Open Access
Add to BibSonomyAbstract
Multiphoton excitation fluorescence microscopy is the preferred method for in vivo deep tissue imaging. Many biological applications demand both high imaging speed and the ability to resolve multiple fluorophores. One of the successful methods to improve imaging speed in a highly turbid specimen is multifocal multiphoton microscopy (MMM) based on use of multi-anode photomultiplier tubes (MAPMT). This approach improves imaging speed by using multiple foci for parallelized excitation without sacrificing signal to noise ratio (SNR) due to the scattering of emission photons. In this work, we demonstrate that the MAPMT based MMM can be extended with spectral resolved imaging capability. Instead of generating multiple excitation foci in a 2D grid pattern, a linear array of foci is generated. This leaves one axis of the 2D MAPMT available for spectral dispersion and detection. The spectral-resolved MMM can detect several emission signals simultaneously with high imaging speed optimized for high-throughput, high-contents applications. The new procedure is illustrated using imaging data from the kidney, peripheral nerve regeneration and dendritic morphological data from the brain.
© 2014 Optical Society of America
1. Introduction
Multiphoton excitation fluorescence microscopy has inherent 3D resolution due to the nonlinear dependence of excitation efficiency on incident light flux [1–3]. Multiphoton excitation can be localized in a femtoliter region at the focal point of a high numerical aperture objective lens reducing specimen photodamage and photobleaching. Multiphoton microscopy further can excite many fluorophores with infrared radiation enhancing excitation light penetration depth in turbid specimens where scattering is reduced. Equally important, unlike confocal microscopy, no confocal pinhole is required for depth resolution allowing scattered emission photons to be collected effectively thereby greatly enhancing image signal to noise ratio (SNR) in turbid specimens. Consequently, multiphoton excitation fluorescence microscopy has become the preferred tool for deep imaging of highly scattering, within a few scattering mean free paths, tissue specimens [4]. In many applications, high speed imaging is critical. For example, high speed imaging is useful for studies of neuronal plasticity that require high resolution mapping of the whole dendritic arbor, covering almost 1 mm3 in volume, while minimizing the period of anesthesia to reduce animal stress [5–10]. One of the methods to improve the imaging speed is multifocal multiphoton microscopy (MMM) [11, 12]. With a lenslet array or diffractive optical element (DOE) [13, 14], a number of foci are generated simultaneously in a specimen and scanned together. Each excitation focus scans a fraction of the whole imaging area; the final image is synthesized by montaging the data from each focus. Within the limit of the available laser power, the number of foci can be maximized and the imaging speed can be improved in proportion to the number of foci. The early MMM used imaging detectors, such as CCD cameras, to record emission signals from multiple foci simultaneously. However, the turbidity of typical biological specimens results in emission photons being scattered to neighbor pixels resulting in a blurred image with degraded SNR. This difficulty was recently partly circumvented by introducing descanned MMM using a detection scheme based on using a multianode photomultiplier tube (MAPMT). With this approach, the scattered emission photons are much more effectively collected by the large area anodes that correspond to tens of micron size areas on the specimen plane providing much better immunity to SNR degradation due to emission photon scattering [15]. Figure 1 shows the schematic of a MMM system. The excitation laser light is divided to several beamlets by the DOE, and they are delivered to the scanning mirrors with different angles by two lenses in the 4-f location. All the beamlets are scanned by the scanning mirrors together, and expanded to slightly overfill the back aperture of the objective lens. The beamlets enter the objective lens with different incident angles, are focused in the sample, and generate the multiple excitation foci distributed into a square 2D array. The emission signals from the foci are collected by the same objective lens, travel along the same beam path as the excitation light, and are descanned by the same scanning mirrors. As the arrival time of the emission photons is much faster than the dwell time for one pixel, the emission photons from each focus are descanned and become stationary. The emission photons are separated by the dichroic mirror from the excitation laser light, and focused at the cathode locations that correspond to the center of each anode of the MAPMT. With the large area of the each anode, much of the scattered emission photons can be effectively collected in the designated channels. Therefore, the deleterious effect of emission photon scattering can be effectively suppressed.
Many biological imaging applications require not only high imaging speed but also the simultaneous detection of multiple fluorophores to delineate structural relationships between different cellular and tissue constituents. Recently, this approach has been used to study synapse formation and dissociation in the context of dendritic remodeling in a mammalian brain [16–18]. Spectral detection is essential to simultaneously monitor the multiple fluorophores marking dendrites, excitatory synapses and inhibitory synapses. At the same time, high imaging speed remains critical in order to image the large dendritic arbor at high resolution so that individual synapses can be reliably resolved. Going beyond the study of a few neurons, the use of color barcoding method has been proposed for mapping large neuronal networks, but would require high throughput spectral-resolved imaging [19]. In addition to morphology determination, high throughput spectral-resolved MMM will also find application in imaging multiple spectral ranges to assay tissue metabolic and biochemical states. For example, spectrally resolved multiphoton imaging has been used to study signaling processes including Ca2+ and Hg2+ ratiometric imaging and protein-protein interactions based on fluorescence resonance energy transfer [20–22], and the throughput of these studies can be greatly improved with the development of spectral-resolved MMM systems.
