1. Introduction

Over the past two decades, there has been an increased interest in the development of microscopy systems to monitor cellular activity over large three-dimensional (3D) fields of view (FOV) with high temporal resolution. Several strategies have been developed to perform high speed volumetric imaging for both camera-based and scanning microscopes [1].

An advantage of modern camera-based techniques is that they exploit parallelized sampling, yielding remarkably high effective pixel rates on the scale of gigapixel/second [2–4]. For example, multiple focal planes can be acquired in rapid succession using fast axial scanning mechanisms such as electrically tunable lenses (ETLs) [5,6], deformable mirrors [7], tunable acoustic gradient (TAG) lenses [8,9] or by remote focusing [10,11]. Alternatively, to achieve faster imaging still, multiple focal planes can be acquired simultaneously with multiple cameras [12,13], or by spatially distributing different axial focal planes onto a single camera with diffractive optical elements [14,15] or specialized beam splitters [16,17].

Such camera-based techniques, however, are based on standard widefield imaging and are thus inherently susceptible to out-of-focus background that can undermine contrast and signal-to-noise ratio (SNR). The effects of background can be partially mitigated by post hoc numerical deconvolution [17–19], or, better yet, by reducing the generation of background pre hoc with structured or targeted illumination [20–22], or with multiplane light-sheet techniques [23–25]. However, when imaging in thick tissue, these background reduction approaches begin to fail, and depth penetration becomes compromised.

Alternatively, multiplane imaging can be approached with scanning microscopy. An advantage of scanning microscopy is that it can provide optical sectioning by the use of a detection pinhole to physically reject out-of-focus background (confocal microscopy), or by nonlinear signal generation which avoids the generation of out-of-focus background (multiphoton microscopy). For example, confocal microscopy can be axially parallelized to enable multifocus detection using a micro-mirror array [26] or reflecting pinholes [27]. In turn, multiphoton microscopy can be parallelized to enable multifocus excitation by spatiotemporal multiplexing [28–31]. The speeds of these techniques, however, are limited by the speeds of the mechanical beam scanners, which, for large fields-of-view (>500 µm), tend to be no more than video rate.

Here we describe a microscopy technique that provides multiplane and optically sectioned imaging over relatively large FOVs and at frame rates of hundreds of hertz. Our technique, called line-scan multi-z confocal microscopy, is a line-scan version of multi-z confocal microscopy [27] and is based on the same two main principles. The first principle is to use low-numerical aperture (NA) illumination to create an axially extended illumination over a large range of depths along with a high-NA detection to achieve high signal collection efficiency. The second principle is to parallelize the readout to simultaneously detect multiple signals from this extended depth range using multiple reflecting slits that are axially distributed. Since the slit apertures are fully reflecting (as opposed to beamsplitters), no signal is lost and the signal collection efficiency remains high. The key advantage of using line scanning rather than point scanning is that our imaging rate is increased by more than an order of magnitude. We describe our microscope and demonstrate that it can be used for different biological imaging applications such as in vivo calcium and voltage imaging, with spatial resolution matched to cell somas.

2. Experimental setup

2.1 Microscope layout

A schematic of our line-scan multi-z confocal microscope is shown in Fig. 1(a), which is in essence a standard line-scan confocal microscope but with differences in both the illumination and detection optics. The illumination beam is produced by a blue laser diode (488 nm/ 200 mW, Omicron LuxX). A Powell lens (LOCP-8.9R10-1.2, Laserline Optics) shapes the beam into a uniform line [32] and two cylindrical lens (LJ1695RM-A and LJ1703RM-A, Thorlabs) shape the line into a desired aspect ratio. A single-axis, large-aperture galvanometric mirror (15 mm S1 8330k, Cambridge Technology) scans the laser line across the sample in the $x$ direction. To achieve low-NA illumination, we significantly underfill the back pupil aperture of the objective (HR LWD 20x, NA 0.6, Optem or XLUMPLFLN-W 2020 NA 1.0, Olympus) using a pair of achromats as an afocal beam compressor. With NA$_{ill}$ $\approx$ 0.1, the axial extent, or Rayleigh range, of the line focus is on the order of 100 µm.

