Informational Lesions: Optical Perturbation of Spike Timing and Neural Synchrony Via Microbial Opsin Gene Fusions

Synchronous neural activity occurs throughout the brain in association with normal and pathological brain functions. Despite theoretical work exploring how such neural coordination might facilitate neural computation and be corrupted in disease states, it has proven difficult to test experimentally the causal role of synchrony in such phenomena. Attempts to manipulate neural synchrony often alter other features of neural activity such as firing rate. Here we evaluate a single gene which encodes for the blue-light gated cation channel channelrhodopsin-2 and the yellow-light driven chloride pump halorhodopsin from Natronobacterium pharaonis, linked by a ‘self-cleaving’ 2A peptide. This fusion enables proportional expression of both opsins, sensitizing neurons to being bi-directionally controlled with blue and yellow light, facilitating proportional optical spike insertion and deletion upon delivery of trains of precisely-timed blue and yellow light pulses. Such approaches may enable more detailed explorations of the causal role of specific features of the neural code.


INTRODUCTION
It has long been debated to what extent synchronous or preciselytimed neural activity contributes to neural computation and behavior. Synchronous neural activity within or between brain regions has been observed during, or associated with, many brain functions including timing-dependent plasticity, global stimulus feature processing, visuomotor integration, emotion, working memory, motor planning, and attention (e.g., Gray et al., 1989;Roelfsema et al., 1997;Brivanlou et al., 1998;Donoghue et al., 1998;Steinmetz et al., 2000;Fries et al., 2001;Tallon-Baudry et al., 2001;Froemke and Dan, 2002;Perez-Orive et al., 2002;Seidenbecher et al., 2003;Courtemanche and Lamarre, 2004;Buschman and Miller, 2007), as measured with multielectrode recording, electroencephalography (EEG), magnetoencephalography (MEG), and local fi eld potential (LFP) analysis. Furthermore, abnormal patterns of neural synchrony have been associated with a variety of neurological and psychiatric disorders such as Parkinson's disease, epilepsy, autism, and schizophrenia (e.g., Bragin et al., 2002;Uhlhaas and Singer, 2006;Berendse and Stam, 2007;Brown, 2007;Huguenard and McCormick, 2007;Orekhova et al., 2007;Gonzalez-Burgos and Lewis, 2008;Roopun et al., 2009). Computationally, synchrony has been implicated in multiple processes, including grouping neurons into 'cell assemblies' that can more effectively represent information to downstream neural networks, acting as a clock for gating or multiplexing information, coordinating information fl ow within small neural networks, selecting stimuli for attention, and performing pattern recognition (e.g., Hopfi eld and Brody, 2001;Friedrich et al., 2004;Tiesinga and Sejnowski, 2004;Borgers et al., 2005Borgers et al., , 2008Sohal et al., 2009). However, determining the causal Informational lesions: optical perturbation of spike timing and neural synchrony via microbial opsin gene fusions and Halo activation are separated by over 100 nm (Figure 1A), making it possible to express both molecules in the same neuron ( Figure 1B), thus conferring bi-directional control of the voltage of that neuron by blue and yellow light (Figures 1C,D) (Han and Boyden, 2007). As a consequence of this two-color bidirectional voltage control, precisely-timed blue and yellow light pulse trains, delivered to a neuron expressing both ChR2 and Halo, can be used to insert spikes into, and delete spikes from, the ongoing activity experienced by that neuron (Figure 1E), resulting in precise perturbations of spike timing without changing overall spike rate (Figures 1F,G). However, the delivery of two separate genes to precisely the same cell population can be a daunting proposition. Furthermore, it is important that ChR2 and Halo be expressed in a predictable proportion to one another, so that the insertion and deletion of spikes can be balanced; separate delivery of the two genes, or independent expression of the two genes, would not accomplish this. Recently we and others have presented gene fusions between ChR2 and Halo that enable them to be expressed in the same expression cassette, bridging ChR2 and Halo with an optimized form of the 'self-cleaving' 2A peptide sequence from foot-and-mouth disease virus or other picornaviruses (Han et al., 2008;Tang et al., 2009). The 2A sequence encourages a ribosomal skip between a glycine and proline within the 2A sequence, resulting in separate proteins being formed from the sequences on either side (Ryan et al., 1991;Ryan and Drew, 1994). In the Tang et al. paper, the fi rst peer-reviewed presentation of opsin fusions using a 2A peptide bridge, the authors showed functional expression of both opsins in the same cell, in vivo, after delivery via AAV. We here quantitate the proportional expression of ChR2 and Halo in the same cell when they are expressed separated by a 2A bridge, and discuss the consequences for the specifi c diffi cult experimental goal of perturbing synchrony and precise spike timing. Such proteins will also be useful for exploring a large set of systems neuroscience experiments in which bi-directional control of the activity of a given neuron population is desired.

