Clathrin-Independent Trafficking of AMPA Receptors

Membrane trafficking of AMPA receptors (AMPARs) is critical for neuronal function and plasticity. Although rapid forms of AMPAR internalization during long-term depression (LTD) require clathrin and dynamin, the mechanisms governing constitutive AMPAR turnover and internalization of AMPARs during slow homeostatic forms of synaptic plasticity remain unexplored. Here, we show that, in contrast to LTD, constitutive AMPAR internalization and homeostatic AMPAR downscaling in rat neurons do not require dynamin or clathrin function. Instead, constitutive AMPAR trafficking is blocked by a Rac1 inhibitor and is regulated by a dynamic nonstructural pool of F-actin. Our findings reveal a novel role for neuronal clathrin-independent endocytosis controlled by actin dynamics and suggest that the interplay between different modes of receptor endocytosis provides for segregation between distinct modes of neuronal plasticity.


Introduction
AMPA-type glutamate receptors (AMPARs) mediate most fast excitatory neurotransmission in the CNS, and their trafficking underpins synaptic function and dysfunction. AMPARs undergo continuous basal trafficking to and from synapses through constitutive exocytosis/endocytosis, and two distinct forms of synaptic plasticity, long-term depression (LTD) and homeostatic downscaling, rely on AMPAR internalization . Of these three processes, much is known about rapid removal of synaptic AMPARs during LTD, which requires clathrin-mediated endocytosis (CME) and a GTPase dynamin (Anggono and Huganir, 2012). In contrast, endocytic pathways involved in constitutive AMPAR trafficking and homeostatic downscaling remain poorly understood.
Mammalian cells possess a diverse array of noncanonical endocytic pathways that do not require dynamin and/or clathrin (Doherty and McMahon, 2009). In neurons, clathrin-independent endocytosis (CIE) is implicated in several modes of presynaptic vesicle retrieval (Jockusch et al., 2005;Chung et al., 2010) and growth cone dynamics (Bonanomi et al., 2008;Joset et al., 2010). To date, however, postsynaptic CIE has not been characterized.
To test the hypothesis that CIE may be involved in constitutive AMPAR trafficking and non-Hebbian plasticity, we used a number of approaches to perturb dynamin-mediated endocytosis and CME in neurons. We find that, in contrast to activity-dependent AMPAR internalization, neither constitutive AMPAR trafficking nor homeostatic downscaling require dynamin or clathrin function, although they do use the same intracellular trafficking compartments after internalization. Instead, constitutive AMPAR cycling is regulated by Rac1 and a nonstructural pool of F-actin that controls the rate of membrane turnover in the neuron. This novel mechanism highlights the diversity of membrane trafficking operating in the neuron and suggests that the distinct functional outcomes of different modes of AMPAR trafficking may be defined by differential endocytic sorting.
Cell culture and transfection. Dissociated hippocampal and cortical cultures were prepared as described previously (Rocca et al., 2008). Transfection of neuronal cultures was performed at 10 -13 d in vitro using Lipofectamine2000 (Invitrogen) according to the instructions of the manufacturer with minor modifications.
Biotinylation of cell surface protein. Live cortical neurons (15-21 d in vitro) were surface biotinylated on ice using the membraneimpermeant Sulfo-NHS-SS-Biotin according to the previously published protocol (Martin and Henley, 2004). Bands were quantified using NIH ImageJ software and normalized to the total receptor band intensity and internalization in the control sample.
