Filamin A Promotes Dynamin-dependent Internalization of Hyperpolarization-activated Cyclic Nucleotide-gated Type 1 (HCN1) Channels and Restricts Ih in Hippocampal Neurons*

Background: HCN channels influence neuronal excitability. Results: Filamin-A (FLNa) facilitated selective, reversible dynamin-dependent internalization of HCN1 and reduced Ih. In hippocampal neurons, dominant-negative FLNa enhanced native HCN1, and decoy peptides disrupting HCN1-FLNa binding reduced channel clustering and augmented endogenous Ih. Conclusion: FLNa modulates neuronal excitability via dynamin-mediated HCN1 endocytosis. Significance: Novel roles for FLNa in mature neuronal function are presented. The actin-binding protein filamin A (FLNa) regulates neuronal migration during development, yet its roles in the mature brain remain largely obscure. Here, we probed the effects of FLNa on the regulation of ion channels that influence neuronal properties. We focused on the HCN1 channels that conduct Ih, a hyperpolarization-activated current crucial for shaping intrinsic neuronal properties. Whereas regulation of HCN1 channels by FLNa has been observed in melanoma cell lines, its physiological relevance to neuronal function and the underlying cellular pathways that govern this regulation remain unknown. Using a combination of mutational, pharmacological, and imaging approaches, we find here that FLNa facilitates a selective and reversible dynamin-dependent internalization of HCN1 channels in HEK293 cells. This internalization is accompanied by a redistribution of HCN1 channels on the cell surface, by accumulation of the channels in endosomal compartments, and by reduced Ih density. In hippocampal neurons, expression of a truncated dominant-negative FLNa enhances the expression of native HCN1. Furthermore, acute abrogation of HCN1-FLNa interaction in neurons, with the use of decoy peptides that mimic the FLNa-binding domain of HCN1, abolishes the punctate distribution of HCN1 channels in neuronal cell bodies, augments endogenous Ih, and enhances the rebound-response (“voltage-sag”) of the neuronal membrane to transient hyperpolarizing events. Together, these results support a major function of FLNa in modulating ion channel abundance and membrane trafficking in neurons, thereby shaping their biophysical properties and function.

The actin-binding protein filamin A (FLNa) regulates neuronal migration during development, yet its roles in the mature brain remain largely obscure. Here, we probed the effects of FLNa on the regulation of ion channels that influence neuronal properties. We focused on the HCN1 channels that conduct I h , a hyperpolarization-activated current crucial for shaping intrinsic neuronal properties. Whereas regulation of HCN1 channels by FLNa has been observed in melanoma cell lines, its physiological relevance to neuronal function and the underlying cellular pathways that govern this regulation remain unknown. Using a combination of mutational, pharmacological, and imaging approaches, we find here that FLNa facilitates a selective and reversible dynamin-dependent internalization of HCN1 channels in HEK293 cells. This internalization is accompanied by a redistribution of HCN1 channels on the cell surface, by accumulation of the channels in endosomal compartments, and by reduced I h density. In hippocampal neurons, expression of a truncated dominant-negative FLNa enhances the expression of native HCN1. Furthermore, acute abrogation of HCN1-FLNa interaction in neurons, with the use of decoy peptides that mimic the FLNa-binding domain of HCN1, abolishes the punctate distribution of HCN1 channels in neuronal cell bodies, augments endogenous I h , and enhances the rebound-response ("voltage-sag") of the neuronal membrane to transient hyperpolarizing events. Together, these results support a major function of FLNa in modulating ion channel abundance and membrane trafficking in neurons, thereby shaping their biophysical properties and function.
Filamin A (FLNa) 2 is an actin-binding protein that stabilizes the cytoskeleton by cross-linking filamentous actin (1,2). FLNa has been studied extensively at early stages of brain development, when it plays crucial roles in structural organization and neuronal migration (3,4). However, despite its wide distribution in the adult brain (5,6), very little is known about the function of FLNa in mature neurons.
Interestingly, interaction of FLNa with several ion channels and receptors has been described, leading to altered trafficking and function in several cell types (7). Here, we explored the nature, mechanisms, and relevance of FLNa to neuronal function by focusing on its interaction with the hyperpolarizationactivated cyclic nucleotide-gated (HCN) channels.
HCN channels are a distinct class of ion channels that are widely distributed across the mammalian brain. By conducting the sub-threshold current I h , these channels regulate neuronal functions such as signal integration in dendrites, oscillatory behavior, and synaptic release (8 -12). In view of their pivotal influence on neuronal excitability, it is perhaps not surprising that trafficking and expression of HCN channels are highly regulated. Distinct subcellular distribution patterns of HCN channels characterize different neuronal populations (13,14), and up-or down-regulation of HCN channel abundance and membrane expression are central to processes of neuronal plasticity (15)(16)(17)(18)(19) as well as to pathological conditions (20 -22).
In cell lines, a direct, specific interaction between FLNa and the HCN1 channel isoform (the most abundant HCN subunit in the forebrain) has been described. This interaction occurs via a 22-amino acid sequence in the channel CЈ terminus, which is absent in all other members of the HCN channel family (23).
Overexpression of HCN1 in melanoma cell lines devoid of FLNa resulted in higher I h amplitude and facilitated gating, implying a restrictive role for FLNa on HCN1 channels (23). Whereas the regulation of HCN1 channels by FLNa in neurons should constitute a novel signaling mechanism between the actin network and ion channels, the presence of such regulation, its mechanisms, and its physiological relevance remain unknown.
Here, we examined the cellular pathways that govern neuronal HCN1 regulation by FLNa and their physiological implications to neuronal function. Our findings support a novel role for FLNa in neurons. By promoting selective dynamin-dependent internalization, FLNa acts to remove HCN1 channels from the membrane and direct them to the endocytic pathway, thereby constraining I h and altering membrane responses to voltage fluctuations. Thus, FLNa emerges as a potent modulator of neuronal properties by controlling the molecular composition and dynamics of select ion channels on the membrane.