 

Fig. 1. (a) Schematic of line-scan multi-z confocal microscope. (b) Expanded view of the detection path indicated by the red dashed rectangle in (a). Images of the three focal planes are spatially distributed across the same camera frame, and simultaneously recorded. (c) Transverse ($xy$) and axial ($xz$) PSF measured with a 0.5 µm fluorescent bead, and corresponding $x$ and $z$ profiles. Scale bar, 10 µm. (d) Normalized bead intensity measured by each detection channel at different $z$ positions.

Download Full Size | PDF

The generated fluorescence signal is epi-detected through the full NA of the same objective, ensuring both maximized collection efficiency and a constrained detection depth-of-field. The fluorescence signal is spectrally filtered with a multi-band dichroic mirror (ZT405/488/561/640 rpc, Chroma) and emission filter (ZET405/488/561/635m, Chroma) and relayed onto three reflecting slits (Optical Filter Source), which are axially distributed in the conjugated image space. Each reflecting slit consists of a 36 $\times$ 25 mm B270 glass slide with anti-reflective coating, onto which a 126 µm $\times$ 25 mm line of protected aluminum is lithographically patterned at the center (note that because of their tilt, the effective cross-sectional widths of the reflecting slits are $\approx 90$ µm). As in standard line-scan confocal microscopy, the purpose of the slits is to select (here reflect) in-focus fluorescence while rejecting (here transmit) out-of-focus fluorescence, leading to optical sectioning. The difference with standard line-scan confocal microscopy is that three image planes are optically sectioned simultaneously, rather than only one (Fig. 1(b)). In the image space, the separation between the slits $\Delta Z$ is given by approximately $M^2\Delta z/n$ [33], where $M$ is the lateral magnification from the object to the slits ($M$ = 22.2 for the Olympus objective and $M$ = 20 for the Optem objective), $n$ is the refractive index of the sample medium and $\Delta z$ is the distance between focal planes in the sample, selected here to be $\Delta z$ = 32 µm (at $M$ = 22.2). The additional (de)magnification from the slits to the camera is given by approximately $\div 4.9$.

The in-focus fluorescence selected by the slits is, in turn, re-imaged onto a high-speed camera (ORCA-Flash 4.0 V3 Digital CMOS, Hamamatsu) using a pair of lenses and a second identical single-axis galvanometric mirror. As the galvanometric mirror in the illumination path scans the illumination line across the sample, the galvanometric mirror in the detection path re-scans images of this line across the camera sensor, thus producing three spatially separated images of the different focal planes, which are recorded simultaneously. Both galvanometric mirrors are synchronized to the camera clock and composite images $512\times 2048$ pixels in size are recorded at a frame rate up to 400 Hz using Hamamatsu image software.

We note that a design challenge in our setup comes from the aberrations occasioned in the detection path beyond the slits, specifically field curvature and astigmatism, which stem primarily from the off-axis imaging of the slits probing the deepest and shallowest focal planes and the short focal length of the camera lens. To partially compensate for these aberrations, a Plössl scan lens is used to reduce field curvature and an additional weakly focusing correction lens (LA1908-A, Thorlabs) is inserted just after the middle slit (lens with dashed outline in Fig. 1(a)) to ensure that all three slits are properly focused onto the camera sensor plane. A detailed description of the optical components is provided in Supplement 1.

2.2 Resolution and optical sectioning

It is well known that both the resolution and optical sectioning capacity of a line-scan confocal microscope are inherently weaker than those of a point-scan confocal microscope [34].

