Optical control of NMDA receptors with a diffusible photoswitch

N-methyl-D-aspartate receptors (NMDARs) play a central role in synaptic plasticity, learning and memory, and are implicated in various neuronal disorders. We synthesized a diffusible photochromic glutamate analogue, azobenzene-triazole-glutamate (ATG), which is specific for NMDARs and functions as a photoswitchable agonist. ATG is inactive in its dark-adapted trans-isoform, but can be converted into its active cis-isoform using one-photon (near UV) or two-photon (740 nm) excitation. Irradiation with violet light photo-inactivates ATG within milliseconds, allowing agonist removal on the timescale of NMDAR deactivation. ATG is compatible with Ca2+ imaging and can be used to optically mimic synaptic coincidence detection protocols. Thus, ATG can be used like traditional caged glutamate compounds, but with the added advantages of NMDAR specificity, low antagonism of GABAR-mediated currents, and precise temporal control of agonist delivery.

I onotropic glutamate receptors mediate fast excitatory synaptic transmission and are ubiquitously expressed in the central nervous system. They can be separated into three major classes: a-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs), kainate receptors and N-methyl-D-aspartate receptors (NMDARs). The latter are involved in the induction of synaptic plasticity, the cellular correlate of learning and memory 1 . NMDARs, of which a functional tetrameric receptor structure has been reported lately 2,3 , have also been implicated in a variety of neurological diseases and dysfunctions including ischaemiarelated cell death, Alzheimer's, Huntington's and Parkinson's Diseases, as well as schizophrenia and autism spectrum disorders 1 . NMDARs are heterotetramers whose subunit composition varies based upon brain region, maturation and synaptic activity. Subunit differences result in variations in receptor kinetics, which confer different computational properties on the receptors 1,4 . Such variations in kinetic behaviour of NMDAR subunits have been examined on extrasynaptic and recombinant receptors using outside-out patch clamp recordings and fast application of agonist 5,6 .
The study of glutamate receptors in their native environment has been facilitated by the development of optical tools, namely light-sensitive caged agonists and antagonists, which take advantage of the temporal and spatial precision that light provides. Compounds including caged glutamate, caged NMDA and caged MK-801 have proven very useful for finely tuned, non-invasive studies of NMDARs [7][8][9][10][11] . For example, MNI-glutamate uncaging in a diffraction-limited laser illumination volume enables synapticlike activation of both AMPARs and NMDARs [12][13][14][15] . Larger illumination volumes can be used to quantify activation and desensitization kinetics of synaptic receptors 15 . However, the study of deactivation of receptors must be done following rapid removal of agonist, a feat not possible with caged glutamate because of slow clearance by diffusion or uptake 15 . This is particularly challenging in functional networks within brain tissue.
Molecular photoswitches provide an interesting alternative to caged compounds since they can be rapidly and repeatedly switched on and off and do not generate photochemical byproducts. Over the past decade, several photoswitches have been developed including both photoswitchable tethered ligands and freely diffusible photochromic ligands (PCL) that allow for the optical control of transmembrane receptors and, by extension, neural systems 16 . GluAzo is a photochromic version of glutamate that functions as a PCL for the kainate receptors GluK1 and GluK2 17 , and ATA-3 is a PCL selective for AMPARs 18 (Fig. 1a). Both GluAzo and ATA-3 are active in the thermally relaxed, dark-adapted state, thus their use with networks of neurons is challenging as the whole preparation must be illuminated to prevent tonic activity.
We now report a unique photochromic agonist that targets a different family of glutamate receptors. This compound, azobenzene-triazole-glutamate (ATG), complements ATA-3 and GluAzo since it selectively activates NMDARs (Fig. 1b). In addition, it possesses an important functional advantage that distinguishes it from our previously developed PCLs: it is inactive in the dark-adapted trans-isoform, but quickly converts into its active cis-isoform when irradiated with ultra-violet (UV) light (370 nm). As such, it is not excitotoxic when applied to neural networks in the absence of light, and it becomes an agonist of NMDARs with millisecond precision when irradiated. In addition, we show that ATG is the first diffusible PCL that can be precisely controlled with two-photon excitation (740 nm).

