Open Access
Add to BibSonomy
Abstract
Single molecule localization microscopy (SMLM) usually requires long image acquisition time at the order of minutes and thus suffers from sample drift, which deteriorates image quality. A drift estimation method with high precision is typically used in SMLM, which can be further combined with a drift compensation device to enable active microscope stabilization. Among all the reported methods, the drift estimation method based on bright-field image correlation requires no extra sample preparation or complicated modification to the imaging setup. However, the performance of this method is limited by the contrast of bright-field images, especially for the structures without sufficient features. In this paper, we proposed to use differential phase contrast (DPC) microscopy to enhance the image contrast and presented a 3D drift correction method with higher precision and robustness. This DPC-based drift correction method is suitable even for biological samples without clear morphological features. We demonstrated that this method can achieve a correction precision of < 6 nm in both the lateral direction and axial direction. Using SMLM imaging of microtubules, we verified that this method provides a comparable drift estimation performance as redundant cross-correlation.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Single molecule localization microscopy (SMLM) is a major type of super-resolution microscopy that includes stochastic optical reconstruction microscopy (STORM) [1], photoactivation localization microscopy (PALM) [2] and other variants [3,4]. To generate a super-resolution image with 10-20 nm resolution [5], SMLM needs tens of thousands of raw image frames, which typically require several minutes or even longer to acquire. During this minute-scale period, sample drift may be accumulated to be several hundreds of nanometers [6], resulting in an obvious resolution reduction in final super-resolution images. Therefore, a drift correction method with high precision is essential for SMLM.
In a drift correction method, drift estimation and drift compensation are two basic steps [7]. If we can precisely estimate sample drift during the imaging process of an SMLM experiment, we can easily use a nanopositioning device to actively compensate the drift, and thus enable active microscope stabilization [8]. In the last decade, a number of methods have been developed to estimate sample drift. The most commonly used method is based on tracking the positions of reference markers attached to the surface of coverslips or dishes, including randomly placed gold nanoparticles [9] and fluorescent beads [10], and fiducial micro-pattern [11]. The localization precision of these reference markers can be estimated to be 1 nm in lateral direction and 1-10 nm in axial direction [12], ensuring excellent performance in drift correction. However, these methods need extra sample preparation, and random distribution of markers makes it difficult to achieve a proper concentration in the region of interest.
Another popular method for drift estimation introduces a near infrared (NIR) LED or laser light into the imaging system. The drift in axial direction is monitored by detecting the total internal reflection of the NIR light at the interface between the coverslip and the embedding medium [13]. No extra reference markers are required in this method, but it cannot estimate lateral drift. And it requires additional efforts for realizing and maintaining total internal reflection of NIR light.
Morphological features of biological samples can also be used to estimate sample drift. For example, Fan et al. utilized naturally abundant circular structures inside fixed cells as a reference to estimate 3D drift [14], but these circular structures are not always accessible. Chen et al. quantified sample drift by correlation analysis of speckle patterns formed by backscattered laser light from samples [15]. In this method, emission filter needs to be removed to acquire speckle pattern, which brings inconvenience to imaging process. Wang et al. calculated sample drift by applying a redundant cross-correlation (RCC) algorithm to super-resolution images reconstructed from subsets of localizations [16]. But this method has not been used for active drift correction or z drift correction. Mcgorty et al calculated sample drift by the correlation of bright-field (BF) images [6]. This method requires no extra efforts in sample preparation. And its optical setup can be achieved by a small hardware modification to existing SMLM system: simply replacing the halogen lamp with a LED. However, this method has higher values of drift correction precision, which is typically 10 nm in lateral direction and 20 nm in axial direction, than other methods. Its performance depends on the contrast of BF images, while sufficient features of samples for presenting high-contrast BF images are not always accessible, especially in the scenarios with small field of view. Therefore, an active drift correction method with high precision, universal application, and easy implementation is in need.
In this paper, we introduced a drift correction method based on correlation of differential phase contrast (DPC) images (hereafter called DPC-correlation method). By replacing the halogen lamp with a LED pair, we can obtain high-contrast DPC images, analyze their correlation, and estimate the 3D drift for correction. We obtained an active drift correction precision of < 6 nm in both lateral direction and axial direction with samples marked with fluorescent beads. And this precision can be guaranteed for different structures, reflecting the universal application of our method. We verified that our method is more robust than the method based on the correlation of BF images (hereafter called BF-correlation method) when applied to different structures, and presents comparable performance with the popular redundant cross-correlation method (hereafter called RCC method) with SMLM experiments.