Confocal imaging. Internalization assays for transferrin and anti-AMPAR subunit antibodies were performed as described previously (Rocca et al., 2008). Each experiment was performed using the same batch of coverslips and neurons cultured under identical conditions. Briefly, hippocampal neurons were live labeled for 20 min at 12-15°C, washed, and allowed to internalize for 0 -30 min before fixation. Surface label was revealed by overnight staining with an appropriate Cy3-conjugated secondary antibody, whereas internalized label was stained by an appropriate Cy5-conjugated secondary antibody after permeabilization. Images were acquired on LSM510 and LSM710 confocal microscopes (Zeiss) and analyzed using NIH ImageJ software. To minimize bias, quantification of internalization was performed by collecting random fields from various regions of the coverslip. The following formula for quantification of internalization was used: where I Cy5 and I Cy3 refer to intensities in Cy5 and Cy3 channels, respectively. Correction coefficient k was calculated as follows: where I Cy5t0 and I Cy3t0 refer to intensities in Cy5 and Cy3 at the 0 time point. To correct for the inherent variability in intensity of surface labeling between different cultures, resulting intensities were normalized so that the average intensity of the control sample was set to 1.
The fraction of internalized AMPARs (iAMPARs) was quantified as loss of surface staining (sAMPARs) relative to the 0 time control. For colocalization analysis, surface remaining AMPARs were blocked with the 4 Figure 1. CIE of AMPARs. A, Neurons were incubated with Tf-555 and anti-GluA2 antibody for the indicated times, fixed, permeabilized, and stained for iGluA2. Arrows denote overlapping puncta visible at 20 min, but not 5 min, of internalization. n ϭ 3. B, C, Neurons were transfected with CHC shRNA or an empty GFP vector and live labeled with Tf-555 or anti-GluA1 for 30 min 4 d after transfection. Control, Empty GFP vector. CHC shRNA blocks uptake of Tf-555 but not iGluA1. Arrows denote puncta of internalized label. n ϭ 4. D, As above but with AP180Cmyc construct. Control, Untransfected cells. n ϭ 3. E, Treatment with 20 M Pitstop 2 blocks Tf-555 but not iGluA1. n ϭ 4. Control, 0 M Pitstop2 (DMSO). Scale bars, 10 m. excess of unconjugated secondary antibody overnight. The amount of colocalization was quantified using an ImageJ/Fiji Colocalization plugin (http://fiji.sc/Coloc_2).
Slice preparation. Acute hippocampal slices were prepared from anesthetized P13-P15 male Wistar rats. Brains were dissected in ice-cold artificial CSF (aCSF) equilibrated with 95% CO 2 and 5% O 2 . Horizontal hippocampal slices (400 -500 m thick) were cut with a VT1200 vibratome (Leica) and incubated in aCSF at 36°C for 30 min and then at room temperature before use. The connections between CA3 and CA1 were cut before transfer to the submerged recording chamber. The procedures were performed in accordance with the United Kingdom Home Office and University of Bristol guidelines.
Electrophysiology. Whole-cell recordings were made from CA1 pyramidal cells visualized with infrared differential interference contrast optics (Olympus BX-51 microscope) in a recording chamber perfused with aCSF (35°C) containing 50 M picrotoxin. Patch electrodes (4 -5 M⍀) were filled with intracellular solution (in mM: 117 CsMeSO 3 , 8 NaCl, 10 HEPES, 5 QX-314, 4 MgATP, 0.3 NaGTP, 0.5 EGTA, and 0.1 spermine), set to pH 7.4, 280 -285 mOsm. Cells were voltage clamped at Ϫ70 mV, and the recorded currents were filtered at 4 kHz and digitized at 10 kHz using a CED Micro 1401 mk II board and Signal 2 software (Cambridge Electronics Design). Cells with series resistance Ͼ30 M⍀ or showing Ͼ20% change were discarded from subsequent analysis. Baseline synaptic responses were evoked at 0.2 Hz with a tungsten bipolar electrode (100 k⍀, 119 m tip spacing; FHC) placed in the stratum radiatum. Consecutive EPSCs were averaged online every minute, and their amplitudes were normalized offline to the first minute of recording.
For synaptic plasticity experiments, EPSCs were recorded from two independent pathways. LTD was induced at 15-20 min after break-in by a pairing protocol (1 Hz ϫ 300 stimuli at Ϫ40 mV) delivered to the test pathway, whereas the control pathway was paused. One-minute average EPSC amplitudes (see above) were normalized to the average value over 5 min before LTD induction.