Results
Synthesis and Photophysical Characterization of ATG. ATG was synthesized in a few steps from the known glutamate derivative 4(R)-propargyl glutamate (1) using click chemistry (Fig. 1c). In brief, 1 was treated with azobenzene azide 2 in the presence of a copper catalyst to afford triazole 3. Global deprotection then yielded ATG. A thermally stable stilbene analogue of cis-ATG, termed cis-STG, was synthesized in a similar fashion by treating 1 with 4 and deprotecting the resultant triazole 5 (Fig. 1d). Details of the synthesis and full characterization can be found in Supplementary Fig. 10, Supplementary Note 1 and in the Supplementary Methods.
ATG behaves as a regular azobenzene that can be converted to its cis-isoform with UV-A light. Conversion into the thermodynamically favourable trans-isoform requires irradiation with violet light (Fig. 1 and Supplementary Fig. 1a). Thermal relaxation into the trans-isoform is very slow in physiological buffer solution in accordance with the 'regular azobenzene' nature of ATG (Supplementary Fig. 1b).
Photopharmacology of ATG. To evaluate ATG as a photoswitchable agonist in neurons, we performed electrophysiological recordings in mouse layer 2/3 cortical neurons in acute coronal slices, while continuously perfusing 200 mM ATG in artificial cerebrospinal fluid (ACSF) (Fig. 2, Supplementary Fig. 2a,b). Using the whole-cell voltage-clamp configuration, we examined the spectral sensitivity of cis-ATG-evoked currents from 350 to 410 nm (Fig. 2a, Supplementary Fig. 2a). Maximal ATG-elicited currents were observed in response to 360 nm light, (3.75 mW mm À 2 ; Supplementary Fig. 2c), whereas above 390 nm cis-ATG-mediated currents were negligible (o À 10 pA). This action spectrum corresponds to the maximal conversion to cis-ATG after 365 nm irradiation, as determined by UV-VIS spectroscopy ( Supplementary Fig. 1a). Switching back to trans-ATG was fastest using 425 nm light (t off ¼ 0.17 s ± 0.03; n ¼ 16, Fig. 2b and Supplementary Fig. 2b). According to these data, ATG is cis-active, which is in sharp contrast to other photoswitchable glutamate receptor agonists previously developed 17,18 . Using our illumination system, the dose-response curve of ATG indicated an EC 50 value of 185 mM under 370 nm light (Fig. 2c). In whole cell current-clamp recordings of mouse cortical layer 2/3 neurons, ATG triggered action potential (AP) firing under 370 nm light (Fig. 2d). Using 420 nm light, AP-firing could be quickly silenced. Thus, ATG photoswitching can be used to control neural activity with light.
Using heterologous expression strategies we next examined whether cis-ATG can activate NMDA receptors containing different isoforms. Cis-ATG-mediated currents were not detectable in HEK cells expressing different subunit combinations. In Xenopus oocytes, however, the high level of receptor expression facilitated the detection of cis-ATG-mediated currents ( Supplementary Fig. 4). To normalize for different expression levels we compared the steady-state amplitude of cis-ATGmediated currents in oocytes expressing recombinant diheteromeric NMDARs with those evoked by saturating NMDA concentrations (1 mM, Supplementary Fig. 4). We found that photoactivation of ATG is able to activate all subunit combinations between GluN1-1a and either GluN2A, B, C or D. The observation that cis-ATG-mediated currents were smaller than those elicited by superfusion of NMDA (o5%), may be because of increased light absorption resulting in incomplete photoconversion in the oocyte setup.
To further demonstrate the selectivity of ATG for NMDARs, we recorded current-voltage (I-V) relationships comparing NMDA (200 mM) puff application with ATG photoswitching (Fig. 2f). Because of their magnesium sensitivity, NMDARs are partially blocked at resting membrane potentials, imparting a J-shaped I-V relationship 20 (Fig. 2f, right, black), which we observed for both NMDA application and cis-ATG-mediated currents (Fig. 2f). As expected for non-selective cation channels, the reversal potential was close to 0 mV. In the absence of external Mg 2 þ the I-V relationship was found to be linear, as expected for NMDARs (Fig. 2f,   We further examined whether the thermally stable cis-ATG analogue (cis-STG, Fig. 1d), which does not photoswitch, has similar pharmacology and specificity for NMDARs. When puff applied, the stilbene cis-STG indeed elicited APs in mouse layer 2/3 cortical neurons ( Supplementary Fig. 5a). The J-shaped I-V relationship ( Supplementary Fig. 5b) indicates that cis-STG also targets NMDARs. As such, cis-STG represents a new structural class of agonist for these receptors.
