An optimized fluorescent probe for visualizing glutamate neurotransmission

Sensor engineering and characterization

Designing intensity-based fluorescent sensors^21,22 requires optimization of two parameters: the insertion site of cpGFP into the binding protein and the composition of the residues adjoining the two (“linkers”). In previous work^21,22, we used quantitative structural analysis of local conformational changes that occur upon ligand binding to choose acceptable insertion positions, and high-throughput screening to optimize the linkers. Although no crystal structures of E. coli GltI are available in the PDB, a high-resolution structure of the 99% identical GltI from Shigella flexneri has been solved in the glutamate-bound state²³, but not in the ligand-free state. We hypothesized that the structural commonalities of our two previous sensors could be generalized to other PBPs, and that based on the global structural homology of Shigella GltI (PDB 2VHA) to MBP, we could apply these generalizations to the selection of insertion sites in GltI (Fig. 1a, Supplementary Figs. 1-7).

Placement of the cpGFP insertion point close to the inter-domain hinge region (after residue 253 of GltI, Supplementary Figs. 1 & 7) followed by screening of mutations to both linkers yielded a variant (“GltI253.L1LV/L2NP”) with (ΔF/F)_max of 4.5 ± 0.1 (s.d., n = 5) (Fig. 1b, Supplementary Fig. 8). Titration of purified protein in vitro (Fig. 1b) indicates an affinity of 107 ± 9 μM (s.d., n = 5) for glutamate, and 145 ± 18 μM (n = 3) for aspartate (which has been identified as a co-neurotransmitter with glutamate in some neurons²⁴), and a pK_a of 6.5 in the glutamate-bound and 7.0 in the ligand-free state (Supplementary Fig. 9). It has no detectible affinity for a panel of decoy L-amino acids (glutamine, asparagine, cysteine, arginine, histidine, serine, proline, tryptophan, β-alanine, taurine), neurotransmitters (glycine, GABA, acetylcholine, serotonin, D-serine, dopamine and its metabolic precursor L-DOPA), pharmacological glutamate receptor agonists (AMPA, NMDA, kainate) or antagonists (philanthotoxin PhTx-74, D-AP5, NBQX, CNQX, DNQX, CPP), or a glutamate transporter inhibitor (TBOA) (Supplementary Fig. 10), nor does their presence in a glutamate titration significantly affect sensor function (Supplementary Fig. 10). The on-rate of binding was faster than could be detected by our stopped-flow fluorimeter (Supplementary Fig. 11). The protein is bright and photostable under 2-photon excitation (Supplementary Fig. 12), and similar to 1-photon excitation, has a (ΔF/F)_max of 4.5 (Supplementary Fig. 13), but with no change in lifetime (Supplementary Fig. 13). As a preliminary test of the sensor on the surface of mammalian cells (HEK293), the GltI253.L1LV/L2NP gene was cloned into the pDisplay vector (Invitrogen) for expression under control of a CMV promoter. This construct (Supplementary Fig. 14) encodes an N-terminal mouse immunoglobin κ-chain leader sequence, which directs the protein to the secretory pathway; a hemagglutinin A (HA) epitope tag; the GltI253.L1LV/L2NP protein; a myc epitope tag; and at the C-terminus, a platelet derived growth factor receptor transmembrane helix, which anchors the protein to the plasma membrane, displaying it on the extracellular side. The HA tag significantly decreases (ΔF/F)_max in glutamate titrations of soluble protein in vitro (Supplementary Fig. 15). Thus a new version of the sensor, (“iGluSnFR”), lacking the HA tag, was cloned into the backbone of the pDisplay vector.