EXPERIMENTAL PROCEDURES
Plasmid cDNA Constructs-A plasmid cDNA construct containing the EGFP sequence at the C terminus of the mouse HCN1 DNA (inserted between amino acids 885 and 886, see Refs. 24, 25) was a gift from Dr. Chetkovich (Northwestern University). The NЈ-terminal EGFP-fused mouse HCN1 and HCN2 cDNA constructs (HCN1 GFP-NЈ and HCN2 GFP ) were gifts from Drs. Santoro and Siegelbaum (Columbia University). An HCN1 GFP channel construct lacking the filamin-binding domain (HCN1(⌬22) GFP ) was generated by deleting the 22-amino acid sequence ( 694 SPPIQSPLATRTFHYASPTASQ 715 ) in the mouse HCN1 CЈ terminus that was previously reported to form the interaction domain (23). The final construct was confirmed by sequencing and yielded functional I h upon expression in HEK293 cells. Human FLNa with an N-terminal fusion of monomeric DsRed (FLNa DsRed ) (26), as obtained via pDsRedmonomer-C1 plasmid (Clontech), was a gift from Dr. Nakamura (Harvard Medical School). This FLNa DsRed plasmid was used as a template to isolate Ig-like domains 23 and 24 of FLNa by PCR with the following primers: 5Ј-GTGCTCGAGGGGACC-CAGGCTTGGTGTC-3Ј (FLNa Ig23 forward; possessing an XhoI site, underlined before the start of Ig-like domain 23); 5Ј-CTTCAATTGAATTCAGGGCACCACAACGCGG-3Ј (FLNa Ig24 reverse; containing an EcoRI site, italic, at the regular filamin A stop codon, in bold, followed by an MfeI site, underlined).
The corresponding PCR fragment was restricted with XhoI and MfeI and inserted into the FLNa DsRed plasmid that had been previously cut with XhoI and MfeI. This resulted in plasmid FLNa(23-24) DsRed where monomeric DsRed (225 amino acids) is coupled via a two-amino acid linker (Ser-Arg) to the C-terminal 220 amino acids of human FLNa (encompassing Ig-like domains [23][24]. All cDNA constructs used in this study and their nomenclatures are summarized in Table 1. HEK293 Cell Culture and Transfection with Plasmid cDNA-Human embryonic kidney 293 (HEK293) cells were maintained in minimum essential medium supplemented with 100 g/ml penicillin/streptomycin, 2 mM glutamine, and 10% fetal bovine serum. The cells were kept in a humidified atmosphere, at 37°C and 5% CO 2 , refreshed every 2-3 days, and passaged upon con-fluence. All culture reagents were from Invitrogen. Transfection of HEK293 cells with cDNA constructs was performed using the TransIT-LT1 method (Mirus), following the manufacturer's protocol. Briefly, 1-2 days prior to transfection, cells were plated on a 12-mm glass coverslip. Cells were co-transfected with HCN and FLNa constructs using 1.5 l of Transit-LT1 reagent and 0.3 g of plasmid cDNA per construct per coverslip (resulting in a 1:1 DNA ratio). All experiments were performed 24 -48 h post-transfection. In a subset of experiments, cells were transfected using the calcium-phosphate precipitation method as described previously (27), with the same amounts of plasmid cDNA as detailed above. Notably, the expression patterns of HCN channels and their influence by FLNa were reproducible using either transfection protocol.
Primary Hippocampal Neurons-Primary hippocampal neurons were prepared from brains of postnatal day 0 (P0) Sprague-Dawley rat pups as described previously (25). Following decapitation, hippocampi were dissected and incubated with the protease papain (Worthington) for 30 min at 36°C. Papain was removed in a series of washes in the presence of the protease inhibitor ovomucoid (Sigma), followed by mechanical trituration. Dissociated cells were plated on 12-mm glass coverslips at a density of 400 -600 cells/mm 2 and grown at 36°C (5% CO 2 ). The cultures were maintained in neurobasal medium (NBM) supplemented with B27 (Invitrogen), which was preconditioned for 24 h in glial culture. All experiments were in compliance with National Institutes of Health and University of California at Irvine animal care regulations.
Transferrin Uptake Assay-To label recycling endosomes, HEK293 cells expressing HCN1 and FLNa constructs were first starved of transferrin by a 15-min incubation with serum-deficient minimum essential medium, followed by a 1-h incubation with 50 g/ml Alexa-conjugated transferrin (Invitrogen) at 37°C. Cells were quickly rinsed with ice-cold PBS, and fixed in 4% paraformaldehyde as described above.
Lentiviral Transduction of FLNa (23)(24) DsRed -To generate lentiviral particles, FLNa(23-24) DsRed was excised using AgeI and EcoRI and subcloned into the lentiviral vector plasmid FCK(1.3)GW (29) from which the GFP cDNA had been removed using AgeI and EcoRI. The resulting vector plasmid was co-transfected with pCMV-⌬8.9 and pCMV-VSVg constructs (29) in HEK293T cells using the calcium-phosphate method. Two days post-transfection, the virus-containing supernatant was collected and purified using the Fast-Trap purification kit (Millipore). Hippocampal neurons were transduced by incubation with 10 l of purified virus for 5 h and fixed at 8 -13 days post-infection (Ͼ21 DIV).
Confocal Imaging and Analyses-Confocal imaging was performed using an LSM-510 confocal microscope (Zeiss) equipped with an Apochromat 63ϫ oil objective (numeric aperture ϭ 1.40). Samples that contained more than one fluorophore were scanned sequentially using the "multi-track" mode with separate excitation beams as follows: an argon laser at a wavelength of 488 nm for EGFP imaging and a He/Ne laser at 543 nm for DsRed imaging. In immunocytochemical experiments that required a third channel, a He/Ne laser beam at 633 nm was used for far-red imaging, and a two-photon tunable Ti:Sapphire excitation beam set at 760 nM was used for illumination at the UV range. To exclude any possibility for signal bleed through originating from the FLNa DsRed construct, experiments were repeated while expressing EGFP-fused HCN constructs with a myc FLNa construct that does not fluoresce, resulting in similar effects on HCN1 distribution as observed for FLNa DsRed . Images were digitized at 12 bits, in a frame size of 1024 ϫ 1024 pixels. Optical slices were scanned at a thickness of Ͻ1 m. Final image adjustments were performed using ImageJ software (National Institutes of Health, version 1.44); images were background-subtracted using the "BG subtraction from ROI" plugin (M. Cammer and T. Collins) set to a threshold of two standard deviations from the mean pixel intensity. For presentation purposes, a mild median filter (radius ϭ 0.2 pixels) was applied.
