1.

Introduction

The mammalian neocortex plays an important role in higher brain function, including sensory perception, cognition, associative learning, and goal-directed motor control.¹ Tangential to the cortical surface, the neocortex is organized in such a way that it is both highly specialized, with defined areas dedicated to specific functions and/or sensory modalities, and highly integrative, with each area receiving converging inputs from different thalamic nuclei, other cortical areas, and several neuromodulatory systems.^1,2 Another form of cortical organization is clearly evident orthogonal to these tangentially distributed maps.² The neocortex is classically divided along its depth into six anatomically defined layers, from superficial layer 1 to deep layer 6. Each layer contains distinct classes of cells that project cortically or subcortically, along with GABAergic interneuron types.^2,3 These local neocortical microcircuits (six layers of interconnected excitatory and inhibitory neurons) will process and integrate all the area-specific inputs from different thalamic nuclei, other cortical areas, and several neuromodulatory systems.

In primary sensory cortices, information from the periphery primarily relayed by the thalamus mainly impact the layer 4 (L4), also known as the granular cell layer. L4 stellate cells mainly excite layers 2/3 (L2/3) pyramidal neurons, which in turn excite neurons in layer 5 (L5), and to the whole cortical column.^4,5 The deep infragranular L5 and layer 6 (L6) are the main source of cortical outputs to subcortical structures (such as the thalamus, striatum, and brainstem).¹ Cortical layers 1, 2, and 3 form the superficial layers, also known as the supragranular layers, which are likely the most integrative layers, gathering sensory information, and contributing an important source of projections to other cortical areas.² Anatomical studies have shed light on the synaptic architecture of cortical microcircuits.^6,7 The difficult challenge now is to relate the wiring diagram of these neuronal networks to their functional operation.

There has been rapid technological progress over the last decade in measuring and perturbing neuronal activity in the mouse neocortex in vivo. Based on blood oxygen level-dependent (BOLD) contrast, functional magnetic resonance imaging (fMRI) has revolutionized human cognitive neuroscience.^8,9 Attempts to apply fMRI findings in mouse neocortex are prohibitively challenging, since the small size of the mouse neocortex ($\sim 1\text{-}\mathrm{mm}$-thick in depth) necessitates exceptionally high signal-to-noise ratio (SNR) and spatial resolution.¹⁰ Electrophysiological methods that detect functional synapses, including paired recordings and glutamate uncaging-based methods, have been applied to map local circuits within mouse cortex. However, most of these electrophysiological techniques estimating the layer-specific connectivity in cortical circuits are limited to ex vivo brain slices.^4,11–15 In vivo simultaneous recordings across the entire depth of the primary auditory cortex¹⁶ and visual cortex¹⁷ have been demonstrated to study the laminar specificity in spatiotemporal dynamics by inserting a sharp linear array microelectrodes, whereas these methods are labor-intensive, invasive (potentially damaging to the cortex) and can only record the action potentials of limited neural cells next to the extracellular electrodes.¹⁸

Optical methods have been proven to be useful in monitoring neural responses in the brain, as they offer advantages for studying functional organization.¹⁹ However, most of the current optical imaging setups involving charge coupled device (CCD) cameras can only provide two-dimensional (2D) information and cannot detect depth-resolved neuronal activation. In order to investigate the three-dimensional (3D) layer-specific interaction in mouse cortex and between different cortical areas in vivo, an imaging system with large field of view (FoV), depth-resolved ability, and high imaging speed is needed. Laser-scanning microscopy (LSM), such as confocal and two-photon microscopy, has revolutionized biomedical research with high resolution, whereas current implementations have limited imaging depth and FoV.^20–22 Moreover, different layers have to be scanned sequentially to image 3D volumes in LSM, which makes it difficult to acquire the 3D functional responses in the brain with high temporal resolution.^20–22 Utilizing a multitude of sources and detectors with an inversion algorithm to reconstruct the underlying tissue properties in 3D, diffuse optical tomography (DOT) has been applied to achieve 3D reconstruction of cerebral hemodynamics in small animals.^23–25 High-density speckle contrast optical tomography has also been developed for in vivo 3D imaging of blood flow in small animals.²⁶ However, these systems were designed to cover subcortical tissues in mouse brain thus suffered from poor spatial resolution especially in axial direction ($>1\mathrm{mm}$).

