Abstract
ORIGINAL RESEARCH ARTICLE published: 03 April 2014 NEURAL CIRCUITS doi: 10.3389/fncir.2014.00029 Simultaneous imaging of neural activity in three dimensions Sean Quirin*, Jesse Jackson , Darcy S. Peterka* and Rafael Yuste Department of Biological Sciences, Columbia University, New York, NY, USA Edited by: We introduce a scanless optical method to image neuronal activity in three dimensions Edward M. Callaway, The Salk simultaneously. Using a spatial light modulator and a custom-designed phase mask, Institute for Biological Studies, USA we illuminate and collect light simultaneously from different focal planes and perform Reviewed by: calcium imaging of neuronal activity in vitro and in vivo. This method, combining structured Hollis Cline, The Scripps Research illumination with volume projection imaging, could be used as a technological platform for Institute, USA Timothy E. Holy, Washington brain activity mapping. University, USA Keywords: three-dimensional imaging, calcium imaging, volume imaging, spatial-light-modulator, brain Bernardo Sabatini, Howard Hughes activity map Medical Institute, USA *Correspondence: Sean Quirin and Darcy S. Peterka, Department of Biological Sciences, Columbia University, 901 NWC Building, 550 West 120th Street, PO Box 4822, New York, NY 10027, USA e-mail: [email protected]; [email protected] INTRODUCTION by taking advantage of programmable three-dimensional illumi- nation with spatial light modulators (SLM) to simultaneously Optical imaging of neural activity has several advantages over excite neurons located at different focal points, together with alternative strategies such as patch electrodes or electrode arrays. a wavefront coded imaging approach that enables us to collect First, it is minimally invasive and allows for the monitoring of light from all focal points simultaneously. Our technique there- large ensembles of neurons with single-cell resolution (Yuste and fore supersedes scanning by illuminating and collects light from Katz, 1991). In addition, it is compatible with a large variety the neurons of interest in parallel. of functional sensors (voltage, calcium or metabolic indicators, either exogenous or genetically encoded), and can be used for MATERIALS AND METHODS chronic imaging of defined cell populations in vitro or in vivo WAVEFRONT CODING FOR VOLUME PROJECTION MICROSCOPY (Grynkiewicz et al., 1985; Kralj et al., 2012; Chen et al., 2013). However, some key weaknesses remain for functional imag- While SLM based multi-site excitation has been coupled to wide- ing of neuronal cell activity. From the very first microscope field microscopy before, it has been limited to 2D or planar designed by Leeuwenhoek, image acquisition is typically limited acquisition (Nikolenko et al., 2008; Anselmi et al., 2011). In con- to a single plane, while nearly all biological structures are three- trast, we use a phase-only SLM to create multiple well-defined dimensional, thus requiring sequential scanning for volumetric beamlets of light that optically sample many neurons through- imaging. As a result of this sequential scanning, the sampling rate out the sample volume. The wavefront coded imaging system then over the volume is slow relative to neuronal activity (1–100 ms). creates a 2D projection of this extended sample volume onto the Moreover, in highly scattering tissue, although two photon exci- detector plane. Our wavefront coded imaging system is physi- tation has afforded single cell resolution and imaging below cally realized by a phase mask placed between an image relay superficial layers (Horton et al., 2013), images are still mostly gen- of the microscope pupil (Figure 1A) and is optimized for high- erated by serially scanning a single beam (for exceptions see Gobel speed, three-dimensional data acquisition in what we call Volume et al., 2007; Cheng et al., 2011; Katona et al., 2012). Temporal Projection Imaging (VPI) (Dowski and Cathey, 1995; Quirin focusing techniques can deliver high-fidelity illumination pat- et al., 2013). Cells are no longer restricted to a single plane and can terns into scattering tissue within a limited field-of-view but still be freely distributed throughout the three-dimensional sample require axial scanning with a moving objective, potentially cou- (Figure 1B). We note that although in some samples (transpar- pling mechanical motion into the sample (Schrödel et al., 2013). ent or effectively sparse), single photon excitation is sufficient, Despite recent advances to increase the image acquisition speed, we regard two-photon capabilities as essential in turbid envi- few proposals are scalable toward meeting the ultimate goal of ronments and for allowing precise modulation of activity when fast, population voltage imaging with single cell resolution in coupled with optogenetics or caged compounds (Papagiakoumou scattering tissue—a task that would require millisecond temporal et al., 2010; Packer et al., 2012). resolution across wide spatial areas (Alivisatos et al., 2012). Here Operationally, in our method, high spatial-resolution struc- we present an alternative to traditional scanning-based imaging tural volume data is first acquired via conventional single beam Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 1 Quirin et al. Simultaneous imaging in three dimensions FIGURE 1 | Optical layout and characteristics of 3D projection-based results using a transparent fluorescent slab demonstrate that the volume imaging. (A) The optical configuration comprises an illumination path which projection imaging path results in clearer separation of the region-of-interest incorporates a SLM for 3D structured illumination and a modified imaging signals when compared with the conventional imaging path. Note that the path using a phase mask (PM) to suppress the imaging effect of defocus. contrast conserves the number of photons in each image. (E) Wide-field (B) An example 3D illumination pattern using a 20x/0.5NA objective. (C) Ideal imaging results comparing the conventional and volume projection imaging surface profile modulation of the phase mask. (D) Experimental imaging techniques. Note that each image is normalized to the respective peak signal. 