Quantitative assessment of signal distribution and clustering (Figs. 3, C-F, 6B, and 8D) was preformed with the Van Steensel's cross-correlation (CCF) method, using the JACoP plugin in ImageJ (30,31). Pearson's correlation coefficient was calculated for each cell by comparing each image to its own duplicate, while shifting one image across the x axis (␦x ϭ Ϯ20 pixels). When ␦x ϭ 0 (no displacement), CCF values were always equal to 1. However, upon shifting the image across the x axis, the degree of change in CCF values indicates the level of signal homogeneity, with punctate images having a steeper decline in CCF values compared with images with diffuse signal distribution (31). Statistical comparison of CCF values was performed on a ␦x value of Ϫ20 pixels, using one-way ANOVA with Holm-Sidak correction for multiple comparisons or Student's t test.
Analysis of somatic HCN1 immunoreactivity in hippocampal neurons (Fig. 7) was performed on z-stacked, backgroundsubtracted and thresholded images (see above) by outlining the somatic region of pyramid-shaped neurons and extracting the mean intensity value per soma (ImageJ, National Institutes of Health). For each acquired image, the normalized intensity ratio (Fig. 7D) was calculated by dividing the mean intensity of the channel's immunoreactive signal in FLNa(23-24)positive neurons by its mean intensity in FLNa(23-24)-negative neurons.
Quantitative determination of HCN1 signal co-localization with endosomal markers was performed using the Manders' coefficient method (32,33). This coefficient describes the degree of signal overlap between two images (with 0 representing no co-localization and 1 representing 100% co-localization) and is particularly suitable for the analysis of two molecules with varying distribution patterns and intensities (32). The analysis was performed using the JACoP plug-in in ImageJ (31). To ensure the specificity of the analysis, control experiments were carried where the Manders' coefficient was calculated for pairs of unrelated images within the same condition. These control studies yielded substantially smaller coefficient values (0.08 Ϯ 0.01; n ϭ 92), which represent the random, basal co-localization of signals in our experiments.
Total Internal Reflection Microscopy (TIRF-M)-TIRF-M imaging was performed using an inverted Nikon Ti microscope with a TIRF attachment, equipped with an oil immersion 60ϫ Apo TIRF objective (numeric aperture ϭ 1.49). Fluorescent excitation was employed using an argon/krypton laser at wavelengths of 488 and 568 nm for EGFP and DsRed imaging, respectively. EGFP and DsRed signals were selected using bandpass filters at wavelengths of 525/50 and 625/50 nm, respectively. Images were acquired using an Andor iXon 897 CCD camera.
Dynamin Inhibition-Inhibition of dynamin in HEK293 cells was carried out by a 4-h incubation with 80 M of the selective dynamin inhibitor dynasore (Sigma) (34) at 37°C. As a vehicle treatment, the organic solvent (0.26% DMSO) was applied for a similar duration.
Myristoylated Peptides for HCN1-FLNa Interference-An N-myristoylated peptide composed of the 22-amino acid sequence that forms the FLNa-binding domain in HCN1 was synthesized alongside a scrambled control peptide (Thermo-Fisher; see Fig. 8A for peptide sequence). The peptides were dissolved in H 2 O to yield stock solutions of 1 mM for long term storage (Ϫ80°C). In experiments testing the acute effects of FLNa on HCN1 distribution (Fig. 8), stock solutions of intermediate concentrations were prepared on the day of the experiment to yield varying concentrations as detailed in Fig. 8. Primary hippocampal neurons were incubated with the peptides for 4 h in NBM ϩ B27 culture medium (at 37°C and 5% CO 2 ) prior to fixation. In electrophysiology experiments ( Fig. 9), the peptides were diluted in artificial cerebral spinal fluid (see composition below) on the day of the experiment, to a final concentration of 100 nM. Hippocampal neurons were preincubated for Ͼ30 min prior to forming a patch seal.
HEK293 Cells-I h was recorded from transfected HEK293 cells using the whole-cell voltage clamp configuration as recently described (27). The recording chamber was continuously perfused with an extracellular recording solution containing the following (in mM): 110 NaCl, 5 KCl, 2 CaCl 2 , 1 MgCl 2 , 10 HEPES, (pH set to 7.30). Recording pipettes (pulled from borosilicate glass) had a resistance of 2-5 megohms when filled with an intracellular solution containing (in mM) 105 potassium gluconate, 30 KCl, 2 Mg-ATP, 5 EGTA, 0.5 CaCl 2 , 0.5 cAMP, 10 HEPES (pH set to 7.30). Series resistance was compensated for at least 75%. Only traces with voltage errors smaller than 2 mV were considered for analysis of channel gating and kinetics. Currents were acquired with an Axopatch 200B amplifier (Molecular Devices), controlled by a customwritten program in MATLAB (Mathworks). Currents were low pass Bessel-filtered at 2 kHz and sampled at 5 kHz. I h was evoked by a series of hyperpolarizing steps from a resting membrane potential of Ϫ50 mV in decrements of 10 mV. I h amplitude was calculated by subtracting the steady-state current at the end of the voltage step from the instantaneous current at its beginning (25,27). I h activation kinetics in HEK293 cells were best fit to a monoexponential function (27), and only traces with goodness-of-fit of Ͼ0.985 were included in the final analysis. The voltage-dependent activation of I h was derived by fitting a Boltzmann equation to I h conductance values of individual traces at different potentials, as recently described (27). Liquid junction potentials (calculated at 13 mV) were corrected for off line.
