INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Fast excitatory synaptic transmission in the CNS is mediated by -amino-3-hydroxy-5-methyl-4-isoxazlepropionate receptors (AMPARs) and N-methyl-D-aspartate type glutamate receptors (NMDARs). The AMPARs are assembled of four to five homologous subunits (Ferrer-Montiel and Montal 1996; Mano and Teichberg 1998; Rosenmund et al. 1998). There are four types of AMPA receptor subunits termed GluR1-4 (or GluRA-D). Different combinations of these subunits determine the functional properties of AMPA receptors (Dingledine et al. 1999). In heteromeric receptors, the GluR2 subunit is dominant in determining both Ca²⁺ permeability and rectification properties. The unique properties of GluR2 are attributed to a single amino acid referred to as the Q/R site in the pore-lining M2 segment (Burnashev et al. 1992; Hume et al. 1991; Mishina et al. 1991). This site is subjected to nuclear RNA editing. In the edited GluR2, a positively charged arginine (R) is present, whereas in the unedited form, a neutral glutamine (Q) is present (Seeburg 1996). Although the subunit composition of AMPA receptors varies, they must contain at least one edited GluR2 subunit to be Ca²⁺ impermeable (Jonas and Burnashev 1995). Results from combined whole cell patch-clamp recording and reverse-transcription (RT) PCR amplification experiments indicate that Ca²⁺ permeability of AMPARs are inversely correlated with the relative abundance of GluR2 mRNA (Bochet et al. 1994; Geiger et al. 1995).

The edited GluR2 subunit makes up more than 99% of GluR2 subunits in rat brain at all developmental stages (for reviews, see Jonas and Burnashev 1995; Seeburg 1996), indicating that the majority of AMPA receptors in the adult CNS are Ca²⁺ impermeable. However, there are small populations of cells lacking the GluR2 subunit and expressing Ca²⁺ permeable AMPARs, for example, Bergmann glia cells in cerebellum and most interneurons of the hippocampus and cerebral cortex (Jonas and Burnashev 1995; Petralia et al. 1997).

Given the importance of the AMPAR in processes ranging from plasticity to cell death (Dingledine et al. 1999), approaches that directly evaluate the function of AMPARs at single synapses are required. Previously NMDA receptor function has been assayed at single synapses using Ca²⁺ imaging techniques (Kovalchuk et al. 2000; Mainen et al. 1999; Malinow et al. 1994; Muller and Connor 1991; Murthy et al. 2000; Petrozzino et al. 1995; Segal 1995; Yuste and Denk 1995). In this study, we have used cortical cultures from GluR2 null animals (Jia et al. 1996), which have Ca²⁺-permeable AMPARs on all neurons, combined with Ca²⁺ imaging techniques to assess the feasibility of imaging AMPAR function. Although, the morphology of neurons in culture may change, they provide a tractable system in which rapid pharmacological manipulations can be made. For example by applying selective antagonists we can isolate MSCTs mediated by either AMPA or NMDA receptors in cultures prepared from the GluR2 null animals. AMPAR-mediated MSCTs were attributed to direct entry of Ca²⁺ through the receptor. By alternating between conditions specific for NMDA and AMPA receptor-mediated MSCTs, we provide data suggesting that the degree of NMDA and AMPA receptor activation is proportionally scaled at individual synapses in a manner independent of the frequency of miniature release.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Animals and cell culture

Cortical neuron and glial cell cultures were derived from E19 to 20 GluR2 knockouts or wild-type (control) littermate mice. The GluR2 knockout animals originated from a colony of CD1 × 129 crosses generated at the Samuel Lunenfeld Research Institute, Mount Sinai Hospital, Toronto (Jia et al. 1996). The animals used for these experiments were from two to seven generations held at University of British Columbia animal facility. Our animal protocol was approved by the Committee on Animal Care University of British Columbia (animal-care certificate A99-0042) and followed Canadian Council on Animal Care guidelines.

Culture methods and conditions were modified from Murphy and Baraban (1990) and Mackenzie et al. (1996). Briefly, cortices and hippocampi were dissected from each fetus and placed in separate dishes containing ice-cold phosphate-buffered saline. A small tissue sample was taken from each carcass for extraction of genomic DNA, and PCR was used to verify the genotype of each fetus. Our experiments indicated that the cultures were best if they were produced from embryonic animals. In this case, it was not possible to genotype each embryo before culturing it. Therefore each embryo was dissected and processed separately. To speed the process up (typically 5-6 fetuses were done) and increase the yield from each animal, we combined the cortex and hippocampus neurons. The cortices and hippocampi of each fetus were placed (dissociated together) in conical tubes containing 10 ml papain solution (20 units/ml) in Earle's balanced salt solution (EBSS) for 15 min at 37°C to dissociate the cells. The dissociated cells (mixed hippocampal and cortical) were then pelleted for 3-4 min and the supernatant discarded. The pellets from each tube were transferred to fresh conical tubes containing 5 ml DNAase (0.005%) solution containing 1 mg/ml bovine albumin and trypsin inhibitor type II and triturated several times with fire-polished glass pipettes to further separate the cells. The DNAase digestion was stopped by slowly dripping 3-5 ml 0.1% wt/vol bovine serum albumin and 0.1% wt/vol trypsin inhibitor type II in EBSS down the side of the conical tube to create a DNAase- and papain-free layer at the bottom of the tube. The cells were then pelleted by low-speed centrifugation for 3-4 min. The supernatants were discarded and the cells from each conical tube were re-suspended in 8 ml plating medium and seeded (2 ml per dish; density = about 1.2 million cells/ml) into poly-D-lysine-coated 35-mm glass-bottomed dishes (MatTek, Ashland, MA). The plating medium was composed of minimal essential medium and 10% fetal calf serum, and 10% heat inactivated horse serum. Some cells were plated on poly-D-lysine-coated 35-mm aclar 33C (a nonfluorescent plastic substrate 0.127 mm thick)-bottomed dishes. Aclar was obtained from Proplastics (Linden, NJ). Cultures were kept in vitro for at least 10 days before use in imaging experiments.