One drawback of traditional caged glutamate compounds is that they are often antagonists of GABA A R-mediated synaptic currents 21 . We therefore tested the effect of trans-ATG on GABA A R-mediated inhibitory postsynaptic currents (IPSCs) in hippocampal CA1 pyramidal neurons. Using 400 mM bath application of ATG, we observed no detectable alteration of spontaneous IPSCs, but observed a 38 ± 6% (n ¼ 10 cells) block of evoked IPSCs and an increase in coefficient of variation, consistent with a presynaptic target ( Supplementary Fig. 6). Nevertheless, this is less than the 50% block by RuBi-glutamate (300 mM 21 ), the 55% block by CDNI-glutamate (400 mM 22 ) and the 83% block by the commonly used MNI-glutamate (300 mM 21 ).
ATG-mediated photoswitching of NMDAR gating. We next considered the possibility that ATG photoswitching could be used to perform temporally precise agonism of NMDARs. We used fast digitally controlled diode lasers at 375 nm to switch to cis-ATG and 405 nm to preferentially switch to trans-ATG (Fig. 3a). The 375 nm laser light was focused to a near-diffractionlimited spot (full-width half-maximum (FWHM): 300 nm; Supplementary Fig. 7) and the 405 nm laser beam was adjusted to form a 4 mm spot (full-width half-maximum; Supplementary  Fig. 7). NMDAR currents were recorded from CA1 pyramidal neuron dendrites within 100 mm of the soma at a holding potential of -30 mV. In the first set of experiments, 200 mM ATG was bath-applied ( Fig. 3b). Laser illumination of 100 ms at 375 nm (B150 mW) was required to see significant cis-ATGmediated currents (average peak response: -87 ± 12 pA; n ¼ 6 cells). Illumination for 5 s with 375 nm laser light resulted in large currents (-494 ± 56 pA, n ¼ 18 cells) that rose over the light pulse duration ( Supplementary Fig. 8a) and decayed over tens of seconds ( Supplementary Fig. 8a,b). The reduction in cis-ATGmediated current was maximal when using 50 ms illumination of 405 nm laser light (11 mW), but incomplete (56 ± 3 % reduction, n ¼ 14 cells; Fig. 3b). The residual current was similar in amplitude to that when delivering a 50 ms 405 nm light pulse alone (P ¼ 0. small but measurable amount of cis-ATG ( Supplementary  Fig. 1a). This could explain why 405 nm illumination reduces current following 375 nm illumination, but evokes current when not preceded by 375 nm illumination (Fig. 3b).
When we examined the spatial dependence of cis-ATGmediated currents, we saw little change in the peak amplitude even if the 375 nm laser was positioned up to 100 mm away from the dendrite. This finding, along with the slow rise and decay times, led us to the hypothesis that we had a very large effective photoactivation volume because of ATG photoswitching out-of-focus-an effect observed for bath application of caged compounds 23 . To further improve the kinetics of the NMDAR response, we tested local perfusion of ATG at 100 mM. Under these conditions, the decay of light-evoked cis-ATG-mediated currents was accelerated over sixfold, and we observed a halving of the peak current at only 5 mm away from the dendrite ( Supplementary Fig. 8d). Local perfusion also permitted the use of higher concentrations of ATG to improve the activation time of NMDAR currents to o100 ms, similar to that of the widely used caged compound MNI-glutamate at similar concentrations ( Supplementary Fig. 9a). Additionally, local application of ATG resulted in an enhanced fractional reduction of cis-ATGmediated currents upon 405 nm light illumination ( Fig. 3c and Supplementary Fig. 9b), up to 77 ± 3% reduction, 100 mM (n ¼ 9 cells). Finally, and most importantly, local application of ATG resulted in faster decay of NMDAR currents following 405 nm illumination (ATG off ): the half-decay of 33±4 ms (n ¼ 9 cells; Fig. 3d) was over 12 times faster than NMDAR current decays recorded in response to MNI-glutamate uncaging (422±72 ms, n ¼ 5 cells, P ¼ 0.01). Interestingly, the weighted decay time