All-optical mapping of barrel cortex circuits based on simultaneous voltage-sensitive dye imaging and channelrhodopsin-mediated photostimulation

Abstract. We describe an experimental approach that uses light to both control and detect neuronal activity in mouse barrel cortex slices: blue light patterned by a digital micromirror array system allowed us to photostimulate specific layers and columns, while a red-shifted voltage-sensitive dye was used to map out large-scale circuit activity. We demonstrate that such all-optical mapping can interrogate various circuits in somatosensory cortex by sequentially activating different layers and columns. Further, mapping in slices from whisker-deprived mice demonstrated that chronic sensory deprivation did not significantly alter feedforward inhibition driven by layer 5 pyramidal neurons. Further development of voltage-sensitive optical probes should allow this all-optical mapping approach to become an important and high-throughput tool for mapping circuit interactions in the brain.


Introduction
A central challenge in studying circuit function within the somatosensory (barrel) cortex and other brain regions is to understand the spatial and temporal patterns of information flow between different columns and layers. In the case of the barrel cortex, measurements of single-neuron responses are of limited value when studying a circuit composed of many neurons of multiple cell types. 1 A more powerful approach comes by considering circuit function at the population level, while monitoring the contributions of individual neuron types. For example, it would be useful to activate specific neuronal subtypes in different columns or layers, while simultaneously recording the activity of populations of neurons from all regions. Here, we adopt such an approach.
For detecting the activity of populations of neurons, voltagesensitive dyes (VSDs) have enabled reliable and fast reporting of neuronal membrane potential. Unlike calcium imaging, which mainly detects action potential firing, VSD imaging provides an unbiased measure of all electrical signals, including postsynaptic potentials and other subthreshold responses. The high spatial and temporal resolution of VSDs permits the tracking of circuit activity and establishing connectivity within and between all columns and layers of the barrel cortex, both in vivo [2][3][4] and in vivo as well. 1,[5][6][7] Over the past few decades, many types of VSDs have been synthesized, including dyes based on absorption or birefringence and fluorescence. 8,9 Among these, the fluorescence-based styryl dyes have been most successful for imaging activity in the mouse or rat brain. [10][11][12][13] These dyes are useful because their vastly enhanced fluorescence in lipid membranes, in comparison to aqueous environments, 14 improves signal-to-noise when imaging neuronal membrane potential changes. Furthermore, these dyes allow imaging of neural activity with relatively low photobleaching and minimal toxicity. 10 In particular, the synthesis of improved long-wavelength styryl dyes 15,16 has greatly enhanced the application of VSD imaging to cortex. 17,18 To probe circuit function, VSD imaging is often combined with electrical stimulation of circuit elements. However, such experiments are limited because extracellular electrode stimulation is unspecific: it is usually difficult to define the neurons that are being stimulated and this makes it difficult to interpret circuit responses. Hence, there is a need for improved techniques that combine selective activation of neuronal populations with VSD imaging of circuit activity.
Here, we have combined VSD imaging with channelrhodopsin-mediated photostimulation to probe circuit function in the barrel cortex. Our experiments are based on a recent report that red-shifted styryl VSDs are spectrally compatible with the lightactivated cation channel, channelrhodopsin-2 (ChR2). 19 The ability to use this channel to photostimulate makes ChR2 a very powerful tool for temporal control of neuronal membrane potential. [20][21][22] Most importantly, ChR2 expression can be genetically targeted to specific subtypes of neurons via transgenic lines, viral delivery, or electroporation. 23 In this way, it is possible to use light to both activate defined neuronal populations and to image responses of their downstream partners.
Here we show that ChR2, in conjunction with a digital micromirror array system, can be used to specifically photostimulate layer 5 neurons across different cortical columns. In addition, all-optical mapping can be done by using VSDs to image responses of various layers and columns while sequentially activating different regions of the somatosensory cortex. By mapping out the contributions of both excitatory and inhibitory circuits in whisker-deprived mice, we found that sensory deprivation during early development does not significantly alter feedforward inhibition mediated by layer 5 pyramidal cells. This all-optical mapping approach provides an important highthroughput technology for determining interactions between somatosensory cortex circuit elements.