To evaluate the resolution of our microscope, we imaged 0.5 µm fluorescent beads over a 200 µm axial range. Using our $20\times$ Olympus objective, we obtained a transverse resolution of $\delta x$ = 4.2 µm and an axial resolution of $\delta z$ = 13.8 µm (Fig. 1(c)). An evaluation of the spatial uniformity of our resolution across different planes and for both objectives is provided in Supplement 1. We note because of the 6.5 µm pixel size of our camera, we would expect a Nyquist-limited resolution of roughly 3 µm. The reason our observed resolution is somewhat worse than this, particularly off axis and in the outer planes, is likely due to the aberrations in the detection path discussed above.

To evaluate the optical sectioning capacity of our microscope, we imaged a thin, uniform fluorescent slide over a 300 µm axial range. We integrated the total fluorescence intensity captured by each detection channel as a function of the slide axial position (Fig. 1(d)). The optical sectioning profiles for the three detection channels are shifted by $\delta z$ = 32 µm, as expected. The full widths at half maximum (FWHMs) of the profiles, from deepest to shallowest, are 37.2 µm, 21.2 µm and 23.8 µm. Importantly, they are confined axially, illustrating optical sectioning, with only small overlap between planes. This optical sectioning capacity was further demonstrated when we imaged a fixed fluorescent sample of a common mold, Aspergillus conidiophores (#297872, Carolina Biological Supply Co.). From the three images acquired simultaneously (Fig. 2(a)), we can clearly observe different features and branching of the filaments. Moreover, the signals in each plane produce little background in adjacent planes.

 

Fig. 2. (a) Simultaneous imaging of different axial planes of Aspergillus conidiophores. Scale bar, 100 µm. (b,c) Normalized intensity images acquired with our line-scan multi-z microscope (b) and a commercial widefield microscope (c) at the same axial focal plane. Scale bar, 100 µm. (d) Normalized intensity plots across the dotted horizontal lines in (b) and (c), illustrating the improved signal-to-background of our microscope over widefield. The vertical dotted line corresponds to the location of a fine filament indicated by the arrow in the images.

Download Full Size | PDF

Finally, we compared images of Aspergillus conidiophores acquired with our line-scan multi-z microscope (Fig. 2(b) – central multi-z plane) and with a standard widefield fluorescence microscope (WFM) (Nikon TE2000-U) (Fig. 2(c)). We imaged the sample when focused to approximately the same axial plane with both microscopes and normalized the intensity of both images for better comparison. As expected, we observe that the out-of-focus background haze is noticeably less prominent in the line-scan multi-z image than in the widefield image. Example profiles of the normalized intensity are shown in Fig. 2(d), where we note that fine sample features (e.g. thin filament indicated by arrows in Figs. 2(b) and (c)) that are readily apparent in the line-scan multi-z image can be completely overwhelmed by out-of-focus background in the WFM image (corresponding vertical dotted line in Fig. 2(d)).

3. Methods

3.1 Mouse preparation and imaging

All animal procedures were approved by the Boston University Institutional Animal Care and Use Committee. Both male and female mice were used in this study (Charles River, Wilmington, MA) and were 8-12 weeks old at the start of the experiments. All injections and surgical procedures were carried out as previously described [35]. Briefly, animals were injected with 1 µl AAV9-Syn-GCaMP7f virus obtained from Addgene (titer: 6.9 e12 GC/ml: 104488-AAV1), targeting the left striatum under stereotaxic conditions (AP: +0.5, ML: −1.8 mm, DV: −1.6). Following complete recovery, animals underwent surgery for the implantation of an imaging window. The window consisted of a stainless steel cannula (OD: 3.17 mm, ID: 2.36 cm, height, 2 mm) with a circular coverslip (size 0; OD: 3 mm). The overlying cortical tissue was carefully aspirated away to expose the corpus callosum, which was then thinned until the dorsal striatum was exposed. The imaging window was centered above the striatum injection site. During the same surgery, a custom aluminum head-plate was attached to the skull, anterior to the imaging cannula, which allowed head fixation during imaging. Prior to imaging, animals underwent handling and habituation to running on a spherical treadmill platform while head-fixed for 1-2 weeks. During imaging, animals were head-fixed under the 20$\times$ Optem objective while freely running on the treadmill. A single imaging session lasted approximately 2.5 minutes and an average laser power output of 4 mW was used.

