|Home | About | Journals | Submit | Contact Us | Français|
In vivo 2-photon calcium imaging would benefit from the use of multiple excitation beams to increase scanning speed, signal-to-noise ratio, field of view, or to image different axial planes simultaneously. We adapted a spatiotemporal multiplexing approach to circumvent the problem of light scattering ambiguity in deep tissue inherent to multifocal 2-photon microscopy. We demonstrate 2-photon calcium imaging at multiple axial planes in the in vivo mouse brain to monitor network activity of large ensembles of cortical neurons in three spatial dimensions.
Optical probing of neuronal activity with fluorescent calcium indicators is a powerful approach to study information processing in the brain1. In particular, calcium imaging with 2-photon laser scanning microscopy (2PLSM)2 is an ideal tool for recording network activity in vivo because it is less invasive than electrophysiology, more cells can be monitored simultaneously, and one can identify the cells being recorded3. Recently, several approaches have been introduced to overcome the traditionally slow raster scanning in 2-photon calcium imaging (2PCI), including acousto-optic deflectors (AODs) or targeted path scanning with closed loop scanning mirrors5–7. Thus, as better fluorescent probes and methods for enhancing excitation and detection become available, temporal resolution in 2PCI will improve further and reporting of action potential firing will be possible for large groups of neurons and volumes of tissue in vivo.
Another way to improve temporal resolution in 2PCI is to scan the sample with multiple beams in parallel8–9. In fact, because time resolution, signal to noise (SNR) and field of view (FOV) are closely related in 2PCI, multifocal scanning would lead to improvements in all of these areas. For instance, in cases where non-linear photodamage is a limitation10, scanning with multiple beams simultaneously and increasing the rate at which laser pulses are delivered to the sample will increase SNR11. Importantly, another potential advantage of multifocal 2PLSM would be the capability to perform calcium imaging of multiple axial planes simultaneously.
Because the brain is a heterogeneous medium that greatly scatters and absorbs light, multifocal 2PCI in the intact brain has not yet been feasible at depths of > 100 μm. The deep-tissue imaging capability of 2PLSM requires non-descanned detection of scattered fluorescence2, relying solely on excitation for spatial contrast. Existing commercial multifocal 2PLSM systems use cameras that generate spatial contrast through detection8, which is subject to much greater scattering than excitation for a given depth12. This leads to a substantial loss of spatial resolution and signal strength with depth (Supplementary Fig. 1). The use of de-scanned detector arrays such as multi-anode photomultiplier tubes (PMTs) can decrease this effect9, but this method is still subject to scattering ambiguity in the form of emission cross-talk and is limited to <70 μm tissue penetration.
A solution to the problem of extending multifocal raster scanning with 2PLSM to deep tissue is to use spatiotemporal excitation pulse multiplexing and demultiplexed readout13. The feasibility of this approach to image fluorescent samples at multiple focal planes has been demonstrated with custom lasers13–15. It was hypothesized that this method would allow multiple beams to be used in deep tissue without scattering ambiguity14. Here, we adopted this approach for 2PCI in vivo using traditional non-descanned detection to collect all scattered light and attribute it to individual beams. We demonstrate that this technique can be applied to achieve multifocal 2PCI in the intact mouse neocortex, at depths greater than the visible wavelength mean free path length and in multiple axial planes across several cortical layers.
Our multifocal 2PCI approach takes advantage of the shorter fluorescence decay time of synthetic fluorescent calcium indicator dyes (typically in the ns range) relative to the laser repetition periods (tens of ns) in order to interleave beams in time. This permits fluorescence from spatially distinct beams to be distinguished, because fluorescence detected at different times is assigned to different portions of the image (Fig. 1). The number of beams that can be implemented is ultimately determined by the laser repetition rate and the fluorophore’s decay time Using astandard80 MHz Ti:Al2O3 laser, this translates into simultaneously scanning four beams that are separated in time by ~3 ns, which works well for the commonly used dye Fluo-4 (~1 ns lifetime; Supplementary Note). Many more beams could be added (and more fluorophores accommodated) by using methods to reduce the repetition rate of commercial lasers (Supplementary Note). In order to achieve short beam spacing and analog readout, we used a prototype hybrid photodetector16 with a short single-photoelectron pulse width and transit time spread (see Online Methods). Nevertheless, a variety of detection schemes could be used15.