Hippocampal Neurons-Electrophysiological measurements in primary hippocampal neurons were performed using the whole-cell voltage and current clamp patch clamp configurations. A 12-mm glass coverslip containing 21-22 DIV neurons was placed in a recording chamber continuously perfused with a CO 2 -saturated artificial cerebral spinal fluid. The artificial cerebral spinal fluid solution contained the following (in mM): 125 NaCl, 25 NaHCO 3 , 20 glucose, 4 KCl, 2 CaCl 2 , 1.25 Na 2 PO 4 , 1 MgCl 2 . To isolate I h from other intrinsic and synaptic conductances, the following blockers were added to the bath solution (in mM): 10 tetraethylammonium, 0.5 BaCl 2 , 0.1 NiCl 2 , 0.02 DL-2-amino-5-phosphonopentanoic acid, 0.01 6-cyano-7nitroquinoxaline-2,3-dione, 0.0005 tetrodotoxin, 0.005 bicuculline. Recording pipettes had a resistance of 3-5 megohms upon filling with an intracellular solution that contained the following (in mM): 105 potassium gluconate, 20 KCl, 2 Mg-ATP, 5 EGTA, 0.5 CaCl 2 , 10 HEPES (pH set to 7.35). Currents were acquired with a Multiclamp 700A amplifier (Axon Instruments) and digitized with a Digidata 1322A converter (Axon Instruments). Series resistance was compensated 30 -40%, reducing maximal voltage error to an average of 1.2 Ϯ 0.1 mV (n ϭ 29). Currents were low pass Bessel-filtered at 2 kHz and sampled at 10 kHz. Signal acquisition and data analysis were performed using pCLAMP 9.0 software (Axon Instruments). In voltage clamp experiments, I h was recorded and analyzed as detailed above for HEK293 cells. Cell capacitance (C in ) and resistance (R in ) were calculated based on a ϩ10-mV test pulse as described previously (25,35). Current density was calculated as the whole-cell I h amplitude, normalized by the cell capacitance. Resting membrane potential was determined immediately following breaking into the whole-cell configuration and switching into current clamp mode. In current clamp experiments, voltage sag was recorded by eliciting a series of hyperpolarizing current injections from 0 to 300 pA in 50 pA increments (step duration ϭ 500 ms). Sag ratio was calculated as 100⅐(V ss Ϫ V peak )/(RMP Ϫ V peak ) (%), where V peak is the minimal potential achieved at the beginning of the current step; RMP is resting membrane potential, and V ss is the steady-state potential at its end.
Differences between groups were determined using Student's t test, nonparametric Mann-Whitney test, or two-way ANOVA, as appropriate. A cutoff value of p Ͻ 0.05 was assumed to indicate a significant difference.

Selective Interference with the FLNa Interaction Site of HCN1
Augments I h and Alters Channel Gating Properties-To study regulation of HCN1 by FLNa, we generated an EGFP-fused HCN1 channel construct (HCN1(⌬22) GFP ) lacking the 22 amino acids that form the FLNa-interacting domain (Table 1 and Fig. 1). Removal of the FLNa-interacting domain was not detrimental to the delivery of HCN1 to the cell membrane or to its function; stepwise membrane hyperpolarization of HCN1(⌬22) GFP -expressing HEK293 cells yielded slowly activating, noninactivating inward currents typical of HCN1-mediated I h , which were virtually absent in mock-transfected cells ( Fig. 2A). In line with previous studies (24, 25, 36 -38), the fusion of EGFP to either the CЈ-or the NЈ terminus of the channel did not compromise its delivery to the membrane and did not influence its biophysical characteristics (data not shown).
Having established a functional HCN1 channel construct lacking the FLNa-binding domain, we employed it to study the consequences of HCN1 regulation by FLNa. Whole-cell voltage clamp recordings from HEK293 cells that co-expressed DsRedfused FLNa (FLNa DsRed ) together with either HCN1 GFP or HCN1(⌬22) GFP revealed a 2-fold increase in I h density in HCN1(⌬22) GFP -expressing cells as compared with HCN1 GFPexpressing cells (75 Ϯ 16 versus 149 Ϯ 28 pA/pF upon voltage  Fig. 2G). Together, these data support the notion that HCN1 regulation by FLNa occurs via a subunit-specific sequence in the HCN1 CЈ terminus and acts to constrain I h amplitude and modulate its gating properties without influencing its reversal potential.
FLNa-mediated Suppression of I h Is Accompanied by HCN1 Channel Clustering-In light of the established roles of FLNa in cytoskeletal dynamics and membrane trafficking (39), we next examined whether the observed inhibition of I h by FLNa is related to altered HCN channel distribution and/or membrane expression. We used confocal imaging of HEK293 cells transfected with cDNA constructs that selectively target HCN1-FLNa interaction domains ( Table 1). Transfection of HCN1 GFP alongside an empty plasmid vector in HEK293 cells resulted in a largely punctate distribution pattern of the channel protein (Fig. 3A). Whereas HEK293 cells express endogenous FLNa (Fig. 3B), the levels of endogenous protein may not suffice to engage a large number of HCN1 molecules generated by overexpression strategies. Therefore, we co-transfected cells with both HCN1 GFP and FLNa DsRed . In these co-transfected cells, the punctate distribution of HCN1 GFP was accentuated (Fig. 3, A, C and G). In contrast, co-expression of FLNa DsRed with either HCN1(⌬22) GFP or HCN2 GFP (both not expected to interact with FLNa) resulted in a largely diffuse distribution (Fig. 3, A,  C-E, and G). The robust clustering of HCN1 by FLNa may be either a direct result of conformational changes of the channel induced by FLNa binding or of more complex signaling pathways involving other cytoskeletal elements. To discern between these possibilities, we employed a DsRed-fused, FLNa truncation mutant (FLNa(23-24) DsRed ), which possesses the HCN1binding region (CЈ-terminal Ig-like domains 23-24) but not the actin-binding regions ( Table 1). The distribution of FLNa(23-24) DsRed in HEK293 cells was diffuse and nonspecific, as opposed to the filamentous distribution observed with FLNa DsRed and endogenous FLNa (Fig. 3B), consistent with lack of actin binding capacity. In contrast to the clustered HCN1 distribution in the presence of full-length FLNa (Fig. 3A), co-expression of the truncated FLNa(23-24) DsRed leads to a homogeneous HCN1 GFP distribution pattern (Fig. 3, A and F), which resembled the expression pattern of HCN1(⌬22) GFP (Fig. 3, A and D). These opposing effects of FLNa and FLNa(23-24) on HCN1  FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 clustering were reproduced using an HCN1 construct with an NЈ-terminal fusion of the EGFP (HCN1 GFP-NЈ , not shown).