Genotype PCR assay

The PCR genotyping procedure was provided by Dr. Franco A. Taverna. To identify wild-type GluR2 allele, we used the following primers FTR2A (CAGCAGATTTAGCCCCTACG) and FTR2B (CCTCACAAACACACCATTTCC) to generate a 623-bp band. To identify the GluR2 null neo insertion, we used the FTR2B and FTneoA (GGATGATCTGGACGAAGAGC) primers to generate a 1,022-bp band. For each 20 µl reaction, we used 1 µg genomic DNA (in Tris-EDTA buffer), 2 µl 10 × buffer, 0.4 µl dNTPs (10 mM stock, 200 µM final), 200 ng primer (FTR2A, FTR2B or FTneoA), and 0.2 µl Taq DNA polymerase (1 U, Boehringer Mannheim GmbH, Germany). The balance of the solution contained milliQ H₂O (added first to ensure that the DNA did not precipitate). A drop of DNAase-free mineral oil was placed on top of the reaction solution. The PCR cycles are 94°C for 40 s, 60°C for 30 s, 72°C for 1 min, 30 cycles, and a final 72°C for 7 min.

Experimental procedures

In all experiments, cultures were continuously perfused in a bathing medium containing (in mM) 137 NaCl, 5.0 KCl, 2.5 CaCl₂, 1 MgSO₄, 0.34 Na₂HPO₄(7H₂O), 10 NaHEPES buffer, 1 NaHCO₃, 0.0003 tetrodotoxin, 0.1 picrotoxin, and 22 glucose (pH 7.4 and approximately 315 mosM). MgSO₄ and CaCl₂ concentrations were altered as indicated. A double-barreled theta application tube was placed near the recording site to rapidly change solutions during imaging. One barrel was filled with a AMPAR-isolating condition solution that contained the bathing solution with the following changes: 5 mM CaCl₂, 0 mM MgSO₄, 5 µM glycine, 20-100 µM cyclothiazide, and 100 µM of the NMDA receptor blocker D,L-2-amino-5-phosphonovaleric acid (D,L-APV). The other barrel was filled with an NMDAR-isolating solution containing the bathing solution with the following changes: 5 mM CaCl₂, 0 mM MgSO₄, 5 µM glycine, 20-100 µM cyclothiazide, and 3 µM of the AMPA receptor blocker CNQX. The cells were alternately perfused with the two isolating solutions using solenoid-controlled valves (solutions changed at approximately 1-min intervals). In some experiments, the AMPA solution contained an additional 100-200 µM CdCl₂ to block voltage-gated Ca²⁺ channels. The desired solution was perfused for about 2-3 s before simultaneous imaging, and voltage-clamp recording were started to provide adequate perfusion. The application of antagonists such as CNQX or APV for just 2-3 s before imaging helps to reduce potential nonspecific actions (such as a reduction in resting calcium level driven by ambient glutamate).

The perfusion was stopped immediately after a trial of imaging and electrical recording was complete (10 s). Before each experiment, the theta tube was filled with bath solution containing phenol red to monitor the perfusion area and the extent of solution exchange. Once an adequate perfusion and solution exchange position were determined the theta tube was left in the same position throughout the experiment. All experiments were conducted at room temperature (~23°C).

Whole cell voltage- and current-clamp experiments (Hamill et al. 1981) were conducted using an Axon Instruments Axopatch 200B amplifier. Electrodes were pulled from 1.5-mm glass pipettes (Warner) and had a tip resistance of 6-9 M when filled with intracellular solution containing (in mM) 0.5 fluo-3 K⁺ salt, 0.7 mag-fura-2, 122 K⁺MeSO₄, 20 NaCl, 5 Mg-ATP, 0.3 GTP, and 10 HEPES (pH = 7.2). In some cases, K⁺ was substituted with Cs⁺ for better clamp control. A calculated liquid junction potential of 12 mV was not corrected for (Neher 1995); therefore all membrane potentials and holding potentials are expected to be more negative than reported.

Imaging of neuronal Ca²⁺ transients was performed with wide field microscopy using an Axiovert TV inverted microscope (Zeiss, Germany) equipped with an intensified CCD camera (Stanford Photonics, Palo Alto, CA) as previously described (Mackenzie et al. 1996; Murphy et al. 1995). We used an ×100 oil 1.3 NA objective (Zeiss, Germany) for optical recordings. The camera acquired data at 30 frames/s (pixel size = 0.2 µm) and the images were captured to a PC using a frame grabber (EPIX, Buffalo Grove, IL). Mag-fura-2, a low-affinity Ca²⁺ indicator with high fluorescence at basal [Ca²⁺]_i at 380 nm excitation (Raju et al. 1989), was included in the pipette solution (0.7 mM) to resolve the fine processes under resting conditions. For each optical recording trial, we collected 10 s of Fluo-3 images (300 frames; at 490-nm excitation) and then switched to 380 nm excitation to collect 1 averaged (1 s) Mag-fura-2 image for normalization (see following text). The excitation light was delivered to the microscope using a flexible fiber optic cable coupled to a Stanford Photonics DX-1000 optical switch/HBO 100 W arc lamp. On the emission side we used a Fura/Fluo-3 dichroic mirror (Chromo Technologies No. 74000) combined with a 540/40-nm emission filter. Under these conditions, we found that the F₄₉₀/F₃₈₀ signal was linear over the range of ratio values associated with MSCTs (calibration procedure using voltage-gated Ca²⁺ current) (Umemiya et al. 2001). We also observed a strong positive correlation (by linear regression) between the integral of the NMDAR component of the miniature excitatory postsynaptic current (mEPSC) and the peak MSCT amplitude for coincident MSCTs/mEPSCs that were mapped to single sites (Umemiya et al. 1999). These data indicated that measurement of the MSCT peak amplitude does provide a measure of mEPSC amplitude despite the fact that the imaging technique has a considerably lower sampling rate. Thus although the peak [Ca²⁺]i reached would also reflect buffering and other factors, it is nonetheless an indicator of the degree of receptor activation. Relatively low excitation light levels were used, and we did not observe detectable bleaching as determined by no significant decline in basal fluorescence during trials without miniature events (see Fig. 2C). To permit adequate filling of the cells, we usually delayed recording by at least 5 min after seal rupture.

Fluo-3 and mag-fura-2 were purchased from Teflabs (Austin, TX) or Molecular Probes (Eugene, OR). D,L-APV were purchased from Precision Biochemicals (Vancouver, BC). Cyclothiazide was purchased from Sigma RBI (Oakville, Ontario). Tetrodotoxin and other chemicals were purchased from Sigma Chemical (St. Louis, MO).

