High-Resolution and High-Speed Atomic Force Microscope Imaging.

The advent of high-speed atomic force microscopy (HS-AFM) over the recent years has opened up new horizons for the study of structure, function and dynamics of biological molecules. HS-AFM is capable of 1000 times faster imaging than conventional AFM. This circumstance uniquely enables the observation of the dynamics of all the molecules present in the imaging area. Over the last 10 years, the HS-AFM has gone from a prototype-state technology that only a few labs in the world had access to (including ours) to an established commercialized technology that is present in tens of labs around the world. In this protocol chapter we share with the readers our practical know-how on high resolution HS-AFM imaging.


Introduction
The atomic force microscopy (AFM) [1] is a powerful tool for direct visualization of biological samples in an aqueous solution with sub-molecular resolution. It allows non-invasive imaging in physiological conditions of unlabeled biological samples, such as nucleic acids [2][3][4] and cell membrane proteins [5,6]. AFM has rapidly emerged as an effective structural analysis tool, complementing atomic structure data acquired by other techniques such as X-ray crystallography, NMR and electron microscopy. In contrast to these latter methods, which heavily rely on ensemble averaging, AFM exhibits a superior signal-to-noise ratio, enabling molecules to be observed individually under aqueous environment. To acquire a topographic image, a sharp tip attached to the free end of a flexible cantilever is brought into contact and scanned over a sample and its profile height is recorded over the selected scan area. The feedback control allows adjusting the force applied to the sample to avoid damaging fragile biological structures. As a result of the imaging process, during which the distance between the sample stage and the AFM probe varied following the sample topography, a 3D profilometric image of the sample surface is obtained. With the advantage of operating in nearly physiological conditions, AFM is an ideal platform to study biological dynamic processes at the molecular scale. Nevertheless, the framerate of one image every few minutes of conventional AFM is not sufficient to visualize most dynamic processes of biological molecules that take place in its majority at sub-second time scales. The introduction of the high-speed atomic force microscopy (HS-AFM) in 2001 opened the door to the visualization of biological molecule dynamics [7]. The technical development of the HS-AFM was performed by the laboratory of Toshio Ando in the University of Kanazawa, Japan.
HS-AFM allows imaging at maximum rates of 5-10 frames per second [8][9][10][11]. The operating principle of the HS-AFM is based on the miniaturization of the moving components of the AFM (cantilever and scanner) to increase their velocity by 1000 times, in the interest of achieving reaction speeds of tens of microseconds. For the HS-AFM probe, the gain in speed does not suppose a lost in performance. In the case of the scanner, the X,Y,Z displacement range is decreased ~1:10 over conventional AFMs. With this technique, centrosome [12,13], DNA origami [14], micelles [15] and many other biological systems have been imaged at high speed and high resolution.
In this book chapter, we describe in detail our HS-AFM imaging procedure for high resolution and high speed imaging, applicable to any version of the Ando-type HS-AFM system "SS-NEX".

High-Speed AFM
The protocol here described is adapted for any Ando-type HS-AFM setup "SS-NEX", commercialized by the Research Institute of Biomolecule Metrology (RIBM), Japan. A picture of the HS-AFM hardware of our "SS-NEX" HS-AFM is presented in Figure 1, where the cantilever is mounted up-facing in a cantilever holder that comes with a pool of 110 µl while the sample is mounted on top of the scanner facing down. In general, imaging experiments are performed in tapping mode where the tip is oscillated at constant amplitude. For that, the cantilever is excited at its resonance frequency with the help of a miniaturized piezoelectric actuator. The sample can move in the XY-plane and also in the Z vertical direction thanks to three small multilayer piezoelectric actuators. The Z-piezo the one displaced at the highest frequency (nominal resonance frequency ~ 600 kHz and ~ 150 kHz once glued on the scanner). The deflection of the cantilever is monitored by an optical system. A laser beam with a 20x optical microscope objective is focused on the backside of the cantilever and the reflected beam tracked with a 15 MHz bandwidth photodiode which sends the signal to a Lock-In amplifier (Fourier Analyzer). The signal goes to the Proportional Integral Derivative (PID) controller which compares the amplitude readout with the value settled by the operator (set point amplitude, A) to feed 1. Glue a cleaned glass-rod on the Z-piezo by using nail polish and let it dry for at least 20 min. Glass-rods can be cleaned by sonication (f ~ 35 kHz) in acetone and stored in a plastic box until use.