Filamin A Regulates Neuronal HCN1 Channel Internalization
To obtain a quantitative picture of the FLNa-induced clustering, we performed a CCF analysis of signal distribution (see "Experimental Procedures") (30). Pooled data from 20 to 23 cells per experimental group (obtained from at least three separate experiments) demonstrated a steeper CCF profile for HCN1 ϩ FLNa-expressing cells, indicative of a more clustered distribution (Fig. 3, C-G).
FLNa Alters HCN1 Distribution Pattern at Both the Cell Surface and Intracellularly-FLNa-mediated clustering of HCN1 channels can explain the inhibitory effects of FLNa on I h if it derives from reduced surface expression and accumulation of the channels within intracellular organelles. To study HCN1 channel distribution on the cell surface, we employed TIRF-M, a technique that allows the visualization of only a very thin section of the cell (50 -150 nm), representing membrane and peri-membranous domains (40). The clustered pattern of HCN1 channels in the presence of FLNa was discernible also in the TIRF plane. It stood in contrast to the more homogeneous surface expression pattern of HCN1(⌬22) and HCN2 channels in the presence of FLNa, as well as to the smooth distribution pattern of HCN1 in the presence of the truncated FLNa(23-24) (Fig. 4A). Close examination of TIRF and wide field images captured from the same cell allowed us to evaluate the subcellular location of HCN1-containing clusters. We found that a significant proportion of HCN1 puncta was present only in the wide field images, i.e. localized within the cell interior (Fig. 4B). Thus, combined TIRF and wide field imaging demonstrated that although FLNa may regulate HCN1 at the cell surface, it The amplitude of the currents was normalized to maximum value to allow for visual comparison of the time course. Only a portion of the initial response is shown. E, activation kinetics of I h in HCN1 GFP -and HCN1(⌬22) GFP -expressing cells. HCN1(⌬22) GFP -mediated I h was faster than HCN1 GFP . F, I h traces representing the protocol used to determine the reversal potential (E rev ) of the current. Upon the initial step to a near-maximal activation voltage of Ϫ110 mV, membrane potential was stepped back to a range of voltage values. Because I h deactivates relatively slowly, the instantaneous peak values of the tail current (inset on the right) represent predominantly the changes in driving force. G, current-voltage (I-V) curves, plotted based on the protocol shown in F. No difference was found between E rev of I h in HCN1 GFP -and HCN1(⌬22) GFP -expressing cells, indicating that the FLNa interaction site does not control ion permeation and selectivity in HCN1 channels (note: standard error in many data points was too small to be visualized). *, p Ͻ 0.05, Student's t test.

Filamin A Regulates Neuronal HCN1 Channel Internalization
also facilitates accumulation of the channels within intracellular clusters.
FLNa Facilitates Dynamin-dependent Internalization and Targeting of HCN1 to Endocytic Compartments-FLNa-induced intracellular clustering of HCN1 channels raises questions regarding the cellular mechanisms involved. Because FLNa regulates trafficking and endocytic processing of several membrane proteins (26,(41)(42)(43), we tested whether FLNa promotes targeting of HCN1 channels to endocytic compartments. Immunolabeling of HCN1/FLNa co-expressing HEK293 cells with an antibody directed against the lysosomal protein LAMP1 revealed substantial co-localization of HCN1-containing puncta with LAMP1-positive organelles (Fig. 5, A and B). In contrast, HCN1(⌬22) GFP was homogeneously distributed and was not preferentially localized to LAMP1-positive organelles (Fig. 5, A and B). To quantify these distinct distribution patterns, we employed the Manders' coefficient method for analysis of signal co-localization (see "Experimental Procedures") (32). As shown in Fig. 5C, there was a significant difference between the co-localization coefficients of LAMP1 with HCN1 and HCN1(⌬22) (n ϭ 16, 21 cells, respectively, based on three independent experiments; Student's t test, p Ͻ 0.001). These data demonstrate a selective FLNa-mediated lysosomal accumulation of full-length HCN1 channels.
Using similar imaging and analysis strategies, we extended our experiments to other types of endocytic organelles. In addition to its presence within lysosomes, HCN1 channel protein accumulated in subpopulations of early endosomes and recy-FIGURE 3. Filamin-dependent clustering of HCN1 channels. A, confocal images of HEK293 cells expressing HCN1 GFP with an empty vector plasmid, as well as HCN1 GFP , HCN1(⌬22) GFP , or HCN2 GFP alongside the FLNa constructs, as indicated. Co-expression of HCN1 GFP with an empty vector resulted in a clustered appearance, which was accentuated upon co-expression of HCN1 GFP and FLNa DsRed . In contrast, the distribution pattern of both HCN1(⌬22) GFP and HCN2 GFP (co-transfected with FLNa DsRed ) was diffuse. Co-expression of HCN1 GFP and FLNa(23-24) DsRed abolished the punctate distribution of HCN1 GFP . B, comparison of the subcellular distribution of endogenous FLNa, FLNa DsRed , and a FLNa dominant-negative mutant (FLNa(23-24) DsRed ) that possess the HCN1-binding domain (Ig domains 23 and 24) but not the actin-binding domains. Note the homogeneous distribution of FLNa(23-24) DsRed, consistent with lack of actin-binding capacity. C-F, quantitative comparison of HCN-channel signal distribution in different conditions, using the van Steensel's CCF. The clustered distribution of HCN1 in the presence of FLNa (shown in all four graphs for comparison purposes) was manifested as a relatively steep change in CCF upon shifting the image on the x axis (see "Experimental Procedures"). In contrast, the CCF distributions of HCN1 (C), HCN1(⌬22)ϩ FLNa (D), HCN2 ϩ FLNa (E), and HCN1 ϩ FLNa(23-24) (F) were shallower, indicating a more homogeneous distribution of the channels. G, comparison of HCN channel distribution CCF across conditions (one-way ANOVA, F(4, 102) ϭ 8.019; *, p Ͻ 0.05; ***, p Ͻ 0.001; Holm-Sidak's test). CCF analysis in C-G was performed with n ϭ 20 -23 cells per condition, taken from Ͼ3 independent experiments. All images are a z-stack representation of confocal optical slices at Ͻ1 m thickness. Scale bars, 10 m. FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 cling endosomes, as evident by increased co-localization of HCN1 GFP with the early endosomal marker EEA1 and with transferrin-labeled recycling endosomes (n ϭ 22-25 cells per experimental group; taken from three independent experiments; Student's t test, p Ͻ 0.001; Fig. 5, D-I).