As a modified version of DOT, laminar optical tomography (LOT) was initially developed to image absorption contrast for hemodynamic changes in mesoscopic range.^27–30 Soon after, it was adapted to fluorescent molecular imaging, termed either fluorescence laminar optical tomography (FLOT)^30–33 or mesoscopic fluorescence molecular tomography.^34,35 Similar to DOT,³⁶ LOT uses an array of photon detectors or a CCD camera to collect photons emitted from locations at different distances away from the illumination position, enabling simultaneous detection of scattered light traveling through different depths in the tissue. By applying much denser spatial sampling, FLOT can achieve a resolution of $\sim 30$ to $100\mu \mathrm{m}$ with millimeters imaging depth.^28,33 Recent studies demonstrated that angled illumination or detection configuration can further improve both resolution and depth sensitivity.^33,37

In order to study the specific function of each layer in the neocortex, development of technologies to regulate the type-specific cells is key to understand how they contribute to local network activity and overall brain function in vivo.³⁸ This control has become feasible with the advent of optogenetics, in which single-gene encoding light-activated ion-conductance regulators or biochemical signaling proteins are introduced into targeted cells.^38–40 This approach provides a reliable method for stimulating or suppressing neural activity in mammalian tissues,^12,41–47 and researchers can now control activity in defined neuronal populations and projections while examining the consequences on behavior and physiology.⁴⁰

In this paper, we demonstrate the depth-resolved capability of a mesoscopic optical imaging technique with angled illumination configuration for in vivo brain functional imaging in mouse cortex. Rather than optical intrinsic contrast, we applied voltage-sensitive dye (VSD), an exogenous fluorescent dye to provide specific optical contrast. VSD can report changes in membrane potential by binding to the neural membrane and converting changes in transmembrane voltage into the signal changes of the emitted fluorescence light.^48,49 Compared to hemodynamics imaging based on absorption contrast (slow signal changes at the second time scale), voltage-sensitive dye imaging (VSDi) provides a reliable and direct measures of neural activity in the brain with relatively high spatial and temporal resolution (at the millisecond time scale).^48,50 We first probed the neural connection in primary sensory cortex in Thy1-ChR2-YFP transgenic mice. We further imaged the layer-specific functional projections between sensory cortex (S1) and motor cortex (M1) in mice in vivo following single whisker stimulation. The results prove this mesoscopic imaging technique has the potential to serve as a useful tool in investigating the layer-specific neural connectivity in the mouse cortex in vivo.

2.

Materials and Methods

All in vivo experiments were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication No. 80-23), approved by the Institutional Animal Care and Use Committee at the University of Maryland, Baltimore and College Park campuses.

2.2.

Imaging System

Figure 1 shows the schematic diagram of the mesoscopic imaging system. A 637-nm laser diode was used as the light source. The light was first collimated (O1) and coupled (O2) into a single-mode fiber to shape the light beam.³³ Light coming out from the fiber was collimated by an objective lens (O3) and the collimated light was expanded into line-field illumination using a cylindrical lens (CL) with a full-line-width at the half-maximum of $\sim 26\mu \mathrm{m}$ at the focal plane. An iris was used to control the length of the line illumination.³³ The emitted fluorescent light was collected back through an objective lens (O4), a dichroic mirror (650 nm, single-edge dichroic beam splitter; FF650-DiO1-$50\times 70\mathrm{mm}$; Andover Corporation), an emission filter (695 nm, 695FG07-50, Andover Corporation), another objective lens (O5), and finally imaged to a high-speed CCD camera (MiCAM02-HR, SciMedia Ltd.). The illumination angle was set at 45 deg, rendering $\sim 30\text{-}\mathrm{deg}$ transmission angle in tissue (with $n\sim 1.33$).³³ The CCD camera was vertically placed to record both fluorescence and reflectance images by changing the emission filter (no filter for reflectance recording and with emission filter for fluorescence imaging). A motor stage was used to translate the sample laterally in scanning direction (perpendicular to the line illumination direction).⁵⁷ To photostimulate layer 5 pyramidal neurons in Thy1-ChR2-YFP mice, blue laser light (473 nm, MBL-FN-473 nm, Changchun New Industries Optoelectronics Tech. Co., Ltd.) was coupled into a single-mode fiber. The fiber was mounted on an $XYZ$ manipulator and targeted at the cortical surface for photostimulation as shown in Fig. 1. To study the layer-specific functional projections between S1 and M1 in B6 mice, a glass pipe (1.0-mm in diameter) fitted onto an $XYZ$ manipulator was aimed at the facial C2 whisker (not shown in Fig. 1). Air-puff stimulus was applied through a Picospritzer pressure valve connected to the glass pipette.^33,56 The stimulation system (either the 473-nm laser or the air-puff stimulus system), the 637-nm illumination, the CCD camera, and the motor controller were synchronized through the synchronization controller box in the MiCAM02-HR imaging system (SciMedia, Ltd.).