To test whether multifocal 2PLSM with excitation-emission multiplexing can be used for calcium imaging in the intact brain, we combined our 4-beam system with a resonant scanning mirror (although any scanning method could be used; Supplementary Note). This allows acquisition of 2-D images with 500 × 500 pixel resolution at up to 250 Hz, corresponding to a maximum field of view of ~400μm × 400μm using a 40× objective (Figs. 2 and and3,3, and Supplementary Fig. 2). We used standard Fluo-4 AM bulk loading protocols to record the activity of layer(L)2/3 neurons in the somatosensory cortex of isoflurane anesthetized mice. To identify cell contours for analysis, we used a morphological analysis of the time-dimension zero-lag cross-correlation image (Fig. 2b and Online Methods). For iterative identification of action potential-dependent calcium transients in neurons we used a ‘peeling algorithm’7, Online Methods), and constructed raster plots of discrete neuronal firing events (Fig. 2). Simultaneous patch-clamp recordings in cell-attached mode showed that the simulated calcium traces convolved from the electrophysiology recordings closely resembled the actual calcium traces (Supplementary Fig. 2). Thus, our prototype 4-beam time-multiplexed microscope performs well for in vivo 2PCI of the activity of L2/3 neurons, and provides the expected 4× larger FOV (or 4× greater time resolution) than conventional single beam 2PLSM systems.
A current limitation of 2PCI is that one cannot easily record network activity at different depths simultaneously. Using AODs with chirped acoustic signals, one can achieve inertia-less random-access axial focusing at kHz rates. However, 2PCI in multiple depths with AODs has only been demonstrated in brain slices with a modest ± 25 μm range6. The approach also requires four AODs, which introduces significant optical losses, as well as detrimental temporal and spatial dispersion. An alternative is to use a piezoelectric objective focusing unit to rapidly move the objective lens5, which allows volumes to be scanned at rates of ~10 Hz, but the approach is ultimately inertia-limited. Importantly, neither method captures full imaging data, precluding the use of image registration methods for motion correction or image processing algorithms for cell detection a posteriori.
To image calcium signals in 3-D with spatiotemporal multiplexing, we implemented slight modifications in the optics of our system in order to scan separate imaging planes with different beams (Supplementary Fig. 3). Beams were arranged quasi-collinearly with a lens added for each beam to allow for focus offset in the axial direction while maintaining uniform back aperture filling. The beams were spaced 30 μm apart from each other in the z-axis for a total pitch of 90 μm. The total axial spacing is ultimately limited by spherical aberration of the objective lens (Supplementary Note). This allowed us to monitor the activity of ~100–200 neurons distributed over a cortical volume spanning from L1 to L3 (Fig. 2h–j).
As a demonstration of the potential applications of multifocal 2PLSM with spatiotemporal multiplexing, we conducted 2PCI experiments at multiple axial depths to determine whether spontaneous activity within L2/3 of barrel cortex can spread in a columnar fashion, as previously reported17. We recorded neuronal activity with Fluo-4 AM in postnatal day (P) 15-P21 anesthetized mice. In order to maximize the number of L2/3 cells imaged, we arranged the beams such that only 2 planes (spaced at 50 μm pitch) were imaged simultaneously, each with 2 beams (Fig. 3). Next, we identified cells that participated in synchronous bursts of activity, previously referred to as peaks of synchrony17 (see Online Methods). We also constructed activity-derived connectivity diagrams and plotted the distribution of peak correlation coefficients amongst pairs of cells in separate depths within L2/3 (Fig. 3d). We find that only a minority of these bursts had an axial (depth) and radial (lateral) spread that would be consistent with columnar connectivity (Fig. 3e) and the relationship between peak correlation coefficient and the axial spread (ΔR) of a cell pair was indistinguishable for cells in the same imaging plane or in different planes (Fig. 3f). These data suggest that spontaneous activity within L2/3 does not tend to propagate along columnar boundaries.
