We first created a fusion protein comprising the mammalian codon-optimized form of N. pharaonis
halorhodopsin (Halo), with EGFP added in-frame at the C-terminus for ease of visualization (see Methods
for details). When expressed in cultured hippocampal neurons using the CaMKII promoter, which targets excitatory neurons of the forebrain 
, Halo-EGFP fluoresced brightly, and appeared evenly distributed around the neuron (). When exposed to ~10 mW/mm2
yellow light (from a xenon lamp, filtered by a standard Texas red excitation filter from Chroma, bandpass 560±27.5 nm), voltage-clamped hippocampal neurons expressing Halo experienced outward currents with rapid onset, stable steady-state amplitude, and abrupt shut-off after cessation of illumination. No supplementation of culture or recording media with the essential halorhodopsin cofactor all-trans retinal was necessary for strong currents to be elicited, consistent with prior work that demonstrated high enough levels of all-trans retinal in mammalian culture and brain to enable type I opsin functionality 
. Light pulses elicited pulse amplitudes of 88.7±32.8 pA (mean±st. dev.; n
22 neurons; ). Repeating a 1-second pulse of yellow light twice, spaced by 1 second of darkness, resulted in similar pulse amplitudes each time (p
>0.50, paired t-test), although during each light pulse, a slight perceptible decay was visible (analyzed later in the manuscript). This relatively stable current amplitude is consistent with what is known about the halorhodopsin photocycle, which can fully complete within tens of milliseconds 
. The light-elicited current amplitude did not vary significantly with holding voltage when assayed at-70 mV, −30 mV, and+10 mV (F
>0.95, ANOVA with factor of holding voltage), nor did any measured kinetic parameters vary across this voltage range, such as onset or offset times of current pulses (F<0.6, p
>0.55 for all comparisons, ANOVA; ). The onset and offset times of elicited currents were strikingly rapid, ~10–15 ms at all holding voltages tested (). When held in current clamp, hippocampal neurons underwent peak hyperpolarizations of 32.9±14.4 mV (n
19 neurons) in response to pulses of yellow light, with no difference between the peak hyperpolarizations achieved by two 1-second pulses separated by a 1-second pause (p
>0.85, paired t-test; ). Furthermore, as expected from the current-clamp experiments, these large voltage changes were quite rapid, with onset and offset times of 68±57 and 73±39 ms respectively for these large voltage swings (). Thus, Halo was capable of reliably mediating hyperpolarizations of large magnitude, with fast onset and offset times at the beginning and end of light exposure.
Millisecond-timescale Halo-mediated neuronal hyperpolarization, elicited by pulses of yellow light.
Since it is important to evaluate whether a new technology has unanticipated side effects, such as altering basal cell physiology, or increasing the propensity for cell death, we conducted several control experiments. First, we characterized the basal state of Halo-expressing neurons electrophysiologically when no light was present. When measured in darkness, there was no difference in the resting potentials of neurons expressing Halo, and those of neighboring wild-type neurons (p
>0.20, t-test; n
19 Halo-expressing cells, n
19 wild-type cells; ). Similarly, membrane resistance was not significantly different between the Halo-expressing cells and the wild-type cells (p
>0.70; ). This result suggests that basal neural activity would be little affected by the presence of Halo in the absence of light. As an independent assay for unanticipated effects on cell health, we assayed whether Halo expression for one week in cultured hippocampal neurons could lead to apoptosis, using the membrane-impermeant DNA stain ethidium homodimer-1 to detect any cell membrane breakdown that would accompany apoptotic cell death 
. We found no difference in cell death between Halo-expressing and control wild-type neurons: 16/308 (5.2%) control neurons counted, and 1/22 (4.5%) Halo-expressing neurons counted, were labeled by ethidium homodimer-1, indicating that Halo was not toxic over the timecourse of a week of expression (χ2
>0.85; ). Thus, along multiple axes, Halo proved to be well-tolerated by mammalian neurons, perhaps as expected given its structural similarity to the well-tolerated photostimulation protein Channelrhodopsin-2 
Safety of Halo in cultured hippocampal neurons.