High speed imaging and spectral detection can, in principle, be simultaneously accomplished by spectrally separating the emission light using dichroic mirrors and directing the different spectral components to multiple MAPMTs. However, MAPMTs are relatively bulky and a system with multiple detectors with their associated optics will have a very long detection optical path resulting in substantial loss of scattered emission photons [15]. Several research groups have investigated various types of spectral-resolved MMM. A MMM system with temporal and spectral resolution has been developed based on a custom-designed streak camera [23–25]. The same group later has developed a spectral-resolved MMM with a CCD camera with a one-axis descanned configuration [26]. In addition, a spectrometer-integrated MMM using an optical multiplexer has been developed [27–29]. However, these previous investigations require specimen scanning or the use of a pulse multiplexing scheme that limits image speed improvement. Here we introduce a spectral-resolved MMM by implementing the spectral detection in a descanned MMM system using a MAPMT. Previous designs of MAPMT based MMMs with descanned configuration utilized 2D array of the excitation foci mapping to the physical anode matrix of the MAPMT. Instead of producing a 2D array of excitation foci, spectral-resolved MAPMT based MMM generates a linear array of excitation foci. The spectral information can be acquired in the orthogonal axis of the MAPMT simultaneously. This approach does not require extension of the detection light path that often suffers from significant signal loss.
2. Method
2.1 System configuration
Figure 2 shows the optical configuration of the spectral-resolved MMM. The excitation light path is the same as the one described in a previous publication [15]. The femtosecond light source used was a Chameleon Ultra II (Coherent, Santa Clara, CA). The excitation laser beam was split into 4 × 1 beamlets with a DOE (customized, Holo/Or, Rehovoth, Israel). The beamlets were sent to the scanning mirrors (6215H, Cambridge Technology, Lexington, MA) through the lens L1 (f 300 mm, singlet, KPX232AR.16, Newport, Irvine, CA) and lens L2 (f 75 mm, doublet, AC508-075-B-ML, Thorlabs, Newton, NJ) in a 4-f configuration and scanned together. The beamlets were expanded by the lens L3 (f 35 mm, doublet, AC254-035-B-ML, Thorlabs, Newton, NJ) and lens L4 (f 175 mm, singlet, KPX196AR.16, Newport, Irvine, CA) to slightly overfill the back aperture of the objective lens (W Plan-Apochromat, 20 × , 1.0 NA, Zeiss, Thornwood, NY). The 4 × 1 excitation foci were generated in a sample with 85 μm separations, and each focus scanned 85 μm × 340 μm area, so the total imaging size was 340 μm × 340 μm. The number of pixels, pixel size and the total acquisition time (frame rate) can vary according to specimens, so each setting is provided for each specimen in the following result section. The emission photons were collected by the same objective lens, delivered along the same optical path as the excitation beam path, and descanned by the scanning mirrors. The stationary emission light after descanning was separated from the excitation laser light by the dichroic mirror (Chroma Technology, Bellows Falls, VT), relayed by lens L5 (f 62.9 mm, singlet, KPX085AR.14, Newport, Irvine, CA) and lens L6 (f 150 mm, doublet, AC508-150-A-ML, Thorlabs, Newton, NJ), and spectrally separated by a prism (PS853, Thorlabs, Newton, NJ) with anti-reflection coating (Chroma Technology, Bellows Falls, VT). For the spectral decomposition, a diffraction grating is an alternative choice. However, the grating generally suffers from significant light loss and significant variation of diffraction efficiency across the spectral range. In contrast, a prism has greater efficiency and more uniform sensitivity over a broader spectral range. However, prisms are less dispersive than gratings. Therefore, in order to achieve the designed spectral dispersion power in the instrument we have utilized two prisms. The spatially separated and spectrally dispersed emission light was focused onto a MAPMT (H7546B-20, Hamamatsu, Bridgewater, NJ) through lens L7 (f 100 mm, doublet, AC508-100-A-ML, Thorlabs, Newton, NJ). The double prisms and the lens L7 were designed to collect the range of the emission spectra of typical fluorescent proteins (from teal to red fluorescent proteins). A BG39 (Chroma Technology, Bellows Falls, VT) and a shortpass filter (ET680sp-2p, Chroma Technology, Bellows Falls, VT) were installed in front of the MAPMT to block the excitation IR and pass only the emission photons in the designed spectral range. Though the MAPMT has 8 × 8 channels, only 4 rows were used for the spatial separated foci while all 8 columns were used for 8 channels in spectral-resolved detection. While we chose to use only 4 excitation foci, a different DOE can be fabricated to generate 8 foci to utilize the full 8 × 8 channels of the MAPMT. Further, within the available laser power, the use of multiple MAPMTs can potentially further improve the imaging speed. For example, placing two 8 × 8 MAPMTs side-by-side will allow 16 foci scanning in parallel with 8 simultaneous spectral channels. A single-photon avalanche diode (SPAD) array can be another choice with its large number of pixels and higher quantum efficiency for further imaging speed improvement and more spectral channels [30–32]. However, it should be noted that the SPAD array has typically a low fill-factor (around 10%) resulting in significant loss of emission photons. This limitation may be overcome in the future with the use of a matching micro-lenslet array that maps each lenslet into the corresponding active area of each SPAD pixel.
2.2 Spectral decomposition
Spectral decomposition is routinely done by linear unmixing on a pixel by pixel basis.
where [I] is an n × 1 matrix representing the concentration of n fluorophore species that is to be determined, [S] is an m × n matrix representing the known spectra of the n species distributed in m detected spectral channels, and [J] is an m × 1 matrix representing the measured spectrally mixed signals of all the fluorophores collected in the mth spectral channel. If [S] is known, retrieving [I] can be simply done by matrix inversion.
