1. Introduction

The wide popularity of fluorescence microscopy in biological research benefits greatly from its unrivaled molecular specificity [1–3]. However, the application of fluorescence microscopy to complex biological systems is hindered by its limited multiplexing capability owing to the intrinsically broad spectra [4]. Common approaches with dichroic mirrors and bandpass filters allow sequential imaging of 3-4 fluorophores. Characteristic changes in the fluorescence intensity, e.g., bleaching and recovery kinetics [5], excitation-induced fluctuations [6], and photostability [7], offer new possibilities to distinguish multiple (3-4) fluorophores. However, dedicated optical setups and/or fluorophores are required, and these emerging approaches are generally time-consuming. Multiplex imaging also plays a key role in multiphoton fluorescence microscopy, creating yet another level of challenge for the simultaneous detection of multiple cellular and molecular structures in live samples [8–10].

Fluorescence spectral microscopy provides a powerful way to discriminate highly overlapping fluorescence signals from different probes based on their spectral signatures [11–17]. This strategy, when combined with linear spectral unmixing, resolves 4-6 probes in the same sample. In typical approaches, spectrally resolved images are acquired by locally dispersing the fluorescence emission from a single spot, e.g., in a confocal setting [18,19]. The need to scan the sample in two dimensions, however, places substantial constraints on the imaging speed. As an alternative, spectral data is sequentially acquired as a series of wide-field images (λ-stack) with tunable thin-film filters, resulting in inefficient use of the scarce signal [20].

In contrast, by sequentially exciting the sample at different wavelengths for the entire imaging field and recording all the emission at each excitation wavelength, excitation wavelength-resolved images can be acquired in the wide field with high throughputs [13,21–25]. Recent developments demonstrated the real-time visualization of 6 subcellular organelles in one image at 66 ms/frame using 6 continuous-wave (CW) lasers spanning the visible range [21]. However, this implementation suffers from the limited excitation channels with the discrete CW laser lines, thus causing high (∼50%) target-to-target crosstalks. To address this problem, we recently showed that with frame-synchronized fast scanning of the excitation wavelength at ∼10 nm resolution, e.g., by combining an acousto-optic tunable filter (AOTF) with a broadband light source (white lamp), up to 6 fluorophores can be simultaneously imaged in the same cell at low (∼1.3%) crosstalks and high temporal resolutions [22]. In this previous work, a single fluorescence band is used, limiting spectral scanning to 8 preset wavelengths in the 460-565 nm range. It thus remains challenging to push the limit of color barrier over six targets that can be simultaneously imaged in one cell.

In this work, we expand excitation spectral microscopy to the entire visible range, thus making full use of the broadband light source. The resultant super-multiplexed spectral microscope features 24 preset excitation wavelengths in the range of 420-650 nm, achieved by sequential imaging with 3 filter cubes for 3 spectral bands (blue, green, and red).

