Cl-out is a novel cooperative optogenetic tool for extruding chloride from neurons

Chloride regulation affects brain function in many ways, for instance, by dictating the GABAergic reversal potential, and thereby influencing neuronal excitability and spike timing. Consistent with this, there is increasing evidence implicating chloride in a range of neurological conditions. Investigations about these conditions, though, are made difficult by the limited range of tools available to manipulate chloride levels. In particular, there has been no way to actively remove chloride from neurons; we now describe an optogenetic strategy, ‘Cl-out', to do exactly this. Cl-out achieves its effect by the cooperative action of two different component opsins: the proton pump, Archaerhodopsin and a chloride channel opsin. The removal of chloride happens when both are activated together, using Archaerhodopsin as an optical voltage clamp to provide the driving force for chloride removal through the concurrently opened, chloride channels. We further show that this novel optogenetic strategy can reverse an in vitro epileptogenic phenotype.

N ormal brain function arises from a delicate, and continually fluctuating, interplay between excitatory and inhibitory synaptic forces, and is therefore very sensitive to anything that affects the balance between excitation and inhibition. Chloride regulation in neurons is particularly important in this respect, because the major forms of synaptic inhibition, including GABA A , GABA C and glycine receptors, are chloride-conducting channels [1][2][3] . Consequently, changes in intraneuronal chloride levels can markedly alter the synaptic effects of both GABA and glycine. Chloride balance further influences neuronal function by modulating transmitter loading into vesicles 4,5 , by affecting neuronal acid-base balance through interactions with bicarbonate 6,7 , and cell volume through osmotic effects 8 .
The level of intraneuronal chloride fluctuates widely both over time [9][10][11] and in different parts of the neuron 12,13 , suggesting that this may play a key role in dictating the functional state of the neuron. Very low levels of intraneuronal chloride means that the GABAergic events are strongly hyperpolarizing, and the inhibitory effect is very strong. Raised intraneuronal chloride, on the other hand, arising either by excess chloride entry or reduced clearance, shifts the GABAergic reversal potential (E GABA ) in a positive direction, and GABA can even become excitatory if E GABA exceeds action potential threshold. Such chloride-dependent disinhibition is thought to be a major factor in several brain disorders [14][15][16][17] , most notably epilepsy [18][19][20][21] .
The relative expression levels of two different cation-chloride co-transporters, KCC2 and NKCC1 (ref. 22), as well as local impermeant anions are reported to influence the baseline chloride balance 23 . At other times, intense synaptic bombardment, especially when there is concurrent glutamatergic and GABAergic activation 24 , can overwhelm the capacity of neurons to regulate chloride levels, and intraneuronal chloride rises sharply at these times 25,26 .
The true extent of these physiological and pathological dynamic changes, however, is not known, partly because we lack the tools for manipulating chloride. Optogenetics has introduced many new tools for regulating neuronal function 27 , including the ability to load chloride into neurons using the optogenetic chloride pump, Halorhodopsin 28 . We used this strategy to demonstrate how raised intraneuronal chloride leads to an acute increase in network excitability 29 and alterations in spiketiming, consistent with patterns of activity that have been recorded in epileptic animals 30,31 . Similarly, the specific KCC2 antagonist, VU0463271, also rapidly induces epileptiform discharges in brain slice preparations 32 . The influence of chloride balance on spike-timing, mediated through its effect on GABA A inhibition 29 , also has implications for a wide range of neuronal activity, including spike-time dependent plasticity, neuronal clustering and information transfer through networks with weak glutamatergic synaptic connections.
Reducing intraneuronal chloride, on the other hand, has proved difficult. As such, we only really have the means to manipulate intracellular chloride in an upwards direction. If we were able to reduce intracellular chloride, this would not only facilitate investigations into chloride regulation and its consequences, it may also provide a means of treating the various neurological conditions associated with raised chloride. The diuretic, bumetanide has some efficacy in this regard, by blocking NKCC1, and has shown promising antiepileptic action in animal models 33 . Its effect, however, is likely to be weak, acting only on chloride entry and not promoting chloride extrusion at all. A recent clinical trial to repurpose bumetanide for treatment of childhood epilepsies was terminated early because of otolaryngeal side-effects 34 . Significantly though, there are no means currently to remove chloride from neurons. We therefore decided to create one using optogenetics (Fig. 1), by coupling the hyperpolarizing proton pump, Archaerhodopsin (Arch) 35 , with one of the newly described optogenetic chloride channels 36,37 . We demonstrate the action of this new optogenetic strategy using dissociated neuronal cultures, and then provide a proof of principle of its network effect, using it to reverse a simple, in vitro model of epilepsy.