[S] can be readily estimated in two ways. First, imaging each fluorophore with our system can provide the required spectral distribution information (Fig. 3(a)). This method gives perfect estimation, but it requires multiple calibration measurements performed in advance. The second method is to estimate [S] with the published emission spectrum of each fluorophore. For typical organic dyes or fluorescent proteins, their emission spectra are well known. With these known spectra, the required spectral distribution of each fluorophore can be estimated using the known wavelength dependent detection efficiency of our microscope taking into account transmittance of optics (e.g. dichroic and barrier filters) and the quantum efficiency of the MAPMT (Fig. 3(b)). The second method clearly has some additional uncertainties due to potential errors in published spectra and in estimating the optical parameters of our system. In either case, we normalize the estimated spectral distribution such that the sum of the coefficients is one so that the total photon count is preserved after unmixing.
Fig. 3 Estimation of spectral distribution by (a) imaging the single-color samples with 8 channels or (b) based on the known emission spectrum of each specimen with the spectral response of the instrument. The graph in (b) shows the total detection efficiency as a function of wavelength. The spectral calibration is discussed in section 3.1.
3. Results
3.1 Detection channel calibration
The spectral calibration of the eight channels was achieved by imaging several known samples: Quinine Sulfate (emission peak (Em): 460 nm, R-14782, Molecular Probes, Eugene, OR), Fluorescein (Em: 513 nm, R-14782, Molecular Probes, Eugene, OR), yellow-green beads (Em: 515 nm, F8859, Molecular Probes, Eugene, OR), orange beads (Em: 560 nm, F8833, Molecular Probes, Eugene, OR), red beads (Em: 605 nm, F8842, Molecular Probes, Eugene, OR), and several second harmonic generation signals from collagen samples (475 nm to 537.5 nm with an increment of 12.5 nm from the laser that generates the maximum output wavelength of 1075 nm). With these calibration samples, the detector was installed to collect from 485 nm to 605 nm with every 15 nm wavelength window.
3.2 Three color fluorescent beads image
We first evaluated the performance of the spectral-resolved MMM with a three color fluorescent bead sample. Three different fluorescent polystyrene microspheres (F8859: yellow-green, F8833: orange, and F8842: red, Molecular Probes, Eugene, OR) were mixed and immobilized in 3D by 2% agarose gel (UltraPure Low Melting Point Agarose, Invitrogen, Carlsbad, CA). Different sizes of beads (4, 10, and 15 μm diameter) were intentionally chosen for each color bead so that the correct species assignment based on spectral unmixing can be readily verified based on bead size. The excitation wavelength was 910 nm, the laser power was about 5mW per focus, and the dwell time was 40 μs per 0.5 μm pixel resolution. Figure 4 shows the three bead image. The acquired raw images for eight wavelength channels are shown in (a). Individual raw images do not directly correspond to the different type of beads due to spectral overlap caused by the broad fluorescent spectra of the fluorophores. Since the spectrum of each bead is distributed differently into the eight spectral channels, the identity of each bead can be determined based on the spectral unmixing algorithm as described in section 2.2. Figure 4(b) shows the spectral distributions of the three color beads over the eight channels. The fractional intensities of each spectrum on each spectral channel are normalized such that the area under each curve is unity preserving the original photon count in the spectral unmixing process. From the acquired coefficients, the eight raw images were processed with matrix calculation for the three decomposed images shown in Fig. 4(c). The final processed image shows that the spectral-resolved MMM can successfully image the three color bead sample four times faster than single focus scanning and provide accurate spectral information that can be used to distinguish the beads. The full-width at half-maximum (FWHM) of the three different beads expectedly showed good agreement with their specified sizes (4.1, 10.1, and 15.2 μm for each 4, 10, and 15 μm bead) demonstrating the spatial resolution of the system. For over hundreds of beads imaged, there was no mis-assignment in the spectral unmixing process.
Fig. 4 Three color fluorescent bead image. (a) Eight channel images in pseudo colors for display purpose. The raw image of each channel contains only combined intensity information. (b) The spectrum of each bead on the eight channels. (c) Processed image in pseudo colors. The image size is 340 μm × 340 μm.
