Optimizing the performance of multiline-scanning confocal microscopy

Abstract

Line-scanning confocal microscopy provides high imaging speed and moderate optical sectioning strength, which makes it a useful tool for imaging various biospecimens ranging from living cells to fixed tissues. Conventional line-scanning systems have only used a single excitation line and slit, and thus have not fully exploited benefits of parallelization. Here we investigate the optical performance of multi-line scanning confocal microscopy (mLS) by employing a digital micro-mirror that provides programmable patterns of the illumination beam and the detection slit. Through experimental results and optical simulations, we assess the depth discrimination of mLS under different optical parameters and compare it with multi-point systems such as scanning disk confocal microscopy (SDCM). Under the same illumination duty cycle, we find that mLS has better optical sectioning than SDCM at a high degree of parallelization. The optimized mLS provides a low photobleaching rate and video-rate imaging while its optical sectioning is similar to single line-scanning confocal microscopy.

1. Introduction

Fluorescence confocal laser scanning microscopy (CLSM) has been an essential tool in biomedical research [1] due to its optical sectioning and capability for high-resolution three-dimensional (3D) imaging [2, 3]. With the help of specific fluorescent labels [4], CLSM has enabled us to investigate subcellular structures as well as the interactions and dynamics of biomolecules in cells and tissues. Recent advancements on pixel reassignment techniques have substantially improved spatial resolution of CLSM, which makes it more popular than ever before [5]. However, the requirement of a high excitation intensity to obtain sufficient photons during a short pixel dwell time causes severe photodamage and thus limits its use in long-term live-cell imaging and single-molecule imaging [6,7].

Parallelized illumination and detection can overcome the aforementioned problems. For example, Nipkow-type spinning disk confocal microscopy (SDCM) that generates numerous points by a microlens and pinhole array has shown to reduce photobleaching rate and increase imaging speed [8, 9]. However, the strong crosstalk between adjacent pinholes has restricted its use to only relatively thin samples [10, 11]. Recently, line illumination has emerged as an effective means of parallelization, although it was suggested three decades ago [12, 13]. In line-scanning confocal microscopy (LS) the fluorescence signal is generated by a line-shaped beam, filtered by a slit and imaged onto array detectors [14–17]. Due to its simplicity, LS has been successfully applied in various areas including fast live-cell imaging [18], endoscopy [19], super-resolution imaging [20], single-molecule fluorescence in situ hybridization [21], high-throughput histopathology [22] and fast multi-focal imaging [23]. In principle, the degree of parallelization in LS can be increased by illuminating with multiple excitation lines. Higher parallelization is beneficial to minimize photobleaching and phototoxicity due to the reduced excitation peak intensity [21, 24]. However, there are only a few reports demonstrating a multiline-scanning system [25, 26] and there have been no studies regarding the optimal degree of parallelization in LS.

Previously, we demonstrated a dual line-scanning confocal system with an inclined beam [21]. It featured single-molecule sensitivity and moderate optical sectioning. Importantly, as the number of lines increased from one to two, reduced photobleaching was clearly observed. A beam splitter and a scientific complementary metal oxide semiconductor (sCMOS) camera conveniently provided two uniform illumination lines and two software-controllable slits, respectively. However, in order to increase the degree of parallelization further and to assess its optical performance, it is highly desirable to employ programmable confocal systems [25, 27–30]. A digital micro-mirror device (DMD) has played a prominent role acting as a configurable pattern generator for both illumination and apertures (pinholes and slits). The DMD is also useful to compare the performance between several parallelized confocal imaging systems.

Here, we investigate the optical sectioning strength of multiline-scanning fluorescence confocal microscopy (mLS) using a DMD. We calculate and measure sea-level responses [10] under different optical parameters and configurations to estimate the maximum degree of parallelization. We show that mLS is superior to multipoint-scanning confocal system at high parallelization in terms of optical sectioning under the same illumination duty cycle. In addition, we demonstrate that mLS enables multicolor and video-rate imaging with reduced photobleaching by employing lasers or a light-emitting diode (LED).