3.2 Cardiac samples preparation and imaging

Human induced pluripotent stem cells (iPSCs) (I.D. Personal Genome Project, PGP1) were maintained in mTeSR1 medium (Stem Cell) and differentiated as 2D monolayers into cardiomyocytes (CMs) using a small molecule differentiation protocol [36] with CHIR99021 (Tocris) and IWP4 (Tocris) in RPMI 1640 Medium, GlutaMAX (Gibco) supplemented with B-27 minus insulin (Gibco). Once beating was observed, cells were metabolically selected in no glucose, RPMI 1640 Medium supplemented with 4 mM DL-lactate (Sigma), then replated and maintained in RPMI 1640 Medium, GlutaMAX supplemented with B-27. Human mesenchymal stem cells (hMSCs) were isolated from Human Bone Marrow Mononuclear Cells (Lonza), and maintained in DMEM, low glucose medium (Gibco) with 10% Fetal Bovine Serum (Sigma) and 1% Penn-Strep. hMSCs were used between passages 4-7. Cells were maintained at 37$^{\circ }$C and 5% CO2. Polydimethylsiloxane devices with wells containing two micro-pillars were cast from 3D printed molds (Protolabs). Prior to seeding, devices were plasma treated and then treated with 0.01% poly-l-lysine (ScienCell), then treated with 0.1% Glutaraldehyde (EMS). Prior to tissue seeding, devices were UV light sterilized and then treated with 5% Bovine Serum Albumin (Sigma) and 2% Pluronic F127 (Sigma) to prevent cell adhesion. CMs and hMSCs were then dissociated and suspended into a 4 mg/mL human fibrinogen (Sigma) and 10% Matrigel (Corning) hydrogel; 0.4 unit of thrombin (Sigma) per mg of fibrinogen, 5 µM Y-27632 (Tocris), and 0.033 mg/mL aprotinin (Sigma) were also added. 60k cells (90% CMs, 10% hMSCs) per tissue were then added to this hydrogel mixture and pipetted into each well. After seeding, gel was polymerized for 5 minutes. Tissues were maintained in DMEM (Corning) 10% Fetal Bovine Serum (Sigma), 1% Penicillin-Streptomycin (Gibco), GlutaMAX Supplement (Gibco), MEM Non-Essential Amino Acid Solution (Gibco), and 0.033 mg/mL aproptinin (5 µM Y-27632 was added for the first 48 hours). Samples were imaged between Day 7 and 9 post-seeding. For imaging, CM monolayers and tissues were stained for 30 minutes, using FluoVolt (Invitrogen) following manufacturer’s protocol. Samples were imaged in Tyrode’s Salt Solution (Sigma) with the 20$\times$ Olympus objective. Average laser powers of 5 mW and 10 mW were used for the monolayers and tissues respectively.

3.3 Image processing and data analysis

A uniform fluorescent plastic slide (#92001, Chroma Technology) was imaged to generate a reference mask to crop the FOV of each focal plane on the camera. We acquired a z-stack of an Aspergillus conidiophores glass slide over a 250 µm axial range (see Visualization 1) to obtain a maximum-intensity projection (MIP) image of the sample at each detection channel. We then estimated elastic transforms for the deepest and shallowest focal planes (referenced to the middle plane), to correct for magnification differences and any residual astigmatism due to off-axis imaging. A visual comparison of the same sample plane acquired by each detection channel, after applying the elastic transforms, is provided in Supplement 1.

For calcium imaging in the mouse brain, we used the Moco plugin [37] in ImageJ to correct for motion artifacts during acquisition. We used a constrained non-negative matrix factorization (CNMF) algorithm [38] in Matlab (MathWorks) to segment the neurons at each focal plane. For each segmented region of interest, a temporal component $F(t)$ was obtained and we calculated $\Delta F /F = (F(t) - F_0) / F_0$, where $F_0$ is the average of the spatiotemporal background for a given region of interest (ROI).