CELL CULTURE
All procedures involving animals were conducted in accordance with the National Institutes of Health Guide for the care and use of laboratory animals and approved by the Massachusetts Institute of Technology Animal Care and Use Committee. Hippocampal regions CA3-CA1 of postnatal day 0 Swiss Webster or C57 mice from Taconic or Jackson Labs were isolated and treated with trypsin (1 mg/ml) for ∼12 min. Digestion was stopped by Hanks solution supplemented with 20% fetal bovine serum and trypsin inhibitor (Sigma). Tissue was dissociated with Pasteur pipettes and centrifuged at 1000 rpm at 4°C for 10 min. Dissociated neurons were plated on glass coverslips pre-coated with Matrigel (BD Biosciences) at a rough density of approximately two to four hippocampi per 20 coverslips. Neurons were transfected using a commercially available calcium phosphate transfection kit (Invitrogen), at 3-5 days in vitro. GFP fl uorescence was used to identify successfully-transfected neurons, indicating a net transfection effi ciency of 1-10%. All images and electrophysiological recordings were made on 9-15 day-in-vitro neurons (approximately 6-10 days after transfection).

MOLECULAR BIOLOGY
The ChR2GFP-2A-HaloYFP-N1 plasmid ('ChR2-2A-Halo') was constructed by fi rst inserting the extended N-terminus 2A sequence (DNA sequence AAGAAACAGAAAATTGTGGCACC AGTGAAACAGACTTTGAATTTTGACCTTCTCAAGTTGGCG GGAGACGTCGAGTCCAACCCTGGGCCC, translated peptide sequence KKQKIVAPVKQTLNFDLLKLAGDVESNPGP) into the BsrGI and NotI sites in the pEGFP-N1 vector, followed by inserting ChR2-GFP without stop codon into the KpNI and BsrGI sites in front of the 2A sequence, and then by inserting Halo-YFP without start codon into the XbaI and EcoRI sites downstream of the 2A sequence. The pEGFP-N1 vector drives gene expression from the CMV promoter.
Light-induced membrane photocurrents were induced by brief light pulses separated by periods in the dark, in neurons currentclamped or voltage-clamped, respectively. Light pulse trains were either input directly into pClamp software (Molecular Devices) or synthesized by custom scripts written in MATLAB (Mathworks), then played back through a Sutter DG-4 light source via a Digidata 1440 (Molecular Devices). Light was refl ected into the sample off of a 700DCXRU (Chroma) dichroic in a Leica DMI6000B inverted microscope. In the DG-4, a yellow fi lter (Chroma, bandpass 575 ± 25 nm) was used to deliver light to activate Halo, and a GFP excitation fi lter (bandpass 470 ± 20 nm) was used to activate ChR2. Powers out the 40× objective were approximately 10 mW/ mm 2 in irradiance.

DATA ANALYSIS
Amplitude and timing data were analyzed using Clampfi t 10 (Molecular Devices) and custom analysis scripts written in MATLAB.