2. Experimental Methods

2.1. Optical setup and DMD operation

A multiline-scanning confocal microscope system (mLS) was built around an Olympus IX73 body as shown in figure 1A and 1B. Here m denotes the number of lines. Lasers (λ = 638/561/488 nm; Cobolt) were coupled into a single mode fiber and collimated with an achromatic lens (L1, f = 100 mm, AC254–100-A; Thorlabs). Alternatively, an LED (λ = 470 nm; M470F3, Thorlabs) was coupled into a multimode fiber (M107L01; Thorlabs), collimated with L1 and spectrally filtered with a bandpass filter (ZET488/10x, Chroma). The illumination beam size was adjusted via an iris. A mirror (M1) and dichroic mirror (DM, Di03-R405/488/561/635-t3–25×36; Semrock) directed the beam onto a DMD (V-7000; ViALUX), which was placed at the primary conjugated imaging plane. The beam was relayed through an achromatic lens (L2, f = 300 mm, AC508–300-A; Thorlabs), fold mirrors (M2 and M3) and sent to an objective lens (UPlanSApo 100×/1.4, oil; Olympus). The collected fluorescence from the objective was reflected by the DMD and sent to a modified Offner system made of a pair of a concave (M4, f = 250 mm, 20DC500ER.1; Newport) and convex mirror (M5, f = −100 mm, 87–680; Edmund) to correct aberration generated by the tilted surface of the DMD [30, 31]. The fluorescence signal filtered by a multiband filter (FF01–446/523/600/677–25, Semrock) was imaged onto an electron multiplying charged-coupled device (EMCCD, iXon Ultra 897; Andor) by a tube lens (L3, f = 150 mm, AC254–150-A; Thorlabs). A piezo stage (Z-Insert.100, Piezoconcept) was used to acquire z-stack images. Micro-Manager was used for image acquisition. A total magnification of our imaging system was ~151 and the field-of-view was 54×54 μm².

Figure 1.

DMD-based multiline-scanning confocal imaging system. (A) DMD patterns for 8LS and 8PS confocal microscopy. Inset, enlarged views of a line and a point on the DMD with 4 pixels width. w, illumination width or line width; p, period; L, field-of-view; N, number of the lines or points. (B) Experimental setup of the mLS imaging system. DM, dichroic mirror; DMD, digital micro-mirror device; F, filter; L1–3, lenses; M1–3, mirrors; Obj, objective lens; SMF, single mode fiber; TL, tube lens. Inset, an enlarged side view of the sample plane when illuminated with 4LS. (C) Synchronization schematic for the mLS.

We used 640×640 DMD pixels for our experiments. To project DMD patterns, black and white images were created and uploaded onto a DMD controller. An output signal from the camera was used to trigger a function generator (DG1022; RIGOL) which generated N pulses for projecting N patterns sequentially on the DMD. For example, in the case of eight line-scanning confocal imaging (figure 1A and 1C), we generated (i) 80 patterns (images) with eight repeated features consisted of 4 pixels ON and 76 pixels OFF and (ii) a pattern with all pixels OFF to avoid unwanted illumination between each frame. In total N = 81 pulses were generated. For Epi illumination, all pixels of DMD were set to ON. The pixel size of the DMD was 13.7 μm, and 1 pixel of the DMD corresponded to ~82 nm on the sample plane.

2.2. Parallelization and illumination duty cycle

We investigated mLS where the number of lines (m) ranged from 1 to 16, denoted as 1LS, 2LS, ... 16LS. The degree of parallelization was assessed by the illumination duty cycle (DC) which was defined by DC = Number of ON pixels/total pixel number. We used 4 pixels (328 nm ≈ 0.72 Airy unit) as an illumination line width (ω) and 640×640 pixels as the field-of-view (FOV) of the DMD unless specified otherwise. For example, the duty cycles of 1LS, 8LS and Epi-illumination are 0.625%, 5% and 100%, respectively. We compared the optical sectioning strength between multiline- and multipoint-scanning systems. For a fair comparison, we set the duty cycles of the two systems as close as possible. Instead of a spinning disk array with a spiral pattern, we used a multipoint scanning system (mPS) with a rectangular pattern, where m denotes the number points along the x (ory) axis given a FOV. For example, 36PS has 36×36 illumination points and exhibits 4×4 ON pixels as a pinhole in a unit cell of 18×18 pixels (figure 1A), yielding a 4.94% DC. Hereinafter we will use mPS and SDCM interchangeably. It should be noted that the number of lines or points we used is only meaningful when a certain FOV is used.

