Generating intravital super-resolution movies with conventional microscopy reveals actin dynamics that construct pioneer axons

ABSTRACT Super-resolution microscopy is broadening our in-depth understanding of cellular structure. However, super-resolution approaches are limited, for numerous reasons, from utilization in longer-term intravital imaging. We devised a combinatorial imaging technique that combines deconvolution with stepwise optical saturation microscopy (DeSOS) to circumvent this issue and image cells in their native physiological environment. Other than a traditional confocal or two-photon microscope, this approach requires no additional hardware. Here, we provide an open-access application to obtain DeSOS images from conventional microscope images obtained at low excitation powers. We show that DeSOS can be used in time-lapse imaging to generate super-resolution movies in zebrafish. DeSOS was also validated in live mice. These movies uncover that actin structures dynamically remodel to produce a single pioneer axon in a ‘top-down’ scaffolding event. Further, we identify an F-actin population – stable base clusters – that orchestrate that scaffolding event. We then identify that activation of Rac1 in pioneer axons destabilizes stable base clusters and disrupts pioneer axon formation. The ease of acquisition and processing with this approach provides a universal technique for biologists to answer questions in living animals.


Standalone DeSOS Microscopy Program Tutorial.
1. Take two confocal or two-photon images of your sample. These images should be taken at two different excitation powers. The excitation powers you choose should not obviously saturate your sample. a. Be sure to record the excitation powers, in watt, used to take your images. A power meter can be used to do this. The powers should be measured at the focal plane or the back aperture of the objective.
i. Note that many commercial microscopes use percentages, e.g., 10%, 75%, to quantify the excitation powers in their controlling program settings. In this case, power measurement is still needed because the relationship between the percentages and the actual powers may not be linear, i.e., the ratio between the percentages and their corresponding excitation powers is not a constant.
However, if the measurements show that the relationship is linear and the ratio is a constant, then the percentages can be directly used as the power parameters in the DeSOS program.
b. In addition, you should record the following image parameters of your system. These are required for the DeSOS program to process your images.

SUPPLEMENTARY MATERIALS AND METHODS
Guide to generating DeSOS images using the DeSOS Microscopy Program.

Standalone DeSOS Microscopy Program.
Executable files can be downloaded to install an application capable of generating DeSOS images. The program and example images can be downloaded from the following links: 3. You will see a window with empty axes as shown below. Development: doi:10.1242/dev.171512: Supplementary information