2. Materials and methods

2.1 Super-multiplexed excitation microscopy setup

We originally realized excitation spectral microscopy by scanning 8 preset wavelengths in the 460-565 nm range [22]. For extension of its multiplex capability, in this study we upgraded the imaging system with extended excitation wavelengths (Fig. 1(a)). Briefly, white light from a plasma lamp (HPLS301, Thorlabs) was collimated and then linearly polarized through a wire grid polarizer (WP25M-VIS, Thorlabs). The polarized beam entered an acousto-optic tunable filter (AOTF) (EFLF100L1, Panasonic), and the 1^st order diffracted beam, with its polarization direction rotated by 90° versus the incident beam [26], was cleaned up with a second polarizer (WP25M-VIS) mounted perpendicular to the first polarizer. This excitation beam was coupled into an Olympus IX73 inverted epifluorescence microscope and focused at the back focal plane of an oil-immersion objective lens (Olympus, UPLSAPO100XO, NA 1.40). To extend the excitation wavelength, the microscope was equipped with three filter cubes for 3 spectral bands (blue, green, and red) to cover the visible spectrum (Fig. 1(c)). For the blue band, the excitation filter, dichroic mirror, and emission filter used were BSP01-532R-25, FF526-Di01-25 × 36, and FF01-519/LP-25, respectively, from Semrock. For the green band, the filters were FF01-505/119, FF573-Di01, and FF01-630/92, respectively, from Semrock. For the red band, the filters were RPE-650SP (Omega), FF660-Di02-25 × 36 (Semrock), and FF01-708/75-25 (Semrock), respectively. In each band, the AOTF unit was driven by an 8-channel RF synthesizer (97-03926-12, Gooch & Housego) to scan for 8 preset wavelengths at 10∼15 nm spectral resolution (Fig. 1(b) and Table 1). The total 24 excitation wavelength profiles at the different applied RF frequencies were measured using a visible spectrometer (USB4000-VIS-NIR, Ocean Optics). Typical excitation power, which was determined by a photodiode power meter (S120VC, Thorlabs), was ∼6 µW at each wavelength, corresponding to a power density of ∼0.16 W/cm² at the sample, which allowed adequate signal and minimal photobleaching. The RF power was preset to different values for the different wavelengths to compensate for variations in the lamp power density and the AOTF transmission efficiency, ensuring that the power arrived at the sample was identical for the 24 wavelengths. Wide-field fluorescence images were continuously recorded using an sCMOS camera (Zyla 4.2, Andor) at an effective pixel size of 130 nm (after 2 × 2 binning) for 512 × 512 pixels (67 × 67 µm² field of view) at 10 fps. The camera was internally triggered, and the “fire” (exposure) output trigger signal was read by a multifunction I/O board (PCI-6733, National Instruments), which in turn modulated the RF synthesizer for the synchronized control of the excitation wavelength in each camera frame.

 

Fig. 1. Excitation spectral microscopy with 24 excitation wavelengths for super-multiplexed fluorescence imaging. (a) Schematic of the setup. Full-frame spectral micrographs are obtained by the frame-synchronized modulation of the excitation wavelength λ in 24 consecutive frames across three excitation/detection bands, with the switch of filter cube every 8 wavelengths/frames. P polarizer, L lens, F bandpass filter, DM dichroic mirror. (b) Measured AOTF output (microscope input) spectral profiles for the 24 excitation wavelengths used in our experiments across the three bands (blue, green, and red), spanning the visible spectrum at 10∼15 nm spectral resolution. (c) Spectral transmission profiles of the excitation filter (F1), emission filter (F2), and dichroic mirror (DM) in each band.

Download Full Size | PDF

 

Fig. 2. Simulated excitation spectra of ten common fluorophores with the 24 excitation wavelengths on our setup. (a) Manufacturer-provided absorption and fluorescence emission spectra of the 10 fluorophores used in the simulation: LipidSpot488 (LS488), SYBR Gold, CF514, Atto532, LysoBrite Orange (LB Orange), MitoTracker Orange (MT Orange), CF583R, CF620R, CF633, and AF647. (b) The simulated excitation spectra for the 10 fluorophores on our setup, for the 24 excitation wavelengths recorded with the 3 filter cubes. Results are obtained by multiplying the power-normalized spectral profiles in Fig. 1(b) with the absorption spectra of the fluorophores in (a, left) and then scaled by the integration of the emission spectra (a, right) over the effective spectral range after accommodating the emission filtering in each band in Fig. 1(c).

Download Full Size | PDF

Table 1. The preset AOTF driving frequencies and excitation wavelengths

View Table

2.2 Image acquisition and analysis

Images were sequentially acquired for the red, green, and blue bands by switching the filter cube turret on the microscope. For each band, the excitation wavelength was scanned for eight preset wavelengths in consecutive camera frames. The preset AOTF driving frequencies and the corresponding excitation wavelengths for all three bands are shown in Table 1. Full-frame excitation-spectral images were thus obtained with 24 camera frames in a few seconds (Fig. 1(a)). Before multitarget imaging, the reference excitation spectrum of each fluorophore was obtained from singly labeled fixed cells by recording images at the above preset excitation wavelengths on our setup under the same acquisition conditions.