Data analysis

To identify dendritic regions showing MSCTs, we constructed difference images by first averaging fluorescence over 1-s periods (30 images) to obtain 10 averaged images (total sampling epoch is approximately 10 s, 300 images). Images of fluorescence changes (difference images) were then produced by subtracting consecutive averaged images (image pairs) and determining their absolute value. The difference images were then averaged and a single image was created that reflected sites that exhibited greatest changes in fluorescence over the 10-s sampling epoch. These potential sites of MSCTs were further evaluated using plots (pixel value vs. time) from approximately 2.0 µm² areas of interests (7 × 7 pixels; see Fig. 2). The change in Fluo-3 fluorescence intensity F of each area of interest was divided by the Mag-fura-2 fluorescence (F₃₈₀) to adjust for differences in dye filling and focus along the dendritic tree (F₄₉₀/F₃₈₀).

To improve the signal-to-noise ratio of the images and to reduce the file sizes, we averaged two sequential frames of raw data (an averaged image was 66 ms). The sites of MSCT origin were defined as previously described (Murphy et al. 1995) by monitoring the rising phase of the Ca²⁺ transient and selecting the dendritic region with the earliest rise. The initiation time of the Ca²⁺ transient was defined as the first point of three consecutive measurements (66-ms interval) that was more than 2 SDs above the mean baseline fluorescence intensity. Because three consecutive points must be greater than 2 SD above the noise we estimate that spurious MSCTs would be detected with a frequency of less than once every 50 trials. The mean baseline fluorescence intensity was calculated by averaging 20 consecutive points before the initiation of a MSCT or a randomly selected control period. The amplitude of MSCT was measured by selecting the maximum fluorescence value within five consecutive points following the MSCT initiation point. Usually the fluorescence peaked within the first three data points. In some cases (in general less than 5%) slow transient elevations in [Ca²⁺]_i were observed (more than 500-ms rise time); these types of events were not considered MSCTs. In trials that lacked MSCTs, baseline noise was calculated by selecting random sequences of frames and determining their peak fluorescence change as described above for MSCT responses. A random number generator was used to prevent the possibility that regions of baseline might be arbitrarily selected that contained more or less noise than others (see Fig. 2C for an example of baseline variation in CNQX). To estimate the baseline fluorescence deviation (noise), we took the maximum fluorescence value occurring within five consecutive points of a randomly selected number. Because the absolute changes in fluorescence levels due to MSCTs could be up to 50%, significant increases in shot noise (photon counting noise) could occur. In this case, baseline variance estimates would not necessarily be a measure of the total MSCT variance due to the higher light levels reached. Because variance due to shot noise is proportional to the mean fluorescence, we scaled the estimated baseline variance (to account for higher shot noise associated with MSCTs) by multiplying it by the fold-change in raw fluorescence attributed to MSCTs. In scaling the entire baseline variance, we have assumed that the dominant source of noise in this system is "shot noise" as previously described by us and consistent with other data (Mackenzie et al. 1996). We then subtracted this estimated baseline variance (corrected for shot noise) from the measured MSCT response variance for each site examined. This manipulation gave us an estimate of MSCT variance free from variance due to shot and dark noise (Sabatini and Svoboda 2000). In general we found that the total fold change of Fluo-3 fluorescence was relatively small and that shot noise correction only reduced coefficient of variation (CV, SD/mean) estimates by a small amount (0.35-0.34 for AMPA responses and 0.30-0.28 for NMDA responses).

Data analysis was performed using custom routines written the IDL (Research System, Boulder, CO) programming language on a Pentium-processor-based computer. Statistical analysis was done by Prism V.3.0 (GraphPad Software, San Diego, CA) and Origin V.3.5 (Microcol Software, Northamptom, MA). Results are presented as the mean ± the SD unless indicated otherwise.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Calcium imaging was performed on cultured cortical neurons under conditions that isolate miniature synaptic activity. To isolate the postsynaptic Ca²⁺ response attributed to the AMPAR-mediated mEPSC, we added picrotoxin and APV to block both GABA and NMDA-type receptors. To slow the decay phase of the AMPAR-mediated mEPSC and to increase charge transfer to facilitate Ca²⁺ transient detection, we used cyclothiazide (20-100 µM) to remove desensitization (Bertolino et al. 1993; Partin et al. 1993). Although cyclothiazide and its analogues may effect the gating of AMPA receptors and the affinity for agonist (Arai et al. 1996), most of the properties of AMPAR mEPSCs appear to be preserved after cyclothiazide including considerable amplitude variability (Atassi and Glavinovic 1999). Cyclothiazide sensitivity can also vary with AMPAR subunit combination (Partin et al. 1994, 1995). However, most synapses on hippocampal neurons appear to be sensitive to cyclothiazide (or its analogs) because the bulk of the mEPSC decay times are lengthened rather than a subpopulation being effected (Atassi and Glavinovic 1999; Vyklicky et al. 1991). Our results suggest that after cyclothiazide treatment mEPSC decay times can be described by a single normal distribution, arguing against a significant proportion of less sensitive synapses (Kolmogorov-Smirnov test; P > 0.05, n = 4 neurons). Under the conditions we used mEPSCs were reversibly suppressed by adding the antagonist CNQX to the bathing media indicating that they were attributed to AMPARs (Fig. 1, A and B). In low-density cultures that had reduced frequencies of mEPSCs, it was possible to correlate the local appearance of MSCTs with AMPAR mEPSCs recorded from the cell soma as we have previously done for NMDAR mEPSCs and MSCTs (Murphy et al. 1995; Umemiya et al. 1999, 2001). In the example shown, a mEPSC (Fig. 1, C-E) with a 19-ms single exponential time course of decay occurs within one frame (33 ms) of the MSCT. The prolonged decay time course of the mEPSC was associated with the use of cyclothiazide to remove desensitization (Atassi and Glavinovic 1999). In the low-density cultures, a significant positive correlation was observed between the integrated mEPSC and the MSCT peak amplitude (R = 0.67, P < 0.01, n = 14 coincident MSCTs/mEPSCs at n = 9 synapses), indicating that Ca²⁺ imaging provides an estimate of the AMPA synaptic current amplitude as observed previously for NMDAR mediated mEPSCs and MSCTs (Murphy et al. 1995; Umemiya et al. 1999, 2001). Although the low-density cultures permitted the correlation of mEPSCs and MSCTs, it was difficult to collect a large number of MSCTs from a variety of different locations due to the lower frequency of mEPSCs. We thus have performed all additional experiments in high-density cultures allowing us to maximize MSCT frequency.