Results
Optogenetic manipulation of intraneuronal chloride levels. We first considered trying to reverse the orientation of Halorhodopsin so that it pumps chloride out of cells, but this strategy is problematic because the period of chloride extrusion would be accompanied by intense depolarization of the neurons. A second strategy presented itself with the publication of two different optogenetic chloride channels, ChloC 36 and the anion channel rhodopsins (ACRs) 37 . We reasoned that if we could control the membrane potential, then we would also have control over the direction of chloride movement through these channnels. We demonstrated this proof of principle by performing gramicidin perforated patch clamp recordings of dissociated cultured neurons, expressing either ChloC or GtACR. This allows a non-invasive measurement of GABAergic reversal potentials induced by brief puffs of the GABA A agonist muscimol delivered from a nearby micropipette. We measured E GABA before and after activation of the optogenetic chloride channels, while holding the cell at a relatively depolarized, or a hyperpolarized level. A voltage ramp was delivered close to the timing of the peak GABAergic conductance. In both cases, E GABA shifted towards the holding membrane potential: when the cells were held at À 50 mV, there was a positive shift in E GABA for both ChloC (avg ± s.e.m.: 2.69 ± 0.55 mV, n ¼ 6) and GtACR (7.93 ± 1.38 mV, n ¼ 5); when instead the cells were held at À 80 mV, there was a negative shift in E GABA (ChloC: À 1.93 ± 0.61 mV, n ¼ 5; GtACR2: À 3.43 ± 0.64 mV, n ¼ 5).
We then investigated whether the optogenetic proton pump, Arch, could provide the hyperpolarizing voltage clamp to drive chloride extrusion through a concurrently activated chloride channel. Since the effect depends critically on the Arch opsin altering the cell's membrane potential, these assessments could not be done using voltage clamp. Instead, we recorded cells in current clamp, again using gramicidin patch pipettes, and measured the shift in the inhibitory synaptic potential ('GABA À DV'), induced by a brief puff of muscimol, as a proxy for a direct, voltage-clamp measure of E GABA (Fig. 1c).
Co-transfection strategies, using separate vectors for the two opsins, produced a mixed population with many neurons only expressing one of the opsins, and only some expressing both. Recordings of these doubly-transfected neurons suggested that this strategy could be successful. Critically, however, we showed that cells expressing ChloC or GtACR in isolation tended to produce a small positive shift in GABA À DV (Fig. 1d), so any transfection strategy resulting in networks where significant numbers of singly-transfected neurons expressing these channels are likely to produce deleterious effects, rather than improvements in network behaviour. We therefore incorporated the expression of both opsin elements into a single-expression system. We used two different co-expression methodologies linking two different pairs of opsins, to yield four different constructs in total (Fig. 2). The first co-expression method used a 'self-cleaving' T2A linker peptide, derived from Thosea signa virus, which allows two separated proteins to be expressed under a single promoter 38 . The second method used a b-HK linker sequence 39 , to create a single-fusion protein by concatenating the two opsins. Because the function of these paired expression systems depend on the cooperative action of both opsin components, we term these new proteins 'co-opsins', and this specific co-opsin, we have called 'Cl-out', named for its cellular effect of removing chloride from the cytosol.
Co-opsin expression and function. The constructs produced good membrane localization (Supplementary figure 2), without affecting either baseline input resistance (Cl-out4 constructs; non-transfected cells, n ¼ 5, R N ¼ 1.30±0.06 MO; transfected cells, n ¼ 5, R N ¼ 1.74 ± 0.26 MO; p¼ 0.176, NS.), or E GABA (non-transfected cells, n ¼ 5, E GABA ¼ À 51.8±2.8 mV; transfected cells, n ¼ 5, E GABA ¼ À 47.2 ± 2.8 mV; P ¼ 0.249, NS). Importantly, the distinct functions of both individual optogenetic elements were readily seen in whole-cell patch clamp recordings using a high chloride electrode filling solution (Fig. 2b). Blue (488 nm) illumination activates preferentially the chloride channel, producing a strong depolarization in cells filled with chloride, whereas green illumination (561 nm) preferentially activates Arch leading to a hyperpolarization. When both opsins were simultaneously activated, in cells recorded through gramicidin perforated patch clamp, it induced a rapid shift in shape of the muscimol-triggered post-synaptic potential ( Fig. 2c-i). Notably some recordings even showed a reversal of the sign of the post-synaptic potential (see Fig. 4a, lower trace), demonstrating that the change is unlikely to be explained by a change in conductance, but rather by a change in E GABA . This was further confirmed by voltage-clamp ramp recordings, showing no change in the slope (Supplementary figure 3; percentage conductance change ¼ À 3.5±2.5%, n ¼ 3). We concluded therefore that Cl-out activation produces its effect on GABAergic events by reducing intracellular chloride levels. In contrast, activation of the individual opsins singly had little or no effect on subsequent GABAergic events (Fig. 1c, Fig. 2c-ii).