constant of ATG off (t weighted of 102±42 ms; n ¼ 9 cells; see example fit in Fig. 3c inset) was intermediate to the decay values for recombinant receptors containing either GluN1/GluN2A (t weighted ¼ 29 ms) and GluN1/GluN2B (t weighted ¼ 193 ms) 24 . We therefore considered the possibility that the rapid ATG off decay could be used to estimate relative contributions of GluN2A and GluN2B to dendritic NMDAR activation. In mature CA1 pyramidal neurons (4P21), NMDARs are mostly comprised of tri-heteromers of GluN1, GluN2A and GluN2B 4,25 whose kinetics are dominated by the rapid deactivation of GluN2A 26,27 . We therefore compared the decay of ATG off in WT and GluN2A KO mice. To get a better estimate of the ATG off kinetics without contamination of 405 nm-induced cis-ATG-mediated currents, we subtracted a scaled curve fit to the 405 nm-induced current of each individual cell (Fig. 4a,b). The half-decay of subtracted ATG off currents was 30±4 ms (n ¼ 9 cells) similar to the decay of NMDAR EPSCs recorded at the same age (40±3, n ¼ 13 cells; P ¼ 0.18; Steel Dwass all pairs nonparametric multiple comparison test; Fig. 4c). To determine whether the decay of ATG off currents was influenced by GluN2A expression, we examined ATG off decays in GluN2A KO mice (P29-P55). Surprisingly, the decay of ATG off remained largely unaltered (KO half-decay: 31 ± 5 ms, n ¼ 5, P ¼ 0.99 Steel Dwass all pairs nonparametric multiple comparison test), but the evoked synaptic responses were nearly three times slower (KO half-decay: 113±7 ms, n ¼ 10, P ¼ 0.0004, Steel Dwass all pairs nonparametric multiple comparison test; Fig. 4c), confirming the kinetic influence of GluN2A expression. Thus, the ATG off decay is insensitive to subunit composition of NMDARs that are known to alter channel deactivation 28 .
Two-photon activation of ATG. One strategy to achieve very localized photoactivation is to use two-photon (2P) excitation.
Combining ATG photoactivation with Ca 2 þ imaging. The calcium permeability of NMDARs links their activity to postsynaptic biochemical alterations, such as spine morphology and glutamate receptor expression, which are associated with synaptic plasticity 20 . Since ATG specifically acts on NMDARs, we examined the possibility of imaging its effect with a Ca 2 þsensitive fluorescent dye (Fig. 6). To this end, we incubated hippocampal slices with Quest Fluo-8-AM, a membranepermeable Ca 2 þ indicator that is compatible with the activation wavelength of ATG because of its excitation wavelength of 490 nm (Fig. 6b). We applied tetrodotoxin citrate (TTX) (1 mM) and felodipine (40 mM), which prevents opening of voltage-gated Ca 2 þ channels to limit the response to NMDAR-mediated calcium entry (Fig. 6 right half). Indeed, upon irradiation with 370 nm in the presence of ATG, Mimicking synaptic coincidence detection with ATG. One of the most intriguing properties of NMDARs is their voltage-sensitive Mg 2 þ block at resting membrane potential. Accordingly, the presence of a neurotransmitter and concomitant postsynaptic depolarization are required to activate the ion channel, rendering NMDARs coincidence detectors of pre-and postsynaptic activity 33 . To test whether cis-ATG-mediated NMDAR activation can replace the synaptic stimulation necessary for coincidence detection, we designed a stimulus protocol, which couples antidromic stimulation (electrical stimulus of the axon hillock; Fig. 7, black line) of the postsynaptic cell with light activation (Fig. 7, purple line) of NMDARs at defined intervals. The electric stimuli were given at subthreshold intensities and did not induce AP firing by themselves. By varying the interval between the two stimuli, we showed that only coincident activation with light and current injection led to suprathreshold signals that generated APs. If the stimuli did not coincide or the delay between stimuli was too long, the number of spikes generated was significantly reduced (Fig. 7d).