View larger version (36K): [in this window] [in a new window]   Fig. 1. AMPA receptor (AMPAR)-mediated miniature excitatory postsynaptic currents (mEPSCs) and miniature synaptic Ca2+ transients (MSCTs) in GluR2 knock out neurons. A: whole cell voltage-clamp recording showing consecutive sweeps of AMPAR-mediated mEPSCs at 65 mV holding potential. The AMPAR-mediated mEPSCs were reversibly blocked by 6-cyano-7-nitroquinoxalene-2,3-dione (CNQX, 3 µM) applied at 1-min intervals in every other trial (100 µM cyclothiazide and 100 µM D,L-APV are present). B: expanded time and current scale showing the time course of the mEPSCs shown in A. C: image indicating a short segment of dendrite where MSCT imaging was performed in the presence of 100 µM D,L-2-amino-5-phosphonovaleric acid (D,L-APV) and 20 µM cyclothiazide to isolate and enhance AMPAR MSCTs in a GluR2 knockout neuron (different neuron than A and B, from lower density culture). D: record of fluo-3 fluorescence normalized to Mag-fura-2 from the indicated site (see arrow) with corresponding whole cell current record (made from the cell soma) indicates that the MSCT shown (top trace) is preceded by a mEPSC that decays with a single exponential time course (19-ms decay time constant). The scale bars for this figure appear in E (time calibration = 100 ms). Plots from multiple regions of interest as well as analysis of images indicated that the indicated MSCT originated from the area shown. E: 6 MSCT events were observed at this site and were aligned by their rising phase and over-plotted with corresponding mEPSC records shown (time calibration = 250 ms). Fast AMPAR mEPSCs were observed to coincide with the MSCTS, note only a small segment of dendrite was imaged so that mEPSCs that were not coincident with MSCTs likely occurred at other synapses.

To characterize GluR2 knockout cultures, we first examined the current-voltage relationship of AMPA mEPSCs and compared them with the wild-type neurons. The cells were held at different membrane potentials ranging from 70 to 50 mV, and mEPSCs were recorded and averaged at each holding potential. As expected, the mEPSCs recorded from GluR2 knockout neurons showed inward rectification (Geiger et al. 1995), which was not observed in the mEPSCs from wild-type neurons. Most of the knockout neurons showed no clear mEPSCs at positive potentials larger than the noise level of the traces. Among five GluR2 knockout neurons only three AMPAR-mediated mEPSCs were detected at +30 mV, in contrast, at 70 mV they were readily observed. In the wild-type neurons, we saw clear outward currents at +30 mV (n = 6 cells). We compared the ratio of mEPSC amplitude at +30 mV and 70 mV holding potentials to assess the degree of rectification. The ratio of event amplitude at +30 and 70 mV for knockout neurons was 0.14 ± 0.30 (2.1 ± 4.5 pA at +30 mV, 27.1 ± 8.5 pA at 70 mV); whereas the ratio for wild-type, it was 0.55 ± 0.16 (13.4 ± 0.3 pA at +30 mV, 20.7 ± 4.2 pA at 70 mV). The miniature current frequency of knockout neurons at +30 and 70 mV were 0.004 ± 0.009 and 12.4 ± 3.9 Hz, respectively. For the wild-type neurons, the mEPSC frequency was 0.67 ± 0.20 Hz at 30 mV and 7.9 ± 2.4 Hz at 70 mV. The relatively larger reduction in miniature frequency at +30 mV in knockout neurons (than wild-type) is consistent with the AMPAR's strong rectification. The apparent lower frequency of AMPAR mEPSCs at positive potentials in wild-type neurons may be the result of some rectification due to the presence of a subpopulation of homomeric GluR1 receptors. Or alternatively (more likely case), increased noise at positive potentials precluded the measurement of smaller mEPSCs, thus reducing their apparent frequency.

To assess where MSCTs attributed to AMPA and NMDA receptors were occurring within dendrites, we used Ca²⁺ imaging. An example of fluo-3 Ca²⁺ responses associated with conditions that isolate miniature synaptic activity mediated by AMPARs from a dendrite of a cultured GluR2 knockout neuron is shown in Fig. 2. A single approximately 1.0 µm² dendritic spine indicated in Fig. 2A showed a MSCT (normalized to basal mag-fura-2 fluorescence) under conditions that isolated AMPA-mediated mEPSCs. A fluorescence ratiometric image demonstrated that the Ca²⁺ transients were largely limited to the dendritic spine (Fig. 2A, inset). Plots of Ca²⁺ transients at this spine over twenty consecutive trials are shown in Fig. 2B. In some trials we observed multiple MSCTs over the 10-s sampling period. Addition of the AMPA antagonist CNQX in alternate trials (even numbered) reversibly suppressed these local changes in Ca²⁺. The Ca²⁺ transient rose to the peak within one of the sampling points (66 ms), and then decayed back to baseline over 1-1.5 s (Fig. 2B). In this study due to relatively high-frequency of basal miniature synaptic activity (approximately 10 Hz, see above) and the use of a 15-Hz sampling rate (to improve signal to noise parameters), we did not investigate the relationship between miniature current and calcium transient amplitude within the high-density cultures. Previous data (Umemiya et al. 1999) suggests that mEPSCs associated with MSCTs are not significantly different in amplitude from those belonging to the general population and therefore the imaging alone provides an accurate estimate of mEPSC activity.

