Abstract
New innovations in single-molecule localization microscopy (SMLM) have revolutionized optical imaging, enabling the characterization of biological structures and interactions with unprecedented detail and resolution. However, multi-colour or hyperspectral SMLM can be particularly challenging due to non-linear image registration issues, which affect image quality and data interpretation. Many of these arise as a consequence of differences in illumination optics (beam profile, power density, polarization, point spread function) for the different light sources. This is particularly acute in evanescent-wave based approaches (TIRF) where beam shape, decay depth, and power density are important. A potential useful approach would be to use a single excitation wavelength to perform hyperspectral localization imaging.. We report herein on the use of a variable angle tunable thin-film filter to spectrally isolate far-red emitting fluorophores. This solution was integrated into a commercial microscope platform using an open-source hardware design, enabling the rapid acquisition of SMLM images with ~ 15-20 nm spectral resolution.. By characterizing intensity distributions, average intensities, and localization frequency through a range of spectral windows, we identified an optimal fluorophore pair for two-colour SMLM. Fluorophore crosstalk between the different spectral windows was assessed by examining the effect of varying the photon output thresholds on the localization frequency and fraction of data recovered. The utility of this approach was demonstrated by hyper-spectral super-resolution imaging of the interaction between the mitochondrial protein, TOM20 and the peroxisomal protein, PMP70.
I. INTRODUCTION
With the advent of single-molecule localization microscopy (SMLM), new limits of optical resolution have been realized, and the range of biological questions that can be addressed with fluorescence imaging has increased dramatically. Techniques such as stochastic optical reconstruction microscopy (STORM/dSTORM)1 and photoactivated localization microscopy (PALM)2 are able to provide lateral resolution of 10-20 nm with a corresponding axial resolution of 10-100 nm.3, 4 This enhanced resolution has facilitated the sub-diffraction limit visualization of cellular structures such as organelle ultrastructure and protein spatial distribution in both fixed and live cells, 5–7 Characterization of heterotypic protein-protein interactions often requires the use of multi-colour imaging strategies; however, the diverse photophysical properties of many dSTORM fluorophores8–10 and PALM fluorescent proteins2, 11 can lead to unpredictable experimental outcomes, including unequal localization precision across channels in multi-colour imaging.
Several of the far-red emitting photoswitchable fluorescent molecules commonly used in dSTORM imaging have been shown experimentally to exhibit favourable and reliable photophysical properties in a buffered reducing environment.10, 12 These include an inducible dark state transition, high duty cycle, moderate to high photon output, and sensitivity to 405 nm excitation.13 Each of these factors contributes to the blinking behaviour of the fluorophore and thus the molecular density and structural integrity of the final image. Interestingly, small molecule non-far-red emitting fluorophores often do not consistently exhibit characteristics appropriate for localization microscopy. This is problematic since conventional multi-channel imaging relies on the acquisition of data from non-overlapping spectral regions, It can therefore be challenging to collect and process multi-colour super-resolution data. Furthermore, multi-channel SMLM data is very sensitive to chromatic aberrations, which can become the limiting factor in determining the extent of colocalization uncertainty between structures imaged in two separate spectral channels. For these reasons, several groups have recently focused on developing methods for spectrally resolving overlapping far-red emission spectra.14, 15 In one instance, this was accomplished by using a long-pass emission filter while another relied on a prism-based technique, which enabled simultaneous collection of fluorescence intensity and spectral information. While the former offers a proven and reproducible method for resolving SMLM fluorophores based on intensity ratio, it relies on a fixed emission cutoff and requires a split-view or multiple cameras.16 The second approach offers true colour super-resolution and the ability to combine more than two fluorophores, but involves a non-trivial optical imaging setup. Hyperspectral imaging has also been accomplished using acousto-optic tunable filters (AOTFs) and liquid crystal tunable filters (LCTFs).17 Both of these particular approaches offer excellent speed and specificity, yet suffer from a significant attenuation of emission intensity, which renders them unsuitable for SMLM. A alternative approach is to use a thin-film tunable filter (TTF), which features exceptionally high transmission and very steep spectral edges.17 Varying the angle of incidence of emitted light on the filter enables hyperspectral imaging with ~ 15 – 20 nm spectral resolution. In this study, we examined the performance of a TTF in distinguishing between a pair of far-red SMLM fluorophores with overlapping emission spectra using a single excitation laser line and full-field hyperspectral imaging on a single EMCCD camera.