3.3 Mouse kidney image
We further evaluated the performance of the spectral-resolved MMM with a fixed mouse kidney sample labeled with three fluorescent dyes (FluoCells® Prepared Slide #3, F24630, Molecular Probes, Eugene, Oregon). Red-fluorescent Alexa Fluor® 568 phalloidin was used to label the filamentous actin. Green-fluorescent Alexa Fluor® 488 wheat germ agglutinin was used to label glycoproteins or glycolipids in the membrane. Finally, the nuclei were counterstained with the blue-fluorescent DAPI. The excitation wavelength was 780 nm, the laser power was about 11 mW per focus, and the dwell time was 40 μs per 0.5 μm pixel resolution. The acquired raw images are typically not uniform (the center area is brighter than the edge area) because while the x and y scanning mirrors are closely spaced, they cannot be both at the eye-point of the scanning system. In addition, the four foci are under different aberration condition, so the center two sub-images are brighter than the top and bottom sub-images. To avoid the non-uniformity in the raw images, they were normalized with rhodamine solution images. A 1 mM rhodamine 6G solution was imaged with each spectral channel by moving the MAPMT to record the same peak intensity signal onto each channel, and the acquired raw images were normalized with the rhodamine solution images. Similarly to the three color bead sample described in section 3.2, the three emission signals of the mouse kidney sample were recorded in the eight channels. In the case of the fluorescent bead sample, the spectral unmixing is straightforward since the beads are sparsely distributed in space where each individual pixel contains contribution from only one species. However, in the mouse kidney sample the three fluorophores are distributed within the complex tissue structures with significant spatial overlaps resulting in considerable spectral overlap within each pixel. We first estimated the spectrum of each fluorescent dye based on their published emission spectra (Molecular Probes, Eugene, Oregon) and the wavelength dependent detection efficiency of this MMM system as described in section 2.2. The three colors were then decomposed with the estimated spectral distribution, and presented in one image. Figure 5 shows that the system can successfully image the sample with details in the three different fluorophores. The morphology shown in the acquired image matched well with the images of similarly labeled samples acquired with different spectral-resolved methods [33]. The distribution of the three unmixed species corresponded well to the expected morphology of the underlying tissue components. The blue arrow in Fig. 5 points to a representative ellipsoidal nucleus labeled with DAPI. The green arrows in Fig. 5 point to tufts of capillaries forming the glomeruli in the kidney with the endothelial cell plasma membrane labeled with wheat germ agglutinin. The red arrow in Fig. 5 points to filamentous structures of the actin cytoskeleton of cells contained in the glomeruli and the brush border of proximal convoluted tubules. It should be noted that the image size was 290 μm × 340 μm rather than a square shape due to the mismatch of the laser wavelength and the DOE specification. The DOE installed in the system was designed to generate 4 × 4 beamlets with 0.8 ° separation angle at 910 nm wavelength. However, in this case 780 nm wavelength was used for efficient excitation. Therefore, the excitation laser was slightly less diffracted than the designed angle resulting in the shortening of one spatial dimension by 50 μm.
Fig. 5 Spectral-resolved MMM image of a mouse kidney labeled with three fluorophores. Blue: DAPI emission. Green: WGA-Alexa 488 emission. Red: phalloidin-Alexa 568 emission. The red arrow indicates an actin-rich structure labeled with phalloidin. The green arrows indicate glycoproteins in the outer part of glomeruli labeled with WGA-Alexa 488. The blue arrow indicates a cell nucleus.
3.4 Nerve sample image
The capabilities of the spectral-resolved MMM are further demonstrated by imaging ex vivo samples of rat sciatic nerves that had been fully transected, entubulated inside a porous collagen scaffold, and harvested from the sacrificed animal one week later [34]. The nerve tissue was gently fixed by 4% paraformaldehyde, embedded in OCT medium, sectioned, stained using a primary antibody against α smooth muscle actin (αSMA; A5228, Sigma Aldrich, St. Louis, MO) that is detected by an Alexa Fluor 488-conjugated anti-mouse secondary antibody, and counterstained by the nucleic acid stain DAPI, and TRITC-conjugated phalloidin (binds to F-actin) [35]. The laser excitation wavelength was 780 nm, power per focus was about 18 mW, and the dwell time was 40 μs per 0.5 μm pixel. The image field-of-view was 290 × 340 μm, same as the mouse kidney image shown in section 3.3. Non-uniform excitation over the image is corrected by normalizing the acquired images with the rhodamine solution image at each channel as described in the mouse kidney image. The signal contribution of the three fluorescent sources (DAPI, Alexa Fluor 488, TRITC) were estimated based on the estimated emission spectra over the eight channels as described in section 3.3. Figure 6 shows a representative image of the nerve sample after spectral unmixing. Each color is analogous to the fluorescent emission of one fluorescent source (red: TRITC, green: Alexa488, blue: DAPI). The instrument can successfully distinguish the spatial distribution of the fluorophores in the nerve sample in agreement with the published morphology of injured nerve samples [34]. The image consists of three parts: the nerve tissue (left), the scaffold (right) and a cell-rich “capsule” that forms between the nerve and the scaffold (center). Phalloidin dimly stains sheath-like actin structures around the Schwann cells that surround axons, and tiny blood vessels in the nerve fascicle. A continuous line, which stains strongly for phalloidin, highlights the nerve perineurium, a thin layer of contractile cells around the fascicle. The perineurium is surrounded by epineurium, a collagenous connective tissue that stains dimly for DAPI (corresponds to the weak collagen autofluorescence) and αSMA (non-specific antibody binding). The bright circular blue feature at the center of the capsule is the autofluorescence emission of the stitch added by the surgeon to keep the nerve and the scaffold together. The stitch is surrounded by contractile cells that stain for phalloidin. In the capsule there are also several large blood vessels that stain brightly for both actin and αSMA. The scaffold stains dimly for collagen (dim DAPI signal) and also contains some weak αSMA signal (non-specific antibody binding). A number of contractile cells are bound on the scaffold. These contractile cells always stain strongly for actin and sometimes for αSMA.
Fig. 6 A representative spectral-resolved MMM image of ex vivo peripheral nerves after transaction and entubulation in a collagen tubular scaffold. Blue: DAPI emission, Green: Alexa 488-conjugated antibody that recognizes an anti-αSMA antibody (stains contractile cells). Red: TRITC-conjugated phalloidin (stains F-actin).