View larger version (38K): [in this window] [in a new window]   Fig. 2. AMPA receptor-mediated MSCTs at a dendritic spine. A: sequential images (66-ms intervals) of MSCTs at a dendritic spine from GluR2 knockout animal measured by the Ca2+ indicator Fluo-3 under conditions that isolate mEPSCs mediated by AMPARs. The images shown are a ratio of Fluo-3 to Mag-fura-2 fluorescence (F490/F380). Inset: a Mag-fura-2 image showing the structure of dendrite and spines. B: consecutive traces of MSCTs recorded from the spine shown in A (2.0-µm2 sampling area). MSCTs recorded under these conditions were 6-cyano-7-nitroquinoxalene-2,3-dione (CNQX, thick lines; 3 µM) sensitive, indicating that they were mediated by AMPA receptors. The trials were performed at about 1-min intervals and indicate that the CNQX block is fully reversible. C: expansion of the baseline signal (10× gain) recorded in CNQX for sweeps 10 and 18 indicates little bleaching and no detectable MSCTs.

Although our cultures from GluR2 null animals would be expected to contain Ca²⁺ permeable AMPARs, it is also possible that activation of AMPARs during mEPSCs may lead to sufficient postsynaptic depolarization to activate voltage-gated Ca²⁺ channels. Therefore the optical responses we monitored (proportional to changes in intracellular Ca²⁺) may be secondary to AMPAR-mediated activation of Ca²⁺ channels and not due to Ca²⁺ permeable AMPARs directly. We have addressed this possibility in several ways including the use of hyperpolarization to clamp the membrane potential well away from potentials that would activate voltage-gated Ca²⁺ channels. At 100 mV holding potential, we still observed MSCTs when using pharmacological conditions that would selectively activate AMPARs (data not shown, n = 4 neurons). In these experiments, the MSCTs were sensitive to CNQX, indicating that they were attributed to AMPARs. Other evidence against the voltage-gated Ca²⁺ channels is the observation that these Ca²⁺ transients are usually quite localized and do not appear to invade the surrounding dendrites; this would be expected if there were large local changes in voltage that would not be well confined by the spine (Koester and Sakmann 1998; Svoboda et al. 1997; Zador et al. 1990).

Pharmacological experiments were carried out to further rule out voltage-gated Ca²⁺ channels as the signal for AMPA MSCTs. In the presence of 0.1-0.2 mM CdCl₂, we could greatly suppress all dendritic Ca²⁺ transients associated with step depolarization from 80 to 0 mV (Fig. 3A, 78.4 ± 0.04% reduction, n = 7 cells). Under these conditions robust AMPA MSCTs were still observed. We observed no significant change in the MSCT frequency (0.08 ± 01 to 0.09 ± 0.01; SE) or amplitude (2.2 ± 0.3 to 1.9 ± 0.2 F₄₉₀/F₃₈₀; SE with Cd²⁺) during Cd²⁺ application. (Fig. 3, B and C, Kolmogorov-Smirnov test, P > 0.05). These data argue strongly that direct Ca²⁺ entry through Ca²⁺-permeable AMPARs is sufficient to cause the local Ca²⁺ transients we have observed.

View larger version (24K): [in this window] [in a new window]   Fig. 3. Voltage-gated Ca2+ channels do not contribute significantly to MSCTs mediated by AMPARs. A: sample traces indicate that Cd2+ (100 µM) effectively blocks Ca2+ transients mediated by voltage-gated Ca2+ channels evoked by a voltage step (from 80 to 0 mV) at the end of the trace, while Cd2+ has little effect on spontaneous MSCTs. Group data from 7 cells shows that Cd2+ does not significantly change MSCT frequency (B) or amplitude (C) when compared with untreated control trials performed on the same cells. (Kolmogorov-Smirnov test, P > 0.05).

As previously observed by others using a variety of techniques to map quantal responses to single CNS synapses (Bekkers and Clements 1999; Frerking et al. 1997; Liu et al. 1999; McAllister and Stevens 2000; Tang et al. 1994; Umemiya et al. 1999), we report a high level of amplitude variability in Ca²⁺ transients associated with repeated AMPA mEPSCs occurring at single synapses (Fig. 4A). The corrected CV (for baseline noise) of AMPAR mediated MSCT amplitude was 0.34 ± 0.18 (SD) measured from 31 synapses from five cells, whereas the CV of NMDAR-mediated MSCT amplitude was 0.28 ± 0.11 (SD, 25 synapses, 5 cells). This coefficient of variability was significantly larger than that expected for baseline variation (CV: AMPA baseline: 0.09 ± 0.06 SD, 28 synapses 5 cells, NMDA baseline: 0.11 ± 0.06 SD, 25 synapses, 5 cells; P < 0.0001, Mann-Whitney test). Although the selection criteria for MSCTs we used were strict (see METHODS) for a more stringent selection of MSCTs we also rejected responses smaller than the mean of baseline value plus two SD of baseline noise (corrected for shot noise, see METHODS). These selection criteria produced a clear separation between baseline noise and MSCTs (Fig. 4B). Furthermore, overplotted sweeps indicate that responses are easily resolved from baseline noise.

View larger version (24K): [in this window] [in a new window]   Fig. 4. Variation of MSCT amplitude at single synaptic sites during multiple events and cumulative probability plots of baseline noise and MSCT amplitude. A: sample traces showing trial-to-trial amplitude variability of AMPAR (top)- and N-methyl-D-aspartate receptor (NMDAR, bottom)-mediated MSCTs at single synapses. APV or CNQX were applied to isolate AMPA and NMDA responses, respectively (4 AMPAR- and 5 NMDAR-mediated MSCTs are overplotted from 2 different synapses). B: cumulative probability plot of baseline noise amplitude and MSCT amplitude, indicating a clear separation of responses from the noise. The baseline noise measured under AMPA condition was 0.17 ± 0.16 (SD) F490/F380, whereas baseline noise under NMDA conditions was 0.11 ± 0.13 F490/F380. We used the baseline mean plus 2 SD as a criteria for the minimum acceptable MSCT amplitude. This value was based on noise the AMPA trials (which had higher noise) and yielded a noise cutoff = 0.5 F490/F380 (dashed line) that was applied to all MSCTs. Data acquired from 5 cells, n = 157 AMPAR MSCTs; n = 81 NMDA MSCTs.