Following the end of the illumination, the neurons slowly accumulated chloride once more, but with a substantially longer time constant (VU0463271 bathed neurons, t ¼ 37.5 ± 6.7 s; n ¼ 7; Fig. 4) than of previous measures of the correction of optogenetically altered chloride levels (t ¼ 8-12 s) 28,29 . In cells bathed additionally in bumetanide (10 mM to block NKCC1), the recovery was significantly slower (t ¼ 58.2 ± 11.5 s; n ¼ 7; Po0.025; Fig. 4b,c), an effect that was reversed by washing out bumetanide (t ¼ 33.1 ± 5.1 s; n ¼ 5). These results suggest both NKCC1 and KCC2 contribute significantly to chloride movements across the cell membrane at rest. Cl-out can persistently reduce network excitability. We next examined whether we could use this approach to modulate the activity patterns in large networks of neurons simultaneously (Fig. 5). We incorporated the Cl-out1 and Cl-out4 constructs under the pan-neuronal, hSyn promoter, together with an enhanced yellow fluorescent protein (EYFP) marker sequence, into adeno-associated viral vectors, and injected the virus into the left cerebral ventricle of newborn mouse pups. Both vectors achieved in vivo expression, but levels of Cl-out1 were far better than for Cl-out4, so subsequent network modulation experiments were performed only for Cl-out1 constructs. Animals were allowed to grow to maturity, and at between postnatal 2-4 months age, they were killed to prepare coronal brain slices. We selected brain slices showing evidence of significant EYFP labelling in the hippocampal formation, and recorded extracellular activity using Neuronexus probes in the CA1 pyramidal layer. We recorded baseline periods, before introducing the KCC2 blocker, VU0463271 (10 mM) into the bathing media. As reported previously 32 , VU0463271 perfusion induced a marked increase in the network excitability, with the appearance of large spontaneous network discharges and also a much larger response to electrical stimulation (Fig. 5b). We then delivered a series of 25 s illuminations, at 30 s intervals (83% duty cycle), using both blue (488 nm) and green light (561 nm) to activate the expressed Cl-out1 co-opsin. We tested network excitability in each cycle of illumination, during the 5 s dark period, following the same protocol described previously 29 . In all cases, this rapidly reversed the hyperexcitable state back to baseline levels, as measured both by the near abolition of spontaneous events ( Fig. 5c; normalized frequency following light activation ¼ 15.3 ± 2.8%; highly significant difference from 100%, Po o0.001) and the intensity of the responses to electrical stimulation ( Fig. 5d; Normalized power, VU, 38.4±12.4; VU þ light, 6.1±2.1; Po0.05), even in the continued presence of perfusing VU0463271. We next asked whether these effects could be achieved by activating the individual component opsins (Fig. 6). For the ArchT experiments, we used animals transfected only with that construct (Fig. 6a). For the ChloC experiments, we used Cl-out1 transfections, utilizing the separation of spectral sensitivities of   the two opsins (see Fig. 2b), and contrasted the effects of illuminating only with blue light with combined blue/green illumination (Fig. 6b). For both opsins individually, the intensity of the response to electrical stimulation following illumination tended to increase, and substantially so in two of the ArchT experiments (Fig. 6c). These experiments confirmed that the restorative action of Cl-out1 in reducing network excitability was indeed due to the cooperative action of the two component opsins.
Once the illumination was terminated, the hyperexcitable state was re-established, but only slowly, over many tens of seconds (example traces in Figs 5b and 6b; pooled data in Supplementary  figure 4), consistent with the cellular time constants shown in Fig. 4. This shows that the inhibitory benefit of Cl-out far outlasts its activation, consistent with our cellular measures (Fig. 4). This property represents a significant advance in our optogenetic, antiepileptic armoury because it is directed at reinvigorating the brain's own endogenous protective mechanism, which previous work has shown to be extremely powerful 40,41 , an effect that long outlasts the period of optogenetic activation.