II. EXPERIMENTAL METHODS AND CALIBRATION
A. Optical Setup
All dSTORM imaging was performed on a home-built TIRF microscope. For this platform, a Galilean beam expander was used to expand the excitation laser (170 mW, 643 nm) to one inch in diameter. A beam steerer post was mounted on top of a linear translation stage allowing for imaging in any of three imaging modalities: epifluorescence (Epi), HiLo, or TIRF. A lens used to focus the excitation beam in the back focal plane of the objective was mounted on the steerer such that the distance between the focusing lens and the objective remained constant over the entire operational regime (Epi, HiLo, TIRF). The excitation beam entered the microscope body (dual-deck Olympus IX83, Olympus Canada, Richmond Hill Ontario) through the back port of the upper deck fitted with a standard fluorescence turret. The filter cube for these experiments only contained a 660 nm dichroic beamsplitter. For these experiments, the emission filter was removed, allowing for the full range of possible emission wavelengths. A custom optical plate holding the thin-film tunable filter (TTF: Semrock Versachrome TBP01-697/13) was placed in the lower deck of the IX-83. The TTF mount was fitted to a galvanometer, which controlled the angle of incidence (AOI) between the optical axis of the microscope (objective) and the filter normal. All images were collected on a Photometrics QuantEM EMCCD camera. Image acquisition and galvo scanning were performed with the open source software Micromanager (version 1.4.22). A schematic representation of the optical setup is presented in Figure 1(a). A basic current source was built based on a quadruple half-H driver (SN754410) that followed a TTL signal from an Arduino microcontroller (Figure 1b). The Arduino was configured to act as both the shutter control and configurable emission filter. To this end, each setting in Micromanager activated two output pins in the Arduino board. The first controlled the shutter and the other activated a channel of the half-H drive, selecting for the specific spectral region of the emission spectra.
B. TTF calibration and fluorophore characterization
Modulating the angle of incidence (AOI) from 0° to 60° allowed for the selection of ~ 20 nm wide spectral regions between 615 nm and 700 nm. To calibrate the system and ensure proper selection of the emission bands, polychromatic light was transmitted along the optical path. By varying the driving voltage to the galvanometer, and using a spectrometer (S2000, Ocean Optics, Dunedin, Florida) mounted to the side port of the microscope, a calibration curve was generated that mapped the driving voltage, AOI, and the corresponding 20 nm emission bands with center wavelengths varying from 670 nm to 700 nm. (Figure 2(a)).