Although the trial-to-trial MSCT variability was high at a single synapse, we observed that different synaptic sites could have significantly different MSCT mean amplitudes (Fig. 5). Previous data on NMDAR-mediated MSCTs indicated that MSCT amplitude was not necessarily higher at smaller synapses due to reduced volume and that MSCT amplitude was positively correlated with measures of synapse size (Mackenzie et al. 1999). Presumably, these differences in amplitude reflect the number of AMPARs expressed at these synapses (see DISCUSSION). An ANOVA (1-way) indicated highly significant differences in the mean amplitude of AMPAR-mediated responses between different sites within a cell (n = 8 cells, average P = 0.012 ± 0.013). For n = 28 and 25 synapses under NMDAR and AMPAR conditions, respectively, we calculated a between synapse CV of 0.77 for both conditions. This CV was considerably larger than expected for within synapse MSCT baseline variation for both AMPA and NMDA conditions (CV 0.34 and 0.28 respectively). Analysis of over-plotted traces as well as the use of statistical criteria indicated that differences in average MSCT amplitude between sites were not attributed to baseline variation. These differences in response amplitude were also unrelated to presynaptic characteristics such as the frequency of events (correlation R = 0.023, P = 0.7, 238 synapses from 14 cells). Given that MSCT response amplitude is not necessarily correlated with apparent differences in response frequency (see preceding text) (Mackenzie et al. 2000; Murphy et al. 1994, 1995), we suggest that miniature frequency and postsynaptic responsiveness can be regulated independently. Because the probability of miniature and of evoked synaptic activity is correlated in this system (Prange and Murphy 1999), we suggest that the frequency of AMPAR MSCTs may provide an indication of evoked release probability; however, it is conceivably that scenarios exist in which each may be regulated independently.

View larger version (45K): [in this window] [in a new window]   Fig. 5. Variation of MSCT amplitude and frequency at single synaptic sites. AMPAR-mediated MSCTs were obtained over 12 10-s imaging trials from a single GluR2 / neuron under conditions that isolate AMPAR activity. Only trials showing MSCTs are plotted. To compare the amplitudes of the responses, the MSCTs were aligned by the first point that was significantly above baseline (see METHODS for criteria). The AMPAR-mediated MSCT amplitude varies between different synaptic sites within a single cell. Because all of the MSCTs recorded during this imaging period were plotted, the different numbers of MSCTs observed (overplotted traces) at each site (boxes are 1.6 µm across and indicate putative synapses) reflect heterogeneity in MSCT frequency.

In MSCT-imaging experiments, we tried to restrict ourselves to large spine containing neurons of pyramidal morphology to avoid interneurons that might endogenously exhibit Ca²⁺-permeable AMPARs. The AMPAR-mediated MSCT frequency was significantly lower in neurons from wild-type littermate controls (Fig. 6). In wild-type neurons, the AMPAR-mediated MSCT frequency (per synapse) was 0.0046 ± 0.0065 Hz versus 0.034 ± 0.055 Hz in knockout animals. The amplitude of the relatively small number of wild-type AMPAR-mediated MSCTs observed was significantly smaller than those recorded from GluR2 / neurons (0.83 ± 0.12 F₄₉₀/F₃₈₀ for GluR2 +/+ neuron n = 19, 1.47 ± 0.1 F₄₉₀/F₃₈₀ for GluR2 / neuron n = 238; mean ± SE, P = 0.0002, unpaired t-test with Welch's correction). The NMDAR-mediated MSCT frequency was not significantly different among wild-type and GluR2 knockout neurons (Fig. 6), indicating that the low AMPA MSCT frequency in wild-type neurons was not due to a general large change in the mEPSC frequency or detection. In contrast, for GluR2 / neurons, the AMPAR-mediated MSCT frequency was similar to the NMDAR-mediated MSCT frequency (Fig. 6).

View larger version (47K): [in this window] [in a new window]   Fig. 6. Comparison of AMPAR-mediated MSCT frequency in GluR2 +/+ and / cultures. The frequency of MSCTs mediated by AMPA receptors were significantly higher (P < 0.001, ANOVA) in the GluR2 / animals than controls (control, n = 7 cells; GluR2 knockout, n = 14 cells). The frequency of MSCTs was determined from 10-s imaging trials performed in an APV containing media as described in METHODS. In GluR2 knockout animals, the AMPAR-mediated MSCT frequency was similar to the NMDAR-mediated MSCT frequency (P > 0.05, ANOVA). MSCTs were identified using the criteria described in METHODS.

Previous analysis of NMDAR-mediated MSCTs indicates that large significant differences exist in mEPSC frequency between different synapses within a single neuron (Mackenzie et al. 2000; Murphy et al. 1994, 1995). When the rates of AMPA (mean 0.046 ± 0.036 Hz) and NMDA (mean 0.051 ± 0.037 Hz) receptor-mediated MSCTs were compared at between synapses we observed a significant positive correlation (R = 0.29, P = 0.001). This result suggests that difference in miniature release probability and not postsynaptic responsiveness underlies the apparent differences in MSCT rates between synapses because they are reflected in the postsynaptic response of two different receptor classes.

Previous immunocytochemical data indicate that not all synapses co-express AMPA and NMDARs (Racca et al. 2000) By using selective antagonists, we found that some sites show only AMPAR-mediated MSCTs (41%, 81/197 synapses) where as others showed only NMDAR-mediated MSCTs (21%, 43/197 synapses) and 37% of total synapses expressed both NMDA and AMPA MSCTs (73/197 synapses). It is possible that synapses only exhibiting AMPA MSCTs may be the result of poor sampling. To test this, we excluded synapses exhibiting only one response per experiment and still observed AMPA only synapses (33%, 46/139 synapses) and NMDA only synapses (14%, 20/139 synapses). Further evidence that this was not a random phenomenon came when we observed synapses exhibiting at least eight NMDA or AMPA only responses. When we compared the MSCT amplitude of each type of synapse, we found that there were significant differences between classes of synapses. The average NMDAR-mediated MSCT amplitude from the sites showing only NMDAR mediated responses were significantly smaller (approximately 40-60%) than the NMDA MSCTs from sites showing both AMPA and NMDA or only AMPA responses (Fig. 7). These differences were still significant even when we removed the sites showing only two responses per experiment (data not shown). At the sites that showed both AMPA and NMDA receptor-mediated MSCTs, the amplitude of each type of MSCT was not significantly different (Fig. 7). We also compared the frequency of MSCTs at the different type of synapses and found that there were no significant differences. The frequency of AMPA MSCTs from sites expressing only AMPA responses was 0.047 ± 0.045 Hz, whereas the AMPA MSCT frequency from sites expressing both AMPA and NMDA responses was 0.058 ± 0.066 Hz. The MSCT frequency at NMDA only synapse was 0.041 ± 0.032 and 0.047 ± 0.039 Hz for sites showing both AMPA and NMDA responses. Thus although differences in the frequency of MSCTs between synapses within a single neuron are apparent (Mackenzie et al. 1999), we did not observe any significant difference (unpaired t-test with Welch's correction) in MSCT frequency between synapses that apparently expressed only functional NMDARs or AMPARs versus those that express both classes of receptors, indicating that the probability of mEPSC activity may be similar.