Finally, we examined whether manipulating intraneuronal chloride levels in this way could influence spike timing, because there is extensive evidence that spike time is strongly regulated by GABA A activation 42,43 . We tested, using multiunit recordings, whether Cl-out activation changed the entrainment of spiking to the dominant field oscillation, in the opposite direction to the effect shown previously, when we drove chloride into neurons using Halorhodopsin 29 . We first applied VU0463271, to load chloride generally into neurons, and then assessed spike entrainment before and after Cl-out activation. These experiments showed a marked sharpening of the entrainment of spikes, as indicated by a significant reduction in the half-width index, which is a measure of the skew of the histogram towards a single peak in the oscillation. This reduction only occurred with co-activation of both opsins together, and not with activation of either ArchT or ChloC alone (Supplementary figure 5).

Discussion
We have developed a novel approach to using optogenetics, in which two opsins are used cooperatively to achieve the desired alteration of neuronal function. Whilst others have expressed different optogenetic proteins in single cells before 39 , these previous studies have done so to provide different optogenetic effects brought about by different colour light; their aim was always to activate one or other protein, and not both together, within the constraints offered by their spectral sensitivities. In contrast, the effect we sought was dependent on the cooperative action of both optogenetic elements. This is the first utilization of optogenetics to provide a voltage-clamp effect to move ions unidirectionally through a channel, and in this way, provides the first means to actively extrude chloride from neurons. The dependence of the Cl-out effect on Arch pumping a proton out of the cell may conceivably alter the pH balance slightly, increasing the intracellular pH relative to outside. This, though, is likely to be minimized by the presence of intracellular carbonic anhydrase and the extracellular pH buffers. It is important to note, however, that such a pH shift would make the bicarbonate reversal potential marginally more positive (that is, the opposite of the E GABA shift we actually measured), meaning that the change we measure cannot be ascribed to any pH change, a point that is further proved by our control experiments using Arch activation alone.
Our development complements our recent demonstration that chloride can be driven into neurons acutely using Halorhodopsin 28,29 , so that now we have the facility for bidirectional control over chloride levels. Currently, this could not be achieved in the same cells, because the spectral activation curves of Halorhodopsin and the Arch component of the Cl-out constructs overlap. So the potential is only realized for one strategy in a given cell, but the co-opsin strategy that we describe here will lend itself to other combinations of opsins with shifted spectral sensitivities as they become available. In this regard, it should also be noted that combining GtACR1 with ArchT might yield a co-opsin which could be activated by a single wavelength, but the rather slow kinetics of GtACR1 would then present other problems regarding the temporal matching of the co-opsin effect. Research using these new tools for manipulating chloride will also benefit from developments in imaging chloride levels in cells [44][45][46] , and has the potential to usher in a new era of research into the role of chloride in neurons. As we have mentioned, chloride levels are highly variable in different cells and across different times, and given what we know about the consequences for neuronal function, it seems highly likely that chloride regulation is an important mechanism for dictating neuronal, network and brain states. In particular, by altering the strength of GABAergic control, changes in intraneuronal chloride is hypothesized to influence spike timing and jitter, and this theory can now be tested. Our new tool also has the potential to correct chloride loading in neurons, a common feature of epileptic networks. Accordingly, we have provided a demonstration that Cl-out can reverse an epileptic phenotype in vitro, and also alter spiketiming, paving the way for future explorations of its effect on various spike-time dependent mechanisms, and also on pathological processes, in more realistic animal models and potentially ultimately in man. The same data shown as histograms of the different phases of the cycle (note that the bins from À 2p to -p are replicated (0-p)). Pooled data (6 slices from 4 animals) showed significant shifts in both the number of spikes (c) and the half-width index (d), which is a measure of the entrainment of the spiking to the local field potential 29 . NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13495 ARTICLE Methods Animal Regulation. All animal handling and experimentation were done according to UK Home Office guidelines, according to the requirements of the United Kingdom Animals (Scientific Procedures) Act 1986. rAAV cloning and production. The different vectors were based on pAAV-hSyn-ChloC-2A-tDimer (purchased from Addgene #52495). To construct pAAV-hSyn-ChloC-2a-ArchT-GFP (Cl-out 1), ArchT-GFP was isolated from pAAV-CAG-ArchT-GFP (purchased from Addgene #29777) and subcloned into pAAV-hSyn-ChloC-2A-tDimer after removal of tDimer component using the BamHI and HindIII restriction sites. For the linked protein (pAAV-hSyn-ChloC-EYFP-bHK-Arch), EYFP-bHK-Arch cDNA was custom synthesized by GeneArt Company and subcloned into the vector using KpnI and HindIII restriction sites, excluding the 2A component. GtACR2 cDNA was then synthesized by GeneArt. It was subcloned into the pAAV-hSyn-ChloC-2a-ArchT-GFP and pAAV-hSyn-ChloC-EYFP-bHK-Arch vectors to replace ChloC, using EcoRI and KpnI restriction sites. Sequence confirmation was outsourced (GATC Biotech). All vectors were amplified using recombinant deficient strain of E.Coli (OneShot Stbl3, Invitrogen) and purified using EndoFree Maxi Kit (QIAGEN #12362). Viral vectors were packaged as AAV serotype 2 by University of North Carolina vector core.