Development • Supplementary information
ii. numerical aperture of the objective lens.
iii. refractive index of the immersion oil or air.
iv. image slice pixel width (x-and y-direction), in nanometers.
2. Install the DeSOS program on a Windows PC. example images "Power=75Percent_3D.tif" and "Power=76Percent_3D.tif" are loaded into the program.
a. The program allows for the import of both 2D and 3D images.
b. For DeSOS algorithm, 3D image stacks are recommended because the blind deconvolution works better with 3D image stacks. Though 2D blind deconvolution is also viable, 3D deconvolution allows for the removal of out-of-focus blur, hence providing better performance.
c. If the images are 3D stacks, the user can use the sliders below the images to change the slice that is currently being displayed. The current and total slice number will be shown above the slider. d. (Optional) "P1" and "P2" images can be cropped (on XY plane) simultaneously according to user-defined cropping boundaries, "Xmin", "Xmax", "Ymin", "Ymax".
Specifically, [Xmin, Ymin] and [Xmax, Ymax] are the coordinates of the top-left and bottom-right corners of the cropping region on the original image plane (before Development: doi:10.1242/dev.171512: Supplementary information Development • Supplementary information 4. Click "Load P1" and "Load P2" to load the firstand second-step images that you want to process with the SOS/DeSOS algorithm. Generally, "P1" is obtained with a lower excitation laser power compared to the power for "P2". For example, in the screenshot below, two cropping). Check the box "Preview" to preview the cropping operation. If not satisfied, uncheck "Preview", change the cropping boundaries, and repeat this process; if satisfied, click "Crop P1&P2" to perform the cropping operation. A message box will appear to confirm the operation, click "Yes" to proceed; this operation cannot be undone, so if the user wants to reverse or change the cropping operation, they will need to reload the raw images. e. (Optional) An offset can be applied to all pixel values on both images. This can be used when your images have a constant background value, e.g., 50, on all pixels; i.e., even the pixels appear to be totally black or empty, they still process a value of 50. This is mainly caused by the offset settings on the detector when acquiring these images. Note that this value is different from background noise. In SOS/DeSOS, a constant background offset (like the constant pixel value of 50 here) could introduce artifacts. It could be eliminated by clicking the button "Apply Offset" with a value "-50" filled in the box below. Development: doi:10.1242/dev.171512: Supplementary information Development • Supplementary information f. (Optional) SOS/DeSOS requires that images "P1" and "P2" are taken from the same field of view. In practice, it may be hard to meet this requirement considering that sample drift could happen when acquiring images. To eliminate this problem, the program provides an image registration tool to register "P2" using "P1" as a reference; i.e., it calculates the drift between "P2" and "P1" and then moves "P2" accordingly to cancel this drift. When image registration is complete, there should be no obvious drift between the two images. To perform this operation, click the button "Registration". A message will appear to confirm this operation. Click "Yes" to proceed. Depending on the size of the images, this operation could take a while. A message box showing the registration progress will show up. 5. Input "P1 Power" and "P2 Power" with the excitation powers (measured with a power meter) that you used to obtain images "P1" and "P2", respectively. Alternatively, you can just input either one of "P1 Power" and "P2 Power" and fill in "Power Ratio"; the other power parameter will be calculated automatically. For example, in the screenshot below, we input "76" for "P2 Development: doi:10.1242/dev.171512: Supplementary information Power" and "75" for "P1 Power" because "Power=76Percent_3D.tif" was obtained with a power setting of "76%" and "Power=75Percent_3D.tif" with "75%" (the power percentages of our microscope are linear related to the measured excitation powers).
6. Check "Two-Photon" box if the images are obtained using a two-photon microscope. If the images are obtained using a confocal microscope, uncheck the box. The example images were acquired with a spinning disk confocal microscope; therefore the "Two-Photon" box is unchecked.
7. Click "Two-Step SOS Algorithm" to perform SOS algorithm on images "P1" and "P2". The resulting image "SOS" will show up.
8. Input the blind deconvolution parameters of the raw images "P1" and "P2". Theoretically, no parameters would be necessary for blind deconvolution, because the algorithm itself can estimate the optimal point spread functions (PSFs) and deconvolved images. In practice, however, the iterative blind deconvolution algorithm that we use might require an extensive amount of computational time to find an optimal solution, and sometimes this solution is simply a "local minimum". Therefore, an initial estimation of the PSFs as accurate as possible is recommended. In this program, we calculate the theoretical PSFs based on diffraction theory using the imaging parameters from the user. The calculated PSFs are used as an initial estimation for the iterative blind deconvolution algorithm. These parameters are: a. "Wavelen. (nm)": wavelength of the fluorescence, in nanometers.
b. "Num. Aperture": numerical aperture of the objective lens.
f. "Quality Factor": an empirical factor (≥1, usually 1 to 3) used to account for nonidealities such as spherical aberrations and refractive index mismatch when calculating the initial PSFs. This parameter could be varied to adjust the deconvolution performance. If the deconvolution performance is not satisfactory, for example, the contrast or signal-to-noise ratio is getting even worse than the raw images, the user can change this quality factor to variate the performance. g. "Iterations": the number of iterations of the iterative blind deconvolution algorithm.
More iterations could have better performance but will take longer computational time.
A value of "20" is recommended for most cases. 9. Click "Blind Deconvolve P1" to perform iterative blind deconvolution algorithm on image "P1".
When it is done, click "Blind Deconvolve P2" to deconvolve image "P2". This step might be time-consuming. For a 2D image, it could take a few seconds; for a 3D stack, it could take minutes for an iteration number "N_int" of 20. A message box will appear to show the deconvolution progress. Development: doi:10.1242/dev.171512: Supplementary information 10. Once the blind deconvolution is done for both images, two deconvolved images "De1" and "De2" corresponding to "P1" and "P2", respectively, will show up.
11. Identical to procedure #4, input "De1 Power" and "De2 Power" with the excitation powers corresponding to "De1" and "De2". Alternatively, you can just input either one of "De1 Power" and "De2 Power" and fill in "Power Ratio"; the other power parameter will be calculated automatically. For this example, we input "76" for "De2 Power" and "75" for "De1 Power", identical to the power setting for SOS algorithm.
12. Identical to procedure #5, check "Two-Photon" box if the images are obtained using a twophoton microscope; uncheck the box if the images are obtained with a confocal microscope.