Fig. 1

Schematic of the mesoscopic system. LD: laser diode; O: objective lens; I: iris; CL: cylindrical lens; F: filter; DM: dichroic mirror; LC: laser controller; and SC: synchronization controller. The illumination arm is arranged at 45 deg in air. The inset image shows the line illumination and the 473-nm laser stimulation on the mouse brain surface.

3.

Results

3.1.

Histological Characterization of the Chr2 Expression in Thy1-Chr2-YFP Mice

Histological characterization of the ChR2 expression by imaging YFP signal was performed in Thy1-Chr2-YFP mice. ChR2 expressed in many brain regions including cortical neurons, corpus callosum and in some deep structures of the brain as shown in Fig. 2(a). In the somatosensory cortex [Fig. 2(b)], ChR2 was expressed in pyramidal neurons and other cells in L5, with some expression in L2/3 and L4 neurons as well. However, previous studies have validated that photostimulation was most effective in evoking action potentials in L5 pyramidal cells and less capable of evoking action potentials in pyramidal cells in other layers.⁵²

Fig. 2

(a) Expression of channelrhodopsin-2 (ChR2) in Thy1-ChR2 mice. (b) Preferential expression of YFP-tagged ChR2 in layer 5 pyramidal neurons in the cortex. The image in the zoomed image is from the red dashed square in (a).

3.2.

2D CCD Imaging after Photostimulation in the Somatosensory Cortex

In the photostimulation experiment, 473 nm blue laser light was used to stimulate the cortical neurons in Thy1-ChR2-YFP mice. The brief flash (5 ms) allowed to minimize stimulation artifacts that could interfere with VSD imaging.⁵² The 473-nm blue laser stimulation could also excite the YFP and the VSD, because of this, the responses during the 5-ms blue light stimulation were discarded. Figure 3(b) indicates the location of the 473-nm laser stimulation, and the 637-nm line illumination. Since the line illumination was set 45 deg toward the right side of the image, most of the responses will appear at the right side of the line illumination. With this experimental setup, we used ChR2 to photostimulate cortical neurons and using VSD imaging to detect resulting responses throughout S1. Figure 3(a) shows the 2D CCD measurements of changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] at different time points after the photostimulation (at one representative scanning position). In general, the responses appeared as double bands shape with different distances to the line illumination location, corresponding to different depths in the cortex (the longer distance to the line illumination site represents deeper region in the cortex). The intensity of each band varied as time changed. At 2.5 ms after photostimulation, the deeper band (with longer distance from the illumination site) already appears large response area and strong intensity. After 2.5 ms (5 ms after photostimulation), the intensity and response area of the shallower band dramatically increased while the intensity and response area of the deeper band slightly increased. At 7.5 and 10 ms after photostimulation, the intensity and response area of both bands decreased. Figure 3(c) shows the quantitative change in fluorescence $[\mathrm{\Delta }F/F(\%)]$ in response to the 5-ms 473 nm laser stimulation at different time points. Fluorescence signal was calculated from the ROI (green square and red square: $5\times 5\text{pixels}$) shown in Fig. 3(b). Together with Fig. 3(a), the dynamics of response from both the deeper band and the shallower band can be clearly displayed.

