Primary cell culture of microglia

All the animal experiments in this study were approved and confirmed by our institutional review board for animal research, the Korea University Institutional Animal Care and Use Committee (KUIACUC-20170405-2) and all experiments were performed in accordance with relevant guidelines and regulations. Detail description of pure microglial cell culture was given in ref. 1. Briefly, we prepared primary glia co-cultures from the cerebral cortex of postnatal day 1 ~ 2 Sprague Dawley rat (Charles River, OrientBio Inc.) which is euthanized by decapitation. After cultivating them in a T-75 culture flask (37°C, 5% CO₂) containing DMEM medium supplemented with 10% fetal bovine serum and 1% penicillin streptomycin at an initial seeding density of about 2.5 × 106 cells/flask for two weeks, cells in a T-75 culture flask were shaken at 280 rpm for 20 minutes to harvest microglial cells detached from the surface of glial cells fully grown on a flask substrate. The supernatant was collected and then centrifuged at 1,500 rpm for 5 minutes to obtain microglial cells deposited at the bottom of the tube. Finally, microglial cells were plated onto poly-D-lysine coated coverslip at density of 50 cells/mm² for experiments.

Quantitative phase image with synthetic aperture method and optical diffraction tomography

Biochemical compositions of actin waves are already well characterized with advancements in various fluorescent molecular markers and related imaging tools [15, 22, 23]. However, like any other fluorescent imaging experiments fluorescence-based imaging techniques can suffer from phototoxicity and photobleaching which can interfere with normal cell functions including the actin dynamics. However, the advantage of label-free quantitative phase optics is two-fold: first of all, it is non-invasive to the cell, thus, ideal for a long-term time-lapses imaging with no worry on photo-toxicity or bleaching; and, second of all, it can produce 3D phase tomography. Of course, fluorescent imaging techniques are useful in seeing various relevant biomolecules (genes and proteins) in actions. In contrast, quantitative phase imaging is good for visualizing delicate 3D structures of phase objects like actin waves that are associated with crawling cells. For those reasons, often it is not suitable for long-term (> 1 day) live-cell imaging. Similar issues also exist with the imaging modality such as scanning electron microscopes (SEMs) and atomic force microscopes, with which one can obtain quantitative 3-D images, but only when cells are treated and no longer alive [24, 25]. On the other hand, quantitative phase imaging (QPI) methods have been recently proven to be useful for quantifying physical properties (structures) of translucent samples such as individual living cells or tissues without any exogenous labeling agents. Typically, transmitted phases are measured over specimens and processed afterwards to reveal sub-cellular structures. QPI methods are not only fast enough for measuring dynamic structural changes taking place inside cells, for example, high-frequency membrane fluctuation of red blood cells [26–28] and transmembrane water fluxes in neurons [29, 30], but also sensitive and accurate enough to measure the dry mass of chromosomes [31, 32]. Furthermore, if combined with scanning systems, QPI methods can map out 3-D images of refractive index of living cells to identify its internal structures with a high spatial resolution and sensitivity. In fact, they could be used as an optical biopsy tool for identifying cancer cells inside densely-packed tissues and calcifications in the breast [33]. With these capabilities, QPI can potentially be very useful for quantifying subcellular actin waves in living cells and thereby facilitate a better understanding of their roles in physiological and pathological function of cells.

In this work SAM was used to obtain 3-D shape information of ruffle waves supported by a crawling cell [19]. An interference microscope with a sample and a reference arm was built using a coherent light source (He-Ne laser, λ = 632 nm). Variation of the interferogram introduced by a living cell in the sample plane was then recorded in the detection arm designed as an off-axis configuration. An oil immersion objective lens (Olympus, 100X, 1.4 NA) was used to capture the light transmitted through the sample. A total of 40 images were taken while a 2-D scanning mirror placed at the front conjugate plane of the sample steered the illumination angle ranging from 0 to 0.6 NA for 200 ms. All images were processed into electric field images according to the digital holographic method and superposed in a coherent fashion to produce final synthesized images. In addition, the diffraction noise on top of bare coherent images was considerably suppressed by the synthetic process. To reconstruct 3-D image with ODT, a total of 200 images were obtained with various illumination angles using the 2-D scanning mirror from 0 to 1.4 NA for 1.33 s. Using the reconstruction algorithm based on the Fourier diffraction theorem under the Rytov approximation and the non-negative constraint iteration, 3-D image was then retrieved from the measured optical field involved in refractive index information [18–21].