Materials and Methods
We used VSD imaging and optogenetic photostimulation to characterize synaptic transmission and connectivity within somatosensory cortical circuits.

Animals
All procedures were approved by the Biological Resource Centre Institutional Animal Care and Use Committee. To photostimulate layer 5 pyramidal neurons, homozygous mice expressing ChR2 under the Thy1 promoter were used (line 18; JAX stock number 007612). 24 P32 to 113 mice were used in the course of the study. For the whisker deprivation experiments, P36 to 44 mice were used.

Histology
To characterize the expression of enhanced yellow fluorescent protein (eYFP)-tagged ChR2 in the brains of these transgenic mice, adult mice were anesthetized and euthanized with an overdose of ketamine/xylazine (10 mg∕kg body weight) and transcardially perfused with 0.1 M phosphate buffer saline (PBS, pH 7.4) followed by 4% paraformaldehyde. The brain was removed and subsequently postfixed in 4% paraformaldehyde for an hour before it was transferred to 30% sucrose in 0.1 M PBS and stored at 4°C overnight in the fixative. The brain was then frozen and sectioned into 30-μm thick slices on a cryostat. Fluorescent images were obtained with a confocal microscope (Nikon AR1-A1).

Whisker Deprivation
For sensory deprivation, all large mystacial whiskers (from rows A to E 25 and arcs 0 to 6, and including α, β, γ, and δ whiskers) were removed. At P0, whiskers from the right cheek were first trimmed with tweezers by applying slow, steady tension to the base of the whiskers, followed by the light cauterization of whisker follicles using a hot wire tip of a custom-made cautery device. The neonates were anesthetized by rapid cooling via indirect contact with ice, to avoid frost burns during the deprivation procedure. To simulate any stimulation from whisker trimming, whiskers on the left cheek of pups were sham trimmed by stroking with tweezers as a control. After cauterization, pups were monitored throughout postop recovery in a warm cage until they recovered from the anesthesia and then were returned to the nursing mother.

Slice Preparation
300-μm thick brain slices from P28-P33 mice were prepared in ice-cold cutting solution containing (in mM); 240 Sucrose, 10 Glucose, 25 NaHCO 3 , 1.25 NaH 2 PO 4 · 2H 2 O, 2.5 KCl, 0.5 CaCl 2 , and 7 MgCl 2 , at 320 to 325 mOsm. Slices were then incubated in artificial cerebral spinal fluid solution containing (in mM); 10 Glucose, 126 NaCl, 24 NaHCO 3 , 1 NaH 2 PO 4 · 2H 2 O, 2.5 KCl, 2.5 CaCl 2 , and 2 MgCl 2 (300 to 310 mOsm). The brain was sectioned at 50 deg to the midline in order to obtain across-column slices similar to the procedures detailed in Finnerty et al. 26 Slices were incubated at 32°C for 30 min, and subsequently kept at room temperature (25°C) for another 30 min prior to slice imaging. For VSD imaging, slices were incubated as above before they were extracellularly stained with the VSD di-2-ANBDQPQ 16 (0.9 mM) in a humidified air-liquid interface chamber for an hour at room temperature as detailed in Kee et al. 11 This VSD was obtained from Prof. Les Loew at the University of Connecticut Health Center, United States. Barrels were visualized by transillumination of the slice. All recordings were done at room temperature.