2.3. Sample preparation

2.3.1. Thin dye layer

Atto 488 dye (41698; Sigma-Aldrich) was diluted with 97% 2,2’-Thiodiethanol (TDE, 166782; Sigma-Aldrich) solution which was made of 970 μL of TDE and 30 μL of 1 × phosphate-buffered saline to match the refractive index of dye solution to that of immersion oil. The dye solution was then dropped on a coverslip and covered by another coverslip.

2.3.2. Fluorescent beads

A home-made flow chamber was made similarly as our previous work [32]. A poly-L-lysine solution (P8920; Sigma-Aldrich) was injected into the flow chamber and incubated for 5 minutes. Yellow-green fluorescent beads with a diameter of 100 nm (F8803, Thermo Fisher) were diluted 500 times and sonicated for 5 minutes to avoid aggregation. The beads were injected into the flow chamber, incubated for 5 minutes and washed out by MilliQ water.

2.3.3. Biological samples

A549 cells (CCL-185; ATCC) were grown on coverslips using F-12K medium (30–2004, ATCC) until 70% confluency. The cells were then fixed with 4% (v/v) paraformaldehyde (15710; Electron Microscopy Sciences) for 10 minutes and permeabilized with 0.1% Triton X-100 (93443; Sigma-Aldrich) for 5 minutes at room temperature [32]. To reduce nonspecific staining, the cells were incubated with 1% bovine serum albumin (B9000S; New England BioLabs) for 30 minutes before they were stained with Alexa Flour 647 (AF647) phalloidin (A22287; Thermo Fisher) for 20 minutes. To extend the storage time, the A549 cells were mounted to a glass slide in Prolong Diamond antifade reagent (P36961; Thermo Fisher) before they were sealed with epoxy. A tissue sample was purchased from Thermo Fisher (F24630). For live-cell imaging, U2OS cells (HTB-96; ATCC) were grown in the 8-well Lab-Tek chamber (155411; Thermo Fisher) until 60~70% confluency and then transduced with CellLight Tubulin-GFP reagents (C10509; Thermo Fisher). After 16 hours in the culture incubator, the U2OS cells expressing tubulin-GFP were imaged.

3. Results

3.1. Optical resolution of mLS

We measured the effective point spread function (PSF) using 100 nm fluorescence beads. Here we employed an sCMOS camera (Zyla 4.2 Plus; Andor) to avoid a pixelation error [33] and used an LED as an excitation light source. The width of the illumination lines and confocal slits were set to 4 pixels of the DMD, which was less than 1 Airy unit, to strongly reject out-of-focus background. The measured full width at half maximum (FWHM) of 8LS in the x- and z-axis were 258.3 ± 3.3 nm and 539.2 ± 6.2 nm (mean ± s.e.m. from >20 beads) while those of Epi were 277.9 ± 5.7 nm and 532.4 ± 3.3 nm (figure 2), where s.e.m. denotes the standard error of the mean. Similar to our previous study, the FWHM in the x-axis (the scanning direction of beam) of 8LS was ~8% smaller than Epi. Interestingly, the FWHMs measured with an LED were slightly smaller (~8 %) than those with 488 nm laser presumably due to the lower coherence of the LED.

Figure 2.

Experimental measurement of the effective PSF. (A) Left: Overview image of 100 nm beads in 20×20 μm² FOV. Scale bar, 5 μm. Right: Fluorescence images of beads measured by eight line-scanning (8LS) confocal microscopy and Epi illumination microscopy in xy and xz planes. Scale bar, 500 nm. The images were binned by 2×2×2 in x, y and z-axis. Lateral (B) and axial (C) profiles.

3.2. Optical sectioning performance of mLS

We measured the sea response of our mLS imaging system with different numbers of illumination lines by recording fluorescence images of a thin dye layer (~5 μm thickness) while maintaining the same illumination line width (ω = 4 pixels) (figure 3A). figure 3B shows the line profiles of the illumination line in figure 3A. While moving the sample along the z-axis, the average intensities at each frame were obtained (figure 3C). As expected, the background level increased as the line separation narrowed, i.e. the number of lines increased. Notably, 1LS, 2LS, 4LS and 8LS showed similar sea-level responses while 16LS and 24LS had markedly higher background at the outer region of the sample. The images in figure 3A are somewhat darker in the top and left areas, which may reduce the effective FOV but this can be improved by optimizing our imaging system.