3 3 two-photon raster scanning. This volume data is then processed 2 2 i2πα(x +y ) e if x + y ≤ 1 p x, y = , (1) to identify the cells to be targeted by the SLM. Using these tar- otherwise gets, we generate a hologram (phase pattern) that when displayed on the SLM creates the 3D structured illumination which illu- minates the regions of interest (ROIs) in the volume (see Quirin whereweusedvaluesof α = 17 and 12 for M = 20x/0.5NA and et al., 2013 and references therein). Data is acquired by imaging 40x/0.8NA objectives, respectively, and x, y are the normalized the fluorescent activity of all of the ROIs simultaneously by the coordinates of the microscope pupil (Figure 1C). cubic phase wavefront coded imaging system regardless of their The trade-off for this defocus invariance is a decreased con- locations in 3D space (Cathey and Dowski, 2002; Quirin et al., trast in the focused image (Figure 1D)(Cathey and Dowski, 2002; 2013). The deterministic illumination provided by the SLM pro- Quirin et al., 2013). As a result, the acquired wide-field image vides a priori knowledge which can be used to extract additional exhibits a characteristic blur (e.g., an aberrated Point Spread information from the collected image, as explained below. Function) which is now essentially invariant regardless of the three-dimensional location of the ROI (Figure 1E). This blur For simultaneous 3D signal acquisition, our wavefront coded imaging system increases the standard limited depth-of-field, is nonetheless spatially restricted, and reduction of the out-of- focus PSF size compared with traditional imaging allows for a while maintaining the full aperture, and hence collection effi- ciency, of the objective. This custom imaging system uses a much higher local spatial density of ROIs (Figure 1D). Now, fluorescence signals from throughout the volume can be simul- phase-only optical mask in the imaging path to effectively null defocus effects in the image (Dowski and Cathey, 1995). The taneously acquired without loss of contrast due to out-of-focus effect of this is to greatly reduce the axial dependence of the point collection (Figure 1E). As an additional benefit, the data vol- spread function (PSF) (i.e., the image is invariant to 3D position), ume is projected onto a 2D image plane thus reducing the total which allows us to record an aberrated line-of-sight projection data throughput required during acquisition. In principle, this of the sample volume onto the image plane. This phase mask is image can be digitally restored to diffraction-limited resolution described by the complex amplitude profile, by use of deconvolution methods—however, these steps are not Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 2 Quirin et al. Simultaneous imaging in three dimensions necessary here. Instead we report results using a least-squares reduces the image of the SLM onto a set of galvanometer mirrors optimal signal extraction method described in section Targeting (Cambridge 6210H, 4 mm open aperture). The galvanometer Calibration, Image Acquisition and Signal Reconstruction, which mirrors are located conjugate to the microscope objective pupil had not been applied in earlier methods. of an Olympus BX-51 microscope by use of an Olympus pupil transfer lens (f5 = 50 mm) and the mounted tube lens (fTL = MICROSCOPE SETUP 180 mm). Two microscope objectives were used for the exper- The structured-illumination microscope with VPI uses a sin- iments presented in the manuscript—an Olympus 40x/0.8NA gle laser for exciting fluorescence of either the calcium indica- WI (slice experiments) and an Olympus 20x/0.5NA WI objective tor dye (Fura2-AM) or a genetically-encoded calcium indicator (in vivo experiments), though many other objectives performed (GCaMP5G). A detailed optical layout of the system is given similarly during informal testing. In practice, due to transmis- in Figure 2. The laser source is a Coherent Mira HP (∼140 fs sion losses from the passive and active optical elements, such as pulses, 80 MHz, linear polarization) which can be manually tuned the EO modulator and the SLM (from pixel-cross talk, pixel fill between ∼720 and 1100 nm. At λ = 800 nm, the system provides factor, etc.), and from beam vignetting at the telescope after the 3.8 W and at λ = 920 nm, 2.2 W (for Fura2-AM and GCaMP5G SLM, the overall transmission efficiency from laser to sample was excitation, respectively). The output beam was directed through approximately 30%. an electro-optic (EO) modulation device (Pockels cell) to mod- When operating in conventional two-photon raster scanning ulate intensity on the sample (ConOptics EO350-160). A broad- mode, Olympus Fluoview was used to control the galvanometers band λ/2 waveplate (Thorlabs AHWP05M-980) was located after and digitize the signal from a Photo-Multiplier Tube (PMT— the EO modulator to rotate the polarization state to be paral- Hamamatsu H7422P-40) located above the microscope. During lel with the active axis of the Spatial Light Modulator (Boulder PMT imaging a removable silvered mirror (M3, mounted in Nonlinear Systems, XY-512) located further downstream. A shut- Thorlabs LC6W and