Statistical analysis. Every experiment was performed at least in triplicate (see figure legends). Data were normalized to the control sample, and statistical analysis was performed using GraphPad Prism software (GraphPad Software). The D'Agostino-Pearson omnibus test was used to test for normality of distribution, and, depending on results, a twotailed t test or Mann-Whitney nonparametric test was used. For ANOVA, one-way ANOVA with Dunn's post hoc test was used. Levels of statistical significance were as follows: *p Ͻ 0.05, **p Ͻ 0.01, ***p Ͻ 0.001. Error bars represent Ϯ SEM.

Results
Endogenous surface AMPARs in live cultured rat hippocampal neurons were labeled with antibodies recognizing extracellular epitopes of GluA1 and GluA2 subunits ( Fig. 1 A, B), and their internalization was monitored by immunofluorescence. In agreement with previous reports (Lin et al., 2000;Lu et al., 2007), we observed constitutive internalization of both GluA1 and GluA2, with 31.6 Ϯ 5.5 and 36.5 Ϯ 4.9% of surface receptors, respectively, internalized after 20 min. Although little overlap was observed between iAMPARs and fluorescent CME cargo transferrin Tf-Alexa Fluor 555 (Tf-555) at 5 min, by 15 min, there was significant colocalization between the two proteins (Fig. 1A). This observation raised the possibility that AMPARs and transferrin may internalize via distinct endocytic mechanisms but later converge within the same endosomal compartments. Mutant Dyn2 reduces Tf-555 but not iGluA1. Control, Dyn2wt-expressing cells. n ϭ 4. B, Overexpression of dynamin 1 and 3 mutants has no effect on iGluA1. Control, Dyn1wt and Dyn3wt-expressing cells, respectively. n ϭ 3 each. C, Neurons were pretreated with DMSO (veh) or 75 M dynasore (ds) for 10 min and allowed to endocytose Tf-555, anti-GluA1, or anti-GluA2 for 30 min in the presence of dynasore. Ds blocks Tf-555 but not iGluA1 or iGluA2. Control, Vehicle-treated samples. n ϭ 4. D, Surface biotinylation-based assay for internalization in cortical neurons. Total, Total protein levels in neuronal lysate; Eluate, protein levels eluted from streptavidin-conjugated beads; No strip, nonstripped control; 0Ј, 0 time (no endocytosis) control. Ds blocks internalization of TfR but not of GluA2. Control, Vehicle-treated samples. n ϭ 3. E, Neurons expressing Dyn2wt-GFP or Dyn2K44A-GFP were live labeled with anti-GluA1, fixed, and stained for iGluA1 and either EEA1 or TfR. iGluA1 localizes to EEA1-and TfR-positive puncta (arrows) in both cells expressing wild-type and mutant dynamin. F, Neurons expressing a dominant-negative Rab5Q79L-GFP construct accumulate AMPARs in GFP-positive enlarged endosomes (arrows). Scale bars, 20 m.
To determine whether AMPAR internalization requires clathrin function, we knocked down the heavy chain of rat clathrin (CHC shRNA;He et al., 2008). Expression of CHC shRNA blocked internalization of transferrin but not AMPARs (Fig.  1 B, C) Fig. 1E). Together, these data indicate that constitutive internalization of AMPARs does not require CME. CME and several CIE pathways require the activity of the large GTPase dynamin to mediate scission of the budding endocytic vesicle from the plasma membrane (Doherty and McMahon, 2009). Therefore, we next tested whether dynamin activity was required for constitutive internalization of AMPARs. Although transferrin internalization was markedly reduced by a dominantnegative K44A dynamin 2 mutant (Dyn2-K44A; Damke et al., 1994), iAMPARs ( Fig. 2A) and sAMPARs (Fig. 3D) were unaffected. Similar results were obtained for neuron-specific isoforms dynamin 1 and dynamin 3 (Fig. 2B). Consistent with these results, the dynamin GTPase inhibitor dynasore (Macia et al., 2006) blocked transferrin uptake but had no effect on iAMPARs (Fig.   2C), as did an alternative inhibitor of dynamin function, Dynole34-2 (Hill et al., 2009; data not shown). We also monitored the dynamin dependence of internalization using surface biotinylation in cultured cortical neurons. As expected, internalization of the TfR was inhibited by dynasore, whereas GluA2 was unaffected (Fig. 2D), confirming that dynamin is not required for constitutive AMPAR trafficking.