2. Glue a mica disk on the glass-rod with cyanoacrylate glue and let it dry for 10 min (see Note 3).
3. Cleave the mica disk with adhesive tape to obtain a clean, atomically flat surface.
To facilitate the cleavage, mica can be previously cleaned with acetone (see Note Immerse the sample to wash it, this will remove the excess of molecules that did not adsorbed on the surface. Repeat the procedure at least five times without the need to replace the buffer between rinses.
6. Keep the sample wet with the working buffer solution under the humid hood until it is mounted on the HS-AFM.

Sample Mounting
1. Wash abundantly the cantilever holder with 2% anionic detergent, tap water, acetone, 70% ethanol and ultrapure water. Dry it gently with air or nitrogen flow if possible (see Note 6).
2. Set a small cantilever on the cantilever holder: position the cantilever on the trench of the cantilever holder and fix it with the cantilever holding screws (Figure 3, see Note 7).
3. Mount the cantilever holder on its base by placing the two magnets of the cantilever holder in contact with the micrometer screws of the base. 4. Wash the pool with 110 µl ultrapure water for 10 times. 9. Troubleshooting the laser alignment: If the SUM is too low, then check the position of the cantilever and try to adjust it. If the cantilever is not in focus point of the laser beam, modify its vertical position by using the Z-cantilever holder micrometer screw. The SUM will increase as the cantilever reaches the focus point of the laser beam. If the laser spot is not well centered, use the XY-cantilever holder micrometer screws to adjust the lateral position of the cantilever. If the back of the cantilever is dirty or is an already-used cantilever then the SUM signal could be lower, in that case replace the cantilever. 12. Check the alignment of the glass-rod and the cantilever by looking if the glass-rod is in the same vertical plane of the cantilever. 13. Connect the scanner to the break-out box (see Note 11).
3.3. Tip-sample approach in amplitude modulation mode 1. Before approaching the sample surface and the tip, check in the video monitor (video camera or CCD) that there are no bubbles around the surface or the AFM chip (that would cause abnormal cantilever deflection). Likewise, ensure that in the camera display there are no particles or dust diffusing around in the liquid, due to contamination (see Note 12). The edges of both the sample support (mica, glass, etc.) and the cantilever should look clean.
2. The tip is placed relative to the sample stage around a millimeter away in the Zaxis, as shown in Figure 5A. In the camera display, the surface edge must be blurred or unfocused, ensuring that there is enough distance between the substrate surface and the cantilever ( Figure 5B). Bring the sample close to the cantilever using the micrometer screws in XY-plane, as indicated by the arrow in Figure 5A 5. Using the Fourier Analyzer, apply an AC sinusoidal voltage to the excitation piezo located on the cantilever holder, to excite the cantilever at the measured resonance frequency in point 3.4 (see Note 16).
7. The amplitude set point of the cantilever should be lower than the free amplitude.
In general, for the approach the amplitude set point is set around 20 % -40 % lower than the free amplitude. 8. Ensure that everything is plugged, Z-scanner, isolation table (if used), X-, Y-and Z-drivers and PID are on (see Note 19). 9. Set parameters to start approaching: center the offset for the output voltage of the X-, Y-, and Z-drivers (see Note 20). In the PID controller: the integral gain is set typically at 20 % -40 % of the maximum, the proportional and derivative gains are set to minimum. The tilt for the X-and Y-axes is also centered. If there are two time constants in the controller (slow and fast), switch to slow. 10. Approach automatically, selecting the velocity of the step motor and the percentage of amplitude reduction that will halt the automatic approach (see Note 21).
11. When the tip is very close to the surface, the free oscillation amplitude may vary significantly -usually increasing-due to unspecific interactions. If the oscillation amplitude varies, stop the automatic approach and readjust the set point amplitude 20 % -40 % lower than the free amplitude. Proceed again to approach automatically.