Discussion
To circumvent the limitations of traditional irreversible caged agonists, we synthesized a novel photo-reversible glutamate receptor agonist, ATG. This compound has functional features that make it unique and useful to neurobiological research: It is inactive in the dark, selectively targets NMDARs, can be quickly activated by irradiation with 360-375 nm light (or 725-740 nm light, 2P activation) and quickly deactivated with 405-460 nm light. Using rapid laser illumination, we demonstrated the unique photoswitching between a cis-active and trans-inactive compound, which enabled activation and deactivation of NMDARs on the timescale of their intrinsic gating properties. We also demonstrated that ATG is amenable to combination with other optical techniques, namely Ca 2 þ imaging. Finally, we showed that ATG could be used to mimic synaptic activation in coincidence detection protocols.
In the absence of detailed structural data and molecular dynamics calculations, it is difficult to explain why cis-ATG is the active form and why it is selective for NMDARs. Structureactivity relationship studies indicate that the pharmacological space available for ATG is relatively narrow, which makes it a challenge to develop red-shifted derivatives (D. Trauner, unpublished results). As the physiological activity of the stilbene analogue cis-STG indicates, the diazene unit (N ¼ N bond) is not essential for ligand binding and is probably not engaged in hydrogen-bonding interactions to the ligand-binding domain. This is also apparent in a recent X-ray structure of GluAzo bound to the GluK2 ligand binding domain 34 . Attempts to red-shift the action spectrum of ATG by substituting it with strongly electrondonating substituents, such as a diethylamino group, in the 4 0position have so far yielded inactive compounds, indicating that substituents in this position clash with the ligand-binding domain in its closed state. Other ways to red-shift, for example, with heterocyclic azobenzenes, can be imagined and are under active investigation. Finally, cis-ATG and cis-STG represent a new class of agonists (and potentially antagonists) for ionotropic glutamate receptors that can be rapidly assembled using click-chemistry.
Herein, we show that ATG photoswitching and lightdependent NMDAR activation can be achieved with a monochromator, LEDs or laser light source. In principle, simple LEDs or standard light sources used in fluorescence imaging could also be used. It should be taken into account, however, that the kinetics of photoswitching are strongly dependent on the applied light intensities. ATG exhibits a high molar extinction coefficient (48,778 cm -1 M -1 at 330 nm). Although such a feature enables efficient light absorption, it also reduces the light intensity at the focal point as a consequence of Lambert-Beer's law 23 . We circumvented this limitation by performing local perfusion of ATG to limit the optical path length in which photoswitching occurred. Two-photon excitation also provides a solution to the problem of out-of-focus light absorption while in parallel improving depth penetration in scattering tissue and providing intrinsic localization, properties well-suited for activation of single spines with a PCL. Deactivation at longer wavelengths can still be performed by large 1P illumination areas, which is advantageous for switching any molecules that have diffused out of the diffraction-limited illumination volume. Although the 2P cross-section has not been estimated, the ability of brief laser illumination durations (o1 ms) to evoke synaptic-like NMDAR current amplitudes could render ATG a useful complement to MNI-glutamate. Very recently, MAG 2p a derivative of glutamate that can be activated with 2P absorption, has been introduced 35,36 . This compound, however, is covalently attached to the receptor and not freely diffusible like ATG.
Because diffraction-limited illumination volumes are much larger than a single synaptic vesicle fusion event, the photolysis of caged neurotransmitter in diffraction-limited spots (by one-or two-photon excitation) is unlikely to mimic the rapid decay of neurotransmitter clearance 15 . PCL compounds, however, are ideally suited to overcome this diffusion-limited problem because photoisomerization of azobenzenes can occur on a picosecond timescale, sufficient for photo-inactivation of an agonist before it diffuses away. It remains to be determined if the switching speed is similar when cis-ATG is bound to the receptor. Nevertheless, we were able to demonstrate that ATG photoswitching can accelerate NMDAR current decays which, when using classical caged agonists, would have been limited by diffusional clearance. This requires that the inactive trans-ATG have an affinity low enough to be effectively a non-agonist. Consistent with the relative low-affinity of trans-ATG, the ATG off decays are nearly thirty times faster than the decay of cis-ATG currents. Trans-ATG also appears to unbind NMDARs more rapidly than glutamate, as the half-decay is over twelve times faster than that achieved by glutamate uncaging (Fig. 3). We demonstrate, for the first time, the fundamental ability to switch 'off' channel agonism, thereby speeding the decay of NMDAR currents.