LB4C) redirect the collected fluorescence ter (Vincent Associates, LS6Z2) was placed in the beam path emission into the PMT module. in order to control illumination state condition (i.e., “ON” and To operate in volume projection mode, custom software was “OFF”). A 1:2 telescope (f1 = 50 mm, f2 = 100 mm, Thorlabs developed under MATLAB (The MathWorks, Natick, MA) which plano-convex lenses) scales the optical beam to approximately loads the correct look-up table and generates a hologram for 8mmtofill theactiveareaofthe SLM. Theexpandedbeam the SLM that targets the selected ROIs located throughout the is then redirected to the SLM by a periscope. The SLM has sample volume. When the illumination shutter is opened, the a custom look-up table which was experimentally determined sample is illuminated with a custom light pattern that addresses at both wavelengths reported for use. The angle of incidence the targeted cell somata. For VPI acquisition, the PMT redirection of the illumination beam to the SLM was ∼10 .A4:1tele- mirror, M3, is removed and the fluorescence emission is passed scope (f3 = 300 mm, f4 = 75 mm, Thorlabs plano-convex lenses) through an optical relay system with the custom phase mask. The FIGURE 2 | Optical system configuration. Plan (A) and elevation (B) views provide the relative distances between optical elements for reproducing the microscope configuration. Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 3 Quirin et al. Simultaneous imaging in three dimensions optical relay consists of a 1:1 telescope (f4f = 150 mm, Thorlabs 2 illumination pattern for the SLM (see Quirin et al., 2013,and references therein for details), setting the EO modulator for achromatic doublets) with a custom designed surface relief pat- terned onto a 1 diameter quartz plate that is placed a distance f4f appropriate illumination power, opening the shutter and acquir- behind the first lens. Note that no amplitude modulation is nec- ing time-series images that were saved as 16-bit TIFs (image essary (i.e., the mask is transparent), critical for high sensitivity acquisition under the Andor Solis environment). Because the imaging, and that the surface relief of the mask is described by, magnified image of the neurons was relatively large (∼200 μm per soma) when using an M = 20× objectivecomparedtothe imaging detector pixel size (16 μm), pixel binning of 4×4was arg p x, y h x, y = π (2) routinely used to improve SNR without any appreciable loss of 2 n − 1 spatial resolution performance. Using the calibration matrices described above, an experimentally measured PSF (taken from the calibration process) is convolved with the estimated location where h is the height, p x, y is the complex function describ- of each ROI on the CCD to form a basis set of images—one image ing the pupil (Equation 1), λ is the emission wavelength and n is the refractive index of the substrate (here, we use fused quartz for each target. Each frame from the time-series stack of images was decomposed into a linear superposition of this basis set using with n = 1.462 at λ = 510 nm). In practice, our realized sur- face relief is a discrete 8-level approximation of Equation 2 with a least-squares fitting, plus one image for background estimation. Formally this was accomplished by creating a matrix of N images, α = 200 over an 18 mm diameter, with an optimal λ ≈ 510 nm. A 510 ± 30 nm chromatic filter was located immediately after ⎡ ⎤ the phase mask. An imaging detector (Andor iXon Ultra 897, B (1) ... B (1) i N electron-multiplying CCD) is located in the imaging plane of the ⎣ ⎦ B = ... . .. ... (3) 1:1 relay system. B (m ∗ n) ... B (m ∗ n) i N TARGETING CALIBRATION, IMAGE ACQUISITION AND SIGNAL whereeachcolumnisone m × n image, lexicographically ordered RECONSTRUCTION in a column, representative of the expected pattern from the ROI Precise targeting of the ROIs (neurons) was facilitated by a 3D (known from the deterministic illumination). In principle, these calibration of the SLM projection patterns. A calibration phan- images can be given by simulation or found experimentally. Later tom was created using a 2% agarose mixture with a dye solution acquisition of the experimental image, I(t), at time t (see for (yellow highlighter dye) at a 1:1 ratio. Standard projection grids example, Figure 1E) can then be characterized by, are projected by the SLM at up to 20 z-positions through the vol- ume of interest and the resulting fluorescence image was used I (t) = B · W (t) + n (4) to calculate the affine transformation that maps the SLM-based illumination pattern to the EM-CCD. A 3rd-order polynomial to describe the image formation where n is additive random noise. fit of each element from the axially dependent affine matrix was The least-squares fitting, used to approximate the axial evolution of the transform and stored for later use (Quirin et al., 2013). A pollen grain slide W (t) = min I (t) − B · W (t) (5) W (t) was similarly used to calibrate the axial dependent, affine trans- formation from the EM-CCD image to the PMT image frame. quickly yields the individual fluorescence from each target, w (t). The slide was translated through the volume range of interest and Any systematic movements of the sample would be easily an EM-CCD image and PMT image were acquired for compari- detected as synchronous changes across multiple ROIs, and would son. An automated fitting routing estimated the axial-dependent not be expected to show the characteristic fast rise and slow decay affine transform matrix and, as with the SLM to EM-CCD trans- of calcium transients resulting from action potentials. No such form, a 3rd-order polynomial fit is made to each