig. 3 were tested for in vitro interaction with filamin A. The constructs were fused in-frame to a tag (HA or c-myc) for immunoprecipitation and were radiolabeled during in vitro translation with L-[ 35 S]methionine.

FIG. 5. Specificity of mHCN1-filamin A interaction.
A, co-immunoprecipitation of mHCN1 and filamin A in extracts from filamin ϩ (A7) and filamin Ϫ (M2) cells. Proteins were immunoprecipitated with antifilamin A antibody from cells expressing mHCN1 (left) or from control cells (right). The presence of filamin A and HCN1 was checked by Western blot. B-G, immunofluorescence from A7 and M2 cells transfected with GFP-tagged mHCN1 and hHCN4 channels. B and C, significant levels of mHCN1 fluorescence signal were detected from both A7 (B) and M2 (C) types of cells. D and E, confocal images (single sections) of mHCN1-transfected A7 and M2 cells treated with cycloheximide to block protein synthesis. mHCN1 channels were clustered in filamin ϩ cells but were more evenly distributed on the cell membrane in filamin Ϫ cells. Nuclei were labeled with 4,6-diamidino-2-phenylindole. F and G, confocal images (single sections) of hHCN4-transfected A7 and M2 cells treated with cycloheximide. Even channel distribution is apparent on the membrane of both types of cells. The bar is 20 m throughout. expressed in human embryonic kidney 293 cells (5).
To verify whether the changes in current density and in kinetics in Fig. 6 were actually attributable to the specific interaction between mHCN1 channels and filamin A, we analyzed the properties of hHCN4 channels expressed in A7 and M2 cells. We did not find significant differences in any of the properties of hHCN4 currents; normalized conductances were 0.12 Ϯ 0.04 (n ϭ 5) in A7 and 0.13 Ϯ 0.07 pS/pF (n ϭ 6) in M2 cells, half-activation voltages and inverse slope factors were Ϫ86.0 and 11.8 mV in A7 (n ϭ 5), and Ϫ84.4 and 10.4 mV in M2 cells (n ϭ 9), respectively. Also, the time constants of activation/deactivation were not significantly different in A7 and M2 cells in the range of Ϫ135 to 45 mV (not shown). These data indicate that the protein-protein interaction of mHCN1 channels with filamin A strongly affects the channel kinetic properties. DISCUSSION HCN channels are members of a superfamily of ion channels which include voltage gated potassium and cyclic-nucleotide gated channels. Many channels belonging to this superfamily interact with and/or are functionally modified by accessory proteins, including ␤-subunits able to modulate ion channel activity and structural proteins involved in the organization of channels and interacting elements into macromolecular complexes. For example, it is known that several KCNQ channel isoforms interact with the ␤-subunits KCNE1 or KCNE2 to generate currents similar to those recorded in native tissue (30). Also, Kv1 channels interact with PSD95, whereas Kv4.2 and Kir2.1 channels have been reported to link to filamin A (17,18,31).
Heterologous expression of individual HCN isoforms does not normally result in currents whose properties reproduce entirely those of the native pacemaker (I f /I h ) currents (4,32). Part of this difference may be attributable to the existence of a "context" dependence of channel properties, i.e. a dependence on conditioning mechanisms such as phosphorylation of the channel protein or the interaction with auxiliary/cytoskeletal proteins. Phosphorylation-dependent processes, for example, can modify channel kinetics (33) and the density of expressed channels (34); also, it has been shown recently that HCN channels are modulated by interaction with the KCNE2 subunit (35,36), even though the relevance of this interaction is still uncertain (32). The existence of a "context" dependence of HCN properties is further supported by evidence that some of the kinetic features of a given isoform vary in different expression systems (37).
One way in which intracellular processes can affect channel function is through protein-protein interactions. To find possible partners for HCN channels we used the yeast two-hybrid assay and screened a cDNA library of mouse brain. To select a bait we used the HCN1 subunit because of its extensive expression in various regions of the central nervous system (e.g. neocortex, hippocampus, and cerebellum (11,14)).
Library screening and yeast mating procedures, the latter used to confirm screening data and remove false positives, indicated that the C terminus mHCN1 bait interacts with three proteins: filamin A, Ral A, and GDI (an inhibitor of Ral A activation). Interestingly, all of these proteins are involved in the organization of the actin cytoskeleton and related plasticity processes. For example, it has been reported that Ral A interacts directly with repeat 24 of filamin A (19,27), the same region that we have found to interact with HCN1.
The proximity of the interacting sites on the filamin A molecule may explain our finding that Ral A and HCN1 interact. This result suggests that Ral A, filamin A, and HCN1 are strictly packed in a macromolecular complex. It is also interesting to note that monomeric G-proteins of the Rho family, of which Ral A is a member, are involved in the modulation of ionic channels such as N-methyl-D-aspartate and ␥-aminobutyric acid type A receptors (38,39). We found that the interaction between HCN1 and filamin A is conserved in in vivo conditions as shown by co-immunoprecipitation of the two proteins from bovine brain (Fig. 2).