Fig. 3

(a) 2D CCD measurements of changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation from oblique line illumination. Time period after stimulation is indicated at the bottom of each image. Scale bar: 0.5 mm. (b) Imaging schematic showing the location of the 473-nm laser stimulation and the 637-nm line illumination. (c) Changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation. Fluorescence signal was calculated from the ROIs (green square and red square: $5\times 5\text{pixels}$) shown in (b). The blue column indicates the 5-ms 473 nm laser stimulation. Time points corresponding to those shown in (a) are indicated near the green plot. $N=4$.

3.3.

Combining Photostimulation with 3D Mesoscopic Imaging in the Somatosensory Cortex

After recording the data at all the scanning positions, 3D neural activities were reconstructed following the protocol in Sec. 2.4. As shown in Fig. 4(a), photostimulation initially evoked excitatory responses in deeper areas (corresponding to layer 5) [Fig. 4(a), top left] that peaked within 5 ms after photostimulation [Fig. 4(a), top right]. Excitatory responses subsequently spread up along the column into shallower areas (corresponding to layer 2/3) by 5 ms after photostimulation [Fig. 4(a), top right]. The excitatory responses maintained in both layer 5 and layer 2/3 until 10 ms after photostimulation [Fig. 4(a), bottom right]. Notably, the area between layer 2/3 and layer 5 (corresponding to layer 4) showed very weak signal, resulting to the overall “sandwich” structure along the cortical depth. In contrast, from the conventional 2D planar imaging, the activation pattern at 5 and 10 ms after photostimulation is shown in Fig. 4(b). The responses appeared surrounding the 473-nm stimulation fiber end, with decreased activation at 10 ms after stimulation. The time-course trend was similar to the 3D result, while since the signals from different depths were integrated, the 3D sandwich structure could not be resolved.

Fig. 4

(a) 3D changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation reconstructed by the mesoscopic system. (b) Changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation captured by the conventional 2D planar imaging system. Time period after stimulation is indicated at the bottom of each image. Red arrow indicates the location of the 473-nm stimulation fiber.

To better visualize the depth-resolved neural activities, 2D $XZ$ cross sections of changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] were reconstructed as shown in Fig. 5(a). The neural response dynamics in different layers can be clearly resolved. $XZ$ cross sections provide a new perspective to investigate the neural activities at different time points. Figure 5(b) further shows the quantitative change in fluorescence $[\mathrm{\Delta }F/F(\%)]$ in response to the 5-ms 473 nm laser stimulation at different time points for L2/3 and L5. Fluorescence signal was calculated from the ROI (green square and black square: $5\times 5\text{pixels}$) shown in Fig. 5(a). The results provide us the temporal and spatial dynamics of the neural responses after photostimulating cortical neurons in different regions of the sensory cortex. All the four animals imaged showed a similar trend.

Fig. 5

(a) 2D changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation reconstructed by the mesoscopic system in $XZ$ cross section. Scale bar: $250\mu \mathrm{m}$. (b) Changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to the 5-ms 473 nm laser stimulation. Fluorescence signal was calculated from the ROIs (green square and black square: $5\times 5\text{pixels}$) shown in (a) at 2.5 ms after stimulation. The blue column indicates the 5-ms 473 nm laser stimulation. Time points corresponding to those shown in (a) are indicated near the green plot. $N=4$.

3.4.