Filamin A Regulates Neuronal HCN1 Channel Internalization
The increased presence of the HCN1 channel protein in several types of organelles along the endocytic pathway implies a role for FLNa in the removal of HCN1 channels from the cell membrane. To study the involvement of FLNa in channel internalization, we focused on dynamin, a GTPase that facilitates endocytosis by enabling the fission of newly formed pits bud-ding from the plasma membrane (44). Incubation of HEK293 cells co-transfected with FLNa DsRed and HCN1 GFP in the presence of the selective dynamin-inhibitor dynasore (80 M) led to a marked decrease in HCN1 channel clustering (Fig. 6, A and B). In addition to rescuing the clustering phenotype, dynasore also augmented I h density 2-fold in HCN1 GFP -transfected HEK293 cells (54 Ϯ 10 versus 106 Ϯ 19 pA/pF for vehicle and dynasore treatments, respectively; n ϭ 23, 30 cells; Mann-Whitney test, p ϭ 0.039; Fig. 6, C and D). Notably, dynasore treatment had no effect on I h in cells co-expressing FLNa DsRed and the HCN1 construct lacking the FLNa-interacting domain (HCN1[⌬22])  2nd column) are shown, as well as TIRF images of the EGFP signal obtained from the same cell (3rd column). The far right column represents a magnified region of the TIRF image to its left. Note the clustered appearance of HCN1 GFP when expressed alongside FLNa DsRed under TIRF conditions. Also note the punctate surface expression pattern of HCN1 GFP at the TIRF plane, compared with the smooth pattern obtained in all other conditions. B, close examination of wide field and TIRF images of the same cell reveals that some of the HCN1 GFP clusters are located in proximity to the cell surface (as evident by their presence in both wide field images and the TIRF plane, green arrowheads), whereas others are intracellular (as evident by their virtual disappearance from the TIRF image, light blue arrowheads). Images are representative of data obtained from three separate experiments, with each condition performed at least in duplicate.
In Hippocampal Neurons, FLNa Regulates Native HCN1 Expression Levels and Subcellular Localization-HCN1 channels have crucial influence on hippocampal neuronal function. Therefore, we examined whether FLNa might constrain native HCN1 channel surface expression and distribution patterns in neurons. We infected primary hippocampal neurons with lentiviral particles encoding the FLNa truncation mutant (FLNa(23-24) DsRed ), which was found effective in blocking the effect of FLNa on HCN1 clustering in HEK293 cells (Fig. 3). At an age when neurons are generally considered mature (DIV Ͼ21), expression of FLNa(23-24) DsRed was evident in infected neurons as DsRed signal in somata and neurites (Fig. 7A). Immunocytochemistry for native HCN1 revealed significantly stronger HCN1 immunoreactive signal in neurons expressing FLNa(23-24) DsRed compared with neighboring noninfected neurons (Fig. 7). This is consistent with a dominant-negative effect of the truncation construct, which interferes with HCN1 binding to native FLNa, preventing FLNa-induced channel down-regulation. Importantly, the effect of FLNa(23-24) DsRed was specific to HCN1, because neither the immunoreactivity of HCN2 nor that of the glutamate receptor subunit GluA2 was affected by the expression of FLNa(23-24) DsRed (Fig. 7, C and  D).
To test the consequences of HCN1-FLNa interactions on the distribution of native HCN1 neuronal channels, we manipulated FLNa acutely, aiming to avoid potential compensatory . Discrete clusters representing HCN1 puncta co-localized to endosomes are indicated by white arrowheads. C, F, and I, quantitative analysis of HCN1 GFP co-localization with endosomal markers, as compared with HCN1(⌬22) GFP . This analysis revealed a significantly stronger accumulation of HCN1 GFP in various subpopulations of endosomes (Student's t test, ***, p Ͻ 0.001; n ϭ 16 -25 cells per group, amounting 131 cells in total). All experiments were repeated least three separate times per condition. FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9

Filamin A Regulates Neuronal HCN1 Channel Internalization
alterations that might arise upon chronic manipulations. We designed employed decoy peptides ("myr-22") consisting of the 22-amino acid-long FLNa-binding domain of HCN1, attached to a myristoyl group (for improved membrane-permeability) (Fig. 8A). As control, we used a scrambled peptide ("myr-Scr") with a similar structure. Using high magnification, thin slice laser confocal scanning of mature (Ͼ23 DIV) hippocampal neurons (see "Experimental Procedures"), we characterized in detail the subcellular distribution of the channels. Z-stack confocal reconstruction of optical slices revealed accumulation of HCN1 in intracellular clusters (Fig. 8B), in line with previous observations in primary hippocampal neurons (25). A 4-h exposure to the myr-22 peptide altered this expression pattern, resulting in a relatively homogeneous distribution of HCN1 channels (100 nM; Fig. 8B). Incubation of live hippocampal neurons with the scrambled (myr-Scr) peptide did not influence HCN1 channel distribution, suggesting that the effects of myr-22 on the subcellular localization of HCN1 did not stem from nonspecific effects mediated by the myristoyl group (46). The effects of FLNa in controlling HCN channel distribution in neurons were supported by studies establishing a dose -dependence of the consequences of interfering peptide application (Fig. 8C). Diverse doses of the scrambled peptide had no effects. Cross-correlation function analysis of 200 individual neurons (based on four independent experiments) quantified this dose-dependent effect (Fig. 8D). Thus, acute interference of native HCN1-FLNa interaction in neurons supports a dynamic role for FLNa in localization and clustering of native neuronal HCN1 channels.