Voltage-Sensitive Dye Imaging
The VSD was excited with 602 to 625 nm light from a halogen light source and fluorescence emission greater than 650 nm was captured by a high-speed charge-coupled device (CCD) camera (Micam02, Brainvision, Japan). At the beginning of the experiment, excitation light intensity was adjusted so that sample fluorescence emission was at 30% pixel saturation to prevent saturating the CCD camera and also to maintain similar levels of fluorescence emission across experiments. The exposure time was minimized to yield minimal bleaching during the course of an entire experiment. 96 × 64 pixel images were acquired at 2.2 ms sampling intervals and were subjected to 7 × 7 pixel spatial averaging.
The stimulus-induced change in VSD signal was calculated for each pixel as follows: where F o is the resting fluorescence and FðtÞ is the VSD fluorescence at time t. Depolarization decreases dye fluorescence, therefore, ΔF∕F o will be positive for a membrane potential depolarization. For determination of VSD signals from layers 2∕3, 4, or 5 within a column, responses from 11 × 11 pixels were typically averaged because this covers most of the area of a layer within a column. Data analyses were done with custom-made MATLAB® programs. For some analyses, a Savitzky-Golay filter was applied to reduce high-frequency noise and smooth optical traces. The Savitzky-Golay smoothing filter is a least-squares smoothing filter which is optimal for removing noise without distorting the waveform of fast signals. [28][29][30] In particular, the Savitzky-Golay filters preserve the amplitude and shape of signal peaks, which is desirable for analyzing VSD signals. This is unlike more standard filters such as adjacent-averaging filters, which tend to distort higher frequency features such as action potentials. This property makes Savitzky-Golay filters useful for signal processing applications that require preservation of high-frequency components of a signal. In addition, it is an effective, fast, and simple-to-implement filter which can be adapted for image signal processing in even three dimensions. 31 Pixels from the electrode and non-slice regions beyond the pia were masked out because they were not fluorescent. The Student's t test was used to test for the level of statistical significance for changes in VSD responses.

Combining Photostimulation with Voltage-Sensitive Dye Imaging in the Somatosensory Cortex
In order to photostimulate specific circuit elements within the barrel cortex, we used mice expressing ChR2 under the Thy1 promoter. 24 These mice (line 18) were bred to homozygosity to enhance the sensitivity of ChR2-expressing neurons to light. In these mice, ChR2 was expressed in many brain regions [ Fig. 1(a)]. Histological characterization of ChR2 expression indicated that ChR2 was expressed in pyramidal neurons and other cells in layer 5 of the somatosensory cortex, with some expression in layers 2∕3 and 4 neurons as well [ Fig. 1 To precisely define which neuron types expressed ChR2, we made whole-cell patch clamp recordings from several different types of neurons and examined the responses of these cells to light. We first focused on layer 5 pyramidal cells, which have previously been reported to express ChR2 in this mouse line. 24 Recordings were obtained from 12 pyramidal cells in layer 5; these cells could be identified by their distinctive pyramidal shape as well as their low-frequency firing of action potential trains with frequency adaptation 32  In addition, we recorded from two other interneurons that did not evince fast-spiking properties [ Fig. 2(e)]. 33 Neither of these cells had direct (short latency) photoresponses, indicting the absence of ChR2 expression. However, light did evoke depolarizing IPSPs in these cells [ Fig. 2(f)], presumably originating from photostimulation of presynaptic basket cells. Finally, we recorded from four pyramidal cells in layer 2∕3; all of these produced direct depolarizations in response to light but the depolarization was sufficiently large to evoke an action potential in only one of these four cells. In summary, photostimulation was most effective in evoking action potentials in layer 5 pyramidal cells and less capable of evoking action potentials in pyramidal cells in other layers. With regard to interneurons, roughly half of the basket cells could be photostimulated while other interneurons seemed not to express ChR2.