3D Mesoscopic Imaging of Neural Connections in Sensory and Motor Cortices

Next, we recorded the depth-resolved neural responses in sensory and motor cortices following single whisker stimulation. Figure 6 shows the images of neural responses evoked by contralateral C2 whisker stimulation. As shown in the 2D depth-integrated images in Fig. 6(a), a single-brief passive deflection of the C2 whisker evoked a stereotypical pattern of cortical activity imaged with VSD. The responses occurred first with a latency of 20 ms following whisker stimulation in sensory cortex and was highly localized, exciting the C2 barrel column of the S1 cortex specifically. Over the next 20 ms, the depolarization spread across a large part of the sensory barrel cortex, indicating that neurons from the surrounding barrel columns became depolarized. Approximately 10 ms after the earliest response in S1 cortex, a second localized anteromedial cortical region within the M1 cortex region was depolarized and spread over the following 10 ms. Finally, after $\sim 60\mathrm{ms}$ following whisker deflection, the evoked activity disappeared gradually. As previously reported,⁷⁰ deflection of C2 whisker initiated cortical activities in two clearly separate focal regions, from which propagating waves of depolarization can spread to a large part of the sensorimotor cortex. However, imaging the voltage-dependent fluorescence changes of neocortex stained with VSD using a fast CCD can only reveal the 2D membrane potential dynamics. In contrast, using our mesoscopic imaging system, the 3D spatiotemporal dynamics of cortical activities can be reconstructed as shown in Fig. 6(b). The ability to resolve 3D spatiotemporal dynamics is of importance when considering the layer-specific functional connections between barrel cortex and motor cortex. At 30 ms in Fig. 6(b), we can notice the neural responses in M1 showed a sandwich structure with a distinct band between the two strong response regions. The two strong response regions were within the identified L2/3 and L5 location of the M1. The corresponding 2D fluorescence changes in $XZ$ cross sections were further shown in Fig. 6(c). This 3D imaging method provided a new perspective about the layer-specific neural interactions between different functional areas within the cortex.

Fig. 6

(a) 2D depth-integrated images of changes in fluorescence [$\mathrm{\Delta }F/F(\%)$, ordinate] in response to C2 whisker stimulation; (b) corresponding 3D fluorescence changes in brain; and (c) corresponding 2D fluorescence changes in $XZ$ cross sections. Time period after stimulation is indicated at the left side of each image. Blue arrow indicates the distinct band between L5 and L2/3.

To study the response dynamics, we further plotted the change in fluorescence $[\mathrm{\Delta }F/F(\%)]$ in response to the 20-ms whisker stimulation at different time points for S1, L2/3 of M1, and L5 of M1 from all four mice as shown in Fig. 7(b). Fluorescence signal was calculated from the ROIs (blue, red, and green cubic boxes: $5\times 5\times 5\text{pixels}$) shown in Fig. 7(a). It is clear that the response in L2/3 of M1 and L5 of M1 appeared after that in S1. To quantitatively study the response latency in S1, L2/3 of M1, and L5 of M1, we defined the response latency as when the fluorescence signal reaches the half maximum value in the trial. The response latency in S1 shows a significantly lower value than that in L2/3 and L5 of M1, while there is no significant difference in response latency between L2/3 of M1 and L5 of M1. The latency is $7.5\pm 0.5\mathrm{ms}$ ($n=4$) between the response in S1 and M1. We further defined the full-width at half-maximum (FWHM) of the response as the duration of the activation. The FWHM in S1 shows a significantly higher value than that in L2/3 of M1, while there is no significant difference in FWHM between S1 and L5 of M1, and between L2/3 of M1 and L5 of M1.

Fig. 7

(a) Example image to showing the ROIs selected for calculation. (b) Changes in fluorescence [$\mathrm{\Delta }F/F$ (%), ordinate] in response to the 20-ms whisker stimulation in four mice. Fluorescence signal was calculated from the ROIs (blue, red, and green cubic boxes: $5\times 5\times 5\text{pixels}$) shown in (a). The red column indicates the 20-ms whisker stimulation. (c) Comparison of the response latency in S1, L2/3 of M1, and L5 of M1. The response in S1 shows a significantly lower value than that in L2/3 and L5 of M1. (d) Comparison of the FWHM of the response in S1, L2/3 of M1, and L5 of M1. The response in S1 shows a significantly higher value than that in L2/3 of M1.

4.