element of the matrix. Matrix multiplication of the PMT to EM-CCD and atypical fluorescent changes were detected during our experi- ments, although controls (gently tapping the microscope stage) the EM-CCD to SLM transforms yield the coordinates to load onto the SLM for precise targeting. Though it is known that any showed we can easily detect and distinguish such events. errors in matching the refractive index of the phantom mate- rial to that of the biological sample will result in uncorrected ANIMAL EXPERIMENTS spherical aberration for the structured illumination pattern and All animal experiments were performed in accordance under that real tissue has inherent inhomogeneity that will further aber- approved protocols following the regulations and guidelines of rate the illumination beamlets, in practice we find that both of Columbia University’s IACUC and ICM. Mice experiments were these factors appear negligible when using the above calibration performed in C57BL/6 mice aged P11 to P60 using 400 μmthick prescription for the axial ranges and samples described in this coronal slices loaded with FURA-2AM. Zebrafish experiments paper. were performed at ages P6-P8 and were performed in accordance Images acquired on the imaging detector in volume pro- with the regulations and guidelines of Columbia University and jection mode were processed using custom analysis software the Howard Hughes Medical Institute. The zebrafish sample is written in MATLAB (The Mathworks, Natick, MA). Volume pro- held fixed within a bead of 2% low-melting point agarose to jection acquisition mode consists of generating the hologram reduce motion artifacts. Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 4 Quirin et al. Simultaneous imaging in three dimensions RESULTS <11 mW incident power per spot) and recorded at 55 volume We first demonstrated the system capabilities using calcium indi- projection images per second (VPPS) with an exposure time of cators in vitro, in acute mouse hippocampal slices to demonstrate 16.6 ms per frame (Figures 3B, 4). Individually distinguished flu- single-cell resolution. Sections of the dentate gyrus were selected orescence activity from the dense three-dimensional cluster of for 3D imaging, specifically because this brain structure pro- cells demonstrates that unique signals from individual neurons vides a challenging opportunity with its dense ensemble of cells can be extracted easily, and with high SNR. Before and after one (Figure 3A). Recent work has reported the presence and impor- identified burst, independent activity is observed in each cell body tance of functional cell-cluster activity with 2D imaging in the regardless of proximity of the targets in space—demonstrating hippocampus (Muldoon et al., 2013; Ramirez et al., 2013), how- cellular resolution. ever these events have never been observed in 3D. We first present This is perhaps unsurprising after examining the acquired an example of the dense, three-dimensional packing of cells in cubic-phase images (Figure 4C). Because SLM-based multisite this structure, which was acquired by conventional sequential two-photon excitation preserves the exquisite precision and sen- raster scanning, and was then used to identify the cells for tar- sitivity of two photon targeting (Nikolenko et al., 2008), and the geting (Figure 3A). Every visually identified cell in the FOV was cubic phase mask provides reasonable energy compactness (blur targeted for illumination by creating and loading the associated is small), and is highly efficient (phase-only modulation), even hologram pattern on the SLM. In one representative demon- with dense targeting, simple signal extraction procedures provide stration, data are acquired for 107 such ROIs (an average of high performance. An example of this is shown in Figure 4D, FIGURE 3 | Simultaneous 3D imaging of hippocampal neuronal activity in (B). Despite simultaneous burst activity in 5 of the neighboring cells, in vitro at 55 volume projections per second. (A) 3D structural data independent calcium transients are detectable. The axial location of each cell acquired by two-photon raster-scanning image stack. (B) 3D functional is given in parenthesis behind the respective fluorescence trace. (C) A imaging with single cell resolution, where 9 of 107 total cells are selected fine-temporal resolution view of the burst activity highlights the variability in from the 3D volume [boxed region of (A)] and their respective activity is given both the temporal and amplitude modulation of the calcium transients. Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 5 Quirin et al. Simultaneous imaging in three dimensions are separated by ∼10 μm. These traces show very clean signals, with little to no sign of cross contamination, and show modula- tion nearly identical to the signal extracted using the least squares fitting procedure (Figure 4B, cell #19, also shown in Figure 4D). As a practical note, in our laboratory, we use this as a rule-of- thumb for the overall quality and performance of the imaging system. If the central lobes of the images of each target are iden- tifiable as distinct puncta, we find that simple methods for signal recovery, such as our linear fitting procedure, and in many cases, traditional ROI selection, will be sufficient to extract high quality signatures from each target. A key advantage of this method is that minute differences in the onset timing and calcium dynamics of the 3D ensemble can now be resolved with high SNR even at these high tempo- ral sampling rates (Figures 3C, 4). In slices from older animals, with less dense labeling, we demonstrated this method at sam- pling rates up to 