The fluorescence of HEK293 cells transfected with pCMV.iGluSnFR was measured in our 2-photon microscope, using a perfusion chamber to efficiently wash cultured cells in HBSS/glutamate solutions. The in situ affinity of the sensor on HEK293 cells is 4 ± 1 μM (s.d., n = 4) (Fig. 1b), a serendipitous 25-fold increase in affinity from the soluble protein that is in the range expected to be physiologically relevant for measuring neurotransmitter release²⁵. Although some fluorescent iGluSnFR is seen inside the cell (presumably endoplasmic reticulum-localized), only membrane-displayed sensor responds to glutamate (Fig. 1c).

Characterization in neurons and astrocytes

To determine the suitability of iGluSnFR for detecting synaptic release, hippocampal astrocyte-neuron co-culture was infected with adeno-associated virus (AAV2/1), under either the human synapsin-1 promoter to drive expression of iGluSnFR in neurons (AAV.hSynapsin.iGluSnFR) or the glial fibrillary acidic protein promoter to drive expression in astrocytes (AAV.GFAP.iGluSnFR). Two weeks after infection, confocal fluorescence showed that iGluSnFR was evenly distributed on the extracellular surface of neuronal somata and dendrites (Fig. 2a). Expression of iGluSnFR on the extracellular surface remained fairly stable after 4 weeks (Supplementary Fig. 16). Titration with glutamate in a flow cell indicates an in situ affinity of the sensor on the surface of neurons of 4.9 ± 1.3 μM (s.d., n = 3 ROIs) (Fig. 1b), with (ΔF/F)_max = 1.03 ± 0.15. In astrocytes, the fluorescence of iGluSnFR was uniformly distributed on the membrane of somata and processes, but with some punctae apparent (Fig. 2b).

To test the performance of iGluSnFR in resolving action potential (AP)-evoked glutamate transients, a series of electrical field stimuli at 30 Hz was delivered (1 field stimulus evokes 1 AP, data not shown). iGluSnFR is sensitive enough to detect global glutamate release from single field stimuli, in both somata and processes of neurons (Fig. 2c) and on the surface of astrocytes co-cultured with neurons (Fig. 2d). In neurons, increases in iGluSnFR fluorescence were detected in response to single APs (Fig. 2e, field of view (ΔF/F)_max = 0.14 ± 0.02; s.d., n = 3 trials); on astrocytes, single AP-induced glutamate release from neurons was reliably detected, but with a lower magnitude change in fluorescence (Fig. 2f, (ΔF/F)_max = 0.07 ± 0.01). In both cell types, the peak fluorescence plateaus at higher AP stimuli (20-160 APs) (Fig. 2g, Supplementary Fig. 17). iGluSnFR also showed fast kinetics on both cells (Fig. 2e,f); decay time increases with larger AP stimuli (Supplementary Fig. 17). SNR in neurons is significantly larger than that measured with SuperGluSnFR¹⁹ under similar conditions (Fig. 2e).

To confirm that the observed change in fluorescence is the result of glutamate release, and not an artifact, e.g. caused by a change in pH at the synapse, AAV.hSynapsin.iMaltSnFR²¹, which has a similar pH profile to that of iGluSnFR (Supplementary Figs. 9 & 18) was tested as well, and showed no response (Fig. 2g).

Finally, cultured hippocampal neurons were imaged in a perfusion chamber with 2-photon illumination and subjected to “puffs” of glutamate. After extensive washing with HBSS to restore the sensor to a glutamate-free state (neuronal culture growth media contains 5% FBS and ~100 μM glutamate, Supplementary Fig. 19), a patch pipette filled with 37 μM glutamate and 1 nM AlexaFluor 568 (as a red tracer) was brought near the neurons. A small amount of the mixture was puffed onto the neurons during continuous perfusion of glutamate-free HBSS by briefly applying positive pressure to the patch pipette (< 200 msec). Green fluorescence was observed to increase rapidly and coincidentally with red fluorescence, and decay just slightly slower than diffusion of the red signal (Fig. 2h). Repeated puffs evoked similar fluorescence responses (Fig. 2h).