12. If the software detects a tip-sample contact (when the set point amplitude approximates the free amplitude, detected thanks to the read-out of the amplitude and the PID output), the automatic approach stops automatically. In case of a manual approach, stop approaching when the Z-piezo starts to retract. 13. After the automatic approach stopped, offset the voltage of the piezo-driver until the output voltage in the PID is close to 0 V (see Note 22).
14. Troubleshooting the approach step: several situations may lead to an anomalous engage or a complete approach failure. It can be observed that the Z-piezo does not hold the voltage value (it moves slowly to fully expanded or contracted position, or back and forth): a) Mica is not tightly attached to the glass-rod or the glass-rod is not firmly attached to the piezo stage. Sometimes mica partially exfoliates allowing some fluid leakage between the different layers. All these situations lead to oscillations of the surface while scanning that may prevent a successful approach.
b) Due to a bad cleavage, or an excess of glue between the mica and the glass, the substrate plane is too inclined respect to the tip, hence the contact is not stable.
c) The cantilever is tip-less (see Note 23). In this case, decreasing the amplitude set point also changes the SUM value in the photodetector (negative deflection values). d) There are big aggregates loosely adsorbed on the surface with strong and unspecific tip-sample interactions that vary the oscillation amplitude. Try to approach with lower amplitude set point, higher free oscillation amplitude or a different area; eventually reconsider the sample preparation.
e) The solution is dusty or there is material adsorbing on the cantilever, which may lead to false engaging in the surface. Exchange the buffer solution in the pool or eventually increase the free oscillation amplitude. Eventually reconsider the sample preparation.

HS-AFM imaging
At this point the clearance distance between the sample stage and the HS-AFM probe is sufficiently small for the Z-piezo range to bring sample and the tip in and out of contact.
1. Out of contact, turn off the cantilever excitation, with the AFM probe, observe in the oscilloscope the width of the thermal vibration of the cantilever AThermal. 8. Optimization of oscillation amplitude: Reduce the amplitude excitation signal, the probe will go out of contact, reduce the set point amplitude A to regain tip-sample contact, further reduce the excitation signal, further reduce A… repeat this procedure until you arrive to a cantilever oscillation that is A0 ~5•AThermal.
9. Check again the feedback saturation limit position by increasing the feedback speed of the I gain, next slow down slightly the I gain to avoid feedback resonance, and next increase the feedback speed of the P gain parameter to increase contrast on fine topographic details. 10. Speed increase: progressively increase the imaging speed, evaluate the loss of quality on the image relative to speed increase. Select the speed that provides a fair compromise between image quality and maximum possible speed.
11. If the sample presents height jumps over a few nanometers, the standard PID feedback speed of reaction may be too slow to correctly contour the sample during downhill scanning (so-called parachuting), in this case the HS-AFM dynamic PID should be turned on [19]. ii) Introduce the HS-AFM cantilever probe with its support in the plasma chamber of the plasma. Select the gas to fill in the chamber (see Note 28). Run the plasma chamber for a few tens of seconds, inert gases require longer times. Typical parameters we use are pressure ~ 1 mbar, power ~ 20 W.
iii) Image the sample with the sharpened tip. If resolution is not good enough further repeat steps i) to iii). Figure 6 shows the image quality improvement on Outer Membrane Porin F (OmpF) following consecutive sharpening cycles (see Note 29).

HS-AFM video processing and displaying
HS-AFM produces large data sets of image sequences that are affected by a number of instrument-specific peculiarities (such as drift of the piezoelectric elements, contrast discontinuities along the fast-and slow-scan axes, feedback parameters, etc.) that need to be accounted for before any analysis is performed.
Hereafter we provide some clues and suggestions in order to fix these instrument-specific hitches in what we believe is an accurate and time effective manner. 3. Align all the frames in the stack to a reference image (see Note 31) in order to compensate for X-and Y-axis drift (see Note 32). This can be first tried in an automated way using normalized cross-correlation (or other image registration algorithms) and reiterated throughout all the frames using control flow statements (see Note 33). Visually inspect the aligned image stack and delete -or manually overlay-badly registered frames to the reference image. 4. Remove X-and Y-axis tilting by means of three-point plane fitting (see Note 34).