3.5 Mouse brain image
Lastly, we have performed in vivo imaging of neuron dendrite morphology and inhibitory synapse distribution in a rodent brain with the spectral-resolved MMM. To obtain mice that express fluorescent labels in a sparse subset of layer II/III pyramidal neurons in visual cortex, we electroporated in utero cre recombinase dependent expression plasmids for eYFP (for neuronal morphology) and Teal-Gephyrin (for inhibitory synapses) in timed pregnant C57BL6 mice as previously described in [17]. Between 6 and 8 weeks of age, mice were anesthetized with isoflurane (3% for induction and 2% during surgery) and a 5mm glass coverslip was placed following craniotomy over the visual cortex. Imaging was performed approximately two weeks later on adult mice (>3months) anesthetized with isoflurane (3% for induction, 1-2% for imaging). The head was positioned in a custom made stereotaxic restraint affixed to the microscope stage. Individual eYFP labeled dendrites and Teal-Gephyrin labeled synapses in layers 2/3 of visual cortex of an anesthetized adult mouse were imaged with the spectral-resolved MMM. As a comparison, the same animal was imaged using a single focus scanning two-photon microscope with two spectral channels previously described in [17]. Specifically, the excitation light path is practically identical to the spectral-resolved MMM except only a single focus is generated. The major difference of the single focus system is that the spectral-resolved detection is performed in a non-descanned configuration. The emission photons are collected by the same objective lens (W Plan-Apochromat, 20 × , 1.0 NA, Zeiss, Thornwood, NY), and separated from the excitation beam path by a short-pass dichroic mirror (750dcspxr, Chroma Technology, Bellows Falls, VT). The emission photons are separated again by another dichroic mirror (520dcxr, Chroma Technology, Bellows Falls, VT) at 520 nm, which is the middle of the eYFP and Teal-Gephyrin emission peaks, and mapped onto two PMTs (R7400U-01 and R7400U-20, Hamamatsu, Bridgewater, NJ) locating at the conjugate plane of the back focal plane of the objective with the tube lens of the microscope (AXIOSKOP 2 FS, Zeiss, Thornwood, NY) and another lens (KPX079AR.14, Newport, Irvine, CA) in 4-f location. A BG39 filter (Newport, Irvine, CA) and a bandpass filter (hq485/70m-2p for Teal-Gephyrin and et560/80m-2p for eYFP, Chroma Technology, Bellows Falls, VT) in front of the two PMTs block the excitation IR and keep the photons of the designed spectral range. For in vivo imaging experiments, the excitation wavelength was 910 nm, the laser power was about 50 mW per focus, the dwell time was 40 μs with 0.5 μm pixel resolution, and the image size was 340 μm × 340 μm for both the spectral-resolved MMM and the single focus scanning microscope. For flat field correction, the acquired images were normalized with the rhodamine solution images acquired in each imaging system as described in section 3.3. For calibration measurements to obtain the true spectra for Teal and eYFP, HEK293 cells were singly transfected with plasmids pFUeYFPW (for eYFP expression) or pFUTealW (for Teal expression) using calcium phosphate mediated gene transfer [36]. Transfected cells with each fluorescent label were imaged in both systems. Imaging the cells labeled with the only single color provided the spectral distribution information of each label for the spectral decomposition. After the spectral unmixing the images were processed with background rejection and 2D B3-Spline filter as a low pass filter as done in the previous publications [10, 17] since the dot-shaped synapses are only about one micron size occupying 3 × 3 pixels and the maximum photon count is only around 30, so the raw images contain significant photon shot noise. While simple low pass filtering after unmixing provides fairly satisfactory results, more advanced unmixing algorithms including the Poisson noise removal algorithm based on variance stabilization or maximum likelihood estimation may further improve our ability to identify small synapses [37–39].
Figure 7(a) shows the neuronal image taken by the spectral-resolved MMM and (b) is the same size field-of-view image by the single focus scanning two-photon microscope at 60 μm depth. The bottom images are the enlarged images of the specified area of detail. The red line-shape structures are dendrites labeled with eYFP, and the green dots are synapses labeled with Teal-Gephyrin. The displayed red and green colors are pseudo-colors for better contrast. The spectral-resolved MMM acquired the image 4 times faster than the single focus scanning reducing the acquisition time of a general 1-2 hour brain imaging session [10, 17] to 15-30 minutes. The speed improvement can be maximized further by optimizing the DOE for more foci as described in the method section.
Fig. 7 Representative images of two-color in vivo mouse brain imaging by the spectral-resolved MMM. Images show the contribution of eYFP-labeled dendrites (shown in red) and Teal-Gephyrin-labeled synapses (shown in green) after spectral unmixing. (a) Image acquired by the spectral-resolved MMM. (b) Image acquired by the single focus scanning two-photon microscope. In each case the bottom images show enlarged images of the specified regions of interest. The arrows indicate representative synapses. The large spots of the Teal-Gephyrin channel shown in both (a) and (b) are not synapses, but just bright objects occasionally observed in the brain. It has been confirmed that these objects are not synapses in an independent study with electron microscope imaging [17], and they are in general excluded in our image analysis.