2. Methods
2.1. Experimental setup
The experimental setup was built from an Olympus IX 73 inverted microscope (Fig. 1). A 405 nm laser (LWVL405-200mW, Laserwave, China) and a 640 nm laser (LWRL640-3W, Laserwave, China) were combined with a customized fiber combiner [17] and guided into an objective (PLAPON 60XO, Olympus, Japan) for activation and imaging. The sample was placed on an automatic XYZ piezo stage with a closed-loop resolution of less than 1 nm (P-545.3C7, Physik Instrumente, Germany) and a manual XY stage (XWJ-50R-2, Sanying, China) for active drift compensation and sample movement, respectively. The fluorescence emission collected by the objective was separated from the lasers with a dichroic mirror (ZT488/532/633/830/1064rpc, Chroma, USA), passed through emission filters (FF01-680/42, NF03-405/488/532/635E, Semrock, USA), and finally detected with an sCMOS camera (Flash 4.0 V3, Hamamatsu, Japan).
Fig. 1. Schematic of an SMLM system equipped with components for DPC-based drift correction. ACL: aspheric condenser lens, M: mirror, C: condenser, XYZ: XYZ piezo stage, XY: XY stage, OBJ: objective, DM: dichroic mirror, TL: tube lens, EFs: emission filters.
DPC enhances the image contrast of transparent biological structures by converting invisible sample phase information into detectable image intensity through asymmetric illumination in complementary angles [18]. To equip our SMLM system with DPC imaging capacity, the halogen lamp of the microscope, which is commonly used for BF imaging, was replaced with a simplified programmable LED pair. This LED pair contains two identical LEDs (660 nm, 800 mW, 3030-R, Xincheng Photoelectric, China) symmetrically positioned on a disk (Fig. 1). The light from the LEDs was collimated with an aspheric condenser lens (f = 60 mm, ACL7560U, Thorlabs, USA) and focused with a microscope condenser onto the sample. The lateral position of the condenser was adjusted properly to ensure complementary illumination angles of two LEDs. The wavelength of the LEDs was chosen to match the fluorescence emission wavelength of the fluorescent beads and the fluorophores used in our SMLM experiment, so that BF/DPC images can be captured with the same SMLM setup. Each LED can be controlled independently. We used a custom-written Micro-Manager plugin to control sCMOS detector, LED pair, and XYZ piezo stage for image acquisition, drift estimation and compensation.
A DPC image can be obtained with the following equation:
where II and III denote BF images illuminated with LED1 and LED2, respectively. For calculating a DPC image, the LED pair was controlled to illuminate the sample with LED1 or LED2 alternately, so that III can be recorded right after II. In this work, II and III were called as a pair of BF images. The distance between the two LEDs was set to be ∼ 4 mm to increase the illumination NA [19], which was also limited by the diameter of condenser. DPC images correspond to the phase gradient along the direction of asymmetry of illumination [18], so the symmetry axis of the LEDs was rotated to be ∼ 45 degrees from the horizontal direction. So that the image contrast along both x and y directions can be enhanced simultaneously with a single-axis DPC imaging. Single-axis DPC imaging requires only two BF images to create a DPC image, and thus the time for drift estimation can be minimized.2.2. Theory for drift estimation and correction
The sample drift was estimated by analyzing the correlation of DPC images. Here, the normalized cross-correlation (NCC) of two DPC images was calculated with the OpenCV function matchTemplete as
The estimation of z drift was slightly complicated. First, we studied the relationship of likeness of DPC images on their relative z distance. As shown in Fig. 2(a), the likeness of two DPC images was quantified with normalized correlation peak value, and it decreases when relative z distance of these two images increases. However, one correlation is not enough to determine the direction of relative z distance (positive or negative z). Therefore, we use three reference images (Rf: reference image at the focal plane; Rp: reference image above the focal plane; Rn: reference image below the focal plane) to estimate z drift. Three correlation peak values calculated from the reference images and an experimental image Ii were obtained (Fig. 2(b)) and the variable ζ can be calculated with
which shows an almost linear relationship for small z distances (Fig. 2(c)). This finding is similar with that from BF-correlation method [6].Fig. 2. Schematic diagram for estimating the z drift with DPC image correlation. (a) Peak values of normalized cross-correlation (NCC) of two DPC images as a function of their relative z distance. (b) Correlation of images for estimating z drift. (c) ζ as a function of the z position (red dots) and the linear fitting results (blue dotted line).