Development • Supplementary information
14. The "Export" buttons next to each image can be used to export the corresponding images to 32-bit (single precision) TIF files. The program will save 2D and 3D images as TIF images and stacks respectively.
15. The icons on the toolbar can be used to add cursor, zoom in, zoom out, and pan any images in the program. For example, we can zoom in all images to show the resolution improvement of the SOS/DeSOS algorithm. a. Note that the brightness of SOS/DeSOS images may not be high enough to show its structural details. We suggest users to export the images first, and then view them and adjust their brightness/contrast in another program, e.g., ImageJ, to observe the details. For 3D stacks, a max z-projection is a good way to visualize the resolution improvement. The screenshot below shows the max z-projections of the exported raw image stack "P1" (left) and the resulting DeSOS stack (right). The brightness/contrast has been adjusted to visualize the details. Compared to the raw image, the improved resolution of the DeSOS image can be clearly seen.
In vivo imaging with spinning disk confocal microscopy of zebrafish Embryos were manually dechorionated at 48 hpf and anesthetized with 3-amino-benzoic acid ester (Tricaine). Anesthetized embryos were immersed in 0.8% low-melting point agarose and mounted on their right side in glass-bottomed 35 mm Petri dishes. A spinning disk confocal microscope from 3i technology © was used for all imaging. A Zeiss Axio Observer Z1 Advanced Mariana Microscope was equipped with X-cite 120LED White Light LED System and filter cubes for GFP and mRFP, a motorized X,Y stage, and a piezo Z stage. The microscope has three objectives: a 20X Air (0.50 NA) objective with a working distance of 2 mm, 63X (1.15NA) water objective with a working distance of 0.66 mm, and 40X (1.1NA) water objective with a working distance of 0.62 mm. The microscope also has a CSU-W1 T2 Spinning Disk Confocal Head (50 uM) with a 1X Development: doi:10.1242/dev.171512: Supplementary information camera adapter and andor iXon3 1Kx1K EMCCD camera as well as dichroic mirrors for 446,515,561,405,488,561,640 excitation and a laser stack containing 405 nm, 445 nm, 488 nm, 561 nm and 637 nm with laserstack FiberSwitcher with 250 uS switch time, photomanipulation with vector © high speed diffraction-limited point scanner ablations, and Ablate!TM © Photoablation System (532 nm pulsed laser with a pulse energy 60J @ 200 HZ). Time lapse images were taken every 5 minutes for 24 hours starting at 48 hpf. Adobe Illustrator and ImageJ were used to process images and enhance image brightness and contrast.

Intravital Microscopy of Mice
Intravital imaging of the mouse brain with two-photon microscopy was performed similarly to previously described (Grutzendler et al., 2002). Briefly, Cx3cr1-GFP/+ mice were anesthetized with ketamine and xylazine cocktail by intraperitoneal injection. The mouse's head was secured and fixed in place using a stereotaxic instrument (Stoelting Co), and the skull was exposed with a midline scalp incision. A high-speed microdrill (Ideal Microdrill) equipped with a 0.7mm burr (Fine Science Tools, 19007-09) was used to thin the skull to approximately 30µm in thickness, using light sweeping motions to thin the skull gradually without applying significant pressure to the brain. PBS was applied to the thinned skull to enhance transparency for imaging.
Following surgery, the anaesthetized mouse was immediately imaged on an Olympus FV1000 microscope equipped with tungsten and halogen visible light sources, four lasers (argon 458nm, 488nm, 515nm; HeNe 543 nm; red diode 635; and a Mai Tai DeepSee Development: doi:10.1242/dev.171512: Supplementary information

Development • Supplementary information
titanium-sapphire 690-1040nm), and a cube filter set with the following spectral configurations: 460-500, 520-560, 575-625, 650-700. All imaging employed a 25× objective water immersion objective (XLSLPLN25XGMP, Olympus USA; NA = 1.0 and WD = 8 mm). The laser was tuned to the excitation wavelength for GFP (920 nm) at four step-wise tuned laser powers (9.37 mW, 10.68 mW, 11.95 mW, and 13.25 mW), and a digital zoom was set to (4x). The imaging depth was ∼100µm from the pial surface and when Z-stacks were acquired, the step size was 1 µm.