Discussion

Visualization of evoked and spontaneous neuronal activity in vivo is of importance for understanding brain functions. Sensory pathways in the mammalian brain have been particular targets of imaging studies, because neural networks can be evaluated following specific stimulation of the sense organs, which translate and transmit physical energy to the brain. The neocortex serves as the central part of processing and responding to specific sensory input.⁷¹ Cortical layering is a hallmark of the mammalian neocortex and a major determinant of local synaptic circuit organization in neural systems¹³. Due to the limited thickness of mouse cortex (with 6 layers in $\sim 1\mathrm{mm}$⁶³), it raises a big challenge to study the layer-specific microcircuits in the cortex. Conventional optical methods for functional brain imaging based on planar CCD could provide widefield imaging of the aggregate neural activity⁷² or mesoscopic imaging at single-cell resolution using sophisticated genetic tools⁷³ or structured illumination schemes,⁷⁴ while does not have depth-resolved capability. At the microscopic scale, confocal and two-photon microscopy can provide cellular spatial resolution, while have limited imaging depth, FoV, and temporal resolution especially for 3D functional imaging. Although we need to notice, there are emerging improvements for multi-photon microscopy in imaging depth,^75,76 large FoV,^77–80 and imaging speed.^81–83 At the macroscopic scale, fMRI and DOT have been applied to achieve 3D reconstruction of cerebral hemodynamics in small animals.^{23–25,84–87} The macroscopic imaging methods can provide a large FoV and imaging depth (some can cover the whole mouse brain) while the spatial resolution is limited. Moreover, the relationship between BOLD signals and underlying brain activities remains unclear.⁸⁸ Photoacoustic tomography (PAT), which starts with optical absorption by tissue molecules and ends with ultrasonic emission through thermoelastic expansion, has been demonstrated in brain functional imaging from microscopic to macroscopic scales.^89–91 However, PAT is also an indirect measure of neural activity in the brain, focusing on assessment of hemodynamics and oxygen metabolism in mouse cortical vasculatures. To image neural responses at different cortical layers in mouse brain cortex, an imaging method in the mesoscopic range, which can visualize the layer-specific interactions that are too deep for the microscopy to image and too small for the macroscopy to resolve is needed. LOT is developed to image absorption contrast for hemodynamic changes in mesoscopic range and has been demonstrated to record the vascular compartment dynamics during somatosensory stimulation.^29,92 However, due to the sparse sources/detectors configuration and the absorption contrast, the previously reported mesoscopic methods still have limited spatial resolution and cannot provide the direct neural responses measurements. In this paper, we demonstrated an improved mesoscopic 3D imaging method using an oblique illumination/detection system configuration and high-density sources/detectors (line illumination and CCD camera). Line illumination could speed up the image acquisition compared to point scanning. VSD was combined to this 3D imaging method to visualize the direct layer-specific responses after specific stimulus in mouse cortex.

We first tested the depth-resolved capability of our imaging method using well-characterized Thy1-ChR2-YFP mice. Using transgenic mice and photostimulation, it is possible to stimulate specific cortical neurons. From the raw CCD recording in Fig. 3, we can see that with oblique illumination/detection configuration alone, neural responses at different depths can be resolved. Since the raw measurements do not take photon distribution and detector sensitivity into consideration, the signal quantification is not accurate enough. For instance, there will be more excitation photons in the shallower regions such as L2/3 compared to deep regions such as L5. Through image reconstruction, 3D neural responses in the cortex can be obtained. The nature of reconstruction can help to improve the signal quantification accuracy at different depths by taking photon distribution and sensitivity of the detectors into consideration.⁹³ At 2.5 ms after photostimulation, both the raw recordings (Fig. 3) and reconstructed results (Figs. 4 and 5) indicate that the deeper region (L5) already appears large response area and strong intensity.