Due to the effect of noise, the cross-correlation of two images recorded at the same z plane is slightly smaller than the self-correlation of one image, and this would affect the estimation of slope k. Therefore, we acquired and used two additional images (hereafter called slope-fitting images) to obtain the slope of this linear relationship. These two images were captured above and below the focal plane (denoted as Iu, Id, respectively). And the slope k can be obtained with
where zu and zd are the relative z distances of the two slope-fitting images from the focal plane. Then, the z drift of image In, recorded at given time point tn, can be estimated with where ζ0 corresponds to the image recorded at the start time of experiment (t0). Note that, the XY drift are estimated by tracking the peak position of the correlation between experimental image In and reference image Rf. Hence, five images, including three reference images and two slope-fitting images, should be acquired before starting an SMLM experiment. In our experiments, Rp and Iu were both recorded at 300 nm above the focal plane, Rn and Id were recorded at 300 nm below the focal plane, and Rf was recorded at the focal plane.To avoid overestimation of drift and the consequent sample oscillation, the estimated drift was multiplied with a constant, which was set to be 0.9 empirically in our experiments. Then it was fed back to an automatic XYZ piezo stage to compensate the sample drift actively.
2.3. Sample preparation and imaging
COS-7 Cells were cultured and seeded into 35-mm dishes as described in our previous work [17]. For quantifying the precision of active drift correction, fluorescent beads (F8807, FluoSpheres, Molecular Probes, USA) with peak emission wavelength of about 680 nm were fixed on the dish before cells were seeded. The microtubules were labelled with Alexa Fluor 647 using a typical immunofluorescence method for SMLM imaging. In experiments not requiring blinking or reactivation of fluorescent probes, samples were soaked with PBS. For SMLM imaging, samples were soaked with standard STORM buffer (50 mM Tris, pH 8.0, 10mM NaCl, 10% glucose, 100mM mercaptoethylamine, 500 μg/mL glucose oxidase, 40 μg/mL catalase). Before imaging, samples were washed several times with PBS to remove suspended impurity. Because the distribution of fluorescent bead is sparse enough and the fluorescence intensity of molecules is too weak compared with the intensity of LEDs, BF images were taken without switching off the 640 nm laser to simplify the hardware control.
A cylindrical lens (f = 1000 mm, LJ1516L1-B, Thorlabs, USA) was inserted into the detection path to introduce astigmatism for 3D localization of fluorescent beads and 3D SMLM imaging [20]. In SMLM imaging, the illumination intensity of 640 nm laser was ∼ 8 kW/cm2, and the intensity of 405 nm laser was controlled manually to maintain a proper activation density. The exposure time for both fluorescence images and BF images was 10 ms. For comparison purpose, two sequential experiments were carried out using the BF-correlation or DPC-correlation methods for active drift correction, and the same biological structure was selected to estimate drift. DPC images were calculated with a pair of BF images as described in section 2.1, while BF images used for correlation analysis were taken with LED1 illumination. The fluorescence images were processed with a customized algorithm called QC-STORM [21].
2.4. Contrast analysis for images
To test the robustness of our method, the performance of drift correction using DPC or BF images containing different biological structures was compared. Note that biological structures have a significant impact on the image contrast. For the same imaging mode, structures with sufficient features present higher image contrast than those without sufficient features. In experiments, high/low contrast regions were selected empirically based on BF images. To verify our determination of image contrast of different regions, we analyzed the power spectra of selected regions as follows. First, we calculated two-dimensional power spectrum of an image by taking a square of the centered frequency-domain image, and normalized the power spectrum by dividing the square of the average gray level and the total number of pixels to eliminate the effect from image-to-image brightness variation. Then one-dimensional power-spectrum was obtained by calculating an average power amplitude at a given radius from the zero-frequency center [22,23].