125 VPPS (<20 mW incident power per spot and 6.3 ms exposure times) at single-cell resolution and with high SNR (Figure 5). We have compared the distribution of fluorescent changes and activity patterns collected using the VPI approach to those recorded with a long established imaging method (one pho- ton, laser-illuminated spinning disk confocal, with an EM-CCD, recording at 20 fps) that were accompanied with cell-attached recordings (Figure 6). The recordings also match prior measure- ments from our laboratory, where ground truth sensitivity to single action potential was clearly demonstrated using the orig- inal planar SLM microscope with Fura-2 loaded cortical slices (Nikolenko et al., 2008). Although this is indirect evidence, we believe this indicates that we should be sensitive and have suf- ficient SNR in most instances to record single action potentials. Co-firing dentate gyrus cells likely represent ensembles of hip- pocampal neurons involved in pattern separation (Muldoon et al., 2013; Ramirez et al., 2013). Our technique thus allows for the visualization of these activity patterns with unprecedented spatio- temporal precision. We also reconstructed the in vivo neuronal activity of the larval zebra fish (P7) with GCaMP5G (Akerboom et al., 2012; FIGURE 4 | Fluorescence change vs. time for all targets in Figure 3. Ahrens et al., 2013) to map the temporal record of brain-wide (A) 109 cells have been targeted by 109 focal spots in the 3D volume and calcium transients at high-speed. Different spatial scales can be the relative fluorescence change of each targeted location is displayed. accessed with the VPI technique by simply selecting an alterna- Data has been low-pass filtered by a convolution with a Gaussian filter of tive microscope objective or modulating the strength of the phase σ = 92 ms. (B) A few representative traces indicate the SNR available by this method. (C) Five second average of the raw acquired images, aberration present on the wavefront coding element at the pupil illustrating both the characteristic cubic-phase aberration and the readily (via use of α,inEquation1). As areference,weshowconven- identifiable individual targets. Left image red inset shows area magnified in tional two-photon galvanometric scanning images that contain 4d, right frame is the same image, contrast adjusted to highlight dim brain-wide activity patterns where both individual cells and neu- features. Scale bar is 40 μm in both images. (D) Magnified area of image in (C), showing dense targeting, and again illustrating the cubic-phase ropil are recruited and fluoresce (Figures 7A,E) (see also Ahrens aberration. Scale bar is 15 μm. Red and blue boxes indicate ROIs used to et al., 2013; Panier et al., 2013). In recent work, the temporal res- generate traces shown on the right, using “traditional” DF/F signal olution of such whole-brain activity mapping was limited by the extraction, without fitting. The red trace corresponds to cell #19 in (B),and axial scan speed and the camera frame rate, sampling at 0.8 Hz displays nearly identical absolute modulation and SNR. Note the near (Ahrens et al., 2013) or 4 Hz using selective axial planes (Panier absence of crosstalk between the traces, although laterally, the two sources are separated by ∼10 μm. Traces are low-pass filtered as in (A). et al., 2013). In contrast, the technique proposed here operates simultaneously across axial planes. Moreover, VPI has no mov- ing parts and therefore exhibits no mechanical acquisition speed which displays a small area of the total image, and shows the limitation, nor the complications that can arise from coupling DF/F traces extracted with a small ROI over only the bright cen- motion of the system into the sample. For illustration, we targeted tral lobe of the cubic-phase PSF (not the basis fitting procedure 49 randomly distributed cells within a 350 × 350 × 150 μmvol- described earlier) for two targeted cells, whose lateral projections ume (on average <14 mW incident power per spot). Using these Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 6 Quirin et al. Simultaneous imaging in three dimensions FIGURE 5 | 3D imaging of neuronal activity in dentate gyrus with (A) and their respective activity is shown in (B). Note that neighboring single-cell resolution at 125 volume projections per second. (A) The cells are seen to have independent activity—confirming single-cell location of neuronal cells within the volume of interest is determined by resolution. The axial location of each cell is given in parenthesis behind collecting a two-photon raster-scanning image stack and identifying cell the respective fluorescence trace. 58 total targets were monitored (out bodies. To demonstrate 3D sensing capability with single-cell resolution, of 61 total labeled cells identified via visual inspection) throughout the 7 neighboring cells are selected from the 3D sub-volume highlighted in tissue volume. ROIs, a custom hologram was created and loaded to the SLM the effect of scattering on the PSF used here and believe that (red targets, Figure 7B). With the simultaneous multisite two- this method can be applied up to 2–3 scattering lengths deep photon excitation, multiple waves of large scale near-synchronous providing that a relatively sparse selection of 3D points is used calcium transients are precisely recorded at 30 VPPS (32 ms expo- (Figure 8). However, there are active developments which will sure times per frame). The monitored active cells in these waves extend the absolute depth where this method can be successfully were temporally sorted based on their activity in the first wave, applied. First, ongoing improvements in red