It is not surprising to see this since within the cortex, ChR2 is mainly expressed in the L5 pyramidal neurons and the photostimulation could preferentially activated the L5 pyramidal neurons.^18,51–54 We can see there was also responses from L2/3 at 2.5 ms after photostimulation. One possibility for the responses may be the sparse ChR2-labeled neurons in L2/3 and the apical dendrites of L5 cells as shown in Fig. 2, though the previous study found that photostimulation was most effective in evoking action potentials in L5 pyramidal cells and less capable of evoking action potentials in pyramidal cells in other layers.⁵² Another possibility is that the signal of L2/3 at 2.5 ms after photostimulation was from the excitation of L5 neuron since the excitation of L5 neurons started at 0 ms and we did not have the recording until 7.5 ms after photostimulation started. At 5 ms after photostimulation, the intensity and response area of the deeper region (L5) slightly increased. On the other hand, the intensity and response area of the shallower band dramatically increased at 5 ms. At 7.5 and 10 ms after photostimulation, the intensity and response area of both regions decreased. Since $\mathrm{\Delta }F/F$ was used to display the neural activation in this study, $\mathrm{\Delta }F/F$ is a good parameter to compare dynamics, but its amplitude is not an appropriate index when comparing fluorescence from different depths considering uneven VSD staining. Although one interesting observation about the signal dynamics is that the intensities of L2/3 and L5 were flipped at 5 ms after photostimulation and lasted until the excitation ended. The consistent observed signal inversion (observed in all four mice) could indicate the dynamics of signal transmission. We suspect that the photostimulation preferentially activated the pyramidal neurons in L5, L5 pyramidal neurons excited L2/3 pyramidal neurons, as well as other L5 pyramidal neurons in neighboring columns, which is consistent with the known connectivity pathways of L5 pyramidal neurons.^52,94 Since the pyramidal neurons in L5 directly innervate L2/3 pyramidal neurons, there was very weak signal in L4, which resulted into a sandwich structure. In contrast, the conventional VSDi method can only provide a depth-integrated image since it does not have the depth-resolved capability as shown in Fig. 4(b). Another observation is that that the light evoked duration of activation pattern (Fig. 4, $\sim 10\mathrm{ms}$) lasted shorter than that induced by whisker (Fig. 6, $\sim 40\mathrm{ms}$), which agrees well that previously reported.⁹⁵ The combination of optogenetics with the 3D mesoscopic imaging method could provide a new possibility to study the 3D functional neuronal connections after stimulating/inhibiting neurons of certain locations or certain types. In order to study the layer-specific functional circuitry, we will combine cell type-specific optogenetics with our mesoscopic imaging techniques for next step.

We then imaged the 3D functional neural connectivity between sensory and motor cortices in the rodent whisker-barrel system, one sensory pathway that is highly amenable to experimental manipulations.^2,33 Rodents use their whiskers to locate and identify objects.⁹⁶ Forces acting on whiskers excite sensory neurons in the trigeminal ganglion, triggering activity that ascends through the brainstem into ventral posteromedial thalamic nucleus and L4 neurons in the primary somatosensory barrel cortex (S1).^4,5 L4 stellate cells mainly excite layers L2/3 pyramidal neurons, which in turn excite neurons in L5.^4,5 A subset of L2/3 and L5 neurons in S1 project to primary motor cortex (M1, M1 also has layered structure similar to S1, with different types of neurons in each layer). The S1 and M1 are reciprocally connected, and their interaction has long been hypothesized to contribute to coordinated motor output. However, very little is known about the nature and synaptic properties of the S1 input to M1.⁹⁷ Neuroanatomical tracing experiments have often been used to predict circuits and have shown that L2/3 pyramidal neurons in S1 had the densest innervation of deeper layers 5/6 in M1, whereas the L5 pyramidal neurons in S1 preferentially innervated the superficial layers of M1.^4,5 However, axodendritic overlap is not necessarily a good predictor of functional connection strength.^4,47 It will be of great interest to examine the functional consequences of the layer-specific projections from S1 to M1, which can help reveal the primary loci where sensorimotor associations are formed and will provide the possibility to serve as a tool with which one can learn more about brain disease processes and the effects of treatment.⁴ Since our mesoscopic imaging method uses line illumination to scan the FoV, we can easily adjust the length of the line beam to cover both the sensory and motor cortices simultaneously. The 3D spatiotemporal dynamics of cortical activities in both S1 and M1 can be imaged (Fig. 6). The sandwich structure with a distinct band between L5 and L2/3 in M1 can help to reveal the 3D neuronal distribution in M1 receiving inputs from S1, which is promising considering it is within intact cortex in living mice. The distinct band between L5 and L2/3 has also been demonstrated in the ex vivo brain slice studies,^4,97 and a recent study has indicated that the neurons located in a thin laminar zone at the L3/5A border form the genuine layer 4 in motor cortex.⁹⁸ The ability to resolve 3D spatiotemporal dynamics is of importance when considering the functional connections between barrel cortex and vibrissal motor cortex since it can provide us a clue about the strength of S1 input as a function of cortical layer. The $7.5\pm 0.5\mathrm{ms}$ ($n=4$) latency difference between S1 and M1 activity is consistent with the recordings from the 2D imaging⁷⁰ and consistent with a pyramidal neuron axonal conduction velocity of $\sim 450\mu \mathrm{s}/\mathrm{ms}$.⁹⁹ For $\sim 3.5\mathrm{mm}$ separation of S1 and M1, the time required for action potential propagation is $\sim 7.78\mathrm{ms}$. The duration of the activation in S1 shows a significantly higher value than that in L2/3 of M1, but not L5 of M1. The early response in S1 and feedback loop linking S1 and M1 may explain the longer duration of activation in S1 compared to L2/3 of M1.⁴ Since the L5 neurons in M1 will further interact with the motor centers in the brainstem after sensorimotor integration,¹⁰⁰ it may cause the duration of the activation time vary. We need to note that there are only four mice in this study, which is not enough to come to a solid result. More mice should be included to get a statistical conclusion. In this study, the response in motor cortex is induced by deflecting the C2 whisker on the contralateral snout. Thus it is conveyed by inputs from both L2/3 and L5 in S1.⁵ In order to determine the separate contributions of L2/3 and L5 neurons to activating targets in M1, it is necessary to excite or suppress certain layer in S1 and observe the 3D functional responses in M1. In the future, we will combine the optogenetic control strategy with the 3D mesoscopic imaging method to further determine the separate contributions of L2/3 and L5 neurons to activating targets in M1.