FLNa Constrains Native I h in Hippocampal Neurons-To examine the functional consequences of HCN1 regulation by FLNa in live neurons, we preincubated Ն21 DIV neurons with myr-22 or myr-Scr peptides, followed by whole-cell voltage clamp measurements. In line with a role for FLNa in suppression of I h , myr-22-treated neurons possessed a significantly larger I h density (Fig. 9, A-D). Resting membrane potential of the neurons (V rest ϭ Ϫ52.3 Ϯ 2.0 and Ϫ52.2 Ϯ 1.3 mV; n ϭ 11, 12 cells for scrambled and myr-22 groups, respectively; Student's t test, p ϭ 0.967) or their input resistance (R in ϭ 271 Ϯ 39 and 313 Ϯ 59 megohms; n ϭ 10, 14 cells for scrambled and myr-22 groups; Student's t test, p ϭ 0.597) did not differ.
This suggested that regulation of I h by FLNa may alter basic neuronal properties primarily at hyperpolarized potentials rather than at rest. To examine this possibility, we used the current clamp configuration to record the voltage response of neurons to a series of hyperpolarizing currents. In neurons treated with either myr-22 or myr-Scr, injection of hyperpolarizing currents resulted in a voltage sag, representing the slow voltage-induced activation of I h in response to hyperpolarizing events and the resulting depolarization of the membrane (45). The sag ratio of myr-22-treated neurons was significantly larger than that of myr-Scr-treated neurons (Fig. 9, E-H), consistent with augmented I h .

DISCUSSION
In this study, we employed genetic, pharmacological, electrophysiological, high resolution imaging, and neurochemical approaches to probe an important role for the actin-binding protein FLNa in neuronal function via interaction with native neuronal HCN1 channels. Employing both HEK293 cells and hippocampal neurons, we found that FLNa selectively internalized HCN1-type channels via a dynamin-dependent mechanism. This FLNa-mediated suppression of HCN1 surface expression was associated with dynamin-dependent sequestration of the channels into endosomal organelles, leading to reduced I h density. Reduced I h density, in turn, attenuated the function of this ion channel in homeostatic membrane responses to transient hyperpolarizing events.
A putative role for FLNa in tethering neuronal ion channels to the actin cytoskeleton has been suggested by the presence of the protein in mature mammalian brain (5,6) and by the established binding of the protein to a number of ion channels expressed in the brain (42,(47)(48)(49)(50). However, although such function of FLNa on ion channel trafficking regulation would be profoundly important for neuronal behavior, experimental evidence supporting this potential role has been limited. Among several tested ion channels, FLNa bound selectively to HCN1 channel protein in a yeast two-hybrid screening via a specific 22-amino acid sequence in the channel CЈ terminus (23). Overexpression of HCN1 in a melanoma cell line devoid of FLNa augmented I h and facilitated channel gating compared with overexpression in a cell line that expresses FLNa (23). More recently, association of FLNa with HCN1 was reported in cochlear hair cells (51), although the functional implications of this interaction have yet to be determined.
Here, we examined the functions of FLNa in mature neurons and found robust effects of FLNa on the control of ion channel abundance and function in the cell membrane. Focusing on HCN1 channels, our studies interrogated FLNa channel interactions from several perspectives. Using confocal and TIRF imaging of fluorophore-fused constructs, a salient feature of HCN1 regulation by FLNa in both HEK293 cells and primary hippocampal neurons was robust clustering of the channel protein. Interestingly, FLNa-mediated clustering of HCN1 channels has been observed around the cell rim of melanoma cell lines following pharmacological suppression of protein synthesis, and it was suggested to reflect clustering of the channels on the cell membrane (23). Although our data do not exclude the presence of HCN1 clusters on the cell surface, both TIRF imaging and immunocytochemical labeling of HCN1 clusters identified the substantial presence of HCN1 clusters in the cell interior, particularly within endosomes. Importantly, both HCN1 clustering and I h down-regulation by FLNa were blocked by preventing dynamin-mediated endocytosis. Together, these findings support a role of FLNa in sequestration of surfaceexpressed channels into endocytic organelles.
How might FLNa promote endocytosis of HCN1? Two broad possibilities come to mind. 1) For guiding HCN1 channels to internalization, the transient interactions between HCN1 and FLNa may couple the channels to molecules that direct them to endocytosis or localize them to endocytosis-prone domains within the membrane. For example, in endothelial cells, transient interaction of FLNa with caveolin-1 was associated with enhanced caveolae internalization and was suggested to regulate their clustering and mobility on the membrane (42). FLNa . The presence of the truncated filamin mutant leads to a specific increase in HCN1 immunoreactivity (n ϭ 29 cells from four independent experiments; one-sample t test, ***, p Ͻ 0.0001), whereas no significant difference was found for HCN2 and GluA2, which are not expected to interact with FLNa (n ϭ 28, 32 from four and three independent experiments; p ϭ 0.132, 0.075). FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9

Filamin A Regulates Neuronal HCN1 Channel Internalization
was also found critical for efficient chemotactic cytokine receptor 2B receptor endocytosis (43) and may route furin receptors from early endosomes to the trans-Golgi network (41). These functions strengthen the notion of FLNa as an active transport controller in the endocytic pathway. 2) For preventing stable anchoring on the cell membrane, FLNa may act by physically preventing or altering the interaction of HCN1 with anchoring molecules that would otherwise stabilize the channels on the cell surface. Supporting this possibility, FLNa competes with talin, a modulator of the adhesion receptor integrin, over the same binding site, which suppresses integrin activation (52,53). In the case of HCN1 channels, this scenario is particularly attractive because, along with FLNa, several other proteins bind HCN channels via the channel CЈ terminus (54), including TRIP8b (tetratricopeptide repeat-containing Rab8b-interacting protein), an important modulator of neuronal HCN1 trafficking (27,38).