Subsequently, the emission intensity of 5 far-red emitting fluorophores (Alexa 647, ATTO 647, Alexa 680, ATTO 680 and CF 680) was measured at several points between the 0° and the 20° position of the filter by taking the average intensity of 3 TIRF images per center wavelength (Figure 2(b)). In all cases, the standard deviation of the measurements was negligible. The EMMCD gray level increased and decreased in a predictable and reproducible manner for all fluorophores, with considerable differences in the relative intensity values of the emitters across the spectrum. This initial characterization identified Alexa 647 and CF 680 as the fluorophore pair with the largest change in intensity (ΔI) measured in percentage terms as:
C. Theoretical vs. experimental emission spectra
To further validate the spectral sensitivity and performance of the TTF, two colour imaging with the far-red emitting fluorophores: Alexa 647 and CF 680, was performed. Using the TTF to select spectral windows with center wavelengths of 670 nm (CW 670) and 700 nm (CW 700) respectively, isolated the theoretical emission maxima of each fluorophore (Figure 3(a)). However, in order to obtain more relevant spectral data, the emission of the same five fluorophores was measured experimentally using a side-port mounted spectrometer. The experimental spectra showed a significant deviation from the theoretical emission curves (Figure 3(b)). Consistent with intensity data shown in (Figure 2(b)), the experimental spectra revealed the variation in efficiency of each fluorophore when excited at 643 nm. Alexa 647 yielded the strongest emission in the CW 670 window, while CF 680 yielded the strongest emission in the CW 700 window. The left side of Table 1 summarizes the maximum intensity ratios of Alexa 647 : CF 680 and Alexa 647 : Alexa 680 (and vice versa) in both spectral windows. Via diffraction-limited TIRF imaging, the ratio of Alexa 647 : Alexa 680 vs Alexa 647 : CF 680 was 5.15 vs 2.12. However at CW 700, the reciprocal ratios were 0.87 vs 1.89. These measurements showed that the efficiency of Alexa 647 was such that it produced a signal of higher intensity than Alexa 680, even in the CW 700 region. This behaviour raises a potential issue for intensity-based spectral assignment. CF 680 however, yielded a stronger signal than Alexa 647 at CW 700, with a CF 680 : Alexa 647 intensity ratio of 1.89. The spectral characteristics of these fluorophores was evaluated using the TTF under dSTORM blinking conditions (Figure 3(c,d)). To this end, photoswitching of individual samples of antibody-conjugated fluorophore was induced as previously reported.13 Compared to diffraction-limited imaging, the changes in the intensity ratios for CW 670 (Alexa 647 : Alexa 680, Alexa 647 : CF680) increased by ~ 3 and ~2, respectively. However, at CW 700, the Alexa 680 : Alexa 647 ratio decreased to 0.50, while the CF 680 : Alexa 647 ratio decreased slightly to 1.70 (Table 1).. Taken together, these data and that shown in Figure 2(b) suggested that Alexa 647 and CF 680 represent the most promising dye pair for spectral discrimination using the TTF at CW 670 and CW 700.
D. SMLM intensity and localization frequency across spectral windows
To characterize the intensity distributions of individual Alexa 647 and CF 680 molecules, we performed a series of dSTORM SMLM acquisitions across a range of spectral windows. The TTF angle was adjusted to collect data from 20 nm spectral bands corresponding to CW 670, CW 680, CW 690 and CW 700. Distinct variations in mean intensity and localization frequency in the histogram distribution were observed (Table 2, Figure S1). From CW 670 to CW 700, Alexa 647 displayed a mean intensity (Iavg) decrease of 70% and localization frequency (LF) decrease of 42%. From CW 700 through to CW 670, CF 680 displayed an Iavg decrease of 93% and an LF decrease of 98%. The data revealed that the ΔIavg and ΔLF between Alexa 647 and CF 680 was greater at CW 670 than at CW 700 (Figure 4 (a,b)), suggesting considerable Alexa 647 crosstalk at CW 700.
III. RESULTS
A. SMLM imaging of control samples, thresholding and crosstalk determination
To assess how localization is affected by intensity, super-resolution images acquired for both fluorophores were analyzed at CW 670 and CW 700 over a range of photon count thresholds (Figure 5 (a,b)). As expected, a higher frequency of localizations were found for Alexa 647 at CW 670 than at CW 700. As the photon output threshold (POT) was increased from 1000 to 3000, the number of localizations decreased accordingly in both spectral windows (Figure 5 (c)). Interestingly, the opposite behaviour was observed for CF 680. The localization frequency for this fluorophore was significantly lower at CW 670 than it was at CW 700. (Table 3) However, the frequency decreased as the POT was increased, as expected. The initial frequency of localizations for CF 680 at CW 670 was very low in comparison to the other images, reflecting the extent of the attenuation of this fluorophore’s dSTORM photon output in this spectral region. The fraction of data recovered at each threshold for both fluorophores in both spectral windows was also quantified (Figure 5(d)). For a given center wavelength, the fraction recovered was calculated by dividing the localization frequency at a given threshold by the original localization frequency for the fluorophore undergoing maximal emission in that spectral window, as follows:
For a series of thresholds i,j,k, and original localization frequency LFo
Alexa 647 and CF 680 both exhibited a similar rate of decrease in the fraction of localizations recovered as the POT was increased, as per the following:
Crosstalk values at CW 670 ranged from 0.02% to 0.03% at the three photon output thresholds, again reflecting the small number of CF 680 localizations that were recovered. At CW 700, crosstalk values ranged from 7% to 11% (Figure 5(e)). Plotting crosstalk vs. fraction recovered for each spectral window at a series of thresholds provided a measure of comparison between two opposing goals: Maximizing the localization frequency of the desired fluorophore while minimizing crosstalk (Figure 5 (f)). Interestingly, at CW 700, a higher recovered fraction of CF 680 (i.e. a lower photon output threshold) was largely associated with a lower crosstalk value. While this may appear counter-intuitive, it reflects the fact that, as the POT is increased, the ratio of localized Alexa 647 molecules to localized CF 680 molecules approaches 1:1. Since Alexa 647 is a more efficient photon emitter, its contribution to crosstalk increases with increasing POT.