Ideally, one would like to introduce more than four beams and accommodate a wider variety of fluorophores, including those with longer lifetimes. The simplest solution is to use a laser with a lower repetition rate, while conserving average power (Supplementary Note). In practice, to be able to use a range of fluorescent dyes, beams should be spaced on the order of 10 ns apart, giving a maximum sample pulse rate and pixel rate of 100 MHz. This is a maximum rate at which information can be extracted by fluorescence emission from a region over which the returning fluorescence is scattered and diffused. Ultimately, the pulse energy required depends on tissue scattering, depth, wavelength, pulse width, focus quality, overall system optical transmission, non-linear photodamage, and on the fluorophore of interest; in our case, a pulse energy of 0.1 – 1 nJ after the objective lens was sufficient for in vivo 2PCI. Therefore, using a 5 MHz laser with ~2–3 W average power at 800 nm, imaging with 8–16 beams should be feasible for a wide variety of commonly used fluorophores.
In summary, we have applied a spatiotemporal multiplexing technique to address the fundamental limitation of deep tissue scattering ambiguity in multifocal 2PLSM. In addition to yielding a four-fold increase in frame rate or FOV, we demonstrate simultaneous in vivo 2PCI at up to 4 different cortical depths across ~100 μm. In the future, the introduction of additional beams and greater axial pitch will make it possible to record activity from an entire volume of neocortex. Additionally, multiplexed multifocal 2PLSM could be combined with AOD scanning or targeted path scanning (Supplementary Note). In parallel with improvements in excitation efficiency, electronics, light collection and detectors that result in higher SNR, multifocal 2PCI with multiplexing will allow neuroscientists to explore new questions regarding the spatiotemporal dynamics of neuronal activity over three spatial dimensions.
All chemicals were purchased from Sigma-Aldrich unless otherwise stated. All experiments were performed under animal protocols approved by the Animal Research Committee and the Office for the Protection of Research Subjects at the University of California, Los Angeles.
To construct the spatiotemporal beam multiplexing microscope prototype (Fig. 1a), we used simple optics to split the main beam from a 80 MHz Ti:Al203 laser (Chameleon Ultra II, Coherent) into 4 spatiotemporally multiplexed beams. Polarizing beam splitters were used with half-wave plates to allow for varying power delivery via each beam. We calculated that, for a 3 ns delay between beams, each beam’s path had to be extended by a distance of ~1 m. The delayed beams were then converged on a conjugate plane of the objective pupil at the desired angle. Differences in optical propagation time of emitted fluorescence due to each beam were deemed negligible due to the short differences in distance traveled (millimeters) compared to the optical delays (meters). Importantly, during fluorescence emission from each beam, the photodetector remains sensitive to the entire collection area of the objective, which ensures that the sensitivity and resolution are equivalent to those of a single beam system.
Pulse length is degraded by group delay dispersion (GDD) and, as determined by previous work18 and by the Sellmeier equation, this is negligible for the optical train consisting of ~1 m air delay lines using several dielectric mirrors. The dominant contribution from delay lines is from dielectric mirrors (~20 fs2 at 800 nm18). In the overall microscope system the largest contributors to GDD are the glass elements (506.0 fs2/cm for BK7 glass at 800 nm, 594 fs2/cm for LakL21 at 800 nm), and in comparison delay line dispersion is negligible.
For imaging different planes simultaneously, the beams were arranged quasi-collinearly with a lens added for each beam to allow for focus offset in the axial direction while maintaining uniform back aperture filling (Supplementary Fig. 3). Beams were spaced either 30 μm apart in the z-axis (for a total pitch of 90 μm) under 40× magnification in the case of four plane experiments (Fig. 2), 75 μm apart in the case of two plane experiments under 40× magnification, or 50 μm apart under 60× magnification (Fig. 3).
In order to de-multiplex fluorescence emission, we chose a state-of-the-art hybrid photodetector (HPD; Hamamatsu Photonics), which uses an avalanche photodiode as an electron bombardment target instead of the dynode stages of a photomultiplier tube (PMT). The HPD has a high first stage electron bombardment gain, which allows a low excess noise factor (a measure of the charge fluctuation of each single photoelectron response) of ~1.3, as well as a high speed response (< 1 ns) that is limited by the capacitance of the target avalanche photodiode19. While the HPD’s single photoelectron gain of approximately 105 is an order of magnitude lower than that of a PMT, the difference is recovered with a GHz preamplifier (Supplementary Fig. 4). The HPD we used has a GaAsP photocathode with a 3 mm diameter active area and 45% quantum efficiency at 510 nm. A custom circuit board was designed and fabricated to demodulate multiplexed signals, built around an analog multiplier integrated circuit (ADL5391, Analog Devices). Emission signals from the HPD were de-multiplexed by analog multiplication with 3 ns square pulses generated from the laser monitor signal. This results in four separate signals that are shaped and amplified for subsequent digitization. Thus, our prototype system has the capability to operate as a fluorescence lifetime imaging system when used with a single excitation beam, using 3 ns wide gates in the time domain.