RESULTS
We recently presented a demonstration of how one could use coexpression of separate ChR2 and Halo genes to support optical disruption of the timing of spikes, without signifi cant alteration of spike rate (Figure 1, adapted from Han and Boyden, 2007). In review of this previously-published work (see Han and Boyden, 2007 for detailed results and methods): we cultured hippocampal neurons and then caused them to fi re precisely-timed spike trains by patch clamping them in current-clamp mode, and repeatedly delivering to each neuron the same fi ltered Gaussian white noise current trace (see Figure 1Ei, top, for a fragment thereof). Such noisy currents had been previously found to induce reliable spike FIGURE 1 | Schema for how to alter neural spike timing, using two-color illumination of neurons co-transfected with both ChR2 and Halo in a 1:1 ratio. Adapted from (Han and Boyden, 2007). (A) Action spectrum for ChR2 (blue, adapted from Nagel et al., 2003), overlaid with absorption spectrum for Halo (orange, adapted from Duschl et al., 1990). Each spectrum is normalized to its own peak, for ease of comparison. (B) Co-expression of Halo-GFP (left) and ChR2-mCherry (middle) in a single neuron expressing both (right, overlay). Scale bar, 20 µm. (C) Hyperpolarization and depolarization events elicited in a single representative neuron, by two interleaved 2.5 Hz trains of yellow and blue light pulses (50 ms duration each), denoted by bars of respective coloration below the trace. , injected with fi ltered Gaussian white noise current, either with no light (left) or with concurrent delivery of a Poisson train of yellow and blue light pulses (right). Plotted is mean ± standard error. (G) Cross-correlation between spike trains elicited by the same fi ltered Gaussian white noise current injection, played twice, when either both current injections were performed in the dark (black curve), or when one of the current injections was performed with concurrent delivery of a Poisson train of yellow and blue light pulses (green trace). Data is plotted as mean ± standard error (averaged across n = 7 neurons). trains in current-clamped neurons in acute cortical brain slices (Mainen and Sejnowski, 1995). On some trials, we also illuminated the neuron with the Poisson train of alternating yellow and blue light pulses shown in Figure 1D (see Figure 1Ei, bottom). We found that when delivered alone, the fi ltered Gaussian white noise current trace indeed induced reliably-timed spike trains, across repeated trials (see Figure 1Eii, top, for 20 overlaid traces, and Figure 1Eiii, top, for corresponding spike rasters). When an optically-sensitized neuron was additionally driven by the Poisson train of yellow and blue light pulses, the neuron fi red spikes with different timings than occurred in darkness, but the resultant spikes were nevertheless still similar across repeated trials of current injection + light illumination (see Figure 1Eii, bottom, and Figure 1Eiii, bottom, for overlaid traces and spike rasters respectively, for 20 trials; Figure 1Eiv shows spike histograms). Inspection of this data shows that relative to the spike train elicited by current injection alone, the Poisson train of light pulses could sometimes abolish spikes that were previously reliable (vertical box 1 spanning Figures 1Ei-iv), create new spikes which were not previously present (vertical box 2 spanning Figures 1Ei-iv), or advance or delay the timing of specifi c spikes relative to their original timing in the dark (vertical box 3 spanning Figures 1Ei-iv). We compared mean spike rates for neurons receiving the fi ltered Gaussian white noise current injection alone vs. with the additional Poisson light pulse train, and found no difference in spike rates for these two conditions (p > 0.90, t-test; n = 7 neurons; Figure 1F), indicating that our optical intervention preserved spike rate. However, precise spike timing was altered signifi cantly: cross-correlations of the spike trains elicited from the fi ltered Gaussian white noise current injection alone vs. with the additional Poisson light pulse train resulted in zero-lag cross-correlations that were on average 37% smaller than cross-correlations of pairs of spike trains elicited from the fi ltered Gaussian white noise current injection alone (p < 0.005, n = 7 neurons). This indicates that precise spike timing was indeed disrupted by the activation of Halo and ChR2, even while spike rate was preserved.