3. Results
3.1. Time cost in active drift correction
The sCMOS camera in our setup was used for recording fluorescence images and BF images. So, SMLM imaging and active drift correction were performed alternatively. To minimize the time cost for active drift correction, we selected a region of 90 × 90 pixels from BF and DPC images for drift estimation. This small region (∼ 9 × 9 μm) can also avoid the influence of fixed pattern background arisen from the uneven illumination of LEDs. And we demonstrated that it took 60 ms (BF) and 68 ms (DPC) to estimate the 3D drift. For active drift correction in the SMLM experiments, the entire drift correction process (including drift estimation and compensation) was finished in 183 ms for BF images, and 239 ms for DPC images, respectively.
3.2. Power spectra of images
A total of 200 BF and DPC images of samples at focal plane were captured. Six 90 × 90 regions with high/low contrast were selected empirically according to BF images (Fig. 3(a-f)) and the power spectra were analyzed. As shown in Fig. 3(g), the power spectra of DPC images are much higher than that of BF images. Besides, for the same imaging mode, the power at low frequency (∼ 0.05 to 0.2, as highlighted in Fig. 3(g)) of high-contrast regions is significantly greater than that of low-contrast regions. The average power (AP) at a frequency range from 0.05 to 0.2 was further calculated as marked in Fig. 3(a-f). These findings indicate that the AP at low frequency is possible to quantify the image contrast. And in this work, the AP of BF or DPC images used for drift correction was calculated to verify our determination of image contrast.
Fig. 3. The power spectra of BF or DPC images containing different structures. (a-c) The selected high-contrast (HC) regions from BF images (red box) or DPC images (black box). (d-f) The selected low-contrast (LC) regions from BF images (magenta box) or DPC images (blue box). (g) The power spectra of the selected BF or DPC regions. Each spectrum was averaged from 200 images. The average power (AP) shown in (a-f) is calculated under frequency range from 0.05 to 0.2, as highlighted in (g). Scale bar: (a-f) 2 μm.
3.3. Robustness of z position estimation
A series of images recorded over a number of z planes were used for investigating the robustness of DPC-correlation or BF-correlation methods in z position estimation. The XYZ piezo stage was moved in a step of 30 nm to scan a Z-axis range of ± 300 nm. At each z plane, ten pairs of BF images were recorded, and each pair can be used to present a DPC image. The slope k and ζf corresponding to the focal plane were calculated with images recorded at +300 nm and -300 nm according to Eq. (6). Then the relative z position of each image was estimated according to Eq. (7) and compared with the pre-determined z positions.
We demonstrated the performance of z position estimation when regions with different contrast were used. As shown in Fig. 4, when a high-contrast region was used, a satisfactory agreement between the estimated z positions and the z-stage positions (pre-determined z positions) can be obtained for both BF and DPC images. However, for the BF-correlation method, the estimated z positions from low-contrast BF images were significantly deviated from the z-stage positions, and the estimated errors calculated with images from one same z plane are much larger. In contrast, the DPC-correlation method presented a better and desirable result in estimating z positions with low-contrast region. Therefore, the DPC-correlation method has better robustness in z position estimation than the BF-correlation method, indicating that the former is a better choice in drift estimation.
Fig. 4. The robustness of z position estimation using BF and DPC images of COS-7 cells. (a) A BF image of COS-7 cells recorded at focal plane. (b-e) Zoomed images for the high-contrast (magenta box) and low-contrast (green box) regions. (f-g) Dependence of the z-stage positions and the estimated z positions using high-contrast (f) or low-contrast regions (g). Scale bar: (a) 10 μm, (b-e) 2 μm.
3.4. Precision of active drift correction
To quantify the performance of active drift correction using DPC-correlation or BF-correlation methods, 3D localization imaging of fluorescent beads was carried out. A total of 3,000 fluorescence images were recorded at 10 frames per second (fps) with ∼ 90 ms pause between two frames. One DPC or BF image was recorded per 10 fluorescent frames for 3D active drift correction. The intensity of 640 nm laser was adjusted properly to enhance the signal of fluorescent beads for high-precision localization. The standard deviation of the bead positions from 3,000 fluorescent frames was calculated to quantify the precision of active drift correction (Table 1).