Dissociated neuronal culture. Assessment of the chloride-extrusion effect of the different proteins was performed using dissociated neuronal cultures. Primary neuronal cultures were prepared from rat pups at embryonic day 18-20 in the following way. A pregnant Sprague-Dawley rat was killed by cervical dislocation and a sagittal incision was made in the abdominal area to remove the pups. The neocortex and hippocampal tissue was isolated from the pups, and digested using papain enzyme (Sigma Aldrich) for 40 min. Cells were then dissociated within the growing medium (Neurobasal A, 2% B-27 supplement, 1% of fetal bovine serum, 0.5% glutamate, and 0.5% antibiotic-antimicotic, GIBCO Invitrogen) by gentle agitation using a pipette gun with a 10 ml tip. Transfection was done by electroporation using Rat Neuron Nucleofactor kit (Lonza Amaxa). For each electroporation, B5 million cells were collected by centrifugation (80 relative centrifugal force) for 5 min. Cell pellets were re-suspended in 100 ml nucleofactor solution and mixed with 1 mg of plasmid DNA. Cells were electroporated using an electroporation machine (Amaxa; programme G-013). Cells were then plated on pre-coated coverslips using poly-L-lysine with 1:10 dilution. Medium was changed 3 and 24 h post-plating. All experiments were performed after DIV 10.
Perforated gramicidin patch. Gramicidin perforated patch recordings were made using 3-7 MO pipettes (borosilicate glass; Harvard Apparatus) filled with a high Cl À electrode filling solution. Fresh Gramicidin (Sigma Aldrich, G5002) stock was made daily with 5 mg ml À 1 concentration in dimethylsulfoxide. Gramicidin stock was added to the high Cl À electrode filling solution to achieve a final concentration of 0.1 mg ml À 1 and mixed by 40 s vortexing and 5 s sonication. Gramicidin electrode filling solution was then filtered using a 0.45 millex pore filter and used for patching straight away. Experiments were performed when series resistance had stabilized below 100 MO, which was typically achieved B30 mins after patching. The series resistance was required to be stable (o10% fluctuation) during the time course of data collection.
For voltage-clamp measurements of E GABA (Fig. 1a), recordings were made in voltage-clamp mode, with repeated 200 ms duration test ramps (triangular up-down function; peak À 50 mV; trough À 90 mV; slope±400 mV s À 1 ) applied at baseline and also at close to the peak of GABAergic current (I GABA ). All other assays of E GABA were made in current clamp mode, because we were assessing functionality of different Cl-out constructs in which the effect is critically dependent on a voltage shift provided by Arch. For this reason, voltage clamping the cells would have abolished the effect. We therefore measured the average shift in GABAergic inhibitory post-synaptic potentials (IPSPs) induced by muscimol (100 mM) application delivered close to the recorded neuron, from a patch pipette coupled to a picospritzer (Parker Instrumentation, 10 ms pressure pulses,10-20 PSI)(see Fig. 1c). IPSPs were measured in the presence of 2 mM kynurenic acid, to block all glutamatergic synaptic events. The timing of the puff was coordinated with the illumination (optogenetic activation) via the pClamp software coupled to a Digitimer box. The shift in IPSP was calculated from averages of the membrane potential over 200 ms epochs, centred on the peak of the muscimol induced conductance, for pairs of events before and after activation of the Cl-out constructs. This measure reflects the shift in E GABA , but is not a direct measure, and so we refer to it as GABA À DV, to make this distinction clear. If action potentials occurred, these were curtailed at the peak of GABA event, before the measurement.