Depending on the context, cargoes endocytosed through CIE and CME may either follow separate intracellular trajectories or converge in endosomal compartments such as early and recycling endosomes (Hansen and Nichols, 2009). Given the time-dependent overlap between internalized transferrin and AMPARs (Fig. 1A), we investigated the post-internalization itinerary of AMPARs endocytosed via CIE. AMPARs internalized through CME during LTD are rapidly enriched in EEA1-positive early and TfR-positive recycling endosomes (Ehlers, 2000), and this was also the case for constitutively iAMPARs in both wildtype dynamin 2 (Dyn2wt) and Dyn2-K44A-expressing cells (Fig.  2E). Additionally, iAMPARs colocalized with a GTPase-defective mutant of a small GTPase Rab5 (Fig. 2F ), suggesting that constitutively iAMPARs traffic through Rab5-positive endosomes. Thus, AMPARs internalized via both CME and CIE traffic along the classical endocytic route, suggesting that convergence of these pathways at the level of early/recycling endosomes.
To assess the functional role of CIE in basal trafficking of synaptic AMPARs, we measured excitatory synaptic transmission in acute hippocampal slices. As a positive control, we used the DIP that contains the proline-rich domain from dynamin 1, which rapidly increases AMPAR currents (Lüscher et al., 1999) through blockade of AMPAR internalization (Fig. 3C). Although DIP effectively increased EPSC amplitude, dynasore did not alter EPSC amplitude (Fig. 3A). The discrepancy between the effects of DIP and dynasore suggests that the effect of DIP on basal AMPAR levels does not reflect the outcome of the functional inhibition of dynamin GTPase activity, given that the inhibitory effect of DIP is realized through its perturbation of the Src homology 3-domain interactions of an endocytic adaptor protein amphiphysin (Owen et al., 1998) rather than abrogation of dynamin GTPase activity, consistent with dynamin not being required for constitutive trafficking of AMPARs. In agreement with previous studies, dynasore abolished NMDAR-LTD (Man et al., 2000;Collingridge et al., 2010;Fig. 3B). In contrast, chronic enhancement of excitatory network activity using increased K ϩ , and the GABA A receptor blocker bicuculline reduced sAMPARs in cells with or without dynamin (Fig. 3D). Together, these data indicate that inhibition of the GTPase dynamin function abolishes LTD but does not affect basal levels of synaptic AMPARs or their homeostatic downscaling.
To elucidate the molecular mechanism of CIE-AMPAR, we focused as a candidate on a small GTPase Rac1 that clusters AMPARs in developing dendrites (Wiens et al., 2005) and regulates dynamin-independent macropinocytosis at the growth cone (Joset et al., 2010). Because neurons overexpressing Rac1 mutants exhibited profound morphological aberrations, we instead chose to pharmacologically inhibit Rac1 activation by the drug NSC23766 (Gao et al., 2004), which did not noticeably affect the cell morphology. NSC23766 increased sAMPARs and decreased iAMPARs (Fig. 4A), indicating that Rac1 activation was required for constitutive AMPAR trafficking.
Given that Rac1 regulates membrane trafficking processes through control of actin polymerization, we tested the role of actin in AMPAR internalization. As reported previously (Zhou et al., 2001), treatment with 5 M Latrunculin A (LatA) decreased F-actin and sAMPARs but increased iAMPARs (Fig. 4B). Stabilization of F-actin by Jasplakinolide had the opposite effect (Fig.  4C), suggesting a bidirectional regulation of constitutive AMPAR trafficking by actin dynamics.