Images collected in the red and blue bands were first aligned to that collected in the green band using a 2D affine transformation (imregtform function in MATLAB). Afterward, for each pixel, the excitation spectrum, namely, the recorded intensity under each scanned wavelength, was separately unmixed into a linear combination of the reference excitation spectra (above) of the pertaining fluorophores using least squares regression under non-negativity constraints (lsqnonneg function in MATLAB). The resultant abundances of different fluorophores at each pixel were then used to generate unmixed images, which were then false-colored and/or merged using Image J (Fiji). We also quantified the autofluorescence/background in cell samples under the same conditions and found it to be <1% of the signal from the fluorescent dyes, thus negligible in spectral unmixing. Unmixed images of various organelles were segmented in MATLAB using histogram-based approaches for defining intensity thresholds. In Fig. 5 h and 5i, the optimal threshold values for the nuclei, mitochondria, LDs, and nucleoli were determined using the ‘imbinarize’ function in MATLAB. In Fig. 5 h, the ER, α-tubulin, acetylated α-tubulin, and actin were segmented using a local threshold determined using the MATLAB function ‘adaptthresh’.

To evaluate the performance of our super-multiplexing excitation spectral microscopy, we adopted an experimental approach to estimate the spectral crosstalks between all color channels [21,22,27]. For the simulated results in Fig. 3, we used different patterns in separated regions for different fluorophores. Thus, each pixel in the simulated image is uniquely positive for only one dye, and all counts in this pixel should be assigned to this dye after 10-color unmixing. The residuals in each pixel being misidentified as other dyes thus directly represent the crosstalks. For the experimental results in Fig. 4 & 5, cells were first fixed and singly labeled with each fluorophore. Forcing the same 6-color or 8-color unmixing procedure onto each singly labeled sample enabled us to evaluate how much each given color channel crosstalks into other channels. Analytical models have also been developed based on the Fisher information matrix to predict how accurately a pair of spectrally similar fluorescent labels can be spectrally unmixed [28]. However, such approaches have not been demonstrated for evaluating the crosstalks in complex settings as the linear unmixing of ten spectrally overlapped fluorophores here.

 

Fig. 3. Simulation of super-multiplexed excitation spectral microscopy for 10-fluorophore imaging. (a) Artificially synthesized 24-channel image array of 10 differently labeled structures acquired at the different given excitation wavelengths across the three excitation bands (blue, green, and red): each of the 10 fluorophores was simulated based on the expected excitation spectra in Fig. 2(b) to occupy a different region of the view, and then an uncorrelated Gaussian noise of a standard deviation 5% of the signal was added to every pixel at each wavelength. (b) Unmixing of the 24-wavelength images in (a) into the 10 fluorophore channels. (c) The unmixed abundancy values in different fluorophore channels for the different simulated regions occupied by each of the fluorophore, representing the crosstalk between different colors.

Download Full Size | PDF

 

Fig. 4. Super-multiplexed excitation spectral microscopy for 6-color cell imaging. (a) 24-channel image array of a fixed COS-7 cell sample labeled by the 6 fluorophores described below, acquired at 10 fps (100 ms exposure time) at the different excitation wavelengths across the blue, green and red excitation bands. (b) Unmixed images of the six subcellular targets. Direct dye staining: LipidSpot488: lipid droplets (LDs), SYBR Gold: nucleus, MitoTracker Orange: mitochondria. Immunofluorescence: CF514: α-tubulin, CF583R: nucleolus, AF647: ER. (c) Overlay of the above six images. (d) Unmixed abundancy values in the different fluorophore channels for samples singly labeled by each of the six fluorophores by forcing the same six-fluorophore unmixing procedure. (e) Reference excitation spectra of the six fluorophores in the 24-wavelength excitation scheme, separately measured on the setup using singly labeled samples. (f) Magnification of the white box region in b showing spatial colocalizations between LDs and mitochondria, ER and α-tubulin, respectively. (g) The quantified fractions of LDs contacting mitochondria, ER, α-tubulin, and simultaneously contacting both ER and α-tubulin, respectively. Scale bars, 20 µm (a,), 10 µm (b, c), 5 µm (e).