Several potential improvements of this mesoscopic imaging technique can be identified. First, the distribution of VSD is not homogenous across different depths in the cortex, although we allowed sufficient time for the VSD to diffuse. A possible alternative solution would be using genetically encoded voltage indicators,^101–110 which will make it possible for cell class specific targeting thus enabling non-invasive longitudinal studies. Surface flatness is another factor that limits the reconstruction accuracy especially when imaging a large FoV, since we assumed the surface being flat in our reconstruction model. The exact shape of the brain surface needs to be extracted in future studies. The assumption of homogeneous optical properties in the reconstruction could potentially limit the accuracy of reconstruction results although the absorption and scattering properties of different types of biological tissues are relatively homogeneous.¹¹¹ An iterative procedure that uses determined optical properties to calculate new fits of signal versus effective source–detector separation might allow for better empirical optimization of both optical properties and further improve the accuracy of the reconstruction model.²⁵ Moreover, photon migration estimation using the mathematical models could not be exact, especially for complicated biological tissues.^64,112,113 The path of photons become more difficult to predict as they scatter further. The mesoscopic methods face resolution deterioration of these reconstructed images as a function of depth.^30,33,112 The problem can be alleviated by combining dense spatial data sets with regularization terms like compressive sensing-based methods.¹¹⁴ The quantitative accuracy and penetration depth can also be improved by incorporating the high-dynamic-range method reported previously.^115,116 In addition, the synthesis of improved long-wavelength VSD has the potential to further enhance the imaging penetration depth.^117–119

In this paper, the time-resolved acquisition protocol has been applied to record the fast neural dynamics, while it requires the biological response to be repeatable for each stimulation trial, which potentially limits the scope of potential applications. We should note that the long-time repetitions could possibly induce long-term plasticity in the neural network. There are two reasons for applying the time-resolved acquisition protocol. First, compared to hemodynamics imaging based on absorption contrast (slow signal changes at second time scale), VSDi reports the neural activity in the brain with millisecond time scale.⁴⁸ Second, the amplitude of change in fluorescence for the response is relatively low ($<0.5\%$) using VSD, which makes it necessary to perform tens of averaging for one recording. Genetically encoded voltage indicators mentioned above could be a possible solution assuming that the SNR can be significantly increased compared with the currently used VSD, which can help reducing the averaging times. In the future, we will also explore other fluorescent dyes that can indicate the neural responses by contrast such as changes in ion concentrations (pH-, calcium-, chloride-, or potassium-sensitive dyes).^48,120–123 These fluorescent dyes indicating changes in ion concentrations happen in seconds time scale and can provide better SNR, which has the potential to significantly reduce the acquisition time for our mesoscopic system and improve the reconstruction accuracy.

In summary, we document that our mesoscopic imaging method could map layer-specific functional regions and their connections. In combination with optogenetic control, the ability to image 3D neuronal responses in the mouse neocortex with high temporal and spatial resolution will yield a wealth of information about circuit organization and function. This methodology has the potential to examine the functional consequences of disrupted functional connectivity in diseased animal models (e.g., genetic and surgical disruptions).