It is known that the SNR and the resolution of the MMM may be inferior to those of the single focus scanning two-photon microscope resulting in limited imaging depth [15]. This also holds for the spectral-resolved MMM. However, the typical imaging depth of 100-300 μm in the single focus in vivo neuronal imaging research can also be achieved with MMM with MAPMT in the descanned configuration with no notable image degradation. Figure 8 shows the spectral-resolved MMM images of the same animal used in Fig. 7 at larger imaging depths. The raw images (Fig. 8(a) and 8(b)) showed image artifacts, appearing as ghost sub-images, due to scattering of the specimen from the large imaging depth, so they were processed with ghost image removal algorithm previously described in [15]. The spectral-resolved MMM images at the larger imaging depths showed clear neuronal cell bodies and dendritic morphology (Fig. 8(c) and 8(d)) is comparable with the morphology obtained by the single focus scanning two-photon microscopes Fig. 8(e).
Fig. 8 A mouse brain image labeled with two colors. (a) Spectral-resolved MMM image at 184 μm depth, and (b) at 156 μm depth. The original images are marked in red circles and their ghost images in blue circles. (c) The processed image of (a) with the ghost image removal algorithm, and (d) the processed image of (b). (e) Single focus scanning image at 156 μm depth.
4. Discussion and conclusion
In the past MMM has shown to successfully achieve improved imaging speed using parallelized excitation. For applications that require spectral-resolved imaging, we have developed a spectral-resolved MMM system that provides both high throughput and high content images. This spectral-resolved MMM is designed based on using one dimension of the MAPMT for resolving the signals from a multiple foci linear array while using the other dimension for resolving the spectral content of each foci. Within the available laser power the number of foci can be maximized resulting in maximum imaging speed improvement. This system has been demonstrated to successfully perform spectral-resolved imaging in ex vivo tissue applications and in vivo deep in the mouse brain with comparable SNR as single focus systems but with much higher acquisition speed.
This spectral-resolved MMM based on MAPMT is an interesting addition to other previously developed spectral-resolved MMMs. The use of the MAPMT with single-photon counting circuitry can achieve much higher frame rate compared with the system using imaging cameras [26], especially ones with long dead-time streak cameras [23–25]; however, streak cameras have the advantage of providing lifetime-resolved imaging capability at the same time. However, multiple readouts (equal to the number of lines in an image) are required for these camera systems resulting in very undesirable accumulation of camera readout noise that substantially degrades image SNR. Moreover, MMMs using cameras as detectors are vulnerable to scattering of emission photons in turbid specimens resulting in further degradation of SNR and limited imaging depth [15]. The pulse multiplexing MMMs have the advantage of being immune to emission photon scattering similar to single focus scanning systems [27, 28]. The pulse multiplexing system further has the advantage that spectral detection can be readily accomplished with a grating spectrometer-camera system. A very promising system with two multiplexed beams has been demonstrated [29]. However, these pulse multiplexing MMMs require custom laser sources that are not broadly available. More importantly, the maximum number of time multiplexed beams generated by these laser sources are currently limited to less than ten resulting in a slower imaging system compared with more conventional MMMs that can readily utilizes tens to hundreds of foci with a standard Ti-Sapphire oscillator.
One major disadvantage of MMM system based on MAPMT is the crosstalk among the foci resulting in image artifacts appearing as ghost images in the neighbor sub-images. Therefore, image post processing to unmix signals from the different foci may be required in a highly turbid specimen or deep tissue imaging [15]. More advanced crosstalk removal algorithms based on a maximum likelihood approach has been developed [40].
The current eight channel detection allows us to image a maximum of eight different fluorophores, but in practice, to avoid severe spectral overlap, 3-4 color imaging would be preferable. To image more fluorophores at the same time, multiple MAPMTs or a SPAD array can be used. However, since different fluorophores have different excitation spectra, simultaneous excitation with multiple excitation wavelengths may be required. Since the DOE is wavelength dependent, the use of multiple excitation wavelengths can significantly complicate the design of this type of MMM systems.
One exciting future direction is the incorporation of simultaneous spectral and lifetime resolved imaging capabilities [23–25, 41]. For MMM systems based on MAPMT, high speed electronics for high speed fluorescence lifetime measurement can be readily implemented at the output anodes of 4 × 4 or 8 × 8 MAPMTs. A low cost, highly parallelized time-correlated single photon counting system based on high speed field programmable gate array electronics is currently under development.
For specimens with bright emission signals, spectral decomposition for fluorophores with sufficiently well separated spectra is often straightforward. However, for low level emission signals, the success of spectral decomposition is often limited by the inherent Poisson noise in the image data resulting in significant uncertainty during spectral unmixing [42]. This error results in improper spectral decomposition and potential misinterpretation of the image data. Therefore, additional spectral decomposition strategies such as image processing with the application of Poisson noise removal algorithms will be required [37–39].
Acknowledgments
This research was supported by grant NIH 9P41EB015871-26A1, RO1 EY017656, 5R01EY017656-02, 5 R01 NS051320, 4R44EB012415-02, NSF CBET-0939511, the Singapore-MIT Alliance 2, the Singapore MIT Alliance for Research and Technology, the MIT SkolTech initiative, the Hamamatsu Corp. and the Koch Institute for Integrative Cancer Research Bridge Project Initiative.