Table 1. The precision of active drift correction characterized by the standard deviation of the repeated localizations of a fluorescent bead
When high-contrast region was used, the performance of active drift correction using either the BF-correlation method or the DPC-correlation method was similar, and the sample was well stabilized to its initial positions (which indicates high accuracy of drift estimation, see Fig. 5(g-i)). Furthermore, comparable precisions of drift correction can be obtained with both methods, and a precision of < 6 nm in all three directions can be guaranteed. On the other hand, using low-contrast region, the drift was still well corrected by the DPC-correlation method, indicating by stabilized bead positions close to initial position and a comparable precision in all three directions. However, when the BF-correlation method was applied to low-contrast region, the sample cannot be well stabilized in all three directions during the imaging process, meaning that the drift cannot be accurately recognized and corrected. Although the precision of 4.3 nm in X direction is similar with DPC-correlation method, there is obvious fluctuation in bead position. And the precisions of 8.1 nm in Y direction and 20.1 nm in Z direction are much worse than DPC-correlation method.
Fig. 5. The precision in active drift correction quantified with 3D localization of fluorescent beads. (a-c) A BF image of COS-7 cells, and zoomed images for the high-contrast region (magenta box in a). (d-f) A BF image of COS-7 cells, and zoomed images for the low-contrast region (green box in d). (g-i) The relative XYZ positions of a fluorescent bead under active drift correction from high-contrast region. (j-l) The relative XYZ positions of fluorescent bead under active drift correction from low-contrast region. All the bead positions were moved to initial positions determined by the average values of bead’s position from the first 10 frames. For one experiment, three beads were selected and comparable results can be obtained. Scale bar: (a, d) 10 μm, (b-c, e-f) 2 μm.
3.5. Performance of posterior XY drift estimation in SMLM
We evaluated the performance of the BF-correlation method and the DPC-correlation method in estimating posterior XY drift from SMLM experiments. A series of 10,000 raw fluorescence images of microtubules were recorded at about 100 fps to generate a final SMLM image, and a pair of BF images were recorded per 100 fluorescence images for posterior XY drift correction. For comparison, 3D active drift correction was not applied, and an NIR Focus module was used to stabilize the focus during the imaging process. The RCC method was selected as a standard post-processing method for comparison. Note that the estimated drift values were smoothed firstly using a Hampel identifier and a smooth filter to remove outliers and reduce noise. Later, the drift in each fluorescence image was obtain by interpolation and corrected by changing the molecule localizations. The full width at half maximum (FWHM) of microtubules was selected as a metric to determine the image resolution.
As shown in Fig. 6(a-e), high-contrast (Fig. 6(b-c)) and low-contrast regions (Fig. 6(d-e)) were selected and used to estimate the XY drift. The FWHM of the microtubule was found to be largest when there was no drift correction (Fig. 6(f)), and smallest when RCC was applied (Fig. 6(g)). The DPC-correlation method presents comparable results with the RCC method (Fig. 6(i, k)). For the BF-correlation method, the FWHM resolution is still acceptable when high-contrast region was used (Fig. 6(h)), but is significantly reduced when low-contrast region was used (Fig. 6(j)). Therefore, it is clear that the DPC-correlation method provides robust XY drift estimation, which is comparable to the popular RCC method and is applied to different structures. We note that the performance of the RCC method depends on the number of localizations and a reasonable number (200–400) of localization points should be collected to guarantee a reliable drift estimation [16], thus limiting the minimal time step size for drift estimation. Moreover, RCC method has not been used for active drift correction and estimation of z drift. On the contrary, our DPC-correlation method has no requirement on the time step size, and could be used to perform 3D active drift correction.
Fig. 6. Comparison on the posterior XY drift correction among the BF-correlation, DPC-correlation and RCC methods. (a) A BF image of COS-7 cells. (b-c) Zoomed high-contrast region for the magenta box in (a). (d-e) Zoomed low-contrast region for the green box in (a). (f-k) Super-resolution images (upper) and zoomed regions (middle) and the corresponding FWHM resolution (bottom) from the same microtubule structure (marked with the yellow boxes in the middle panels). Different XY drift correction methods were used in post-processing of 10,000 raw images: no drift correction (f), RCC method (g), BF-correlation with high-contrast region (h), DPC-correlation with high-contrast region (i), BF-correlation with low-contrast region (j), and DPC-correlation with low-contrast region (k). Scale bar: (a) 10 μm, (b-e) 2 μm, (f-k) 10 μm in the upper panel and 1 μm in the middle panel.