Two-photon glutamate uncaging and imaging in brain slices

To characterize the sensitivity and kinetics of iGluSnFR in the context of its applicability for studying mammalian synaptic transmission, we employed two-photon glutamate uncaging and imaging in acute rat hippocampal slices²⁶. 2-4 weeks after viral delivery of AAV.hSynapsin.iGluSnFR into the hippocampus, infected pyramidal neurons in CA1, CA2 and subiculum were patched and filled with AlexaFluor 594. This anatomical dye enabled glutamate uncaging to be targeted at spine heads located on peri-somatic apical oblique and basal dendrites within ~120 μm of the soma. Linescans were recorded at 600-1200 Hz (excited with a second laser at 920 nm), on the spine heads where glutamate was uncaged (Fig. 3). The whole-cell patch electrode recorded cellular electrical responses in current-clamp mode.

Resting membrane potential of iGluSnFR-expressing pyramidal neurons appeared normal (-65.0 ± 0.5 mV, s.e.m., n = 9 neurons). Large, fast iGluSnFR signals were observed at all spines tested (Fig. 3a,b, n = 28 spines from 9 neurons taken from 5 animals), including at uncaging laser intensities that evoked physiologically-sized EPSPs at the soma²⁷. Peak amplitude (Fig. 3c) and time-integrated area (Fig. 3d) of iGluSnFR signals correlated roughly linearly with somatic EPSP amplitude. Single-trial results are shown in Supplementary Fig. 20. Responses to single uncaging pulses were extremely rapid (~5 msec). Strikingly, at some spines iGluSnFR responded to very low uncaging laser intensities that produced no EPSP signals above background (Fig. 3b: light blue trace). Control experiments testing the necessity of MNI-glutamate (Fig. 3e), background glutamate detection (Fig. 3f), and glutamate diffusion (Fig. 3g) yielded the expected results. These experiments indicate that iGluSnFR sensitively and rapidly reports glutamate dynamics at relevant cellular structures within the physiological regime for neurotransmission in intact brain tissue.

Mouse retina in vitro imaging

Glutamate is the predominant excitatory neurotransmitter in the mammalian retina²⁸. In the inner plexiform layer (IPL), ON-type bipolar cells release glutamate onto ON-type ganglion cells following light increments, whereas OFF-type bipolar cells release glutamate onto OFF-type ganglion cells following light decrements. The ON and OFF bipolar cell axon terminals stratify in separate halves of the IPL: ON bipolar terminals stratify in the proximal half, whereas OFF bipolar terminals stratify in the distal half²⁸. The known functional anatomy of glutamate release in the retina was used to test iGluSnFR responses to synaptic activity in an intact neural circuit.

iGluSnFR was expressed in ganglion cells²⁹ of adult mice (> 4 weeks) by viral transduction of AAV.hSynapsin.iGluSnFR. Retinas showed robust iGluSnFR expression in the IPL 14 days post-injection (Supplementary Fig. 21). 14-21 days post-injection, we targeted for whole-cell recording an iGluSnFR-expressing ganglion cell of the OFF-delta type (OFF-Sustained³⁰) (Fig. 4a). The voltage-clamped cell showed robust excitatory post-synaptic currents (EPSCs) in response to light stimulation (Fig. 4b), confirming that iGluSnFR does not interfere with excitatory synaptic transmission. Fluorescence responses simultaneously recorded from the dendritic arbor in the OFF stratum of the IPL (Fig. 4a) correlated with EPSCs during the dark phase of the stimulus (Fig. 4b), consistent with glutamate release from OFF bipolar cells.

To confirm that light-evoked fluorescence changes signaled glutamate release, we tested the effect of selectively blocking ON bipolar cells. ON bipolar cells express the sign-inverting mGluR6 receptor at their dendrites, whereas OFF bipolar cells instead express iGluRs²⁸ . An agonist of mGluR6 receptors, L-AP4, selectively hyperpolarizes ON bipolar cells, blocking their synaptic release, but does not block release from OFF bipolar cells²⁸. L-AP4 selectively blocked iGluSnFR signals in the ON stratum of the IPL, but not the OFF stratum, confirming that the iGluSnFR signal depends on glutamate release from bipolar cells (Fig. 4c).