We were surprised that the ATG off decay was not altered in GluN2A KO mice, especially since faster unbinding rates of glutamate are thought to underlie the faster deactivation rates of GluN2A-containing channels 34 . One possible interpretation of these results is that the affinity of trans-ATG is so low that photoconversion reveals a ligand-independent closing transition. To confirm this, future experiments would be required, especially to confirm rapid switching of cis-ATG when bound to the NMDAR.
Our results show that ATG is a powerful tool that permits precise temporal control of NMDAR gating not otherwise achieved with state-of-the-art caged compounds. They extend the reach of photopharmacology to an important subtype of glutamate receptors and demonstrate that photoswitchable neurotransmitters that are inactive in the dark can be synthesized and used to precisely control receptor activation in their native environment.

Methods
UV-Vis spectra. ATG was dissolved to a concentration of 50 mM in buffer containing (in mM) 138 NaCl, 1.5 KCl, 2.5 CaCl 2 , 1.2 MgCl 2 , 10 Glucose and 5 HEPES, adjusted to pH 7.4. UV-VIS spectra were taken in a 100 ml cuvette with the switching light (monochromator) introduced through a glass fibre from the top of the cuvette, perpendicular to the light path of the spectrometer (Varian, Cary 50). The kinetics of the trans-to cis-conversion were recorded at the maximal absorption wavelength of trans-ATG (330 nm). To achieve fast switching rates, we used high power LEDs at 365 and 460 nm (Prizmatix) for trans-cis and cis-trans isomerization, respectively.
Cortical slice preparation and external solutions. Cortical coronal slices were prepared from C57Bl6JRj mice (postnatal day 10-15, both male and female animals were used without known experimenter bias). Following decapitation, the brain was rapidly removed and transferred to an ice-cold saline solution composed of (in mM) 2.5 KCl, 1.25 NaH 2 PO 4 , 25 NaHCO 3 , 0.5 CaCl 2 , 7 MgCl 2 , 25 glucose, 75 sucrose saturated with carbogen (95% O 2 /5% CO 2 ). Slices (300 mm thick) were made using a Campden vibratome 7,000 smz-2 (NPI Electronic). Slices were incubated at 34°C for 30 min in ACSF composed of (in mM) 125 NaCl, 2.5 KCl, 1.25 NaH 2 PO 4 , 26 NaHCO 3 , 2 CaCl 2 , 1 MgCl 2 , 20 glucose saturated with carbogen (95% O 2 and 5% CO 2 ). After incubation, slices were stored at room temperature from 30 min to five hours before being recorded. Experiments were carried out at room temperature. Unless stated otherwise, ATG was added from a 200 mM dimethyl sulfoxide stock to the ACSF to yield a final concentration of 200 mM. The ACSF was heated to 40°C to improve the solubility of the ATG stock. The solution was not filtered because ATG adheres to filter materials.
For the identification of target receptors and calcium imaging experiments, the iGluR antagonists NBQX (25 mM) and D-AP-5 (40 mM), and the channel blockers TTX (1 mM) and felodipine (40 mM) were bath-applied, whereas MK-801 (50 mM) (all from Abcam) was loaded into the patch pipette. NMDA (1 mM, Sigma-Aldrich) and cis-STG (200 mM) were puff-applied through a glass pipette using a pressure ejection system (PDES, NPI Electronic). For voltage-clamp recordings, TTX was added to the ACSF.
Patch clamp recordings of cortical layer 2/3 neurons. Pyramidal neurons were patched using fire-polished glass electrodes with a resistance of 6-9 MO. Current-clamp recordings were carried out using the following intracellular solution (in mM): 140 K-gluconate, 10 HEPES, 12 KCl, 4 NaCl, 4 Mg-ATP, 0.4 Na 2 -GTP. For whole-cell voltage-clamp recordings, we used (in mM) 110 Cs-gluconate, 15 NaCl, 10 HEPES, 5 TEA, 0.16 EGTA, 4 Mg-ATP, 0.4 Na 2 -GTP. Recordings were made with an EPC 10 USB amplifier, controlled by the Patchmaster software (HEKA). Data was filtered at 2.9-10 kHz and digitized at 50 kHz. Holding potential was corrected for a 14 mV liquid junction potential. Cells were rejected if leak currents were 4200 pA or series resistance 425 MO. Data was analysed using the Patcher's Power Tools (MPI Göttingen) and routines written in IgorPro (Wavemetrics).