B. 2-channel hyperspectral SMLM imaging of mitochondria and peroxisomes
To demonstrate the utility of the approach described herein, hyperspectral super-resolution imaging of the mitochondrial protein TOM20, labeled with Alexa 647 (TOM-Alexa 647) and the peroxisomal protein, PMP70, labeled with CF-680 (PMP70-CF 680) in HeLa cells. Images were reconstructed at CW 670 and CW 700. At CW 670, there was virtually no visible crosstalk, as seen in the in-vitro control and single-labeled samples (Figure S2). At CW 700, significant crosstalk was apparent. The difference in morphology between peroxisomal puncta (yellow arrows) and residual mitochondrial signal (purple boxes) served to illustrate the extent of this phenomenon (Figure 6(a)). When a POT of 300 was applied, close examination of the CW 700 image revealed that crosstalk was not completely eliminated. Increasing the threshold further may attenuate crosstalk to a greater degree, but at the expense of the true CF 680 signal. Continuously increasing the threshold with the goal of attenuating all crosstalk presents a dilemma wherein the reduction of false positives to zero is achieved only by increasing false negatives to an unacceptable level. The result is an image that lacks adequate structural context. In general, after thresholding, the Alexa 647 signal at CW 700 was comprised of sparsely situated localizations that could be only be accounted for with a density filter Figure 6(b)). The parameters of the filter used in this case were a minimum # neighbors = 5, radius = 100 nm. Under these conditions, all localizations which did not have at least 5 adjacent localizations within a 100 nm radius were removed. This served to eliminate the majority of crosstalk in the image. The CW 670 featuring TOM20-Alexa 647 and CW 700 PMP70-CF 680 images were then merged to create a two-colour hyperspectral SMLM image (Figure 6(c)).
IV. CONCLUSION
We have demonstrated the usefulness of a tilt-tunable filter for performing hyperspectral SMLM. This approach allowed for dSTORM SMLM of Alexa 647 and CF 680 using a single excitation source. Photon output thresholding and density filtering were used to reduce spectral cross-talk between these two fluorophores. Building from this platform, there are a number of directions to be pursued. Since biological structures are inherently heterogeneous, the absolute concentration of fluorophores in any given field of view cannot be easily determined. Therefore, although similar results were observed for multiple replicates of the data presented, a more robust statistical treatment of the data may be required to standardize photon output thresholds for specific fluorophore-protein combinations. In addition, fluorophore attribution would be facilitated by the presence of paired localizations in the two spectral windows. Simultaneous acquisition of both signals in distinct spectral windows (for example with a fixed beamsplitter and 2 EMCCD cameras or a split-view configuration) would allow for ratiometric determination of fluorophore identity.14 This would require the TTF to transition between CW 670 and CW 700 at a rate double that of the single localization with the shortest duty cycle in any given frame. Further experiments with optimized galvanometer stability are required to explore this possibility. In addition, modulating both the position of the spectral window in the emission region as well as its bandwidth, by using a combination of long-pass and short-pass tunable filters offers the possibility of achieving enhanced better spectral resolution between two fluorophores.