View larger version (45K): [in this window] [in a new window]   Fig. 7. MSCT amplitude is significantly reduced at synapses expressing only NMDAR-mediated MSCTs. At sites that showed both AMPAR- and NMDAR-mediated MSCTs (pooled data), the amplitude of MSCTs mediated by each type of receptor were similar (n = 62 sites, P = 0.69, unpaired t-test with Welch's correction for all comparisons). At sites that showed only AMPAR-mediated MSCTs, the amplitude of MSCTs were not significantly different from AMPAR MSCTs recorded from sites with both AMPA and NMDA responses (n = 25 sites, P = 0.21). NMDAR-mediated MSCTs from sites showing only NMDA responses were significantly smaller than those occurring at sites containing both (AMPA and NMDA) responses (n = 9 sites, P < 0.01) and sites containing only AMPA responses (P < 0.005). For this analysis, only sites that showed 3 or more responses were included. Error bars represent the SE.

Previous electrophysiological data by Gomperts et al. (1998), Umemiya et al. (1999), McAllister and Stevens (2000), and Watt et al. (2000) suggest that the amplitudes of NMDA and AMPA components of miniature responses at single synaptic sites are scaled proportionally. Although these studies used high-resolution patch-clamp techniques, they were not able to always map a particular mEPSC to a defined synapse. Therefore using our MSCT imaging technique in combination with solutions that isolate AMPAR and NMDAR activity, we have further addressed this issue. At sites that showed three or more MSCTs, we found that there was a positive correlation between the amplitude of AMPA and NMDA MSCTs (R = 0.94), suggesting that the function of these receptors were co-regulated at each synapse (Fig. 8).

View larger version (22K): [in this window] [in a new window]   Fig. 8. Correlation of AMPAR and NMDAR MSCT amplitudes across different synapses suggest proportional scaling of responses. At the sites that showed both AMPAR and NMDAR mediated MSCTs, the MSCT amplitudes were positively correlated (R = 0.94, P < 0.001). The data shown is from synapses with 3 or more MSCTs and each point is the mean ± SE (n = 30 pooled synaptic sites). Linear regression was performed using a weighted fitting based on the size of the x and y error bars.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We demonstrate that by using GluR2 null mutant animals MSCTs mediated by Ca²⁺ permeable AMPARs can be detected at single synapses using Ca²⁺ imaging. In GluR2 null neurons, the frequency of AMPAR-mediated MSCTs is not different from those mediated by the NMDAR. In contrast, in wild-type neurons, the frequency of AMPAR-mediated MSCTs was reduced by more than 85% when compared with those mediated by NMDARs. Interestingly, occasional synapses in wild-type neurons showed MSCTs under conditions that favor AMPARs, suggesting the presence of Ca²⁺-permeable AMPARs as found in interneurons. However, we had restricted our analysis to large spiny neurons of pyramidal morphology to reduce the chance of recording from an interneuron. Pyramidal neurons of the cerebral cortex are positively labeled with antibodies specific to the GluR2 subunit (Petralia et al. 1997) and therefore should contain Ca²⁺-impermeable AMPARs. However, the ratio of GluR2 subunit expression to the other subunits and its subcellular localization may be factors affecting Ca²⁺ permeability. Single-cell RT PCR studies have shown that even in an apparently homogeneous population of neurons, the ratios of different GluR subunit mRNAs can vary (Yin et al. 1999). Because it is the proportion of the GluR2 subunit to other GluR subunits that determines the Ca²⁺ permeability, it is possible that even in wild-type neurons a small number of AMPARs do not have the GluR2 subunit. Recently, Yin et al. (1999) have shown that a subpopulation of hippocampal pyramidal neurons exhibit GluR2 subunit expression in the soma but no labeling in the dendrites. They then used Co²⁺ staining to assess the presence of functional Ca²⁺ permeable AMPA/kainate receptors. Interestingly, they detected Co²⁺ staining in dendritic arbors, whereas little staining was observed in the soma of this population of pyramidal neurons. Ca²⁺ imaging data by Yuste et al. (1999) suggest that some synapses within CA1 pyramidal neurons (that express edited GluR2) can exhibit Ca²⁺-permeable AMPARs. Thus positive labeling for the GluR2 subunit in a neuron does not necessarily mean that all AMPARs are Ca²⁺ impermeable. Furthermore, Wenthold et al. (1996) using immunoprecipitation with subunit-specific antibodies have shown that about 8% of total AMPAR complexes are homomeric GluR1 (presumably Ca2⁺ permeable) in the CA1/CA2. It is conceivable that the subunit stoichiometry of glutamate receptors is regulated, permitting selective pyramidal synapses to express Ca²⁺-permeable AMPARs. Therefore given our and other results, the Ca²⁺-permeable AMPARs might have a wider distribution than previously thought and may not just be limited to interneurons.

The AMPAR-mediated MSCTs recorded from GluR2 null animals are CNQX sensitive. This suggests that either direct Ca²⁺ entry through the AMPAR or secondary activation of voltage-gated Ca²⁺ channels is responsible for the MSCT signals. In general, miniature synaptic activity would be expected to produce little depolarization and would be unlikely to activate voltage gated Ca²⁺ channels. The addition of CdCl₂ effectively blocked 78% of voltage step evoked Ca²⁺ transients (step from 70 to 0 mV), whereas it has no significant effect on either MSCT frequency or amplitude. Thus voltage-gated Ca²⁺ channels are unlikely to play a role in the MSCTs mediated by AMPARs. These results are consistent with previous findings that Cd²⁺ blocks evoked EPSCs, whereas it has no effect on spontaneous miniature synaptic currents or miniature end-plate potentials (Bao et al. 1998; Losavio and Muchnik 2000). The inability of Cd²⁺ to affect the frequency or amplitude of MSCTs also makes it unlikely that these events are dependent on the store-operated Ca²⁺ channels because they would be expected to be potently blocked under these conditions (Hoth and Penner 1993; Nakamura et al. 2000).