We next probed whether the fast response times of Halo could support naturalistic sequences of hyperpolarization events, in response to trains of brief pulses of yellow light. shows three traces of hyperpolarization events elicited in a single neuron, resulting from repeatedly playing back a Poisson train (mean inter-pulse interval, λ
100 ms, 59 pulses), of 10 ms-duration yellow light pulses, to illustratively simulate stochastic inhibitory (e.g., GABAergic) synaptic input. shows three such hyperpolarization traces, taken from different neurons. The variability of such trains was remarkably low in many regards–across ten repeated trials in a single cell, across multiple cells (n
5 neurons), and over time throughout a sustained train of 59 pulses (). Notably, we found that for hyperpolarizations elicited by 10 ms-duration light pulses during a λ
100 ms Poisson train, the mean amplitude was −4.56 mV (averaged across trials and neurons), but the trial-to-trial standard deviation of this amplitude was only 400 µV (averaged across neurons, and , left side). The trial-to-trial jitter of the time the hyperpolarization took to reach its peak value was also small, 1.27 ms (averaged across neurons, and , left side). The neuron-to-neuron variability of amplitude and timing was somewhat larger than the trial-to-trial variability, with standard deviations of 1.45 mV and 1.78 ms, respectively, but these values nevertheless demonstrated that precise inhibitory control of a population of neurons could take place with millivolt and millisecond resolution. Finally, we quantitatively examined the through-train sustainability of light-elicited voltage changes, by comparing the amplitude mean and amplitude variability, and timing variability, between the hyperpolarization events elicited by the first five light pulses in the 59-pulse train, and the last five light pulses in the train (). No difference was seen for any of these statistics, compared between the events elicited at the beginning vs. end of the train (p
>0.10 for all measures, t-test). Identical conclusions held for the λ
200 ms Poisson train with 46 pulses (, and ). The high temporal and amplitude fidelity of Halo-mediated hyperpolarizations suggests that Halo may be an ideal tool for simulating many forms of synaptic inhibition.
Halo-mediated naturalistic trains of inhibitory events.
We next analyzed the ability of Halo to enable rapidly-inducible and rapidly-reversible silencing of neural activity. In patch clamped neurons, we delivered trains of intracellularly-injected somatic currents (~300 pA, lasting 4 ms each), causing the neurons to fire 20 action potentials at 5 Hz with a 100% success rate (, “I-injection”). We then scheduled yellow light pulses to occur at certain phases within the somatic current-injection protocol–specifically, during the times when spikes 7 through 11, and spike 17, would normally be driven (, “Light”). Finally, we presented the light pulses and the somatic current injections simultaneously (, “I-injection±light”, three trials shown), and discovered that strikingly, spiking was blocked precisely during the periods of yellow light exposure, and at no other times. Most remarkably, the rapid onset and offset kinetics of Halo allowed the deletion of even single spikes
–specifically, the second yellow light pulse, timed to silence just spike 17, was able to eliminate spike 17 without affecting neighboring spikes 16 or 18 at all. We repeated this experiment five times, on each of n
6 neurons (). Across all these experiments, the second pulse of yellow light reduced the probability of firing spike 17 to 3.3%, whereas neighboring spikes 16 and 18 fired 96.7% of the time; this latter success probability was not significantly different from the success rate of the first spike in the train, before any light exposure at all (χ2
>0.30). In total, during periods when the yellow light was off, somatic current pulses elicited spikes 98.7% of the time, whereas during periods when the yellow light was on, somatic current pulses elicited spikes only 1.2% of the time. The temporal precision of Halo in silencing spikes therefore offers a novel method of creating ultratransient, precise, and effective inhibition of activity in genetically-specified neurons.
Halo-mediated silencing of neuronal spiking.
Halo-mediated currents and hyperpolarizations appeared to decay slightly during 1-second long pulses of yellow light (), raising the question of how Halo might perform during longer-duration neural silencing experiments. We exposed Halo-expressing neurons to continuous yellow light, and found that the hyperpolarization decayed with a time constant of approximately 16.8±10.4 seconds (n
8 neurons; representative trace shown in ). We tested for recovery of Halo function after 15 seconds of yellow light, with repeated 1-second test pulses of yellow light delivered every 30 seconds, and found that even after four such rest periods, the hyperpolarization peak amplitude remained down by 33% from its original peak amplitude, suggesting that the Halo protein had entered an inactive state ( and , black dots). Earlier studies on halorhodopsin from another archaebacterium, Halobacterium halobium
, found evidence for a similar type of rundown, due to accumulation of an inactive 13-cis
retinal form of the halorhodopsin molecule, which required seconds to recover to the original state 
. However, these earlier studies also found that brief periods of blue light could facilitate rapid recovery of H. halobium
halorhodopsin to the active form, by assisting in the re-isomerization of 13-cis retinal to the all-trans form. For Halo, we found that a brief pulse of moderate-intensity blue light (~10 mW/mm2
400 ms-long), delivered through the GFP excitation filter) could completely restore Halo to its active state (), recovering the hyperpolarizations elicited by the test pulses of yellow light to their original amplitude (p
>0.05 for each test pulse, paired t-test; n
8; , open blue dots). Thus, despite the existence of rundown during long-lasting exposure of Halo-expressing neurons to yellow light, brief periods of blue light exposure will facilitate the optimal performance of Halo during long periods of neural inhibition in vivo
Blue light facilitates optimal Halo function.