shifted indicators and this same ordering was used to display the activity in the sub- and light sources, whose longer wavelengths penetrate more sequentwaves.Weobservedthatorderingisstronglypreserved in deeply through tissue, could allow for significant and sizeable the subsequent waves, with sub-second precision, despite occur- volumetric imaging with high spatial and temporal resolution. ring many minutes later (Figure 7C). It is important to note that With the development of faster SLMs, or external modulation this is not an epileptiform or simple activity pattern—the spa- schemes, temporal coding and multiplexing can be added to tiotemporal profile of the activity is dispersed throughout the augment the selectivity at large imaging depths by imposing an sample with the cells closest geographically not necessarily having a priori temporal structure on each target, allowing for phase the smallest relative offsets in the onsets of activity (Figure 7D). locked detection (Ducros et al., 2013). This could be a critical This is, to our knowledge, the first demonstration of simultaneous improvement, as it would allow for imaging neurons whose pro- three dimensional calcium activity imaging in in vivo zebra fish jections fall within the same pixels of the camera (such as neurons preparations with single cell precision at a temporal resolution directly on top of one another). Additionally, advances in algo- sufficient to resolve the dynamics of neural activity patterns. rithms for signal extraction that jointly consider the recorded spatial and temporal signatures of each source will improve sep- DISCUSSION aration for overlapping and delocalized signals (Pnevmatikakis et al., 2013). Using SLM structured illumination with VPI, we can image the activity of neuronal populations throughout the brain of the lar- Both the illumination and data acquisition are simultaneous and can target multiple ROIs throughout the volume of tissue val zebrafish simultaneously. But, although the larval zebra fish is transparent, most living organisms are not. Our method as allowing for parallel activation and imaging. We have demon- implemented here still relies on direct optical imaging and is strated high SNR recording of calcium transients at speeds of up adversely affected in highly scattering environments. This has to 125 Hz, limited by the camera transfer rate. This frame rate been a challenge for all optical imaging methods in neuroscience, and sensitivity is already well matched to the current generation of and the described method is not immune to it. We evaluated GFP based genetically encoded voltage indicators, such as Arclight Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 7 Quirin et al. Simultaneous imaging in three dimensions FIGURE 6 | Calcium transients in hippocampal slices display similar recording in cell attached are shown (below) with the filtered (0.2–4 kHz) amplitudes and kinetics with confocal microscopy and volume cell attached recording and threshold crossing spikes. Note the ability of projection 3D imaging. (A) Example calcium transients from three the Fura-2AM to resolve single action potential firing, which was usually representative granule cell ROIs, acquired from the dentate gyrus, using accompanied by a 5% dF/F. Therefore, in most experiments, single, or at a spinning disk confocal microscope. The sample preparation was minimum doublet spiking would be detected using calcium imaging. identical to the brain slice experiments described in the main text. Only (D,E) The same as (A,B), but with three examples of ROIs measured a short 2 min segment of the data is shown. Data were acquired at 20 using the 3D volume projection imaging system described in the fps. (B) (Left) dF/F distributions for the three ROIs shown in (A).Onthe manuscript. Data displayed were acquired at 55VPPS, and smoothed with right is the y-axis zoomed histogram, shown to highlight the positively a Gaussian filter to match the sample rate of the confocal images. Note skewed dF/F distributions characteristic of active neuron spiking. (C) A the similar shape calcium transients (D), and dF/F distributions (E).The representative experiment showing the ability of Fura-2AM calcium different size calcium dynamics and the dF/F distributions suggest that transients to faithfully resolve action potential firing in a cell attached single or at minimum doublet spiking can be resolved with the volume recording. The deconvolved calcium dynamics (above) from a cell projection imaging technique. (Jin et al., 2012). Advances in the technologies utilized here, such from the number of targets. This is an important distinction, and as higher power lasers and increased pixel count SLMs, along with deserves attention. Admittedly, the best fast AOD systems are cur- faster cameras will soon accommodate direct imaging of voltage rently capable of high performance, and set a mark by which all activity in 3D. alternative high speed approaches are measured. However fast the In contrast to high speed acousto-optic deflector microscopy performance, we believe they will be difficult to scale significantly techniques, 3D SLM microscopy has the advantage of decoupling beyond their current levels because with very high speed ran- the pixel dwell time (time allotted to collect fluorescence signal) dom access approaches, the total collected fluorescence signal per Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 8 Quirin et al. Simultaneous imaging in three dimensions FIGURE 7 | Simultaneous 3D imaging of neuronal activity of marker is indicative of the amplitude of signal modulation at each zebrafish in vivo at 30 volume projections per