Download Full Size | PDF

 

Fig. 5. Super-multiplexed excitation spectral microscopy for 8-fluorophore imaging in cells. (a) 24-channel image array of a fixed COS-7 cell sample labeled by the 8 fluorophores described below, acquired at 10 fps (100 ms exposure time) at the different excitation wavelengths across the blue, green and red excitation bands. (b) Unmixed abundancy values in different fluorophore channels for samples singly labeled by each of the 8 fluorophores. (c) Unmixed images of the 8 fluorophores: lipid droplets (LipidSpot488), nucleus (SYBR Gold), α-tubulin (CF514), nucleolus (Atto532), mitochondria (MitoTracker Orange), ER (CF583R), acetylated α-tubulin (CF620R), and actin (AF647-phalloidin). Overlays of two species are shown in each image for comparison of the structural arrangements of the different targets. (d) Overlaid image of the 8 unmixed fluorophores. (e) Reference excitation spectra of the eight fluorophores in the 24-wavelength excitation scheme, separately measured on the setup using singly labeled samples. (f) Magnification of the white box region in (d) for five subcellular structures highlighting acetylated α-tubulin overlapping with a subset of microtubule filaments, a subset of actin bundles, and marking ER-mitochondrion contacts. Green and white arrowheads point to α-tubulin only and colocalization of α-tubulin and acetylated α-tubulin, respectively. Cyan arrowheads point to ER-mitochondria contacts marked by acetylated α-tubulin. White boxes indicate the colocalization of α-tubulin, acetylated α-tubulin, and actin bundles. (g) The quantified fractions of mitochondria contacting the ER and acetylated α-tubulin, and simultaneously contacting both ER and acetylated α-tubulin, respectively. Scale bars, 20 µm (a), 10 µm (c, d), 2 µm (f).

Download Full Size | PDF

2.3 Multi-labeling cell samples

COS-7 cells (Cell Culture Facility, UC-Berkeley) were cultured in 8-well LabTek chambered coverglasses in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco 31053-028) supplemented with 10% fetal bovine serum (Corning), 1× GlutaMAX Supplement, and 1× non-essential amino acids, at 37 °C and 5% CO₂. To label mitochondria, MitoTracker Orange (1:1000) was added to the medium for 30 min at 37 °C and washed 3 times with DMEM. Cells were then fixed using 3% paraformaldehyde and 0.1% glutaraldehyde in phosphate-buffered saline (PBS), followed by two washes with 0.1% sodium borohydride in PBS. Cells were blocked and permeabilized in a blocking buffer (3% bovine serum albumin with either 0.5% Triton X-100 or 0.02% saponin in PBS), followed by primary and secondary antibody labeling.

Primary antibodies used were chicken anti-α-tubulin (ab89984, Abcam) for labeling of microtubules, rabbit anti-Nogo (ab47085, Abcam) for labeling of the ER, mouse IgG1 anti-NPM1 (32-5200, ThermoFisher) for labeling of nucleoli, and mouse IgG2b anti-acetylated α-tubulin (T6793, Sigma). The following dye-conjugated secondary antibodies were prepared in house using previously described protocols [29]: goat anti-mouse IgG1 (Jackson 115-005-205)-Atto532 (AD532-31, ATTO-TEC), goat anti-mouse IgG1 (Jackson 115-005-205)-CF583R (96085, Biotium) (Fig. 4), goat anti-mouse IgG2b (Jackson 115-005-207)-CF620R (92194, Biotium), goat anti-Rabbit IgG (H + L) (Jackson 111-035-003)-CF583R (96085, Biotium), goat anti-Rabbit IgG (H + L) (Jackson 111-035-003)-AF647 (A20006, ThermoFisher) (Fig. 4), goat anti-chick (Jackson ImmunoResearch 703-005-155)-CF514 (92103, Biotium). These antibodies (0.3-0.4 mg/mL) were used at 1:60 dilution for immunofluorescence. After antibody staining, cells were incubated in LipidSpot488 (500-1000×, 70065, Biotium), SYBR Gold (3000-10000×, S11494, ThermoFisher) and/or Alexa Fluor 647-phalloidin (500-1000x, A22287, Invitrogen) in PBS to stain lipid droplets, nuclear DNA, and the actin cytoskeleton for 30 min, and the sample was washed with PBS for 10 min × 3 times before imaging in PBS.

3. Results

We start by simulating the excitation spectra of 10 common fluorophores spanning the visible range. Figure 2(b) shows the expected spectra collected at the 24 preset excitation wavelengths (Fig. 1(b)) in the three fluorescence bands (blue, green, and red) of our setup (Fig. 1(c)), generated by multiplying the power-normalized spectral profiles in Fig. 1(b) with the literature absorption spectra of the fluorophores in Fig. 2(a). These results are in general agreement with our experimental observations (Figs. 4 and 5 below); small deviations are attributed to possible differences between the literature absorption spectra and the actual excitation spectra of the fluorophores.