For antidromic stimulation, glass electrodes (5 MO) filled with ACSF were placed within 20 mm of the axon hillock and the stimulus pulse was applied through an isolated stimulation unit (A-M Systems). The stimulation intensity was set to be subthreshold. The temporal pattern of the antidromic and ATG light stimuli were controlled through the Patchmaster software (HEKA).
Photoswitching of ATG and calcium imaging. A Poly-V monochromator (FEI Systems) controlled through the Patchmaster software was used to toggle between trans-and cis-ATG at varying wavelengths (370 and 420 nm unless otherwise indicated). During calcium imaging experiments, the monochromator was controlled by the Live Analysis software (FEI). The calcium indicator Quest-Fluo-8-AM (50 mM MoBiTec) was added to the ACSF, and acute hippocampal slices were incubated for at least 20 min at 37°C to facilitate uptake of the indicator. Calcium changes were recorded at 480 nm, digitized at 10 Hz, background corrected and the DF/F 0 ratio was calculated using IgorPro routines and ImageJ (NIH).
Laser-mediated photoswitching and uncaging. Visually guided patch experiments were performed using differential interference contrast or Dodt contrast. Dendrites were visualized using either a confocal (592 nm excitation of Alexa Fluor 594) or a 2P imaging system (810 nm, Ultima, Prairie Technologies). We coupled a 405 nm diode laser (Model PhoxX 405-120, Omicron Laserage) and a 375 nm diode laser (Model PhoxX 375, Omicron Laserage) into the photoactivation galvanometers of the scanhead (for rapid spot positioning) using single-mode optical fibres (Part QPMJ-A3S,A3S-400-3/125-3-2-1, Oz Optics). For one-photon photoswitching, we used 150 mW (bath application) or 375 mW (local application) of 375 nm and 10.5 mW of 405 nm, measured after the objective. For two-photon photoactivation, a pulsed Ti:Sapphire laser (Chameleon Ultra II, Coherent) tuned to either 725 nm or 740 nm was directed into a photoactivation path. Photoactivation was performed on proximal dendrites within 100 mm of the soma to minimize possible effects of dendritic filtering. Illumination spots were typically placed o0.5 mm from the tip of the spine head, except when longer duration pulses were used (4500 ms), in which case the spot was placed 1 mm away.
Expression and two-electrode recordings in Xenopus oocytes. Frog oocytes were surgically removed from the ovaries of X. laevis (Nasco) anesthetized with ethyl 3-aminobenzoate methanesulfonate (2.3 g l À 1 ; Sigma). The lumps of oocytes were incubated with 300 U ml À 1 (10 mg ml À 1 ) collagenase type I (Worthington Biochemicals) for 3 h at 21°C in Ca 2 þ -free Barth's solution (in mM) 88 NaCl, 1.  3 15 HEPES, pH adjusted to 7.6 with NaOH). Oocytes were maintained in Barth's solution supplemented with 100 mg ml À 1 gentamycin, 40 mg ml À 1 streptomycin and 63 mg ml À 1 penicillin. Intact oocytes of stages V or VI were selected and cRNA was injected with a Nanoliter 2010 injector (WPI) within 8 h after surgery. For expression of GluN1/GluN2 heteromers, 20 nl (4 ng) of cRNA for each subunit were injected. Electrophysiological recordings were carried out 5 days after injection. Two-electrode voltage clamping was performed using a TurboTec-10CX amplifier (npi electronic) controlled by Pulse software (HEKA). For photoswitching experiments, the recording chamber was illuminated using LEDs (365 and 460 nm, Prizmatix) coupled to a light guide, which was placed directly above the oocyte. LEDs were controlled via the TTL outputs of the ADC/DAC (ITC-16, Instrutech). Borosilicate glass capillaries (Harvard Instruments) were pulled to resistances of 0.1-1 MO with a vertical puller (PIP5, HEKA) and filled with 3 M KCl.