For calcium imaging with our prototype 4-beam system, we incorporated either a 16 kHz resonant scanning mirror (SC-30, EOPC) or a 12 kHz resonant scanning mirror (CRS, Cambridge Technology/GSI Lumonics). This is a conventional scan method that has been used previously for 2-photon laser scanning microscopy (2PLSM)20–22. The resonant scanning mirror executes a bidirectional sinusoidal scan, producing a line rate of 32 kHz or 24 kHz, respectively. We chose to compensate for the non-uniform power delivery per angle and depth with an electro-optic modulator (S-350, ConOptics), but one could instead block the edges of the scan range. For the slow scan axis we used a closed-loop scanning mirror (6200H, Cambridge Technology). Function generators were used to generate scanning mirror signals (33210A, Agilent). With these components, our 4-beam system is capable of acquisition rates of up to 250 frames per second with for 512 × 512 images for single plane imaging and 60 volumes per second for multiple plane imaging.
After de-multiplexing and amplification, signals were digitized by a standard four-channel frame-grabber acquisition board (Helios, Matrox; Supplementary Fig. 4). Using two HPDs and a fast analog switch (ZYSW-2-50DR, Mini-Circuits), two colors can be switched or interleaved at the pixel or frame level and digitized by the same board. Signal levels were digitized at 8-bit precision per sample at 80 MHz, using a clock signal derived from the laser. We did not use the HPD in a photon-counting read-out scheme to allow for high peak photon flux, but this is certainly possible. Once digitized, the data were transferred across the PCI-express peripheral bus at sustained transfer rates > 700 MB/s and then displayed in real time by custom software using supplied imaging libraries (Matrox Imaging Library, Matrox). Raw images were averaged for display purposes only (to help visualize Fluo-4 labeled cells), while complete raw data was stored on hard disk. Geometrical horizontal distortion due to sinusoidal scanning distortion was corrected offline by linear interpolation resampling, or online with nearest-neighbor resampling. Imaging data were stored across a custom-designed high-bandwidth hard disk array with a maximum sustained transfer rate of 400 MB/s (Windows XP, Microsoft). This allows for two channels of data to be recorded without interruption at roughly 100 MB/s at 10-bit resolution. Data can be continuously recorded until the disk array space is exhausted (~ 3 hours).
We used male and female juvenile c57Bl/6 mice (n = 12; postnatal days 15–21) for all experiments and followed standard protocols23–25. A modified cranial window surgery, in which the glass coverslip only partially covers the dura, leaving a gap for injecting the calcium indicator was performed under light isoflurane anesthesia (0.5–1% in O2). Sulforhodamine (SR) 101 was also injected to label astrocytes. With 2–4 injections of the calcium dye per animal we can label thousands of L2/3 neurons over a large area (1 mm2). 2-photon calcium imaging (2PCI) at 800 nm of Fluo-4 AM or Oregon green BAPTA-1 AM (OGB) began after a 30–60 min recovery period. We used three different objectives: 40× 1.0 NA (Zeiss), 40× 0.8 NA (Olympus) and 60× 0.9 NA (Olympus). Typically, an average power of 1 W at 800 nm was input into the microscope (Supplementary Fig. 5). The emission of SR101 in the red channel (filter 629/53 nm) was separated from that of the calcium dye in a green channel (filter 525/50 nm) with a dichroic that transmits wavelengths above 562 nm. Calcium imaging was done under isoflurane anesthesia (~0.5–1%). Several 180 second image sequences were recorded (250 frames per second with 16 kHz resonant scanning mirror and single imaging plane; 60 volumes per second for four-plane imaging; and 100 volumes per second for two-plane imaging with 12 kHz resonant scanning mirror). For multiple plane imaging, individual beam powers were carefully adjusted by half-wave plates and polarizing beam splitters to compensate for varying depth of imaging and variations in preparations. Beam cross-talk due to fluorescence lifetime was deemed neglibible (Supplementary Figs. 6 and 7), and no losses due to multiple beam imaging were found (Supplementary Fig. 8).