Based on the expected excitation spectra in Fig. 2(b), we next simulated image patterns made of the 10 fluorophores (Fig. 3(a)). A Gaussian noise of a standard deviation 5% of the signal was added to every pixel at each wavelength. Separation of the simulated 24-wavelength image based on the reference spectra showed unambiguous identification of the 10 fluorophores (Fig. 3(b)). Quantification showed remarkably low crosstalks between the 10 fluorophores at ∼0.5% on average (Fig. 3(c)), highlighting the potential of extended excitation spectral microscopy to the entire visible range.

We next experimentally applied 24 excitation-wavelength spectral microscopy to cell samples (Fig. 4(a)). With a combination of live-cell and fix-cell staining by organic dyes (LipidSpot488, SYBR Gold, MitoTracker Orange, CF514, CF583R, and AF647), 6 distinct subcellular targets (lipid droplets (LDs), nucleus, mitochondria, microtubule, nucleolus, and endoplasmic reticulum (ER)) were well unmixed (Fig. 4(b)) based on their reference excitation spectra (Fig. 4(e)) from singly labeled samples. Meanwhile, forcing the same 6-fluorophore unmixing procedure onto samples singly labeled by each fluorophore, we evaluated the crosstalks between different fluorophore channels (Fig. 4(d)). We thus found that the 24-wavelength excitation spectral microscopy offers superior performance with 1.6% maximum crosstalk and 0.29% average crosstalk between the 6 fluorophores, in comparison to the 3% maximum crosstalk and 1.3% average crosstalk using 8 excitation wavelengths for 6 fluorophores in our previous work [22].

The well-unmixed images clearly visualized the spatial organizations and potential interactions between the different intracellular structures (Fig. 4(c)). For example, for interactions between the LDs and the other organelles (Fig. 4(f)), we quantified the fractions of LDs that contacted mitochondria, ER, and microtubules (Fig. 4 g), and found that only ∼7% contacted mitochondria, but 89% and 73% contacted the ER and microtubules, respectively. More interestingly, 71% of LDs simultaneously contacted the ER and microtubules. The direct visualization of the three-way interaction points to microtubules’ role in establishing and maintaining LDs-ER contacts, as previous studies suggested [21].

To further explore the multiplexing potential of our 24 excitation-wavelength imaging system, we added two more fluorophores (Atto532 and CF620R) and subcellular structures (the actin cytoskeleton and acetylated α-tubulin), and permutated the organic dyes tagged to the above intracellular structures/organelles, highlighting the flexibility of fluorophores choice in a wide range. Especially, we challenged the ability of the super-multiplexed excitation microscopy with spatially overlapped fluorescent labels. For this, we immunostained cells for α-tubulin and acetylated α-tubulin, which substantially overlay each other, with CF514 and CF620R, together with six additional stains for other targets (Fig. 5(a)).

Figure 5(a) presents the full-frame excitation spectral images obtained every 24 camera frames for the different excitation wavelengths across the three excitation bands. Whereas the raw images at each wavelength (Fig. 5(a)) contained substantial contributions from multiple fluorophores due to the large spectral overlaps (Fig. 5(e)), unmixed images (Fig. 5(c)) showed excellent spectral separation at <0.5% average crosstalk (Fig. 5(b)), allowing for easy distinction of each cellular structure. Several expected cellular features were observed, i.e., acetylated α-tubulin was found as segments along microtubules, indicating only a subset (∼34% based on pixel colocalization) of the microtubules were acetylated (white vs. green arrowheads in Fig. 5(f)) [29,30]. Interestingly, we occasionally observed thick actin bundles, but not thin actin filaments, colocalizing with microtubules where the tubulin was preferentially acetylated (white boxes marked regions in Fig. 5(f)), probably representing cross-linking of actin filaments and microtubules through microtubule-associated proteins [31,32]. Additionally, as the mitochondria settled with extensive contact sites with the ER network (Fig. 5(c)), we found most ER-mitochondria contact sites preferentially localized to acetylated α-tubulin (cyan arrowheads in Fig. 5(f)). Quantification showed that ∼80% of the mitochondria-ER contact sites were marked by acetylated α-tubulin (Fig. 5 g), consistent with acetylated α-tubulin’s role in organizing organelles and maintaining inter-organelle contacts [33,34].