To attain this, first filter each frame with a 2D Gaussian blur filter to remove high frequency noise. Afterwards, identify three points that sit on the substrate/background of the image. Make sure these very same points are on the substrate only throughout all the frames. Fit each triad with a plane and subtract it from the aligned video.

Perform a horizontal line-by-line levelling (see Note 35). As the background has
been already subtracted, a median (0 order) levelling will suffice (see Note 36). To avoid levelling artefacts make sure to exclude features not belonging to the substrate/background (usually by means of their height) from the computation.
Define a region of interest over the substrate and fit the histogram with a Gaussian function to verify that the variance of the background has decreased after the procedure. 6. Fine align all the frames in the stack to a reference image using sub-pixel precision. For this purpose a software package that performs template matching by means of zero mean normalized cross-correlation was developed by our group some time ago [20]. Alternatively, use the ImageJ plugin Template Matching and Slice Alignment (see Note 32). 7. (Optional) Filter frame by frame the image stack with a matrix Gaussian filter to remove unwanted high-frequency noise. Be aware that over-application of lowpass filters or use of large matrices will tend to blur the data.

Notes
1. Although AFM image processing and analysis is in general best done with the software supplied by the instrument manufacturer, software packages provided with HS-AFM instruments might be rather limited, and more powerful third party packages are often needed.
2. Alternatively, a free software compatible with many MATLAB scripts is GNU Octave (https://www.gnu.org/software/octave/). All the procedures described in 7. To settle the cantilever, completely remove the plate and position the cantilever, then first secure the screw at the bottom of the plate and afterwards the one in the front (see Figure 3). Do not screw to strong, screws can break easily, but enough to ensure the tighten fix of the cantilever and avoid possible vibration during imaging. Alternatively, slide the cantilever into place with the help of a tweezers to rise up the plate and secure the screws as explained before. Pay attention not to damage the cantilever during this operation. The cantilever must be parallel to the holder otherwise it would be easier to break it when mounted on the setup. For imaging experiment, the small cantilever listed in the "Material" section should be used. 8. Imaging experiments can be performed with the laser diode driver set in constant intensity "I mode" or constant power "P mode", depending on the type of sample and conditions in which to be imaged. For high-resolution imaging, we recommend using the laser diode driver in the "I mode". The "P mode" operates a feedback system that may introduce noise to the image. It can be used to avoid local heating when performing one-day long measurements.
Characteristic values we use are: in the "I mode", set the output between 21.8 and 23 mA, while in the "P mode" set the output between 0.13 and 0.18 mW for the laser diodes used in HS-AFM, although this number may vary with the ageing of the laser diode.
9. To facilitate this task, set the video camera in high light sensitivity ("night shot" mode) so the laser spot is intensified and easily visible on the screen. 13. To prevent the tip crashing on the surface, it is critically to be very attentive to the video camera while performing a coarse approach. Stop the coarse approach when the surface edge starts to define but is still blurred or unfocused. If the tip crashes brutally on the surface it will be most likely damaged.
14. The cantilever position on the surface is important for high-speed imaging. If the cantilever is not placed in the edge but closer to the middle area of the surface, during imaging, the Z-scanner exerts a viscous drag or a hydrodynamic pressure that drags the base or chip of the cantilever, which leads to an increase in the response time and imprecise cantilever deflection [8,21,22]. 15. Usually, AFM operating software programs include this function. 16. The fundamental resonant frequency is convoluted when using piezo acoustic excitation of the cantilever with a myriad of peaks due to 'echoes' in the liquid cell pool, the so-called the forest of peaks. HS-AFM imaging is performed selecting a frequency that 'echoes' in the fluid cell in the range of frequencies of the fundamental resonance of the cantilever [23]. 17. Typically, the free oscillation amplitude is set in concordance of the roughness of the sample; the flatter the sample; the lower the amplitude.