The strategy illustrated in Figure 1 centers around the idea of using blue light to push the voltage of a cell over spike threshold, and using yellow light to push the voltage of a cell under spike threshold. Critical to this strategy is the proportional expression of ChR2 and Halo in a predictable ratio to one another. Ideally, for every spike inserted, another will be deleted. A neuron that expresses high levels of both ChR2 and Halo will have a high probability of generating a spike in response to a blue light pulse, but will also have a high probability of losing a spike in response to a yellow light pulse. On the other hand, a neuron that expresses low levels of both ChR2 and Halo will have a low probability of generating a spike in response to a blue light pulse, but will also have a lower probability of losing a spike in response to a yellow light pulse. Thus to a fi rst approximation, the strategy in Figure 1 should work to disrupt spike timing but not spike rate, independent of the net expression levels of ChR2 and Halo, as long as their proportion is kept constant. Of course, this approximation will break down if the scenario approaches an extreme: for example, a high-fi ring rate neuron may not be responsive to blue light pulses because of a ceiling effect on fi ring rate, but may be greatly affected by yellow pulses; on the other hand, a low-fi ring rate neuron may respond to blue light pulses with increased spiking, but may have few spikes to lose when yellow light pulses arrive. Thus it will be important to choose light pulse trains that deliver light pulses at appropriate rates given the neural population under investigation; the light pulse rate should scale with the fi ring rate of the neural population (and in principle, could even be adapted in real time during a physiology or behavior experiment). Two further comments should be made: given that ChR2 and Halo generate intrinsically different scales of currents (ChR2's peak currents are an order of magnitude greater than Halo's), the ratio of blue to yellow light powers may need to be adjusted to balance the current magnitudes against one another. Finally, in the intact brain in vivo, light powers will fall off with distance from the light source; however, the difference between blue and yellow light power attenuation in brain tissue is not very large over the short distances (∼1 mm) utilized for in vivo optical control scenarios, which are signifi cantly less than the absorption length constant of visible light in tissue (Bernstein et al., 2008b). Tiling the brain with arrays of LED-or laser-coupled optical fi bers may allow for more even light distribution in the brain, than possible with just one inserted fi ber (Bernstein et al., 2008a;Zorzos et al., 2009).
To express both ChR2 and Halo in a stoichiometric ratio, we evaluated a number of candidate strategies. Using two viruses to deliver the two genes could lead to signifi cant variation of the ChR2/Halo ratio across cells. Placing an internal ribosome entry site (IRES) -a long sequence of several hundred base pairs, large compared to the space accorded in many kinds of viral vector -in between two genes often results in much lower expression of the second gene (levels 0-20% those of the fi rst gene) (Mizuguchi et al., 2000;Hennecke et al., 2001;Yu et al., 2003;Osti et al., 2006). We decided upon a methodology using the 'self-cleaving' 2A sequence (Figure 2A) from foot-and-mouth disease virus (Ryan et al., 1991;Ryan and Drew, 1994). When an mRNA encoding for ChR2-2A-Halo is being translated, a specifi c Gly-Pro bond in the 2A sequence would not be enduringly formed by the ribosome (a 'ribosomal skip'), resulting in the separate translation of ChR2 and Halo into separate proteins. We utilized a variant of 2A with additional amino acids appended to the N-terminus which enhances cleavage from 90% to 96-99% (as assessed by an in vitro translation assay) (Donnelly et al., 2001a,b) (Figure 2A, blue and green colored residues). The use of a different 2A peptide to co-express channelrhodopsin-2 and halorhodopsin within a single gene was recently published, including proof-of-principle electrophysiological data resulting from in vivo infection of neurons with the resulting virus (Tang et al., 2009). Here, expression of the ChR2-2A-Halo gene by transfection in cultured hippocampal neurons resulted in excitatory and inhibitory photocurrents in response to blue and yellow light respectively. The photocurrents were smaller than would result from expression of either gene alone (Figure 2C), in part because of overlap in the two molecules' action spectra ( Figure 1A). Bluelight elicited photocurrents, measured in voltage clamp mode, were 157 ± 63 pA (n = 22 neurons; mean ± standard deviation), ∼30% of the value previously reported for raw ChR2 expression in cultured hippocampal neurons (Boyden et al., 2005), and yellow-light elicited photocurrents were 40 ± 24 pA, ∼45% of the value previously reported for raw Halo expression in cultured hippocampal neurons (Han and Boyden, 2007). These currents, nevertheless,  (Donnelly et al., 2001a;de Felipe et al., 2006); blue letters indicate amino acids from the 1D peptide sequence that, when appended to the N-terminus of the 2A sequence, increase in vitro-translated protein cleavage from 90% to 96%, and green letters are part of a sequence from the 1D peptide sequence that boosts in vitro-translated protein cleavage to 99-100% levels (Donnelly et al., 2001a,b). are suffi cient to cause a neuron near the threshold of spiking to fi re an action potential or to be momentarily silent, and routinely resulted, in current-clamped neurons, in effective perturbations of membrane voltage by 5-10 mV ( Figure 2B). Plotting, for each recorded neuron, the blue-light elicited photocurrents vs. the yellow-light elicited photocurrents yielded a signifi cant linear relationship with correlation coeffi cient r = 0.51 (R 2 = 0.26, p < 0.02; Figure 2D). Thus, the 2A expression vector here utilized was able to mediate proportional, functional expression of ChR2 and Halo in the same cell.