Mapping Responses Evoked by Stimulation of Layer 5 Neurons
To produce spatially structured photostimulation of ChR2expressing neurons, we used a digital micromirror array system that could produce spatial patterns of illumination of any shape or size within the microscope field of view. By focusing an image of the micromirror array on the slice, we could photostimulate defined areas [ Fig. 3(a); see also Ref. 19]. This made it possible to map responses to inputs coming from different regions of the barrel cortex, specifically by sequentially photostimulating individual layers of columns A, B, C, and D in a 4 × 3 grid. Such photostimulation was done in a pseudorandom pattern to avoid consecutive photostimulation of adjacent regions. Brief light pulses (3 ms duration, 460 nm) were used to evoke action potential firing [as in Figs. 2(b) and 2(d)]. These brief flashes allowed us to minimize stimulation artifacts that could interfere with VSD imaging. 19 The stimulation artifacts during the brief light flash obscured direct responses to light, but allowed imaging of longer latency synaptic responses. With this experimental arrangement, we could use ChR2 to photostimulate specific regions, while using VSD imaging to detect resulting responses throughout the barrel cortex. Photostimulation of the entire layer 5 [460 nm; Fig. 3  In contrast, electrical stimulation of layer 5 neurons evoked strong responses which spread primarily up the column, with relatively little spread into layers 4 and 5 of neighboring columns [Figs. 4(d) and 4(e)]. The lack of spread presumably is due to the electrode more effectively stimulating multiple types of interneurons, in comparison to photostimulation of only a fraction of basket cells, thereby limiting the spread of excitation between columns. 2,39,40 In particular, disynaptic inhibition between layer 5 pyramidal cells is known to be mediated by Martinotti cells 40 and this could limit lateral spread.
Responses recorded within layer 5 of column C were relatively similar to responses to photostimulation of this layer [lower traces in Figs. 4(c) and 4(e)]. Furthermore, while upward spread of responses within column C was similar for responses to both photostimulation and electrical stimulation, stronger postsynaptic activity was detected in layer 2∕3 following electrical stimulation. This is likely to be caused by antidromic activation of layer 4 neurons by electrical stimulation; these excite layer 5 pyramidal cells, which in turn would excite layer 2∕3 cells. Responses in columns A and E in layer 2∕3 are lower in comparison to responses in column C [ Fig. 4(d)]. There is evidence of surround inhibition mediated by somatostatin interneurons in layer 2∕3, 41 which is consistent with pyramidal cells activating somatostatin interneurons, which, in turn, inhibit neighboring layer 2∕3 pyramidal neurons. These results demonstrate our ability to use focal photostimulation to selectively stimulate a genetically defined population of neurons in layer 5 and not those in layers 2∕3 or 4.

All-Optical Circuit Mapping in the Somatosensory Cortex
Having demonstrated that our approach can visualize responses within the entire somatosensory cortex resulting from photostimulation of layer 5 neurons within column C [ Fig. 4(b)], we could next do the converse experiment by measuring responses in layer 5 of column C while photostimulating many other parts of the slice. By dividing the somatosensory cortex into 12 subregions, each consisting of one layer in one column, we could map the strength and spatial range of input signals that were evoked in layer 5 in responses to photostimulation of each subregion [ Fig. 5(a), top]. To obtain such input maps, excitatory responses evoked in layer 5 cells of column C were measured between 4.4 and 22 ms after a light flash. These signals were integrated to provide a measure of total excitatory drive and improve the signal/noise ratio for the responses, and the amplitude of these integrated signals was mapped (pseudocolor scale) according to the location of the photostimulus that elicited them [ Fig. 5(a), top]. These input maps revealed that the largest responses were evoked within layer 5 of column C, with less input coming from neighboring columns in layer 5 or other layers along column C [ Fig. 5(a), top]. Responses elicited when photostimulating column C reached their peak amplitude at 9.6 ms after initiation of the light flash [ Fig. 5(b)], while photostimulation of layer 5 in the neighboring columns (B and D) resulted in responses with a somewhat longer time to peak (9.6 to 14 ms). These responses are consistent with the anatomy of layer 5 pyramidal cells, 42,43 which have processes extending up within the column and along layer 5 to innervate neighboring columns. Given that a few layer 2∕3 pyramidal neurons expressed ChR2 [ Fig. 1(b)], it is also possible that excitatory responses originated from photostimulated layer 2∕3 pyramidal neurons. We thus conclude that the responses arose at least partly from monosynaptic connections between presynaptic layer 5 pyramidal neurons and their postsynaptic targets.
The amount of light power provided by our LED light source, as well as inefficient coupling between the LED and the micromirror array, limited the amount of light per micromirror element. As a result, we had to bin light output from a number of micromirrors to generate sufficient depolarization to photostimulate individual neurons. Although this limited the spatial resolution of photostimulation, we were able to further refine the input maps by subdividing the barrel cortex into 54 photostimulation regions [ Fig. 5(a), bottom]. In this case, more spatial structure was evident: the range of inputs to layer 5 neurons could more clearly be seen to span all layers and columns. Further, the responses elicited by photostimulating each subregion of a larger element were not similar [ Fig. 5(a), bottom]. For example, activating the four subregions within layer 5 of column C yielded a range of responses that could not be detected in the lower resolution map where all of these regions were photostimulated together [Figs. 5(b) and 5(c)]. Photostimulating the larger area resulted in responses that were larger than the average of the responses to photostimulating the individual subregions (Fig. 5). This probably indicates that synergistic effects occur when more neurons are simultaneously active, for example due to reciprocal excitatory synaptic interactions. In addition, the higher resolution input map indicated much lower connectivity from layer 4 of neighboring columns to layer 5 [ Fig. 5(a)], which is in line with the known synaptic connections between neurons in these regions.