second. (A) location. (E) Comparison of the fluorescence spike time profile between Conventional single plane two-photon raster-scan data acquisition reveals the two imaging modalities shows the signal of the volume projection sequences of coordinated whole-brain activity. (B) 49 targets are technique to be consistent with the two-photon scanning acquisition. distributed throughout an acquisition volume of 284 × 270 × 114 μmto The scanning mode data series was taken from the box marked area in sample activity in 3D. (C) Multiple repetitions of these events occur and (A) while the four volume projection mode series’ were taken from the exhibit similar time-courses. (D) The associated spatial patterns of the x marked areas. Note that the fluorescence signal has been normalized events confirm that the activity has repetitive structure. The size of the in each time series of (C) and (E) for visualization. target is limited by the relatively low duty cycle per location and number of targets is the overall laser power available, which can the maximum emission rate of the fluorophore (a physical charac- be arbitrarily increased until the overall power deposition over teristic of the particular chromophore). Though each target could the entire FOV exceeds the total acceptable heat load for the sam- theoretically receive the full output of the laser, simply increasing ple. We note that with programmable SLMs, one can achieve the the intensity of illumination yields diminishing returns because optimal synthesis of both methods with simultaneous, multisite of fluorophore saturation and continued high intensity excita- random access targeting, offering tremendous flexibility in mon- tion will lead to increased photodamage and bleaching, rather itoring activity. Also, while we describe two photon SLM-based than increased signal, along with a loss of spatial resolution (Hopt multiple beamlet excitation here, any predetermined structured and Neher, 2001). With SLM microscopy, the issue is different. illumination can be used. Each target has very high duty cycle with respect to the sampling In conclusion, we present a technique for fast, simultaneous, rate (true simultaneous multisite illumination), but with lower two-photon optical data acquisition of neuron activity which is average instantaneous power—multiplexing the beam results in distributed throughout three dimensions. This has been demon- diminished fluence per target, with a corresponding reduction strated in different animal preparations, both in vitro and in vivo, in total fluorescence per target. In this case, what limits the total that are relevant for neuroscience. The combination of structured Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 9 Quirin et al. Simultaneous imaging in three dimensions FIGURE 8 | Axial dependence of the imaging PSF vs. depth in τ = 65.7 μm. In an attempt to quantify the loss of the PSF fidelity with INC scattering tissue. Characterization of the imaging performance of volume increasing scattering, a circular region was defined [centered on the peak projection imaging using GCaMP6s labeled tissue at P31 in white matter intensity, insert (B)] and the relative loss of energy was plotted as a tissue. GCaMP6s was chosen to provide near uniform labeling throughout function of depth, yielding τ = 63.7 μm (B). An image of the PSF is PSF the tissue. (A) A focused spot was translated axially via a mechanical given as a function of τ in (C) and the PSF fidelity is observed to INC stage and the resulting collected intensity is reported, yielding a mean degenerate by ∼3τ , at which point the system performance approaches INC scattering length of the incident illumination (λ = 920 nm) of that of a conventional SLM microscope. illumination with volume projection imaging appears to us a Chen, T. W., Wardill, T. J., Sun, Y., Pulver, S. R., Renninger, S. L., Baohan, A., et al. (2013). Ultrasensitive fluorescent proteins for imaging neural activity. Nature promising platform for future work on brain activity mapping 499, 295–300. doi: 10.1038/nature12354 (Alivisatos et al., 2012). Cheng, A., Goncalves, J. T., Golshani, P., Arisaka, K., and Portera-Cailliau, C. (2011). Simultaneous two-photon calcium imaging at different depths with spa- ACKNOWLEDGMENTS tiotemporal multiplexing. Nat. Methods 8, 139–142. doi: 10.1038/nmeth.1552 Sean Quirin, Darcy S. Peterka, Jesse Jackson, and Rafael Yuste are Dowski, E. R., and Cathey, W. T. (1995). Extended depth of field through wave- front coding. Appl. Opt. 34, 1859–1866. doi: 10.1364/AO.34.001859 supported by the NIH Director Pioneer Award (DP1EY024503), Ducros, M., Houssen, Y. G., Bradley, J., Sars, V. D., and Charpak, S. (2013). NIMH (R41MH100895), Keck Foundation and NARSAD. Sean Encoded multisite two-photon microscopy. Proc. Natl. Acad. Sci. U.S.A. 110, Quirin was additionally supported by the NEI (T32EY013933). 13138–13143. doi: 10.1073/pnas.1307818110 This material is based upon work supported by, or in part by, Gobel, W., Kampa, B. M., and Helmchen, F. (2007). Imaging cellular network the U.S. Army Research Laboratory and the U.S. Army Research dynamics in three dimensions using fast 3D laser scanning. Nat. Methods 4, 73–79. doi: 10.1038/nmeth989 Office under contract number W911NF-12-1-0594. The authors Grynkiewicz, G., Poenie, M., and Tsien, R. Y. (1985). A new generation of Ca2+ thank Misha Ahrens and Davis Bennett for assistance with the indicators with greatly improved fluorescence properties. J. Biol. Chem. 260, zebrafish preparations. 