3.6. Performance in active drift correction
The DPC-correlation method was used for active drift correction in SMLM imaging, and its performance was quantified and compared with BF-correlation method. For comparison, in each field of view (FOV), we imaged the same structure sequentially using the DPC-correlation method or BF-correlation method. In each SMLM experiment, a total of 10,000 frames for 2D imaging and 15,000 frames for 3D imaging of fluorescence images were recorded at about 100 fps and 3D drift correction was executed per 100 fluorescence frames. The intensity of fluorescence signals in the two experiments was roughly the same by adjusting the intensity of the 640 nm laser. During the post-processing, the localization tables from the two experiments were filtered randomly to maintain an equal number of molecules. The imaging results from 2D imaging were also post-processed using the RCC method for comparison. The FWHM of microtubules was analyzed for evaluating the image resolution.
We firstly used high-contrast region in the drift correction. We used a same number of localizations (3,115,270 points) to reconstruct super-resolution images for the two methods. The mean values of signal-noise-ratio (SNR) in the raw fluorescence images were 11.4 for the BF-correlation method and 11.6 for the DPC-correlation method, respectively. The BF and super-resolution images of the same sample are shown in Fig. 7(a, d). The FWHM resolution of the same microtubule structure was determined to be 57.5 nm for both the BF-correlation (Fig. 7(e)) and the DPC-correlation method (Fig. 7(g)). Interestingly, applying RCC correction to these super-resolution results cannot improve the FWHM resolution, no matter for the BF-correlation (Fig. 7(f)) or the DPC correlation (Fig. 7(h)) methods. These findings verified that, with high-contrast region, both the BF-correlation method and the DPC-correlation method present a satisfactory performance in active drift correction, where the FWHM resolution cannot be further improved by the RCC method.
Fig. 7. The performance of 3D active drift correction using high-contrast region. (a) A BF image of COS-7 cells. (b-c) Zoomed images of the high-contrast region marked with the red box in (a). (d) A super-resolution image of microtubules. (e-h) Zoomed super-resolution images (marked with the green box in (d)) from different drift correction strategies. The corresponding FWHM resolution of the microtubules marked with the yellow boxes are shown below. Scale bar: (a, d) 10 μm, (b-c) 2 μm, (e-h) 1 μm.
The results of active drift correction using low-contrast region are shown in Fig. 8. We used a same number of localizations (1,857,083 points) to reconstruct super-resolution images for the two methods. The mean SNR values in the raw fluorescence images were 11.6 for the BF-correlation method and the DPC-correlation method. The FWHM resolution of the microtubule was determined to be 65.2 nm when the BF-correlation method was used (Fig. 8(e)), and improved to be 57.4 nm after post-processing with the RCC method (Fig. 8(f)). When the DPC-correlation method was applied in the active drift correction, the FWHM resolution of the microtubule was calculated to be 50.4 nm (Fig. 8(g)), which is significantly higher than the resolution from the BF-correlation method. And there was no improvement in FWHM resolution (50.6 nm, Fig. 8(h)) after post-processing with the RCC method. These findings verified again that the DPC-correlation method can ensure a drift correction performance comparable to or even better than that from the RCC method. Since the performance of the DPC-correlation method is robust for regions with different contrast, this method should be applicative to a wide range of application scenarios.
Fig. 8. The performance of 3D active drift correction using low-contrast region. (a) A BF image of COS-7 cells. (b-c) Zoomed images of the low-contrast region marked with the red box in (a). (d) A super-resolution image of microtubules. (e-h) Zoomed super-resolution images (marked with the green box in (d)) from different drift correction strategies. The corresponding FWHM resolution of the microtubules marked with the yellow boxes are shown below. Scale bar: (a, d) 10 μm, (b-c) 2 μm, (e-h) 1 μm.
Finally, we carried out 3D SMLM of microtubules with 3D active drift correction using BF and DPC images of low-contrast region. A total of 831,089 localizations were used to reconstruct super-resolution images. The mean SNR values in the raw fluorescence images were 8.4 for the BF-correlation method and 8.8 for the DPC-correlation method, respectively. The results are shown in Fig. 9. When the BF-correlation method was used, the FWHM resolution of the microtubule was determined to be 63.3 nm for the lateral direction (Fig. 9(f)) and 132.1 nm for the axial direction (Fig. 9(g)), respectively. Using the DPC-correlation method, the FWHM resolution of the microtubule was calculated to be 54.2 nm and 80.9 nm for the lateral and axial directions (Fig. 9(h-i)), respectively. These results confirmed that the DPC-correlation method outperforms the BF-correlation method in active drift correction of 3D SMLM.