DeSOS microscopy
The blind deconvolution operation can be performed using various commercial or openaccess programs. In this work, the images were blind deconvolved using either AutoQuant Blind or a blind deconvolution algorithm implemented in Matlab. Both blind deconvolution methods utilized iterative maximum likelihood estimation (MLE) algorithms, which can be explained as follows (Biggs and Andrews, 1997;Holmes et al., 2006). The imaging process is modeled as ( ) = ℎ( ) ⊗ ( ) + ( ) , where denotes the 3D spatial coordinate, ( ) the measured distorted image, ( ) the ideal undistorted image, ℎ( ) the PSF of the system, ⊗ the inherent convolution process, and ( ) the noise. MLE iteratively estimates ℎ( ) and ( ) simultaneously which have the highest likelihood of being correct given the measured imaging data. Specifically, the algorithm first guesses an initial ℎ( ), and then estimate which ( ) could have generated ( ). The estimated ( ) is then reblurred by ℎ( ) and compared to the actual image ( ), where the error in this comparison is used to re-estimate ℎ( ) in order to reduce the error in estimation.
These steps are repeated again and again until a convergence is reached or a certain Development: doi:10.1242/dev.171512: Supplementary information stopping criterion is satisfied. For the deconvolution with AutoQuant Blind, the images were imported, deconvolved, and exported using the commercial software; for the deconvolution with Matlab, the MLE algorithm was integrated in our open-access DeSOS application which could perform both blind deconvolution and SOS operations.
The second operation of DeSOS microscopy, SOS, is a saturation-based superresolution fluorescence microscopy technique that can be easily implemented and requires no additional hardware. In general, SOS linear combines conventional (confocal or two-photon) fluorescence images to generate a super-resolved image with a √ -fold resolution improvement compared to the diffraction limit. The method is based on the steady-state solution of fluorescence intensity in a two-level fluorophore model, where is the spatial coordinate, the peak excitation intensity, ( ) = exp(−2 2 / 0 2 ) a Gaussian excitation profile with a 1/ 2 radius 0 2 , the number of excitation photons needed for a fluorophore to emit one photon ( = 1 for confocal, = 2 for two-photon), ( ) the effective PSF of the system, and and the constants related to detection efficiency, cross-section, excitation wavelength, etc.
Note that the excitation profile ( ) is assumed to be Gaussian to simplify the illustration; in practice, ( ) can be of any other mathematical forms that can describe the focus intensity distribution, e.g., Lorentzian, and SOS will still work. SOS utilizes the Taylor expansion of ( ) , which is ( ) = [ ( ) − 2 2 2 ( ) + 3 3 3 ( ) − ⋯ ] .
Considering the (Gaussian) excitation profile ( ), high powers of ( ), such as 2 ( ), deconvolution with Matlab, SOS was integrated in our open-access DeSOS application and the images deconvolved with the MLE algorithm were automatically processed with the SOS algorithm. All output images were exported as TIF files for further analysis.

Quantification and Statistical Analysis
Pixel groupings were completed using two distinct methods. Composite outline renderings of distinct actin populations were created by first dividing the range of pixel intensities into quartiles. Distinct pixel populations were then visualized by adjusting the intensity threshold of the image to reflect a particular quartile. Particle analysis was then completed to develop the outline rendering of each quartile. These renderings were then compiled into a single composite image.
Grouping of pixels to determine the area and intensity of distinct groupings was completed using ImageJ's magic (tracing) tool. First the tolerance of this tool was set to 10% of the maximum pixel intensity in the image. (E.g. If the maximum pixel intensity in the image was 1000 au, then the threshold would be set at 100 au.) This change in tolerance allows for ImageJ to select adjacent pixels that have intensity values within 100 au of each other.
Thorough pixel groupings using the magic (tracing) tool were completed for distinct ROI in the image. The location, size, and area of each grouping were recorded. The location of each group was used to prevent counting pixel groupings more than once. Development: doi:10.1242/dev.171512: Supplementary information