18. To convert the amplitude value from V to nm, it should be calibrated once approached using the optical beam detection sensitivity, which requires performing force curves (approach and retract cycles between the tip and the sample). In general, avoid calibrating the sensitivity before imaging -or before the experiment-to prevent tip damage or tip wearing. The calibration can be performed at the end of the experiment, after imaging. At this step -to approach-use an approximate value: the sensitivity does not change significantly for the same type of cantilevers. 19. When the PID controller is active and the amplitude set point is set lower than the free oscillation amplitude, the Z-piezo fully expands (to positive voltage).
20. If the full voltage range of the Z-piezo driver is 50 V, center it to 25 V. If the full voltage range of the X-and Y-piezo drivers are 100 V, center them to 50 V. 21. A typical percentage of amplitude reduction for halting the automatic approach is ~30 %. Alternatively, manually approach slowly moving the step-motor to bring close the cantilever to the sample surface. This is particularly useful when the surface is rough and/or heterogeneous. 22. Alternatively, slowly up or down with the step-motor. 23. In general for high-resolution imaging, it is very common to use amorphous carbon or carbon nanofibers tips. These materials are brittle and therefore they break easily. Always scan with very gentle tapping.
24. During the first images the risk of tip damage is high, here it is intended to limit it.
25. If the integral I gain is increased too much, saturation of the feedback loop bandwidth occurs; the Z-displacement cannot be done any faster even if the PID controller is requiring it. Then when I saturation is reached the feedback loop breaks, and error accumulates in the integrator-circuit, an I signal overshoot is created. Next, the circuit will create to compensate an overshoot of the I signal in the other direction. This oscillatory sequence of overshoots creates strips in the topographic image of the HS-AFM. Importantly, the I saturation depends on the topography of the sample under the tip. I saturates on the sample topographies that demand faster motions of the z-displacement, the uphill and downhill motions. 26. To avoid tip damage during the withdraw, first retract the tip from the sample using the Z-piezo displacement; for this set point A>A0 before using the motor.
27. Otherwise, as surfaces get charged with plasma, the cantilever can break due to electrostatic repulsions.
28. The nature of the gas injected in the plasma chamber affects the etching rate of the amorphous carbon EBD tip. For example, oxygen oxidizes the carbon or the EBD tip, hence is more aggressive than inert gases such as helium, whose atoms just physically bombard the carbon. As a general rule, the less is the imaging resolution, the more etching power should be applied on the tip apex.
29. The final shape of the sharpened apex depends on its initial shape, high resolution apexes are not achievable always by plasma etching, a minimum of sharpness of the initial apex is required. So, in the case that after a few cycles of sharpening the required resolution is not achieved, the HS-AFM probe should be replaced.
30. This operation can be implemented in few simple steps using a for loop. Routines to perform in MATLAB this and other described processing steps are freely available at https://sites.google.com/view/fm4b-lab/group-members/dr-arinmarchesi.
31. The performance of image registration also depends upon the choice of the reference image (template). To increase the signal-to-noise ratio you shall build your reference image by averaging over several frames (usually [10][11][12][13][14][15][16][17][18][19][20]. For this purpose, we recommend to use the median, which is more robust than the mean.
Cropping is also commonly used to remove unwanted features from the edges of the averaged image.
32. Mechanical and piezo-scanner drifts may result into movies in which the sequential frames are not perfect real-space superposition. Therefore, before any data can be extracted from a HS-AFM image stack, all images must be perfectly aligned with respect to a stable coordinate origin.
33. For this purpose, an excellent option is the ImageJ plugins Template Matching and Slice Alignment (available at https://sites.google.com/site/qingzongtseng/template-matching-ij-plugin). Please note that this routine requires your video to be converted in 16-bit format, which might result in an unwanted loss of information. If this is the case, discard the obtained aligned video and apply the computed X and Y shifts to your original image stack to create a new registered image stack.
34. Nonlinearities of the piezoelectric scanners to the applied voltage can result in curved trajectories of the probe or sample stage (depending on the configuration of your system). If scanner bow artefacts are present, you shall consider fitting your image stack with a second order polynomial surface through five points.
35. As the horizontal axis is usually the fast scan axis, any change in imaging conditions over the time of the scan will lead to horizontal discontinuities in the topographs. It will thus be well accounted for by a horizontal line-by-line levelling.
36. In this routine, for each row of the image, the row median is subtracted from every element in the row.