In vivo loose cell-attached recordings were performed as previously described26, using a patch clamp amplifier (2400, A-M Systems, Sequim, WA) and borosilicate microelectrodes (4–6 MΩ) filled with a potassium gluconate solution containing (in mM) 105 K-gluconate, 30 KCl, 10 HEPES, 10 phosphocreatine, 4 ATP-Mg, 0.3 GTP (adjusted to pH 7.3 with KOH). Cells were targeted using the shadow-patch method27.
Image processing was done using the MATLAB environment (Mathworks). Images were first transformed using linear interpolation to correct for the sinusoidal resonant scanner trajectory. Contours of cells were detected by generating an image of the zero-lag cross correlation (or normalized standard inner product) between a 1 μm neighborhood and a 10 μm neighborhood28–29. The resulting image was thresholded to generate cell contours. Once cells were identified, raw fluorescence intensity traces were calculated for each, and linear drift was then subtracted. A peeling algorithm30 was applied to identify transients by procedurally fitting exponentials with fixed decay length of 1 s and amplitude at least 5% ΔF/F using linear least squares method, accepting fits with average R2 improvement greater than 0.1 standard deviations over the mean. After each successful fit, the detected transient was subtracted from the data and another fit was attempted, until the end of the trace was reached. This resulted in lists of events, their amplitude, and times for each cell.
To identify peaks of synchrony, the event-amplitude was summed over all cells and a trigger was applied over time at a threshold determined by Monte Carlo analysis (n = 1,000 simulations), as previously described31. This consisted of generating Monte Carlo data sets assuming uniform event arrival time while preserving the event amplitude distribution and number, and then identifying the 95% CL threshold for peak total event amplitude per time. Once peaks of synchrony were identified, the spatial distribution of the cells participating in each peak was characterized using the standard deviation of cell position in the radial (i.e., lateral) and axial directions. To identify significantly correlated cell pairs, the peak of the cross correlation function between event rates for pairs of cells was calculated, as was the delay time at which the peak occurred (with a maximum amplitude of 1 s). Monte Carlo analysis involving 1000 trials of randomly permuting of event times was performed to identify 95% CL thresholds for peak cross correlation values for each cell pair.
Supplementary Figure 1. Effect of tissue depth on light scattering as measured by camera-type detection or photomultiplier tube.
Supplementary Figure 2. Multifocal in vivo calcium imaging and simultaneous electrophysiology.
Supplementary Figure 3. Optical configuration for multiple beam imaging.
Supplementary Figure 4. Read-out electronics.
Supplementary Figure 5. Optical transmission through various components of the prototype spatiotemporally multiplexed microscope.
Supplementary Figure 6. Negligible beam cross-talk for Fluo-4 with 4-beam temporal multiplexing.
Supplementary Figure 7. Negligible beam cross-talk with OGB-1 AM with 2-beam temporal multiplexing.
Supplementary Figure 8. Multiple beam scanning increases brightness and scanning speed for constant average laser power per beam.
Supplementary Note. Longer fluorescence lifetimes, improving acquisition rates, and additional fluorophores.
We thank D. Kleinfeld, J.F. Leger, and T. Otis, as well as members of the Portera-Cailliau laboratory for discussions and comments on the manuscript. We are also grateful to M. Suyama and Y. Kawai with Hamamatsu Photonics K.K. for engineering support. This work was supported by grants from the National Institutes of Health (the Eunice Kennedy Shriver National Institute of Child Health & Human Development and the National Institute of Neurological Disorders and Stroke).
Author contributions: A.C., T.G., P.G., K.A., and C.P.-C. conceived the project. A.C. designed and built the microscope and control electronics, and developed the microscope software. T.G. performed in vivo multifocal calcium imaging and simultaneous cell-attached recordings. A.C. analyzed the data. A.C., T.G. and C.P.-C. wrote the manuscript. K.A. and C.P.-C. supervised the project.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interest