Figure 3.

Sea response of mLS. Fluorescence images (A) and line profiles (B) of 4LS (p = 160 pixels) and 8LS (p = 80 pixels) when illuminating a Atto488 thin dye layer. Scale bar, 10 μm. (C) z-response measurement at different number of illumination lines.

Next, we imaged a 16 μm thick mouse tissue section using Epi and mLS imaging systems at an imaging depth of z = 5 μm. The illumination linewidth and line separation were the same as the previous experiments. Similar to the thin dye layer, 1LS, 2LS, 4LS and 8LS yielded high contrast images but 16LS showed pronounced background (figure 4A). To keep the same level of light dose for each imaging method, the exposure time of each line was fixed to 2 ms such that the frame rate of 1LS was 8 times slower than that of 8LS.

Figure 4.

Optical sectioning performance of mLS with different numbers of illumination lines (A) and different line widths (B). Images and line profiles of mouse kidney sections stained with Alexa Fluor 488 wheat germ agglutinin. Line profiles along yellow dashed lines. Scale bar, 10 μm.

To study the effect of the slit width on the sectioning capability of mLS, we imaged the same sample by 8LS but with different line widths. As shown in figure 4B, the out-of-focus background started to increase when the line width was larger than 12 pixels (0.98 μm) while 4 and 8 pixels showed the similar results. Note that a width of 8 pixels (1.45 Airy units) may be beneficial to increase the illumination duty cycle and the efficiency of the excitation and detection; however, it could impair the axial resolution to some extent.

3.3. Comparison with spinning disk confocal microscopy

Taking advantage of the flexibility of a DMD-based imaging system, we directly compared the optical sectioning performance between mLS and SDCM. As described earlier, we chose 8LS and 36PS for comparison because their illumination duty cycles were the same (5 %). The distance between adjacent lines in 8LS was 6.58 μm and the distance between adjacent points in 36PS was 1.48 μm. When imaging tubulin-GFP in live U2OS cells, we observed that the background levels of both approaches were similar but the fluorescence intensity of 8LS was 1.3-times higher than that of 36PS (figure 5A). Importantly, when imaging thicker samples such as tissue specimens 8LS showed a much lower background level (~1.5-fold) compared to 36PS (figure 5B), indicating that 8LS has a stronger sectioning capability than SDCM.

Figure 5.

Images of live U2OS cells expressing tubulin-GFP (A) and mouse kidney sections stained with Alexa Fluor 488 wheat germ agglutinin (B) by 8 line-scanning (8LS), 36 point-scanning (36PS) confocal microscopy and epi-illumination widefield microscopy (Epi). Right: line profiles of 8LS and 36PS. Scale bars, 10 μm.

To gain further insight on the optical sectioning strength of mLS and SDCM, we calculated their sea-response for a fluorescent sample with a thickness d. The sea-response was calculated as: $I_{sea}(z;d)={\int }_{-z-d/2}^{-z+d/2}{I}_z\left(z^{\prime }\right)d{z}^{\prime }$ where I_z(z′) is the axial response to an infinitely thin fluorescent sample at a relative distance z’ [10]. A detailed description of I_z has been provided in the Appendix. We first calculated the sea-response of mLS with d = 5 μm. Whereas CLSM displayed a sharp response at the sample boundary, mLS showed a gradual increase in background as the number of lines increased from 1 to 16 due to crosstalk between lines (figure 6A). mLS showed an elevated background at z = 0 and a slow background tail at a long distance from the sample. In contrast, SDCM showed a sharper drop at the boundary compared to mLS and a slowly varying offset (figure 6B). When the sample is thicker, i.e. d = 20 μm, the background levels of both methods increased due to the substantial out-of-focus background (figure 6D and 6E). Figure 6C clearly depict the depth discrimination of mLS and SDCM. In order to estimate the extent of background level, we defined the offset background as $I_{offset}(d)=\frac{I_{\mathit{\text{sea}}}(z=0;d)}{I_{\mathit{\text{sea}}}(z=0;d\approx 0)}-1$ to indicate how much the out-of-focus background contributes to a total signal at z = 0. The offset backgrounds of mLS were plotted as a function of the sample thickness d in figure 6F. Our simulation results corresponded well with our experimental data. Our results showed that when the number of illumination lines is larger than 4, mLS had less background than SDCM.