In the presence of steady background light, neurites in the OFF stratum of the IPL showed sparse, spontaneous fluorescence events. The glutamate reuptake inhibitor DL-threo-β-benzyloxyaspartate (TBOA, 40 μM), which blocks EAAT1 and EAAT2³¹, increased both the amplitude and the decay time constant of these spontaneous fluorescence events (Fig. 4d,e). The increased response in TBOA is consistent with a reduced rate of glutamate removal from the synaptic cleft.

We tested whether iGluSnFR could be used to measure receptive fields of bipolar cells at the level of their synaptic outputs. In response to drifting sine-wave gratings of increasing spatial frequency (Fig. 4f), modulation of the iGluSnFR fluorescence peaked at ~6 cycles/mm. This tuning is consistent with surround inhibition at low spatial frequencies. At high frequencies (>10 cycles/mm), bright and dark stimulus regions cover the receptive field simultaneously and response modulation is reduced (Fig. 4f,g). These measurements are consistent with the known receptive field size of bipolar cells^32,33 .

In vivo glutamate and calcium signaling in C. elegans

To demonstrate the in vivo functionality of iGluSnFR, it was tested in three animal species: C. elegans, zebrafish, and mouse. The nematode C. elegans has 302 neurons, many of which are glutamatergic. The left and right AVA neurons are command interneurons located in the head lateral ganglia that play an essential role in the reverse locomotion circuit³⁴. AVA is post-synaptic to 40 other neurons³⁵. In the nerve ring, AVA makes only post-synaptic connections; nerve ring iGluSnFR signal from AVA should thus be exclusively post-synaptic. AVA expresses NMDA- and AMPA-type iGluRs as well as a glutamate-gated chloride channel³⁶, but appears not to express EAT-4, the primary vesicular glutamate transporter³⁷. Its role as an integrator of pre-synaptic signals and its lack of glutamatergic vesicles make it a useful neuron in which to test the ability of iGluSnFR to report post-synaptic glutamate.

We simultaneously monitored glutamate input into AVA, and AVA somatic calcium, using iGluSnFR and the recently developed red genetically encoded calcium indicator RCaMP^38,39, both under control of the rig-3 promoter. iGluSnFR expression was strongest along the neuronal process in the nerve ring, and RCaMP1e expression was strongest at the soma (Fig. 5a, b). Fluorescence from the neuronal process in the nerve ring was used to detect glutamate input (from iGluSnFR), and fluorescence from the cell body was used to detect calcium output (from RCaMP). We observed that iGluSnFR responses reliably preceded RCaMP responses (Fig. 5c), consistent with existing evidence that glutamate provides strong excitation to AVA, leading to reversal behavior^40-42.

A delay of a few seconds in the cell body RCaMP1e response likely reflects the indirect relationship between depolarization and calcium, as well as the low affinity and slow rise time of RCaMP1e, which produces a strong fluorescence response only above ~1-2 μM [Ca²⁺]³⁸. A simple integrator-type computational model shows a direct correlation between the presumed glutamatergic input measured with iGluSnFR, and the resulting calcium signal inferred by somatic RCaMP (Fig. 5c). A single leaky integrator yields a good fit to the experimental data in amplitude and relative timing (Supplementary Figs. 22 & 23). This relationship supports a circuit model in which glutamatergic input correlates both with the timing and the magnitude of subsequent AVA output, and is consistent with existing circuit diagram models of AVA and the surrounding nerve ring.