For voltage imaging of CM samples, we manually segmented $50\times 50$ µm ROIs and extracted the voltage traces in ImageJ. A Matlab detrend function with linear fitting was applied to all the raw traces to correct for photobleaching.

4. Results

4.1 In vivo calcium imaging in mouse brain

One of the fundamental goals of neuroscience is to understand the different network dynamics in the brain responsible for cognitive function and behavior. Monitoring the activity of populations of neurons over 3D volumes rather than 2D planes is crucial in achieving this goal [39]. To demonstrate the capability of our system to perform high speed functional imaging over relatively large FOVs across multiple planes, we performed in vivo calcium imaging in the mouse brain.

We imaged the striatum area of a mouse brain expressing GCaMP7f, a genetically-encoded calcium indicator, within a volume of $617\times 645\times 64$ µm$^{3}$ at a 100 Hz frame rate for 2.5 minutes. Figure 3(a) shows the temporal (max-min) projection of the three focal planes after motion correction. Using CNMF, we identified 99, 128 and 93 neurons in each plane separately, from deepest to shallowest. We note that there was often signal overlap between focal planes (Fig. 1(d)) owing to the fact that many neurons resided between adjacent focal planes and also because scattering somewhat undermined the optical sectioning of our microscope. Duplicate neurons were merged, leading to a total of 195 independent neurons that were identified throughout the imaging volume (see Visualization 2 for a complete recording of the average projection of the three focal planes). The $\Delta F /F$ traces of these neurons are shown in Figs. 3(b) and 3(c) for an expanded view, demonstrating that, even when imaging at high speed, our microscope provides ample signal-to-noise ratio (SNR) for neuronal segmentation and calcium spike identification.

 

Fig. 3. High-speed volumetric in vivo calcium imaging of GCaMP7f-labeled mouse brain. (a) Max-min projections of three focal planes recorded in the striatum at 100 Hz over 2.5 min. From left to right, 99, 128, 93 neurons are identified in each plane using constrained non-negative matrix factorization. Scale bar, 100 µm. $\Delta z$ = 32 µm. (b) Activity of the 195 distinct neurons identified throughout the imaging volume. (c) Expanded view of the calcium traces for a subset of neurons indicated by the red rectangle in (b).

Download Full Size | PDF

4.2 Voltage imaging in cardiac monolayer and tissue

A more challenging application is to image voltage dynamics, since these occur on the millisecond timescale [40]. To demonstrate the novelty and benefit of our microscope, we performed voltage imaging in CM samples. We imaged a CM monolayer labeled with FluoVolt (F10488, Invitrogen), a newly derived voltage sensitive fluorescent dye, over a volume of $575\times 645\times 64$ µm$^{3}$ at 400 Hz for 10 seconds. After imaging spontaneous beating in the CM monolayer (see Visualization 3), we extracted voltage traces for manually segmented $50\times 50$ µm ROIs across the three focal planes (Fig. 4(a)). We recorded typical action potential (AP) waveforms with clearly identifiable phases (Fig. 4(b); i: resting potential, ii: peak depolarization, iii: end of repolarization).

 

Fig. 4. High-speed volumetric voltage imaging of iPSC-derived CM monolayer. (a) Max projections of three different focal planes, with manually segmented $50\times 50$ µm ROIs. Scale bar, 100 µm. $\Delta z$ = 32 µm. (b) Optically recorded voltage traces for 10 seconds at 400 Hz. From bottom to top, the traces correspond to ROIs shown from left to right in (a). Voltage traces depict different phases (i - iii) of each AP.