Figure 6.

Sea-response calculations of mLS (A, D) and mPS (B, E) at a sample thickness d = 5 μm and d = 20 μm. (C) Comparison of sea-responses for 4LS and 26PS (blue), and 8LS and 35 PS (green) at d = 20 μm. (F) Sample thickness dependence of the offset background level for various imaging systems.

3.4. Multi-color, video-rate imaging and photobleaching

The mLS imaging system exhibits the main features of line-scanning confocal microscopy, meaning that it is straightforward to realize multi-color imaging (figure 7A) and fast 3D imaging (figure 7B). The main benefit of an increased number of illumination lines was that we could minimize photobleaching rates due to the reduced peak excitation intensity [21]. We obtained time-lapse images of actin labeled with AF647 by 1LS and 8LS with the same light doses and frame rates (figure 7C). The intensity time traces (figure 7D) were fit with a single exponential decay function, and it was observed that 8LS had a 1.8 times slower photobleaching rate than 1LS. A log-log plot of the fluorescence decay rate of each modality versus the excitation intensity in figure 7E gave a slope of 0.27 ± 0.04, indicating that a low peak excitation intensity with a longer dwell time efficiently mitigates photobleaching.

Figure 7.

(A) Two-color images of mouse kidney sections stained with Alexa Fluor 488 wheat germ agglutinin (green) and Alexa Fluor 568 phalloidin (red) imaged by Epi and 8LS. (B) Left: A maximum intensity projection image from 3D stacks imaged by 8LS with a frame rate of 50 Hz. Right: 2D images at z = 0, 3, 6, 9 μm. (C) Time-lapse image of A549 cells stained with AF647 phalloidin imaged by 1LS and 8LS. The frame rate was 3 Hz and the illumination intensity was 3.2 kW/cm² for 1LS and 0.40 kW/cm² for 8LS. (D) Photobleaching time traces of actin in A549 cells. Inset, log-log plots of photobleaching rates for ILS, 8LS and 16LS. Scale bars, 10 μm.

4. Discussion and Conclusions

Here we have demonstrated that the degree of parallelization in multiline-scanning confocal microscopy can be increased up to eight while maintaining the optical sectioning capability. The mLS imaging system showed low photobleaching as well as other desirable features such as multi-color and 3D high-speed imaging. Under the same illumination duty cycle, for example, at 5%, mLS exhibited less background than SDCM when imaging thick specimens. This may not be quite surprising because for 36PS the separation between pinholes was just 1.5 μm so crosstalk between pinholes is substantial. For reference, a traditional SDCM imaging system with a 4% duty cycle has 2.5 μm of pinhole-pinhole distance. It is possible to increase the illumination duty cycle and detection efficiency by using a wider slit when the sample has a moderate thickness.

A DMD-based imaging system is versatile to configure excitation and pinholes/slits [25, 28–30] but it has three major drawbacks. Firstly, it may induce optical aberrations, preventing diffraction-limited resolution. The edge of each micro-mirror in the detection pathway may distort the PSF, and a modified Offner system that relays an off-axis intermediate image of the DMD to a camera may not be able to fully correct the optical aberrations. Secondly, it reduces the fluorescence signal due to the rather low reflectance of the DMD (~65%). This could make it challenging to obtain high signal-to-noise-ratio (SNR) images for weakly fluorescent samples such as single-molecules. Lastly, due to the low illumination duty cycle of mLS, even with 1 W of input power the maximum output power at objective lens will be <25 mW. This is rather low for achieving high spatial resolution for single-molecule localization microscopy [34]. A possible way to overcome these problems is to make use of a combination of galvo mirrors and multiple slits instead of DMD. A similar approach was successfully used in single line-scanning confocal system [13, 20]. High efficiency beam splitters can also improve the light efficiency [21, 35]. It should be noted that SDCM has the latter two issues as well. For example, it was reported that the detection efficiency of SDCM was significantly lower than that of Epi [36] and the light efficiency of the state-of-the-art SDCM was ~1% [35].