Optogenetic expression. For general neuronal expression of the optogenetic proteins, viral vectors (pAAV-hSyn-ChloC-2a-AchT-GFP manufactured by UNC viral vector core using plasmid constructs provided by ourselves; rAAV5/ CAG-ArchT-TdTomato (UNC)) were injected into wild-type mice (C57BL/6 J). Injections were made into postnatal 0-day-old pups. The pups had local anaesthetic (EMLA, AstraZeneca) cream applied to the left top of their head, and were anesthetized subsequently using isofluorane inhalation. A single injection of virus was made using a 10 mL Hamilton syringe, with a bevelled 36 gauge needle (World Precision Instruments) B1 mm anterior to lambda, and 1 mm lateral to the midline into the left hemisphere at 1.7-0.8 mm deep to the skin (four separate 200 nl injections, deepest first). Approximately 0.8 ml (B10 11 -10 12 viral particles) was injected over a 12 min period. Animals were allowed to recover. Subsequent electrophysiological recordings were made between the ages of 5 weeks to 3 months.
Brain slice experiments. Coronal brain slices (350 mm for patch clamp experiments or 400 mm for network activity experiments) were prepared from the injected animals, at age 2-4 months (male), on ice-cold oxygenated (95% O 2 /5% CO 2 ) artificial cerebrospinal fluid (ACSF: 125 mM NaCl, 26 mM NaHCO 3, 10 mM glucose, 3.5 mM KCl, 1.26 mM NaH 2 PO 4 , 3 mM MgCl 2 ). After cutting, the slices were transferred to incubation, submerged or interface chamber (room temperature) perfused with oxygenated ACSF (ACSF: 125 mM NaCl, 26 mM NaHCO 3, 10 mM glucose, 3.5 mM KCl, 1.26 mM NaH 2 PO 4 , 2 mM CaCl 2 , 1 mM MgCl 2 ) for at least 1 hour before transfer to a recording interface chamber (33-36°C), after which time they perfused with slightly lower Ca 2 þ ACSF (1.2 mM CaCl 2 ; ACSF otherwise as before). The divalent cation concentration followed the protocol of Sanchez-Vivez and McCormick (1999). In some experiments, cation-chloride cotransporters were blocked using 10 mM of VU0463271 (Tocris) to block KCC2, and 10 mM bumetanide (Tocris) to block NaKCC1 (minimum drug incubation time of 20 min). Extracellular recordings were made using tetrodes in a diagonal arrangement separated by 50-100 mm (Neuronexus), connected to a 1401-5 Analog-Digital converter (Cambridge Electronic Design, Cambridge) and using Spike2 software (Cambridge Electronic Design) running on a desktop computer (Dell). Multichannel extracellular recordings were digitized at 20 kHz for high frequency bandpass (300-1,000 Hz) and 5 kHz for low-frequency bandpass (1-300 Hz), using 16 linear 4-channel-probes in diamond configuration (Q1x1-tet 'tetrode'; NeuroNexus) connected to a ME16-FAI-mPA-system and MC_Rack software (Multichannel Systems, Reutlingen). The tetrodes were mounted on patchmaster micromanipulators, and recordings were made from the pyramidal cell layer in CA1. Bipolar electrical stimulation (0.1 ms duration) was delivered via tungsten electrodes onto the white matter. Optogenetic proteins were activated by 561 nm, 50 mW solid-state laser (Cobolt) connected to a fibre optic with a fine cannula (400 mm core, 0.20 NA; Thorlabs; after optimizing the optic fibre coupling, the unfiltered light at the cannula tip was measured at 15-30 mW) synchronous with blue epi-fluorescence illumination (460 ± 20 nm excitation filter) controlled using an electrically-coupled, mechanical shutter (Sutter Instruments), and delivered through a 4 Â air objective (0.28 NA, Nikon). The central illumination of the epifluorescent blue illumination was B10 mW mm À 2 .
Code availabilty. Analyses were done using in-house software (available on request) implemented on either Igor (Wavemetrics) or Matlab. Sample sizes were chosen based on the large effect sizes found in our previous studies using Halorhodopsin to increase intracellular Cl À (ref. 29). Recordings and analyses were not blinded.
Data availability. The data that support the findings of this study are available from the corresponding author on reasonable request.