3440–3450. Hopt, A., and Neher, E. (2001). Highly nonlinear photodamage in two-photon REFERENCES fluorescence microscopy. Biophys. J. 80, 2029–2036. doi: 10.1016/S0006- Ahrens, M. B., Orger, M. B., Robson, D. N., Li, J. M., and Keller, P. J. (2013). Whole- 3495(01)76173-5 brain functional imaging at cellular resolution using light-sheet microscopy. Horton, N. G., Wang, K., Kobat, D., Clark, C., Wise, F., Schaffer, C., et al. (2013). Nat. Methods 10, 413–420. doi: 10.1038/nmeth.2434 In vivo three-photon microscopy of subcortical structures of an intact mouse Akerboom, J., Chen, T. W., Wardill, T. J., Tian, L., Marvin, J. S., Mutlu, S., et al. brain. Nat. Photonics 7, 205–209. doi: 10.1038/nphoton.2012.336 (2012). Optimization of a GCaMP calcium indicator for neural activity imaging. Jin, L., Han, Z., Platisa, J., Wooltorton, J. R., Cohen, L. B., and Pieribone, V. A. J. Neurosci. 32, 13819–13840. doi: 10.1523/JNEUROSCI.2601-12.2012 (2012). Single action potentials and subthreshold electrical events imaged in Alivisatos, A. P.,Chun, M.,Church, G. M.,Greenspan,R.J., Roukes,M.L., and neurons with a fluorescent protein voltage probe. Neuron 75, 779–785. doi: 10.1016/j.neuron.2012.06.040 Yuste, R. (2012). The brain activity map project and the challenge of functional connectomics. Neuron 74, 970–974. doi: 10.1016/j.neuron.2012.06.006 Katona, G., Szalay, G., Maak, P., Kaszas, A., Veress, M., Hillier, D., et al. (2012). Fast two-photon in vivo imaging with three-dimensional random- Anselmi, F., Ventalon, C., Begue, A., Ogden, D., and Emiliani, V. (2011). Three- dimensional imaging and photostimulation by remote-focusing and holo- access scanning in large tissue volumes. Nat. Methods 9, 201–208. doi: 10.1038/ graphic light patterning. Proc. Natl. Acad. Sci. U.S.A. 108, 19504–19509. doi: nmeth.1851 10.1073/pnas.1109111108 Kralj, J., Douglass, A. D., Hochbaum, D. R., Maclaurin, D., and Cohen, A. E. (2012). Cathey, W. T., and Dowski, E. R. (2002). New paradigm for imaging systems. Appl. Optical recording of action potentials in mammalian neurons using a microbial Opt. 41, 6080–6092. doi: 10.1364/AO.41.006080 rhodopsin. Nat. Methods 9, 90–95. doi: 10.1038/nmeth.1782 Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 10 Quirin et al. Simultaneous imaging in three dimensions Muldoon, S. F., Soltesz, I., and Cossart, R. (2013). Spatially clustered neu- Ramirez, S., Liu, X., Lin, P.-A., Suh, J., Pignatelli, M., Redondo, R. L., et al. ronal assemblies comprise the microstructure of synchrony in chroni- (2013). Creating a false memory in the Hippocampus. Science 341, 387–391. cally epileptic networks. Proc. Natl. Acad. Sci. U.S.A. 110, 3567–3572. doi: doi: 10.1126/science.1239073 10.1073/pnas.1216958110 Schrödel, T., Prevedel, R., Aumayr, K., Zimmer, M., and Vaziri, A. (2013). Brain- Nikolenko, V., Watson, B. O., Araya, R., Woodruff, A., Peterka, D. S., and wide 3D imaging of neuronal activity in Caenorhabditis elegans with sculpted Yuste, R. (2008). SLM Microscopy: scanless two-photon imaging and pho- light. Nat. Methods 10, 1013–1020. doi: 10.1038/nmeth.2637 tostimulation using spatial light modulators. Front. Neural Circuits 2:5. doi: Yuste, R., and Katz, L. C. (1991). Control of postsynaptic calcium influx in 10.3389/neuro.04.005.2008 developing neocortex by excitatory and inhibitory neurotransmitters. Neuron Packer, A. M., Peterka, D. S., Hirtz, J. J., Prakash, R., Deisseroth, K., and Yuste, R. 6, 333–344. doi: 10.1016/0896-6273(91)90243-S (2012). Two-photon optogenetics of dendritic spines and neural circuits. Nat. Methods 9, 1202–1205. doi: 10.1038/nmeth.2249 Conflict of Interest Statement: The authors declare that the research was con- Panier, T., Romano, S. A., Olive, R., Pietri, T., Sumbre, G., Candelier, R., et al. ducted in the absence of any commercial or financial relationships that could be (2013). Fast functional imaging of multiple brain regions in intact zebrafish lar- construed as a potential conflict of interest. vae using selective plane illumination microscopy. Front. Neural Circuits 9:65. doi: 10.3389/fncir.2013.00065 Received: 03 January 2014; paper pending published: 31 January 2014; accepted: 10 Papagiakoumou, E., Anselmi, F., Begue, A., de Sars, V., Gluckstad, J., Isacoff, E. March 2014; published online: 03 April 2014. Y., et al. (2010). Scanless two-photon excitation of channelrhodopsin-2. Nat. Citation: Quirin S, Jackson J, Peterka DS and Yuste R (2014) Simultaneous imaging Methods 7, 848–854 doi: 10.1038/nmeth.1505 of neural activity in three dimensions. Front. Neural Circuits 8:29. doi: 10.3389/fncir. Pnevmatikakis, E. A., Machado, T. A., Grosenick, L., Poole, B., Vogelstein, J. 2014.00029 T., and Paninski, L. (2013). Rank-penalized Nonnegative Spatiotemporal This article was submitted to the journal Frontiers in Neural Circuits. Deconvolution and Demixing of Calcium Imaging Data.Available Copyright © 2014 Quirin, Jackson, Peterka and Yuste. This is an open-access article online at: http://www.stat.columbia.edu/∼liam/research/abstracts/cosyne- distributed under the terms of the Creative Commons Attribution License (CC BY). 13/eftychios.pdf The use, distribution or reproduction in other forums is permitted, provided the Quirin, S., Peterka, D. S., and Yuste, R. (2013). Instantaneous three-dimensional original author(s) or licensor are credited and that the original publication in this sensing using spatial light modulator illumination with extended depth of field journal is cited, in accordance with accepted academic practice. No use, distribution or imaging. Opt. Express 21, 16007–16021. doi: 10.1364/OE.21.016007 reproduction is permitted which does not comply with these terms. Frontiers in Neural Circuits www.frontiersin.org April 2014 | Volume 8 | Article 29 | 11
Journal
Frontiers in Neural Circuits
– Unpaywall
Published: Apr 3, 2014