Effects of Sensory Deprivation on Cortical Circuits
To demonstrate the utility of our all-optical mapping approach for the study of barrel cortex circuits, we next used this approach to define the role of activity in the development of these circuits. For this purpose, we removed whiskers from mice and then waited 30 days to prepare somatosensory cortex slices. To investigate the effect of sensory deprivation in isolation from potential effects of competition between neighboring deprived and spared barrels on local cortical circuits, we removed all whiskers from the right cheek and left the other side intact as controls. Light cauterization resulted in whisker loss for approximately 1 month, ensuring that no whisker-related activity occurred during the first 2 weeks of postnatal development. At P36 to 44, slight regrowth of a few short and misshapen whiskers was sometimes observed.
To determine whether such whisker deprivation affected our ability to photostimulate pyramidal cells, we measured layer 2∕3 responses evoked by photostimulation of layer 5 with light stimuli of varying intensity [ Fig. 6(a)]. The relationship between light intensity and layer 2∕3 responses measured in control slices was saturable, with maximal responses evoked in control slices by LED light of 4 to 7 mW∕mm 2 . This relationship was very similar in slices from deprived mice [ Fig. 6(b)]. Thus, chronic whisker deprivation caused no apparent change in the ability to photostimulate layer 5 neurons in these ChR2 transgenic mice.
To examine the effects of whisker deprivation on cortical circuitry, we prepared slices from whisker-deprived mice and photostimulated layer 5 of column C [ Fig. 6(a)]. When comparing deprived and control slices, although deprived slices exhibited a trend of slightly increased excitatory responses compared to controls [ Fig. 6(c)] there were no significant differences in excitatory responses in layer 5 [p ¼ 0.62, unpaired t test, n ¼ 6 for each side; Fig. 6(d)]. Excitatory responses in other layers in response to layer 5 photostimulation also were not significantly different [layer 2∕3: p ¼ 0.64; layer 4: p ¼ 0.24, unpaired t test, n ¼ 6 for each side; Fig. 6(d)]. To consider disynaptic inhibition driven by layer 5 pyramidal neurons, as well as monosynaptic responses evoked by direct photostimulation of basket cells, we used a pharmacological isolation approach to examine IPSPs. 27 We treated slices with PTX, which blocked inhibitory input to GABA A receptors and eliminated IPSPs. By subtracting responses measured in the presence of PTX from those measured prior to drug treatment, we could calculate PTX-sensitive IPSPs as a measure of inhibitory network responses. We called these responses compound IPSPs to indicate that they may reflect both monosynaptic and polysynaptic components. They had a slower latency to peak than excitatory responses [Figs. 6(c) and 6(e)], presumably due to both the slower kinetics of inhibitory postsynaptic responses relative to excitatory responses as well as possible polysynaptic components, which would have a longer latency due to the summation of multiple synaptic delays.
Compound IPSP responses were similar between deprived and control slices for all layers [layer 2∕3: p ¼ 0.93; layer 4: p ¼ 0.57; layer 5-6: p ¼ 0.4; n ¼ 6 for each side; Figs. 6(e) and 6(f)]. Given that whisker deprivation late in adulthood is known to transiently decrease ascending inhibition from layers 5 to 2∕3, 44 this result suggests that early experience-dependent plasticity involves interneuron circuits not driven by layer 5 pyramidal cells.
Our all-optical approach allowed us to map the organization of inhibitory circuits by measuring compound IPSPs evoked in different layers of column C (asterisks) in response to photostimulation of surrounding layers and columns (Fig. 7). In control slices, local photostimulation of layer 5 provided the strongest inhibitory inputs to layer 5, both within the column and in neighboring columns [ Fig. 7(a)]. This is consistent with the disynaptic inhibition occurring between layer 5 pyramidal cells mediated by Martinotti interneurons, 40 and is likely to be mediated by that circuit. Most of the compound IPSPs input to layers 2∕3 [ Fig. 7