Fig. 9. The performance of 3D active drift correction using low-contrast region in 3D SMLM. (a) A BF image of COS-7 cells. (b-c) Zoomed BF/DPC images of the low-contrast region marked with the red box in (a). (d-e) 3D super-resolution images of microtubules under the two drift correction methods. (f, h) Zoomed x-y view of the super-resolution images marked with the magenta boxes in (d-e), and the corresponding FWHM resolution (shown in the bottom row) in the lateral direction calculated from the regions marked with the cyan boxes. (g, i) Zoomed y-z view of the super-resolution images marked with the red boxes in (f, h), and the corresponding FWHM resolution in the axial direction (shown in the bottom row). Scale bar: (a, d-e) 10 μm, (b-c) 2 μm, (f, h) 1 μm, (g, i) 200 nm.
4. Conclusion
We present a high-precision 3D drift correction method for SMLM. In this method, we estimate sample drift by the correlation from DPC images rather than BF images, and thus significantly improves the drift estimation precision and robustness. We demonstrated that this new method is capable of providing an active drift correction precision of < 6 nm in both lateral and axial directions, which is comparable with other high-precision methods and should be sufficient for typical SMLM. And this new method preserves the advantages of BF-correlation method, for example: 1) No extra requirements in the sample preparation, 2) Easy modification to the SMLM system, and 3) Active drift correction in 3D without cumbersome imaging process. We verified that the new method reduces the dependency of drift correction precision on sample structures, making the method suitable for a wider range of application scenarios, especially for that cannot provide high-contrast BF images.
Here, 660 nm LEDs were selected as the illumination light to match the emission wavelength of the fluorescent beads and the probes used in our experiments. For other fluorescent probes or even multi-color imaging, we can switch to LEDs with other wavelength or use white light LEDs to capture DPC images with the same setup as the fluorescence images. Besides, in this paper, we simplify the experiment setup by using only one sCMOS camera to record both BF images and fluorescence images, which may reduce the maximum frame rate in capturing fluorescence or BF images and thus the frequency for drift correction. To enable drift correction with higher frequency, we can use a separate path for drift correction. In this case, the LED wavelength should not overlap with the emission wavelengths of fluorescent probes used in SMLM experiments. Nevertheless, we believe the proposed drift correction method using correlation of differential phase contrast images is helpful to reduce the complexity of SMLM experiments, and thus would be easily used in many biological labs.
Funding
Hainan University (KYQD(ZR)-20077); China Postdoctoral Science Foundation (2020M682418); Fundamental Research Funds for the Central Universities (2018KFYXKJC039); National Natural Science Foundation of China (81827901).
Acknowledgments
We thank the Optical Bioimaging Core Facility of WNLO-HUST for technical support.
Disclosures
Pending Chinese patent (202110703134.X).
Data availability
Data underlying the results in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. M. J. Rust, M. Bates, and X. Zhuang, “Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM),” Nat. Methods 3(10), 793–796 (2006). [CrossRef]
2. E. Betzig, G. H. Patterson, R. Sougrat, O. W. Lindwasser, S. Olenych, J. S. Bonifacino, M. W. Davidson, J. Lippincott-Schwartz, and H. F. Hess, “Imaging intracellular fluorescent proteins at nanometer resolution,” Science 313(5793), 1642–1645 (2006). [CrossRef]
3. S. T. Hess, T. P. Girirajan, and M. D. Mason, “Ultra-high resolution imaging by fluorescence photoactivation localization microscopy,” Biophys. J. 91(11), 4258–4272 (2006). [CrossRef]