To gain additional insights into organelle organizations, we quantified organelle distributions by calculating the distance between the center of the nucleus and each organelle pixel (Fig. 6). For the eight organelles we co-imaged in Fig. 5, relatively homogenous, wide spatial distributions of cytoskeleton structures (actin, α-tubulin, and acetylated α-tubulin) and ER were observed throughout the cytoplasm, whereas mitochondria and LDs displayed an asymmetric, narrower distribution toward the cell center (Fig. 6(a)). Interestingly, we further found varying degrees of this asymmetric distribution between different cells. For example, whereas the LDs and mitochondria both distributed tightly to the nucleus zone for the cell shown in Fig. 5, both organelles were more dispersive in the cytoplasm for the cell shown in Fig. 4. By scrutinizing the distribution patterns of LDs and mitochondria of different cells in our 6-fluorophore and 8-fluorophore data, we found that the two distributions were highly coordinated (Fig. 6(b)), whereas the distributions of cytoskeleton structures and the ER remained wide and homogenous in all cells (not shown). While it is known that contacts between LDs and mitochondria occur under starvation [35], our observation of similar distribution patterns between the two organelles suggests possible interactions under normal conditions. Our spectral imaging showing coordination between 2 of the 8 labeled targets thus demonstrates the potential for visualizing interactions in complex biological systems.

 

Fig. 6. Quantification of organelle spatial distributions. (a) Spatial distributions of the eight organelles co-imaged in Fig. 5, based on the distance to the center of the nucleus. (b) The spatial distribution of LDs and mitochondrion from six different COS-7 cells.

Download Full Size | PDF

4. Conclusions and discussions

In this work, we expanded excitation spectral microscopy for super-multiplexed imaging. Utilizing three excitation/emission bands instead of a single band enabled the imaging of more fluorescent labels across the visible range, and the scanning of 24 excitation wavelengths provided good spectral characteristics for identifying different labels in the same cell. We thus demonstrated the unmixing of 10 fluorophores with minimal color crosstalks in simulations. Experimentally, we simultaneously imaged up to 8 fluorescent dyes labeling different intracellular structures, achieving low (<0.5%) crosstalks between the different labels that were even lower than what we previously achieved for 6 fluorophores under 8-wavelength excitation. We thus advanced the level of fluorescence multiplexing in one cell, opening new possibilities to study complex intracellular organizations and interactions. Moreover, our facile approach is based on a low-cost, lamp-operated epifluorescence microscope, and should be straightforward to implement on most lab-existing microscopes.

As with the typical linear unmixing problem in spectral microscopy, theoretically, N spectral channels may unequivocally unmix up to N targets. Therefore, our imaging scheme here with 24 excitation wavelengths across the visible range should be sufficient for most biological needs, potentially for samples concurrently labeled by >10 fluorophores. Our current experimental demonstration of 8-target imaging was limited by the labeling strategy, e.g., the available host species of primary and secondary antibodies and the labeling specificity of organic dyes. Newly established fluorescent probes and labeling strategies may provide superior multiplexity and specificity in identifying cell structures in the future [36,37]. In addition, antibody complex formation technique is promising to extend the number of targets labeled simultaneously in a single sample [38]. The integration of fluorescence contrast with other imaging modalities, e.g., vibration contrast in stimulated Raman scattering (SRS) [39], offers an additional way to alleviate the need for fluorescent probes for further multiplexing. The potential extension of our approach to three-dimensional hyperspectral imaging may be achieved by further combing light-sheet fluorescence microscopy, for which case frame-synchronized fast wavelength scanning may be implemented with supercontinuum light sources. It should also be feasible to visualize biological samples at the tissue level, e.g., organoids and vasculature structures, using low magnification objectives. Whereas in this work we focused on fixed cells, the super-multiplexing capabilities we demonstrated here should also be extendable to live cells. The current 24-wavelength excitation microscopy limits the imaging speed to ∼6 seconds, mostly constrained by the mechanical switching between the three filter cubes. Future developments to synchronize filter wheels with the camera frames may help push 24-wavelength recording to the sub-second time scales.