Functional Connectomes of Barrel Cortex
Because VSD imaging allowed us to simultaneously obtain responses from the whole barrel cortex much quicker and more conveniently than is possible with patch clamp recordings or multielectrode recordings of a similar area, our all-optical approach enables rapid and high-content circuit mapping throughout the barrel cortex. In addition, the flexibility of combining a micromirror array system with VSD imaging allows us to photostimulate and record responses from multiple arbitrary areas depending on the organization of the structure of interest, delivering high information content in a single experiment. For example, a response matrix that displayed compound IPSPs elicited in various layers (2∕3, 4, or 5) of different columns (A to D) in response to photostimulation of each of these regions is shown in Fig. 8. In this matrix, the gray scale represents the strength of compound IPSP inputs measured in each column and layer in response to photostimulation of each column and layer. Such a two-dimensional comparison of connectivity between all elements in the barrel field is not possible with conventional techniques. The response matrix for control slices (Fig. 8, top) indicates that compound IPSPs are strongest in layers 4 and 5, with relatively small responses evoked in layer 2∕3 irrespective of the region that was photostimulated. This is consistent with the distribution of interneurons in the barrel cortex, where PV and somatostatin interneurons are present at highest density in layers 4 and 5. 45 In addition, compound IPSP responses also tended to span approximately two columns on either side of a given column (Fig. 8). Notably, the averaged compound IPSP response matrix for deprived slices was similar to that for controls (Fig. 8, bottom; p > 0.05 for all matrix elements, 2 sample t test, n ¼ 6). This indicated that whisker deprivation has remarkably little effect on inhibitory circuits activated by our photostimuli.

Discussion
We have combined ChR2-mediated photostimulation of pyramidal cells with VSD imaging of resultant responses in  somatosensory cortex. Our results provide insights into both optical circuit mapping technology as well as the effects of experience-dependent plasticity on cortical circuits. We will discuss each of these points in turn.

All-Optical Circuit Mapping
Our laboratory usually relies on optogenetic circuit mapping based on combining ChR2-mediated photostimulation with whole-cell patch clamp recording. 24 This approach provides highly specific targeting of both stimulation and detection, and additionally provides high temporal and spatial resolution. However, the use of an electrophysiological readout limits data throughput. The all-optical circuit mapping described here circumvents this limitation by relying on VSD imaging of neuronal activity, which is technically simpler and can be applied to many neurons and regions in the same experiment. It also facilitates analysis of circuit divergence, which requires monitoring activity in multiple postsynaptic neurons. 19 There have been other all-optical mapping efforts based on combining ChR2-mediated photostimulation with VSD imaging, but these have largely focused on photostimulating a single region. 19,46,47 We show here that all-optical mapping of both circuit inputs and outputs can be done over many neurons and many regions in a highthroughput manner in the mouse barrel cortex (Fig. 8). The consistent columnar and laminar organization of the mouse barrel cortex 48 makes it an ideal target for such a connectomics analysis.
The combination of optogenetic mapping and the ability to image neuronal responses throughout the barrel cortex with high temporal and spatial resolution yields a wealth of information about circuit organization and function. The ability to both stimulate defined areas containing potential presynaptic neurons while simultaneously detecting neuronal activity in multiple regions containing potential postsynaptic neurons allows investigation of circuit interactions between neighboring cortical columns. Electrical stimulation is limited because of the difficulty in interpreting responses due to the lack of specificity in stimulating neurons. Here, we specifically photostimulated groups of layer 5 neurons-primarily pyramidal cells and, to a lesser extent, basket cells-across different cortical columns by using ChR2, in conjunction with a digital micromirror array system.
We also showed that all-optical mapping can be done at a higher resolution, limited only by the ability to reliably photostimulate neurons (Fig. 5). Light delivery to specific regions via our micromirror array is relatively poor, but this can be improved through the use of brighter light sources, more efficient couplers, and more efficient optogenetic actuators. In our application, all cells were extracellularly stained with VSD, making it difficult to identify the cellular sources of signals. Sensory deprivation is known to increase the excitability of layer 5 pyramidal neuron dendrites and to increase the proportion of burst-firing cells. 49 We observed a trend toward slightly increased excitatory responses in deprived slices compared to controls [ Fig. 6(c)], but there was not a statistically significant difference. This may be due to VSD imaging integrating responses over many cell types.
The all-optical mapping approach described here can be further improved via the use of genetically encoded voltage sensors, 50-54 which will allow reporting of electrical activity of defined neuron types. In particular, probes capable of detecting hyperpolarizations associated with inhibitory synaptic activity will be critical for mapping inhibitory circuits. This will be particularly valuable for examining synaptic inhibition, because the pharmacologically subtracted compound IPSPs used in our study are likely an overestimate of synaptic inhibition due to increased network activity following PTX treatment.