We recently showed that one member of the type I opsin family, the light-activated cation channel Channelrhodopsin-2 (ChR2), could mediate neural excitation in response to pulses of blue light (~470 nm, delivered through a standard GFP excitation filter) 
. Furthermore, the excitation peak of ChR2 is spectrally separated far enough from Halo that commonly-available fluorescence excitation filters (e.g., HQ450/50x and HQ590/55×) may be sufficient to enable neural photostimulation and photoinhibition (illustrated in ). We targeted individual neurons simultaneously with both Halo and the human codon-optimized form of ChR2 (with the red fluorophore mCherry 
fused to the C-terminus, for ease of visualization), both under control of the CaMKII promoter (). mCherry was visualized with the same Texas red filter used to stimulate Halo (see 
for detailed spectral properties). Note that due to spectral overlap, mCherry visualization (with yellow light) can result in Halo stimulation, and GFP visualization (with blue light) can result in ChR2 stimulation, so from a practical standpoint using the ChR2/Halo system as here configured requires cell identification and voltage control to occur during distinct experimental stages. We found that such neurons could respond to rapidly-switched pulses of yellow and blue light with hyperpolarizations and depolarizations respectively (). Poisson trains (mean inter-pulse interval λ
100 ms) of rapidly-alternating yellow and blue light pulses elicited rapidly-alternating hyperpolarizations and depolarizations in the same neuron (). The ability to drive excitation and inhibition in the same neuron, using two different wavelengths of light, may open up the ability to answer questions for which no current technology permits resolution.
Bi-directional optical control of voltage with blue and yellow light pulses.
One prominent example of such a long-standing question is whether synchrony or precise timing of neural activity has an important or causal role in neural computation. Such activity has been observed during, or associated with, various brain functions such as timing-dependent plasticity, global stimulus feature processing, visuomotor integration, motor planning, and attention (e.g., 
), and abnormal patterns of neural synchrony have been associated with a variety of neurological and psychiatric disorders 
. To date, however, no generalized strategy has permitted the disruption of neural synchrony or precisely-timed spiking without the confound of altering spike rate. To examine whether Halo and ChR2 together could enable this kind of experiment, we generated precisely-timed spike trains by repeatedly injecting cultured hippocampal neurons with a single filtered Gaussian white noise current trace (see , top, for a fragment thereof; see Methods
for details). Such noisy currents had been previously found to induce reliable spike trains in current-clamped neurons embedded in cortical brain slices 
. On some trials, we concurrently illuminated the neuron with the Poisson train of alternating yellow and blue light pulses shown in (see , bottom). We found that when delivered alone, filtered Gaussian white noise currents indeed induced reliably-timed spike trains (see , top, for twenty overlaid traces, and , top, for corresponding spike rasters). When Halo and ChR2 were concurrently driven by a Poisson train of yellow and blue light pulses, the spike timings were different from those of spikes elicited in darkness, but were still reliable from trial to trial (see , bottom, and , bottom, for overlaid traces and spike rasters respectively). Inspection of spike histograms () clearly shows that relative to the spike train elicited by current injection in darkness, optically driving Halo and ChR2 could sometimes abolish previously reliable spikes, create new spikes, or advance or delay the timing of specific spikes. We compared mean spike rates for neurons receiving filtered Gaussian white noise current injection alone vs. with additional illumination, and found no difference in spike rates for these two conditions (p
>0.90, t-test; n
7 neurons; ), indicating that our optical intervention preserved spike rate. However, precise spike timing was altered significantly: cross-correlations of Gaussian current injections delivered alone vs. delivered with concurrent illumination resulted in zero-lag cross-correlations that were on average 37% smaller than cross-correlations of pairs of spike traces resulting from Gaussian current injections alone (p
<0.005, see Methods
for details of this analysis). This indicates that precise spike timing was indeed disrupted by the activation of Halo and ChR2, even while spike rate was preserved. Thus, Halo and ChR2 are synergistic reagents, together enabling two-color controlled excitation and inhibition of individual, genetically-specified neurons. The ChR2/Halo system constitutes an easy-to-use, yet powerful, optical toolbox for the analysis and engineering of genetically-specified circuit elements in the brain, and also opens up the bidirectional optical voltage control of a variety of electrically excitable cells.
Multichannel optical disruption of precise spike timing, without alteration of spike rate.