What kinds of experiment are enabled by the use of linked microbial opsins for bi-directional control of the voltage of individual neurons? Many 'obvious' but much-desired experiments may be facilitated with use of this reagent. For example, by appending to ChR2 a myosin-binding domain (MBD) that preferentially targets it to cell bodies and dendrites (Lewis et al., 2009), the resultant ChR2-MBD-2A-Halo could enable activation of a population of neurons when their cell bodies are illuminated with blue light, as well as silencing of specifi c projections when their axonal terminals are illuminated with yellow light. The ability to guarantee that a single population of neurons can be both activated and silenced is itself valuable, enabling testing of necessity and suffi ciency of the same population of neurons, to a specifi c neural computation or behavior. However, it is useful to explore conceptually how this technology might be useful for probing complex questions concerning synchrony. We have, in  Fries et al., 2008 for experimental examples of this measure) across the two areas plotted vs. spike frequency, measured in dark; spike-spike coherence across the two areas plotted vs. spike frequency, measured when one site is illuminated. (D) Shows the use of optical perturbation in a closed-loop fashion, triggering yellow-blue pairs of light pulses at one site, off of a feature of the local fi eld potential (LFP) recorded at another site, which is left unilluminated. In this case, the trough of the LFP (bandpass fi ltered to yield a defi ned set of frequencies) at the second site serves as the trigger for delaying spikes at the fi rst site. (i) From top to bottom: spike train trace and local fi eld potential trace recorded at two separate sites; timeline of when yellow or blue light is delivered to the site from which spikes are recorded (in this case, each yellow-blue pair is triggered by the trough of the LFP at the second site; each trigger is indicated by an arrow); spike train trace and local fi eld potential trace under the closed-loop protocol. (ii) From top to bottom: phase relationship between spikes recorded at the fi rst site and local fi eld potential troughs recorded at the second site, in the dark (e.g., see Pesaran et al., 2002;Gregoriou et al., 2009) for examples of this kind of data); phase relationship between spikes recorded at the fi rst site and local fi eld potential troughs recorded at the second site, under the closed-loop protocol.
outlined in schematic form four protocols enabling perturbation of spike timing or synchrony, in order to study their contribution to circuit output. In all four cases, appropriately-timed blue and yellow light pulses are used to disrupt the timing of spikes in a region containing neurons sensitized to light with a 2A construct, thus altering the spike statistics. Because light will be delivered to all the neurons within a defi ned region, one caveat may apply: the optically-sensitized neurons in the region will be exposed to the same temporal pattern of blue and yellow light, and this common input may alter (e.g., increase) the synchrony between optically-sensitized neurons within the illuminated region; this may potentially complicate experiments where altered synchrony within a region is undesired, although again the use of multiple light sources to deliver multiple light patterns to a single region may ameliorate this issue. One protocol is a direct generalization of the strategy outlined in Figure 1, the delivery of a light pulse train to a brain region to increase interspike interval (ISI) variability, turning rhythmic neurons into irregularly-fi ring ones ( Figure 3A). This 'informational lesion' might enable the deletion of the information encoded by the precision of spiking of those cells, without eliminating all activity generated by those cells, and thereby causing a gross disruption of connected circuits (as might occur with a conventional lesion of the cells of interest). A slightly more complex version of this idea is illustrated in Figure 3B, in which light pulses are precisely triggered, in a closed-loop fashion, upon the occurrence of particular recorded spikes, so that the neuronal fi ring rate becomes more bursty (but, once again, without changing overall spike rates). The ability to test the role of bursting in spike signaling may help resolve how the temporal integration properties of neural circuit elements contribute to neural computations. The 2A strategy can also be used to disrupt correlations between multiple regions that exhibit synchrony (Figure 3C), making spike timing more variable in one region so that its activity correlation with activity in a second region decreases, or making spike timing more variable in one region to see if such activity causally infl uences the activity within the second region. In this way it might be possible to understand how regions communicate and process information together in a coordinated fashion. As a fi nal example, the multi-site experiment can also be performed in a closed-loop fashion (Figure 3D), triggering the delivery of a specifi c light pulse train to one region, upon a particular phase of the local fi eld potential (LFP) recorded in a second region, and thus phase shifting spike times in the fi rst region with respect to the LFP of the second. Many other possible protocols exist; the key advance enabled by the 2A construct in such protocols is the ability to lesion information or alter information in the brain, without grossly disrupting the spike rate, as happens with traditional lesions, some forms of pharmacology, or normal activation or silencing of neurons.