4. S. van de Linde, A. Loschberger, T. Klein, M. Heidbreder, S. Wolter, M. Heilemann, and M. Sauer, “Direct stochastic optical reconstruction microscopy with standard fluorescent probes,” Nat. Protoc. 6(7), 991–1009 (2011). [CrossRef]
5. F. Huang, G. Sirinakis, E. S. Allgeyer, L. K. Schroeder, W. C. Duim, E. B. Kromann, T. Phan, F. E. Rivera-Molina, J. R. Myers, I. Irnov, M. Lessard, Y. Zhang, M. A. Handel, C. Jacobs-Wagner, C. P. Lusk, J. E. Rothman, D. Toomre, M. J. Booth, and J. Bewersdorf, “Ultra-High Resolution 3D Imaging of Whole Cells,” Cell 166(4), 1028–1040 (2016). [CrossRef]
6. R. Mcgorty, D. Kamiyama, and B. Huang, “Active microscope stabilization in three dimensions using image correlation,” Opt Nano 2(1), 3 (2013). [CrossRef]
7. W. Colomb, J. Czerski, J. D. Sau, and S. K. Sarkar, “Estimation of microscope drift using fluorescent nanodiamonds as fiducial markers,” J. Microsc. 266(3), 298–306 (2017). [CrossRef]
8. R. Tafteh, L. Abraham, D. Seo, H. Y. Lu, M. R. Gold, and K. C. Chou, “Real-time 3D stabilization of a super-resolution microscope using an electrically tunable lens,” Opt. Express 24(20), 22959–22970 (2016). [CrossRef]
9. P. Bon, N. Bourg, S. Lecart, S. Monneret, E. Fort, J. Wenger, and S. Leveque-Fort, “Three-dimensional nanometre localization of nanoparticles to enhance super-resolution microscopy,” Nat. Commun 6(1), 7764 (2015). [CrossRef]
10. M. Bates, B. Huang, G. T. Dempsey, and X. Zhuang, “Multicolor super-resolution imaging with photo-switchable fluorescent probes,” Science 317(5845), 1749–1753 (2007). [CrossRef]
11. Y. Youn, Y. Ishitsuka, C. Jin, and P. R. Selvin, “Thermal nanoimprint lithography for drift correction in super-resolution fluorescence microscopy,” Opt. Express 26(2), 1670–1680 (2018). [CrossRef]
12. S. H. Lee, M. Baday, M. Tjioe, P. D. Simonson, R. Zhang, E. Cai, and P. R. Selvin, “Using fixed fiduciary markers for stage drift correction,” Opt. Express 20(11), 12177–12183 (2012). [CrossRef]
13. S. A. Jones, S. H. Shim, H. Jiang, and X. Zhuang, “Fast, three-dimensional super-resolution imaging of live cells,” Nat. Methods 8(6), 499–505 (2011). [CrossRef]
14. X. Fan, T. Gensch, G. Büldt, Y. Zhang, Z. Musha, W. Zhang, R. Roncarati, and R. Huang, “Three dimensional drift control at nano-scale in single molecule localization microscopy,” Opt. Express 28(22), 32750–32763 (2020). [CrossRef]
15. S. Y. Chen, R. Heintzmann, and C. Cremer, “Sample drift estimation method based on speckle patterns formed by backscattered laser light,” Biomed. Opt. Express 10(12), 6462–6475 (2019). [CrossRef]
16. Y. Wang, J. Schnitzbauer, Z. Hu, X. Li, Y. Cheng, Z. L. Huang, and B. Huang, “Localization events-based sample drift correction for localization microscopy with redundant cross-correlation algorithm,” Opt. Express 22(13), 15982–15991 (2014). [CrossRef]
17. Z. Zhao, B. Xin, L. Li, and Z. L. Huang, “High-power homogeneous illumination for super-resolution localization microscopy with large field-of-view,” Opt. Express 25(12), 13382–13395 (2017). [CrossRef]
18. L. Tian, J. Wang, and L. Waller, “3D differential phase-contrast microscopy with computational illumination using an LED array,” Opt. Lett. 39(5), 1326–1329 (2014). [CrossRef]
19. L. Tian and L. Waller, “Quantitative differential phase contrast imaging in an LED array microscope,” Opt. Express 23(9), 11394–11403 (2015). [CrossRef]
20. B. Huang, W. Wang, M. Bates, and X. Zhuang, “Three-dimensional super-resolution imaging by stochastic optical reconstruction microscopy,” Science 319(5864), 810–813 (2008). [CrossRef]
21. L. Li, B. Xin, W. Kuang, Z. Zhou, and Z. L. Huang, “Divide and conquer: real-time maximum likelihood fitting of multiple emitters for super-resolution localization microscopy,” Opt. Express 27(15), 21029–21049 (2019). [CrossRef]
22. N. B. Nill, “Objective image quality measure derived from digital image power spectra,” Opt. Eng. 31(4), 813–825 (1992). [CrossRef]
23. S. Koho, E. Fazeli, J. E. Eriksson, and P. E. Hanninen, “Image Quality Ranking Method for Microscopy,” Sci. Rep. 6(1), 28962 (2016). [CrossRef]