References and links
1. W. Denk, J. H. Strickler, and W. W. Webb, “2-photon laser scanning fluorescence microscopy,” Science 248(4951), 73–76 (1990). [CrossRef] [PubMed]
2. P. T. C. So, C. Y. Dong, B. R. Masters, and K. M. Berland, “Two-photon excitation fluorescence microscopy,” Annu. Rev. Biomed. Eng. 2(1), 399–429 (2000). [CrossRef] [PubMed]
3. W. R. Zipfel, R. M. Williams, and W. W. Webb, “Nonlinear magic: multiphoton microscopy in the biosciences,” Nat. Biotechnol. 21(11), 1369–1377 (2003). [CrossRef] [PubMed]
4. F. Helmchen and W. Denk, “Deep tissue two-photon microscopy,” Nat. Methods 2(12), 932–940 (2005). [CrossRef] [PubMed]
5. W. Denk, K. R. Delaney, A. Gelperin, D. Kleinfeld, B. W. Strowbridge, D. W. Tank, and R. Yuste, “Anatomical and functional imaging of neurons using 2-photon laser scanning microscopy,” J. Neurosci. Methods 54(2), 151–162 (1994). [CrossRef] [PubMed]
6. R. Yuste and W. Denk, “Dendritic spines as basic functional units of neuronal integration,” Nature 375(6533), 682–684 (1995). [CrossRef] [PubMed]
7. K. Svoboda, W. Denk, D. Kleinfeld, and D. W. Tank, “In vivo dendritic calcium dynamics in neocortical pyramidal neurons,” Nature 385(6612), 161–165 (1997). [CrossRef] [PubMed]
8. W. C. A. Lee, H. Huang, G. P. Feng, J. R. Sanes, E. N. Brown, P. T. So, and E. Nedivi, “Dynamic remodeling of dendritic arbors in GABAergic interneurons of adult visual cortex,” PLoS Biol. 4(2), e29 (2006). [CrossRef] [PubMed]
9. W. C. A. Lee, J. L. Chen, H. Huang, J. H. Leslie, Y. Amitai, P. T. So, and E. Nedivi, “A dynamic zone defines interneuron remodeling in the adult neocortex,” Proc. Natl. Acad. Sci. USA 105(50), 19968–19973 (2008). [CrossRef] [PubMed]
10. J. L. Chen, W. C. Lin, J. W. Cha, P. T. So, Y. Kubota, and E. Nedivi, “Structural basis for the role of inhibition in facilitating adult brain plasticity,” Nat. Neurosci. 14(5), 587–594 (2011). [CrossRef] [PubMed]
11. J. Bewersdorf, R. Pick, and S. W. Hell, “Multifocal multiphoton microscopy,” Opt. Lett. 23(9), 655–657 (1998). [CrossRef] [PubMed]
12. A. H. Buist, M. Muller, J. Squier, and G. J. Brakenhoff, “Real time two-photon absorption microscopy using multi point excitation,” J. Microsc. Oxford. 192(2), 217–226 (1998). [CrossRef]
13. L. Sacconi, E. Froner, R. Antolini, M. R. Taghizadeh, A. Choudhury, and F. S. Pavone, “Multiphoton multifocal microscopy exploiting a diffractive optical element,” Opt. Lett. 28(20), 1918–1920 (2003). [CrossRef] [PubMed]
14. B. O. Watson, V. Nikolenko, and R. Yuste, “Two-photon imaging with diffractive optical elements,” Front Neural Circuits 3, 6 (2009). [PubMed]
15. K. H. Kim, C. Buehler, K. Bahlmann, T. Ragan, W. C. A. Lee, E. Nedivi, E. L. Heffer, S. Fantini, and P. T. C. So, “Multifocal multiphoton microscopy based on multianode photomultiplier tubes,” Opt. Express 15(18), 11658–11678 (2007). [CrossRef] [PubMed]
16. N. W. Gray, R. M. Weimer, I. Bureau, and K. Svoboda, “Rapid redistribution of synaptic PSD-95 in the neocortex in vivo,” PLoS Biol. 4(11), e370 (2006). [CrossRef] [PubMed]
17. J. L. Chen, K. L. Villa, J. W. Cha, P. T. So, Y. Kubota, and E. Nedivi, “Clustered dynamics of inhibitory synapses and dendritic spines in the adult neocortex,” Neuron 74(2), 361–373 (2012). [CrossRef] [PubMed]
18. D. van Versendaal, R. Rajendran, M. H. Saiepour, J. Klooster, L. Smit-Rigter, J.-P. Sommeijer, C. I. De Zeeuw, S. B. Hofer, J. A. Heimel, and C. N. Levelt, “Elimination of inhibitory synapses is a major component of adult ocular dominance plasticity,” Neuron 74(2), 374–383 (2012). [CrossRef] [PubMed]
19. J. Livet, T. A. Weissman, H. Kang, R. W. Draft, J. Lu, R. A. Bennis, J. R. Sanes, and J. W. Lichtman, “Transgenic strategies for combinatorial expression of fluorescent proteins in the nervous system,” Nature 450(7166), 56–62 (2007). [CrossRef] [PubMed]
20. T. Nagai, A. Sawano, E. S. Park, and A. Miyawaki, “Circularly permuted green fluorescent proteins engineered to sense Ca2+.,” Proc. Natl. Acad. Sci. USA 98(6), 3197–3202 (2001). [CrossRef] [PubMed]
21. A. W. Nguyen and P. S. Daugherty, “Evolutionary optimization of fluorescent proteins for intracellular FRET,” Nat. Biotechnol. 23(3), 355–360 (2005). [CrossRef] [PubMed]