We next asked whether the effect of actin is mediated directly on AMPAR trafficking or indirectly via its structural role at the synapse. The reported existence of distinct dendritic F-actin pools with dif- ferent dynamics and sensitivities to LatA (Gu et al., 2010) prompted us to attempt to pharmacologically uncouple the structural and nonstructural functions of actin. LatA at 0.5 M had no effect on the overall levels of dendritic F-actin (Fig. 4B), suggesting that the structural pool of F-actin remained intact, whereas lack of effect on synaptic AMPARs (data not shown) indicated that limited depolymerization of F-actin did not affect AMPAR distribution, ruling out the homeostatic effect of enhanced neurotransmitter release. Strikingly, 0.5 M LatA increased iGluA2 to the same extent as 5 M (Fig.  4B) and also increased the uptake of a membrane-binding dye  styryl) pyridinium dibromide] but not of transferrin (Fig. 4D), suggesting that the observed increase of AMPAR cycling is likely attributable to the increase in the overall dendritic CIE. Furthermore, 0.5 M LatA abolished the reduction of sAMPARs induced by bicuculline and KCl (Fig. 4E), indicating that this nonstructural pool of actin was required for homeostatic downscaling.

Discussion
To date, the majority of AMPAR endocytosis studies have been focused on LTD (Anggono and Huganir, 2012), and therefore, it has been commonly assumed that AMPARs traffic exclusively via CME. Here, we use a wide variety of acute and chronic loss-offunction approaches to show that basal internalization and homeostatic downscaling of AMPARs proceed normally in the absence of clathrin or dynamin function. Although individual pharmacological and genetic tools may in some cases exhibit offtarget effects (Dutta et al., 2012;Park et al., 2013;Willox et al., 2014) and compensatory upregulation (Damke et al., 1995), CIE of AMPARs is observed regardless of the methodology used. Although the full molecular details of the endocytic pathway remains to be established, our data reveal a surprising fundamental difference between the fast (LTD) and slow (constitutive cycling, homeostatic downscaling) modalities of neuronal receptor endocytosis. Therefore, it is tempting to speculate that differential commitment of the same cargo to distinct endocytic itineraries (Di Guglielmo et al., 2003) may provide a parsimonious mechanistic framework for explaining the subsequent fates of AMPAR molecules internalized during constitutive cycling, homeostatic downscaling, and LTD .
Loss of synaptic AMPARs and decreased synaptic transmission after actin depolymerization has been generally interpreted as an indirect consequence of the structural disruption of the spine and synapse (Zhou et al., 2001;Hotulainen and Hoogenraad, 2010). We show that a dynamic pool of F-actin limits endocytic AMPAR trafficking and total membrane uptake but has no obvious role in the maintenance of the structural integrity of the synapse. This trafficking pool of F-actin has distinct properties from the main dendritic F-actin, allowing for functional uncoupling between structural plasticity and membrane trafficking. These findings indicate that the structural role of actin at the synapse is realized independently from its role in the organization of local membrane trafficking and are in line with recent reports of multiple pools of actin coexisting at the synapse (Gu et al., 2010). It remains to be determined which of the actin regulatory proteins resident at the synapse may define the distinct identities of these pools.
Dynamin-independent internalization of both major AMPAR subunits and the distinct cytoskeletal regulation of bulk membrane uptake from CME suggest that this pathway may potentially function as a nonselective recycling route for any cargo that fails to be sequestered into a adaptor-driven endocytic pathway, such as CME. Indeed, although our study has predominantly focused on AMPAR trafficking, evidence is accumulating that CIE may carry other cargo receptors in other synaptic contexts (Cinar and Barnes, 2001;Kumari et al., 2008;Borroni and Barrantes, 2011). This emerging diversity of synaptic receptor trafficking challenges the extant perception of CME as the predominant endocytic mechanism in neurons and raises new questions about the mechanisms and regulatory pathways that allow neurons to regulate their function.