To demonstrate that the signal detected by iGluSnFR was specifically related to glutamate release, we examined eat-4(ky5) mutants, which are defective in the vesicular glutamate transporter that loads glutamate into synaptic vesicles. AVA calcium transients in eat-4 mutants were not accompanied by increased fluorescence of iGluSnFR (Fig. 5d, Supplementary Fig. 24), suggesting that iGluSnFR specifically detects a glutamate-associated signal, and not general AVA activity. It should be noted that eat-4 mutants exhibited a reduced number but normal amplitude of RCaMP calcium transients in AVA neurons, consistent with previous results demonstrating that excitatory glutamatergic transmission is one of several inputs that can activate AVA and backward movement^40-42.

In vivo imaging in fish and mice

In larval zebrafish, iGluSnFR reliably reported spatial organization of direction-selective synaptic activity in the optic tectum, consistent with but more rapid than GECIs⁴³ (Supplementary Figs. 25-27).

To evaluate iGluSnFR functionality in the intact rodent brain, we delivered it to layer V of motor cortex via infection of AAV.hSynapsin.iGluSnFR in juvenile mice (P20-30) (Fig. 6a). Two weeks after infection, two-photon imaging in vivo revealed densely packed, yet resolvable individual dendrites, which were clearly visible in the superficial layers (Fig. 6b). In post-fixed brain slices, confocal imaging in layer V revealed sparse labeling of cells, and individual somata displayed a ‘halo’-like expression pattern with iGluSnFR clearly peri-membrane (Supplementary Fig. 28). Neuropil was also strongly labeled.

Following iGluSnFR expression, head-fixing posts were implanted and mice were imaged under resting-awake state or running on a free-floating treadmill (Fig. 6c). Since individual dendrites are clearly resolvable under low-density labeling conditions, we tested whether glutamate transients could be detected in the apical tuft dendrites of layer V neurons of awake, behaving animals by two-photon microscopy in primary motor cortex (M1), in the region specific to forelimb representation (spatial coordinates taken from refs. (mixture of ketamine and xylazine) mice, we frequently observed spontaneous, repetitive 44,45). In anesthetized glutamate transients over hours (Supplementary Fig. 29), consistent with the observation by microdialysis that ketamine increases extracellular glutamate⁴⁶. In resting-awake animals, we observed numerous transient glutamate events with large (ΔF/F)_max (~0.2-0.4, Fig. 7c and Supplementary Fig. 30). Additionally, we detected glutamate transients within single dendritic spines over time during head-fixed running (Fig. 6c).

We next imaged iGluSnFR fluorescence in response to motor tasks. Head-fixed animals were subjected to forward/reverse running on a free-floating treadmill. Fast scanning of small cortical ROIs revealed robust and reliable glutamate responses during running (Fig. 6c,d). Peak iGluSnFR responses are restricted to small regions, whose size (~0.25 μm × 0.25 μm), shape and proximity to the dendritic shaft are indicative of spine heads (Supplementary Fig. 31). Moreover, iGluSnFR revealed dendritic segments containing multiple spines sensing many glutamate events during trials of forward running (Supplementary Fig. 31). Spines reporting glutamate events on these dendritic segments were clustered on the same branch, but their responses were uncoordinated. Consistent with the fast time-scale of neurotransmission, line scanning of a spine head and its associated dendritic shaft during running showed fast kinetics of a glutamate transient (only on the spine head) that lasted ~50-100 msec (Fig. 6d,e).

Glutamate events were either running-associated or running-direction-associated (i.e. reverse-only or forward-only; Fig. 6f). They typically commenced upon running onset and stopped upon running cessation (Fig. 6g). These glutamate events were highly reproducible during several running trials. Furthermore, we found that local application of tetrodotoxin via a small craniotomy (Supplementary Fig. 32) effectively blocked running-related glutamate transients along apical tuft dendrites (Fig. 6h). Intra-peritoneal injection of pilocarpine (a non-selective muscarinic receptor agonist) produced seizures in the mice, and doubled the amplitude of observed glutamate events (Supplementary Fig. 33). Thus, iGluSnFR provides a reliable means for investigating glutamate transients in single dendritic spines as well as dendritic branches in awake behaving mice.