Effects of Activity on Somatosensory Cortex Circuitry
By mapping out the contributions of both excitatory and inhibitory circuits after chronic whisker deprivation, we found that whisker deprivation for 4 to 6 weeks after birth did not significantly alter excitatory responses or feedforward inhibition mediated by layer 5 pyramidal cells. Whisker trimming with spared D2 and D3 whiskers is known to cause fast (within 24 h) depression of layer 5 responses to trimmed D1 whisker stimulation, 55 but this depression might not reflect changes in layer 5 pyramidal neuron synapses since the origin of these changes is not clear. Further, it may be caused by competitive effects between spared and deprived columns. Another study using a competitive sensory deprivation paradigm revealed reduced synaptic inputs from layer 2∕3 of the trimmed column to regular spiking layer 5 pyramidal cells. 56 Our observation of no significant change in excitatory responses in all layers when photostimulating layer 5 pyramidal cells is consistent with the fact that sensory deprivation of all whiskers does not change the excitability of pyramidal cell somata. 49 We also saw no significant changes in PTX-sensitive compound inhibition driven by photostimulation of layer 5 cells after whisker deprivation from P0 to P44 (Fig. 6). This contrasts with the effects of whisker deprivation in older mice (age around 8 to 11 weeks), which is known to transiently decrease ascending inhibition, likely mediated by Martinotti interneurons, from layers 5 to 2∕3. 44 Taken together, it appears that inhibitory circuits driven by layer 5 pyramidal neurons may be stable in early development and express experience-dependent plasticity in late adulthood. The amount of inhibition that layer 2∕3 pyramidal neurons receive from layer 2∕3 Gad2-expressing interneurons remains stable when whisker input is deprived at 8 to 11 weeks. 44 As layer 5 pyramidal neurons provide primary excitatory drive for somatostatin-expressing (SOM) Martinotti interneurons, 40,57 this suggests that these SOM interneurons serve as targets of transient activity-dependent plasticity in adulthood while experience-dependent plasticity in inhibitory circuits in early development could involve other circuits such as parvalbumin interneuron circuits. 58,59 The use of mice expressing ChR2 selectively in either the somatostatin or parvalbumin interneurons could be important to differentiate the effects of experience-dependent plasticity mediated by these neurons.
Although no evidence of experience-dependent changes was found in ascending inhibition from layer 5 pyramidal neurondriven SOM Martinotti interneurons (Fig. 6), other SOM circuits are present throughout all layers of the barrel cortex; 45 within layer 2∕3, these neurons are readily recruited by activity in layer 2∕3 pyramidal neurons. 57 In general, the signal-to-noise ratio of VSD imaging makes it difficult to reliably detect small IPSPs in the somatosensory cortex. This is because the distribution of inhibitory neurons in the somatosensory cortex is diffuse, which could result in small IPSPs being masked by strong excitatory responses following stimulation.

Conclusions
Our high throughput, all-optical mapping approach can be a valuable tool to evaluate the functional interactions between circuit elements in the somatosensory cortex and elsewhere in the brain. Development of improved optogenetic voltage sensors will increase the utility of this approach and will yield even more valuable information about somatosensory circuit function and how this changes during development.