Whereas FLNa facilitated internalization of HCN1, it influences additional binding partners (7). For example, interaction of FLNa with K ir 2.1-type potassium channels in vascular FIGURE 8. Redistribution of native HCN1 channels in hippocampal neurons following acute interference with FLNa-HCN1 interaction. A, to acutely interfere with regulation of native HCN1 by FLNa in hippocampal neurons, a myristoylated peptide containing the 22 amino acids that form the FLNa interface in HCN1 (myr-22) was used as decoy. A similar peptide containing a scrambled sequence (myr-Scr) was used as control. B, orthogonal confocal reconstructions of native HCN1 immunoreactive signal in hippocampal neurons. The images represent a single z-slice (XY) and orthogonal views along the YZ and XZ axes. Under control conditions (no treatment), HCN1-positive clusters were visible, often localized to the interior of the soma (indicated by arrowheads). This clustered pattern was greatly reduced following incubation of neurons with 100 nM of the myr-22 peptide (middle panel) but was unaltered following incubation with 100 nM myr-Scr (left). C, dose-dependent effect of myr-22 on HCN1 clustering in neurons. The images depict neurons following incubation with varying concentrations of myr-22 or myr-Scr. Neurons were double-labeled against native HCN1 and the microtubule-associated protein MAP2. The effect of myr-22 on HCN1 distribution was increasingly visible at concentrations higher than 10 nM, whereas no such effect was observed for myr-Scr across concentrations. D, quantitative analysis of HCN1 signal distribution in neurons following incubation of myr-22/myr-Scr peptides at varying concentrations. Note the differences in the cross-correlation function of HCN1 signal at peptide concentrations of 100 nM and 1 M, indicating reduction in HCN1 clustering following myr-22 application. Analysis was based on n ϭ 20 neurons per condition (total of 200 neurons), taken from four independent experiments. All confocal images represent single Ͻ1-m optical z-slices, acquired using a 63ϫ oil objective with a digital zoom of 3.2, at 1024 ϫ 1024 bit. Scale bars, 10 m.
smooth muscles leads to an increased number of functional channels on the cell membrane (55). Similarly, FLNa enhances membrane expression of BK Ca calcium-activated potassium channels (50) and dopamine D2/D3 receptors (48,56), although it is also required for D3 receptor internalization (52). In the case of the -opioid receptor, although FLNa is not essential for receptor expression on the cell membrane, it is critical for agonist-induced internalization (49). Thus, whereas further work is needed to establish these roles in the context of neuronal function, it is conceivable and perhaps even likely that FLNa regulates neuronal channels in addition to HCN1. Distinctive effects of FLNa on different partners may enable the protein to orchestrate ensembles of channels in neuronal membranes, with major effects on neuronal behavior. For example, FLNa may increase surface expression of the potassium channel subunit Kv4.2 (47). Co-regulation of HCN-and Kv4.2-mediated currents in neurons may occur in various contexts and can modulate neuronal firing properties (57,58). Furthermore, like HCN1, Kv4.2 channels are also enriched in dendrites of pyramidal hippocampal neurons, where they act to modulate signal propagation (59) and concurrent regulation of dendritic HCN and Kv4.2 trafficking, and function has been described in the context of activity-dependent alterations in neuronal input (15,16,25,60,61). It is tempting to speculate that by binding both channels, FLNa might be instrumental in this co-regulation.
FLNa regulation of ion channels might also operate in brain pathology. For example, FLNa binds the Alzheimer disease (AD)-related proteins presenilin 1 and presenilin 2 (62), and a mutation in presenilin 1 that is associated with early onset AD resulted in a dramatic increase of hippocampal FLNa expression in a transgenic mouse line (63). Furthermore, increased amyloid-␤ in the brain can lead to enhanced complex formation between FLNa, ␣7 nicotinic acetylcholine receptors, and Toll-like receptor 4, and inhibition of this complex using a high affinity FLNa-binding agent significantly reduced many of the AD-related adverse affects mediated by amyloid-␤ signaling (64). Interestingly, reduction in HCN1 was recently reported both in an AD animal model and in the human brain. This was associated with protein complex formation between HCN1 and amyloid protein precursor and increased amyloid-␤ production (65). As the relationship between neuronal degeneration FIGURE 9. Functional implications of FLNa regulation in hippocampal neurons. To assess functional regulation of native neuronal HCN channels by FLNa, intrinsic I h was measured in hippocampal neurons using the whole-cell patch clamp recording configuration, following incubation with either myr-22 or myr-Scr peptides (100 nM). A, a family of hyperpolarization-activated traces, demonstrating enhanced I h following myr-22 application. B, comparison of single I h traces upon a hyperpolarizing step, extracted from A (indicated by arrows in A). The augmented I h is evident as the increased amplitude between the instantaneous (I ins ) and the steady-state (I ss ) current (indicated by arrows). C, individual values (each obtained from a different cell) representing I h density upon steps to Ϫ130 mV (Mann-Whitney test, *, p ϭ 0.041). D, analysis of I h density across different potentials. E, injection of hyperpolarizing currents resulted in a voltage sag (arrows) due to the activation of I h . F-H, voltage sag analysis revealed enhanced sag response following myr-22 application. F, comparison of single voltage traces extracted from E (indicated by arrows in E). G, sag ratio values obtained from individual cells upon hyperpolarizing current injections of 250 pA (Mann-Whitney test, *, p ϭ 0.034). H, analysis of sag ratios demonstrating elevated voltage sag following myr-22 application across a range of current intensities. I and J, to control for potential nonspecific peptide effects, including those potentially mediated by the myristoyl group, I h was compared between hippocampal neurons that were incubated with the myr-Scr peptide, and naive (untreated) neurons. Both I h density (I) and sag ratio (J) remained unaltered. I, n ϭ 8 -10 cells per data point per group; J, n ϭ 6 -8 cells per data point per group. FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9 and ion channel dysfunction continues to unfold (66), further investigation into the involvement of aberrant FLNa signaling may provide important insights into channelopathies and altered excitability in AD, and potentially in other neurological disorders (21,67).