DISCUSSION
We here discuss the implications of gene fusions of ChR2 and Halo, linked by the 'self-cleaving' 2A peptide, capable of supporting cellular synthesis of both opsins in proportion to one another, towards enabling a set of neurons to be simultaneously sensitized to blue light activation and yellow-light silencing. As an example of computational neuroscience-driven molecular engineering, such reagents may enable novel kinds of perturbation, such as the disruption of spike timing in the absence of altering spike rate. The construct presented here is freely available from the nonprofi t service Addgene (reagent 'ChR2-2A-Halo' at http://www.addgene.org/ Edward_Boyden) and is ready for use for transfection, gene gunning (Zhang et al., 2008), transgenic implementation , or electroporation (Petreanu et al., 2007;Lagali et al., 2008). The current construct, ChR2-2A-Halo yielded, when transfected into neurons, blue-and yellow-light elicited currents that were 30-40% of those obtained when ChR2 or Halo were expressed alone, respectively. Another paper that recently also disclosed the use of a 2A bridge to co-express ChR2 and Halo also showed currents that were smaller than those resulting from expression of individual opsins alone (Tang et al., 2009). These are still useful currents, especially if the goal is to dither the activity of a neuron above and below spike threshold in order to add and subtract spikes (as opposed to strongly driving or inhibiting neurons); however, a question for any technology is whether the effi cacy can be improved further. One solution would be to use stronger opsins to begin with, such as novel molecules like ChIEF (Lin et al., 2009) and Arch (Chow et al., 2009a,b), which can mediate functionally higher currents than ChR2 and Halo, respectively. Other methods for expressing two genes also exist: for example, bi-directional promoters can lead to higher expression levels than can 2A, but at the expense of signifi cantly more variability in the ratio of the protein levels resulting from each gene (e.g., see Fig. 3A in Amendola et al., 2005), due to the greater chance for noise to creep in downstream of the stoichiometry-determining event.
The prospects for using optical control technologies for ultraprecise neuromodulation therapies in clinical settings are exciting, since precise activation and silencing of specifi c cell types may increase effi cacy and reduce side effects of treatments relative to purely electrical methods. However, it is important to consider not just straightforward optical activation and silencing, but also complex perturbations such as desynchronization as here described, which could enable correction of neural dynamics abnormalities such as those found in neurological and psychiatric disorders like Parkinson's and epilepsy. As pre-clinical studies of the safety and effi cacy of optical control technology begin ), it will be increasingly important to derive principles of neural control, prototyping novel therapies that take full advantage of the computational power of these new technologies.

ACKNOWLEDGMENTS
ESB acknowledges funding by the NIH Director's New Innovator Award (DP2 OD002002-01), as well as by the NSF (0835878 and 0848804), the McGovern Institute Neurotechnology Award Program, the Department of Defense, NARSAD, the Alfred P. Sloan Foundation, the Jerry Burnett Foundation, the SFN Research Award for Innovation in Neuroscience, the MIT Media Lab, the Benesse Foundation, and the Wallace H. Coulter Foundation. XH acknowledges the Helen Hay Whitney Foundation and NIH 1K99MH085944. Thanks to Jennifer Raymond and Bob Desimone for discussions.