A variety of imaging and local perfusion approaches have generated data suggesting that the amplitude of repeated miniature events at a single synapse is not necessarily fixed (Bekkers and Clements 1999; Forti et al. 1997; Lin et al. 1998; McAllister and Stevens 2000; Murphy et al. 1995; Umemiya et al. 1999). Consistent with these observations, the coefficient of variation of AMPAR-mediated MSCT amplitude at a single synaptic site was 0.34 and was significantly higher than that expected for baseline variation alone. Liu et al. (1999) have provided evidence suggesting that the major contributor to this variability is synaptic cleft glutamate concentration assuming that AMPARs are not saturated by glutamate released from a single vesicle. Data from Renger et al. (2001) suggest the amount of release per vesicle is developmentally controlled leading to selective activation of NMDARs early in development.

In addition to within synapse variability, AMPA MSCT imaging confirms that the mean MSCT amplitude also varies between synapses as observed by others (Gomperts et al. 1998; Umemiya et al. 1999). The source of this variability is likely due to synapse size. Elegant quantitative immunogold labeling studies by Racca et al. have shown for both NMDARs and AMPARs, synapse size is positively correlated with receptor number (Racca et al. 2000). With regards to functional NMDARs, Mackenzie et al. (1999) have demonstrated that the NMDAR-mediated MSCT is positively correlated with synapse size. In the present study, we found that at synapses that showed both AMPAR- and NMDAR-mediated MSCTs, the amplitude of the MSCTs are positively correlated (Fig. 8). Thus we suggest that the amplitude of AMPAR-mediated MSCT is also positively correlated with synapse size. Previous work by Gomperts et al. (1998); Umemiya et al. (1999), Watt et al. (2000), and McAllister and Stevens (2000) support our finding and suggest that the amplitude of NMDA and AMPA components of miniature responses are positively correlated (between different synapses), suggesting that synapses scale the numbers of both receptors proportionally.

At synapses showing only NMDAR-mediated MSCTs (silent synapses), the MSCT amplitude was significantly smaller than the NMDA MSCT recorded from synapses showing both AMPA and NMDA MSCTs. Perhaps, the smaller MSCT amplitude could be due to the fact that silent synapses are still immature. Immunogold electron microscopy (EM) studies indicate there is no significant relationship (Takumi et al. 1999) or a weak positive correlation between synapse size and the number of NMDARs (Kharazia and Weinberg 1999; Racca et al. 2000). Interestingly, data from Kharazia and Weinberg (1999) suggest that many of the smallest active zones may lack functional AMPARs. Furthermore, Racca et al. (2000) and Takumi et al. (1999) had shown that the AMPAR immunonegative synapses (putative silent synapses) are smaller than synapses containing both AMPA and NMDA receptors. On the other hand, there is a strong positive correlation between synapse size and the number of AMPARs (Kharazia and Weinberg 1999; Racca et al. 2000; Takumi et al. 1999). Our data indicate that synapses containing only AMPARs had significantly larger MSCTs than the synapses containing only NMDA receptors. Interestingly, Kharazia and Weinberg reported that a disproportionate number of the largest excitatory synapses appeared to be immunonegative for NMDAR subunit NR1 thus providing a precedent for synapses without NMDARs (Kharazia and Weinberg 1999). Furthermore, recent time lapse and retrospective immunocytochemistry data by H.V. Friedman et al. (2000) also suggested the presence of synapses bearing only AMPARs. In the case of the GluR2 null neurons, it is possible that NMDARs are not always necessary to provide Ca²⁺-mediated signals to properly outfit synapses given that the AMPAR can provide a local Ca²⁺ source. Data from developing spinal interneurons (Rohrbough and Spitzer 1999) indicate that when Ca²⁺ permeable AMPARs are present, NMDARs are not necessarily co-localized at synapses.

The GluR2 subunit has been suggested to be involved in delayed neuronal death after an ischemic insult or after kainate-induced status epilepticus (Bennett et al. 1996; Friedman 1998; L. K. Friedman et al. 1994, 2000; Gorter et al. 1997; Pellegrini-Giampietro et al. 1992 ; Pollard et al. 1993). Interestingly, studies by L. K. Friedman et al. (2000) have shown that after transient focal ischemia the area destined to undergo infraction shows a down regulation of both GluR2 and NR1 after the insult (L. K. Friedman et al. 2000). The downregulation of NR1 suggests a reduction of Ca²⁺ influx via NMDARs, whereas the down regulation of GluR2 subunit indicates enhanced Ca²⁺ influx through GluR1 and/or GluR3 homo- or heteromeric AMPARs. Thus Ca²⁺ permeable AMPARs that we show can directly elevate Ca²⁺ in spines may play an important role in delayed neuronal death in stroke and epilepsy.

In conclusion, our results indicate that AMPAR MSCT imaging in GluR2 knockout mice can be used to confirm previous electrophysiological and immunocytochemical data indicating heterogeneity of function both within and between synapses (Bekkers and Clements 1999; Liu and Tsien 1995; Murphy et al. 1995; Murthy et al. 1997; Umemiya et al. 1999; Yuste et al. 1999). Although some of our results are confirmatory, they are importantly derived with a different experimental approach and provide additional confidence in these previous findings. We have also addressed issues that were not previously studied using whole cell recordings. For example, we have shown no significant correlation between the estimated amplitude of mEPSCs (from imaging data) and their frequency, suggesting that factors regulating miniature release frequency can be distinct from those regulating amplitude. Given that recent data indicates dynamic regulation AMPA receptor levels in response to either occupancy of the receptor by ligands or stimulation of NMDARs (Lissin et al. 1999a,b; Lu et al. 2001), we suggest that imaging of AMPAR MSCTs also provides an ideal system to localize the sites of these functional changes and a provide a better link between immunocytochemical and electrophysiological experiments (Liao et al. 2001; Shi et al. 1999). Given that prolonged mEPSC stimulation is sufficient to alter the distribution of AMPARs (Liao et al. 2001; Lu et al. 2001), it is possible that we may be able to observe the redistribution of functional receptors in future experiments designed to better detect this phenomenon using our imaging approach.