Oocytes were clamped at À 70 mV. All recordings were performed in continuously superfused with Barium Ringer (BaR, in mM) 115 NaCl, 2.5 KCl, 1.8 BaCl 2 , 10 HEPES-NaOH, pH 7.2). When a stable holding current was attained, the recording protocol was started. To further prevent opening of calcium-induced chloride channels, niflumic acid (NFA, 250 mM) was added to the BaR. All agonist solutions contained 10 mM glycine. In addition, the BaR used for recording photocurrents was supplemented with ATG (200 mM). For recording of cis-ATG mediated currents, the microscope light was switched-off and the photoswitching recording sequence was started. A 5 s pulse of blue light (460 nm) was followed by 5 s of UV light (365 nm) and, again, 5 s of blue light. These protocols were carried out first in the absence of ATG to control for possible artifacts, and then repeated in the presence of 200 mM ATG. For the recording of NMDA-induced currents, 1 mM NMDA in BaR supplemented with 10 mM glycine was perfused. After steadystate currents were achieved, the NMDA was washed out until a stable baseline was reached. The application of 1 mM NMDA was repeated until two consecutive applications resulted in similar steady-state amplitudes. For analysis, light-induced ATG-independent currents were subtracted from cis-ATG-mediated currents. The steady-state amplitude of the corrected cis-ATG-mediated currents were then normalized to steady-state amplitude of NMDA-induced currents.
Data analysis. Data analysis was performed using the Neuromatic analysis package and custom routines within the IgorPro environment (Wavemetrics). Cells were rejected from analysis if the leak current was 4 À 100 pA at À 30 mV holding potential or the series resistance was over 15 MO, except for the GluN2A KO animals in which larger leak currents were accepted. Cis-ATG-mediated current peak amplitudes from one-photon photoswitching were measured over a 1 ms window around the peak. Error bars are presented as the mean ± s.e.m. unless otherwise indicated. Current decays were estimated from a fit with a double exponential equation y ¼ A 1 exp{ À (x À x 0 )}/t 1 } þ A 2 exp{ À (x À x 0 }/t 2 }. The weighted decay time constant (t weighted ) was calculated as Isochronal amplitudes represent averages of the currents in a 200 ms window centred at the time point at which the on-spine response had reached 75% of its peak value.
To better estimate the NMDAR current decay following 405 nm illumination without contamination from partial cis-activation of ATG, we performed a subtraction protocol. 405 nm-induced currents were fit to an empirical function that describes the rising phase and dual exponential decay 38 : The fits of each individual cell were scaled by eye to maximally overlay the slow decay component of 405 nm only and 375/405 nm-induced currents. We justified such a scaling because the population traces overlapped (Fig. 4a,b). The scaled fits for each cell were then subtracted from the associated average current trace induced by 375 nm illumination followed by 405 nm illumination. Cells were only analysed if three or more recordings were performed under each illumination condition (375 nm only, 375 nm followed by 405 nm, 405 nm only). GABAR spontaneous events ( Supplementary Fig. 6) were detected using an event detection algorithm within the Neuromatic analysis software based on initial 10 pA threshold detection. Events were refined using two additional search features: (1) event onset detection using a 1.5 ms backward sliding window to ensure the point of threshold detection was 7 Â s.d. of background noise; and (2) event peak detection using a 0.2 ms forward sliding window to ensure the event peaked within 2 ms of the time point of event threshold detection with a current greater than that of the point of event threshold by at least 2 Â s.d. of background noise. Parameters were determined empirically, but were held constant for all analysis.
Peak amplitudes of IPSCs ( Supplementary Fig. 6) were estimated from a 200 ms window centred at the time point of the peak value of the average of all events for a particular condition (number of events were typically between 50 and 300 per cell). Each synaptic current peak amplitude was corrected for series resistance error, as well as for a slight (B5%) decay in peak amplitude of IPSCs over the duration of the recording, which was not due to ATG activity (estimated from sham experiments, n ¼ 5 cells). The coefficient of variation of the GABAR IPSCs was calculated as ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s 2 peak À s 2 background p peak where s represents the s.d. over a 200 ms window. Background variance was measured as the standard deviation over a 200 ms window immediately before the stimulation pulse.
Statistical analyses requiring multiple comparisons were first examined with a nonparametric one-way analysis of variance (Kruskal-Wallis) followed by nonparametric tests between specific values. We considered comparisons to be significantly different if P-values were o0.05.