Line-scanning confocal microscopy exhibits an anisotropic PSF, i.e. sharper features along the scanning direction. Although the degree of anisotropy is negligible in many cases [21], several methods have been suggested to obtain isotropic resolution, for example, by bi-directionally scanning and subsequent image processing [37, 38]. Alternatively, the anisotropic PSF could be corrected through deep learning techniques, for instance, by isotropic 2D construction from anisotropic raw images [39]. The background level of mLS is moderately higher than that of CLSM. To further suppress the background, one could apply a spatiotemporal phase modulation of the excitation beam [40] or 3D deconvolution [41].

Although mLS imaging system can reduce photo-bleaching, light-sheet fluorescence microscopy is more suitable to long-term volumetric imaging [6, 7, 42]. Nonetheless, the main advantages of mLS are that it can achieve high spatial resolution and high photon collection efficiency with high NA objective, and it can image both thin and thick samples without a specially designed sample holder. We expect that multiline-scanning confocal microscopy will be a good alternative to SDCM and a powerful tool for many applications such as fast high-throughput imaging on thick samples and live-cell imaging with low photodamage.

Appendix

The z response of each imaging system was calculated to be [10]:

$$I_z=\iint {\left|h_{exc}\otimes g\right|}^2\cdot \left({\left|h_{det}\right|}^2\otimes p\right)\otimes gdxdy$$

where h_exc and h_det are the excitation and detection amplitude PSF of the objective lens, $\otimes$ stands for convolution operation, g is the grating function which contains the coordinates information for parallel illumination and p is the pinhole/slits function.

For CLSM and SDCM, h_exc and h_det have the same form [43]:

$$\begin{gathered} h(x,y,z) \\ =A{\int }_0^{\alpha }\sqrt{\cos (\theta )}\sin (\theta )J_0\left(k\sqrt{x^2+y^2}\sin (\theta )\right)\exp (ikzcos(\theta ))d\theta \end{gathered}$$

where A is a constant proportional to beam intensity, J₀ is the zero order Bessel function of the first kind, k is the wave number, θ is the polar aperture angle and a is the maximum half aperture angle of the objective. The grating functions and pinhole function are given by [44]

$$g_{CLSM}(x,y,z)=\delta (x)\delta (y)\delta (z)$$

$$g_{SDCM}(x,y,z)=\sum _{n=1}^N\delta \left(x-x_n\right)\delta \left(y-y_n\right)\delta (z)$$

$$p(x,y)=\{\begin{array}{cc}1,& \sqrt{x^2+y^2}\leqq p_0\\ 0,& \sqrt{x^2+y^2}>p_0\end{array}$$

where N is the total number of parallelized pinholes, x_n and y_n are the offset of x and y from the first pinhole to the N^th pinhole, p₀ is the radius of pinhole in SDCM and is the width of slit in mLS.

For mLS, the detection amplitude PSF is the same as that of CLSM while the excitation amplitude PSF is written by [21, 43]:

$$h_{mLS}(x,y,z)=A{\int }_{-\alpha }^{\alpha }\frac{\sqrt{\cos (\theta )}\exp (-\text{ikxsin}(\theta ))}{\exp (-ikzcos(\theta ))kcos(\theta )}d\theta$$

The grating and slit functions are given by

$$g_{mLS}(x,y,z)=\sum _{n=1}^m\delta \left(x-x_n\right)\delta (z)$$

$$p_{mLS}(x,y)=\left\{\begin{array}{cc}1& \left|x\right|\leqq p_0\\ 0& \left|x\right|>p_0\end{array}\right\}$$

For the z response calculations, we chose the parameters as follows: The effective numerical aperture of the objective lens was 1.4 (oil immersion, n = 1.518). The excitation and detection wavelength were 0.48 μm (λ_exc) and 0.52 μm (λ_det), respectively. A 54 μm × 54 μm of the imaging FOV was divided by a two-dimensional pixel grid, of which the pixel size was 106.2 nm. The step size of δz was 100 nm.