It is known that filamin A interacts with several membrane proteins, among those are the K ϩ channels Kv4.2 and Kir 2.1, which bind to filamin A at their C-terminal ends. Kv4.2 is reported to bind via the sequence PTPP (17,31). Although the sequence PXXP is conserved in mHCN1, we found that it is not responsible for the interaction between filamin A and mHCN1. Using the yeast mating procedure, we restricted the mHCN1 C terminus region interacting with filamin A to a stretch of 22 amino acids downstream from the CNBD (aa 694 -715). This region comprises a diproline sequence (aa 695-696), which could potentially contribute to a protein-protein interaction with filamin A. This was, however, ruled out by the evidence that its mutation (PP-AA) did not disrupt the binding to filamin A (Fig. 3I).
A region homologous to this was not found in the C termini of either the HCN2 or HCN4 isoforms as verified by the lack of interaction with filamin A in the yeast mating assay in Fig. 3, J and K. Although we did not investigate the interaction between HCN3 and filamin A, no region homologous to the HCN1 binding site was found by ClustalW or BLAST screening (not shown).
What is the function of the interaction between filamin A and HCN1? The binding of filamin A could be a factor affecting the clustering of HCN1 channels in restricted regions of the cell membrane. As shown by the confocal images in Fig. 5, D and E, although HCN1 channels are normally distributed in a "hot spot"-like arrangement in cells lacking filamin A, dense concentrated expression areas are lost, and channels are evenly distributed on the membrane. Clusterization was due to the specific HCN1-filamin A interaction, because it was not observed in either filamin Ϫ or filamin ϩ cells transfected with hHCN4 (Fig. 5, F and G). Clusterization was not due to the lower level of mHCN1 protein expression in A7 relative to M2 cells, because no clusterization was observed with hHCN4, which was expressed equally well in A7 and M2 cells but at a much lower level (Ͻ50%) than mHCN1 in A7 cells as quantified Because filamin A binds D2 and D3 receptors and has been proposed to be functionally linked to downstream signaling components (26), our results suggest the intriguing possibility that filamin A contributes to the organization of subcellular macrocomplexes in which receptors, second messengers, and effectors are in close proximity. Additional support to a function in channel localization and clustering comes from evidence that filamin A binds caveolin 1, a structural protein responsible for the organization of caveolae (25). Caveolae represent a type of lipid raft, spatially delimited membrane microdomains known to compartmentalize membrane proteins in macromolecular complexes. For example, proteins of the ␤-adrenergic transduction pathway localize to lipid rafts in cardiac membranes (40).
Of the four HCN isoforms known, HCN1 is the least sensitive to cAMP, and shifts of the channel activation curve caused by saturating cAMP concentrations are of the order of only a few mV (4,11,32,41). Because we found that only HCN1 binds filamin A, this appears to contrast with the hypothesis that interaction with filamin A serves the purpose of concentrating elements of the cAMP second messenger cascade in restricted locations for improved channel cAMP-dependent modulation. However, most tissues expressing HCN channels possess more than one isoform (8), and heteromers can be formed by coassembly of HCN1 and HCN2 (41,42), or HCN1 and HCN4 (32). An intriguing possibility is therefore that HCN1 subunits endow heteromeric channels with an "anchoring" function whose aim is to localize channels in specific membrane subdomains, whereas the "modulating" function would rely on the cAMP sensitivity of HCN2 or HCN4 subunits.
From our data in Fig. 5, C and E-G, it also appears that the interaction with filamin A was not necessary for channel translocation from the endoplasmic reticulum to the plasma membrane. Indeed, the blocking of protein synthesis by cycloheximide did not cause an accumulation of fluorescence in the endoplasmic reticulum of M2 cells indicating that channels can reach the membrane or its proximity even in the absence of filamin A.
The lack of interaction between filamin and mHCN1 caused not only a redistribution of the channel in the membrane but also a change in the properties of the expressed current. In particular, in filamin Ϫ cells the HCN conductance was twice that in filamin ϩ cells, and both channel activation and deactivation were ϳ2 times faster (Fig. 6). In agreement with the view that HCN1 properties are modified by a specific interaction with filamin A, we found no differences in the properties of HCN4 channels expressed in A7 or M2 cells. An increased conductance is consistent with the idea that filamin A concentrates channel expression in specific locally restricted areas, thus limiting random access of channels to the membrane and thus decreasing the mean channel density.
On the other hand, we do not have a ready explanation for the slowing action of filamin A binding to channels. However, there is established evidence that the C terminus is involved in determining the kinetics of HCN channels (4,43,44), and the slowing due to the binding of the C terminus to filamin A might reflect an interference with the mechanism by which the C termini affect gating.
In conclusion, our results showed that HCN1, but not HCN2 or HCN4, binds to an ubiquitous isoform of filamin, filamin A. The interaction involves the C-terminal end of HCN1 and the last two Ig-like repeats at the C-terminal end of filamin A, the site of the protein dimerization domain, whereas the actinbinding domains of the protein remain free. Binding to filamin A slows HCN1 channel kinetics and causes channels to cluster within restricted regions of the cell membrane, thus reducing the density of channel expression and whole-cell conductance. A possible function of HCN1 channel compartmentation is to increase the efficiency of channel control by modulating agents.