22. X. Zhang, Y. Xiao, and X. Qian, “A ratiometric fluorescent probe based on FRET for imaging Hg2+ ions in living cells,” Angew. Chem. Int. Ed. Engl. 47(42), 8025–8029 (2008). [CrossRef] [PubMed]
23. J. Qu, L. Liu, D. Chen, Z. Lin, G. Xu, B. Guo, and H. Niu, “Temporally and spectrally resolved sampling imaging with a specially designed streak camera,” Opt. Lett. 31(3), 368–370 (2006). [CrossRef] [PubMed]
24. L. Liu, J. Qu, Z. Lin, L. Wang, Z. Fu, B. Guo, and H. Niu, “Simultaneous time-and spectrum-resolved multifocal multiphoton microscopy,” Appl. Phys. B 84(3), 379–383 (2006). [CrossRef]
25. L. Liu, J. Qu, L. Wang, Z. Fu, Z. Lin, B. Guo, and H. Niu, “Development of a simultaneously time-and spectrum-resolved multifocal multiphoton microscopy system using a streak camera,” in Proceedings of IEEE Conference on Biophotonics, Nanophotonics and Metamaterials (IEEE, 2006), pp. 186–189. [CrossRef]
26. Y. Shao, J. Qu, H. Li, Y. Wang, J. Qi, G. Xu, and H. Niu, “High-speed spectrally resolved multifocal multiphoton microscopy,” Appl. Phys. B 99(4), 633–637 (2010). [CrossRef]
27. W. Amir, R. Carriles, E. E. Hoover, T. A. Planchon, C. G. Durfee, and J. A. Squier, “Simultaneous imaging of multiple focal planes using a two-photon scanning microscope,” Opt. Lett. 32(12), 1731–1733 (2007). [CrossRef] [PubMed]
28. E. Chandler, E. Hoover, J. Field, K. Sheetz, W. Amir, R. Carriles, S. Y. Ding, and J. Squier, “High-resolution mosaic imaging with multifocal, multiphoton photon-counting microscopy,” Appl. Opt. 48(11), 2067–2077 (2009). [CrossRef] [PubMed]
29. E. V. Chandler, C. G. Durfee, and J. A. Squier, “Integrated spectrometer design with application to multiphoton microscopy,” Opt. Express 19(1), 118–127 (2011). [CrossRef] [PubMed]
30. S. Cova, A. Lacaita, M. Ghioni, G. Ripamonti, and T. Louis, “20‐ps timing resolution with single‐photon avalanche diodes,” Rev. Sci. Instrum. 60(6), 1104–1110 (1989). [CrossRef]
31. H. Dautet, P. Deschamps, B. Dion, A. D. Macgregor, D. Macsween, R. J. McIntyre, C. Trottier, and P. P. Webb, “Photon counting techniques with silicon avalanche photodiodes,” Appl. Opt. 32(21), 3894–3900 (1993). [CrossRef] [PubMed]
32. S. Cova, M. Ghioni, A. Lacaita, C. Samori, and F. Zappa, “Avalanche photodiodes and quenching circuits for single-photon detection,” Appl. Opt. 35(12), 1956–1976 (1996). [CrossRef] [PubMed]
33. Molecular_Probes, “FluoCells® Prepared Slide #3 (Mouse Kidney Section with Alexa Fluor® 488 WGA, Alexa Fluor® 568 Phalloidin, DAPI) ” (Life Technologies Corporation), http://www.lifetechnologies.com/order/catalog/product/F24630?ICID=search-f24630.
34. E. C. Soller, D. S. Tzeranis, K. Miu, P. T. So, and I. V. Yannas, “Common features of optimal collagen scaffolds that disrupt wound contraction and enhance regeneration both in peripheral nerves and in skin,” Biomaterials 33(19), 4783–4791 (2012). [CrossRef] [PubMed]
35. M. Buydash, “Mechanical modeling of tissue response during early stage entubulated peripheral nerve regeneration,” in Department of Mechanical Engineering (Massachusetts Institute of Technology, 2013).
36. F. L. Graham and A. J. van der Eb, “A new technique for the assay of infectivity of human adenovirus 5 DNA,” Virology 52(2), 456–467 (1973). [CrossRef] [PubMed]
37. R. D. Nowak and E. D. Kolaczyk, “A statistical multiscale framework for Poisson inverse problems,” IEEE Trans. Inf. Theory 46(5), 1811–1825 (2000). [CrossRef]
38. B. Zhang, J. M. Fadili, and J.-L. Starck, “Wavelets, ridgelets, and curvelets for Poisson noise removal,” IEEE Trans. Image Process. 17(7), 1093–1108 (2008). [CrossRef] [PubMed]
39. B. Q. Spring and R. M. Clegg, “Image analysis for denoising full-field frequency-domain fluorescence lifetime images,” J. Microsc. 235(2), 221–237 (2009). [CrossRef] [PubMed]
40. J. W. Cha, V. R. Singh, K. H. Kim, J. Subramanian, Q. Peng, H. Yu, E. Nedivi, and P. T. So, “Reassignment of Scattered Emission Photons in Multifocal Multiphoton Microscopy,” Sci. Rep. 4, 5153 (2014). [CrossRef] [PubMed]
41. S. Pelet, M. J. Previte, and P. T. So, “Comparing the quantification of Förster resonance energy transfer measurement accuracies based on intensity, spectral, and lifetime imaging,” J. Biomed. Opt. 11, 034017 (2006).
42. J. Pawley, Handbook of Biological Confocal Microscopy (Springer, 2006).