Download Full Size | PDF

Since engineered CM tissues have optical, structural and functional properties that are more similar to in vivo tissues, they provide better models than simple 2D CM monolayers [41]. We imaged un-paced CM tissue, labeled with FluoVolt, at 400 Hz for 20 seconds. 5mM MYK-461 (Cayman Chemical), a cardiac-specific myosin ATPase inhibitor, was added to attenuate tissue contraction and associated movement-related optical artifacts [42]. Similarly, we extracted voltage traces for manually segmented $50\times 50$ µm ROIs at different depths (Fig. 5(a) and Visualization 4). We observed that the CMs’ APs are highly synchronized (Fig. 5(b) and 5(c)), indicating that even in the absence of an external electrical pacing signal, the cells are electrically coupled within the tissue. An assessment of the similarities and differences in AP waveform morphology can potentially provide insight into the establishment of intercellular synchrony and cellular maturity throughout the tissue. For example, the ability of our microscope to simultaneously monitor APs from multiple CMs at high-speed and over relatively large volumes could aid in the study of in vivo population dynamics in near-natural 3D environments.

 

Fig. 5. High-speed volumetric voltage imaging of µTUG hMSC-CM tissue. (a) Max projections of three different focal planes, with manually segmented $50\times 50$ µm ROIs. Scale bar, 100 µm. (b) Voltage traces of ROIs shown in (a) for an acquisition of 20 seconds at 400 Hz. Traces from bottom to top correspond to ROIs from left to right.(c) Overlay of representative AP from the recorded traces, highlighting both onset synchrony and waveform variability.

Download Full Size | PDF

5. Discussion

In summary, we have demonstrated an augmented variant of line-scan confocal microscopy that provides simultaneous multiplane imaging over a relatively large FOV at high speed–here, hundreds of hertz, compatible with voltage imaging. Our system offers similar advantages to its point-scanning analogue [27]. It is highly light efficient since the signal is detected using the full NA of the objective and with fully reflecting slits. Moreover, it uses the same laser excitation power to image multiple planes simultaneously, thus reducing the amount of excitation light inflicted on the sample to image those planes.

Our system bears resemblance to a light-sheet microscope in the sense that our line-scan illumination effectively corresponds to a scanning light-sheet within the sample, though here oriented vertically rather than horizontally (or obliquely). In particular, light-sheet microscopy has been implemented in confocal line-scan geometries using detection slits either physical [43] or electronic based on camera rolling shutters [44]. The key difference with our technique is that we use multiple slits that are axially distributed, allowing the acquisition of multiple planes simultaneously, rather than a single plane at a time (though see [45,46] where multiple planes are acquired simultaneously, albeit without the advantage of optical sectioning).

Compared to multiphoton multiplane techniques [28–31], which also provide optical sectioning, our system does not provide the same depth penetration, but it is certainly easier to implement and more cost-effective. The speed of multiphoton microscopes is usually limited by the speed of the excitation beam scanners (though see examples of faster variations [47,48]). To obtain frame rates on the order of kilohertz, for example compatible with voltage imaging, this generally limits the imaging to small FOVs [47,49,50]. To date, multiphoton microscopes combining high speed, relatively large FOV, and simultaneous multiplane capacity have not yet been reported.

The speed of our system is currently limited by the speed of our scanners and our camera frame rate. In principle, this speed can be increased by using faster single-axis or resonant scanners combined with a faster camera (the latter are readily available). In principle, also, the sample interplane separation can be adapted to different applications by simply adjusting the total magnification of the system or the physical distance between the reflecting slits. Moreover, the imaging volume can be expanded with no speed penalty by simply decreasing the illumination NA and adding more reflecting slits. Another strategy to augment the axial range is to rapidly change focal depths using mechanisms such as ETLs [5,27], TAG lenses [8,9] or deformable mirrors [7]. However, it should be noted that our method of detection using a single camera and re-scan galvanometer imposed several design constraints that limited our ability to navigate the above parameters while maintaining a reasonably large FOV. In hindsight, perhaps a better design would have been to replace our single area-scan camera with multiple high-speed line-scan cameras, one for each detection slit. Such a design would be limited by the speed of current line-scan cameras to roughly comparable frame rates as our present system. On the other hand, it would allow even more flexibility between FOV, number of planes, interplane separation, etc., while obviating the need for a second re-scan galvanometer and field curvature correction optics. We anticipate implementing such an alternative design in the future.