Fluorescence cell images

For fixation, microglia cultured on a cover glass were incubated in 4% paraformaldehyde at room temperature for 20 minutes and then treated with 0.1% Triton X-100 (Sigma-Aldrich) for 20 minutes at room temperature. After rinsing with phosphate-buffered saline (PBS), cells were incubated with rhodamine-phalloidin (Sigma-Aldrich) at 1:200 dilution and DAPI (4’,6-diamidino-2-phenylindole, Sigma-Aldrich) at a concentration of 0.1 μg/mL in PBS for 20 minutes to stain cytoskeletal F-actin and cell body, respectively. Fluorescence cell images were taken with a confocal microscope (Axio Observer D1; Carl Zeiss, Germany).

Scanning electron microscopy

Microglia were fixed in Karnovsky's fixative (2% glutaraldehyde and 2% paraformaldehyde in 0.05 M sodium cacodylate buffer; pH 7.2) for 2 hours at 4°C. These cells were then dehydrated in stepwise increase of ethanol concentrations from 30% to 100% at intervals of 10%. Subsequently, samples were dried at 60°C for 24 hours and finally coated with platinum (Pt) using a sputter (Sputter E-1030, Hitachi, Japan). Scanning electron microscopy imaging (SEM; FE-SEM S-4700, Hitachi, Japan) was performed at 15 k.

Time-lapse imaging with a phase-contrast microscopy

Substrates containing harvested MG were placed at 37°C with 5% CO₂ regulated in a home-built chamber mounted on an inverted microscope (IX71, Olympus, Japan) stage. Time-lapse images were acquired at an interval of 15 seconds, typically for a time period longer than 24 hours, using a cooled CCD camera (MFcool, ProGres, Germany). Phase contrast microscope (PCM) time-lapse images were acquired inside a temperature (37°C, 5% CO₂) regulated Live Cell Instrument (LCI) customized chamber mounted onto a microscope stage [5].

Micro-patterned substrate

Micro-scale grooves were patterned with photolithography using a positive photoresist (AZ7200, AZ Electronic Materials Ltd.). These patterned micro-scale grooves (20 μm in width) were etched by deep reactive ion etching (DRIE) and the photoresist layer was removed. Our target depth of DRIE was 4 μm. The fabricated wafer was used as a master for PDMS (Sylgard 184 Silicone Elastomer, Dow Corning) replica molding. After that, PDMS 10:1 mixture was poured over the fabricated wafer and salinized using vapor-phase (tridecauoro-1,1,2,2, -tetrahydrooctyl)-1-trichlorosilane (Sigma Chemical Co., St. Louis, MO, USA) under vacuum for 15 minutes before PDMS pouring. After peeling PDMS from the wafer, we rinsed PDMS with acetone, isopropyl alcohol (IPA), and deionized water (DI water) in order.

Image analysis of a cell migration

The detailed descriptions of cell tracing and image analysis were given in ref. 5. To trace out the crawling cell in detail, all acquired images were binarized and then the centroid position of cell body was extracted for each frame. Based on the position data of extracted centroids, first the trajectory was smoothed by the third order polynomial function (as a low-pass filtering) in order to minimize some high-frequency fluctuation noise and image processing errors. The polynomial function (third order) was applied to fit the sliding window of 11 successive points (which corresponds to 150 seconds) locally. Next, the local curvature k was calculated by curvature function at each time step using the filtered trajectory. Finally, a low-pass filter was applied to k(t) with cutoff at 30 seconds, and the local extrema of the filtered k(t) were considered to be a turning point, along each segment the turning angle is defined as α. Cell images were cropped with the typical cell size area and aligned with respect to the centroid positions and turning angle of cell.

Kymographs were generated from these aligned cell images using ImageJ. For defining ruffling regions at the leading edge, the images were binarized with the threshold grayscale value between 70 and 80 (out of 255).

The emergence of biased peripheral ruffle waves leads a turn in crawling MG cells

When microglia are harvested from rat brains and plated on to a culture dish at a very low cell density, they are free from any interactions. Thus, their motilities are solely determined by individual cells themselves. As aforementioned in the introduction, these free individual MG cells exhibit quite an interesting motile behavior as illustrated in Fig 1A. Using low-pass filtering, the trajectory is represented as red line fitted locally from the centroid position data to a third order polynomial function with sliding window of 11 successive points. Next, in this filtered trajectory, the local curvature k is calculated by the curvature function at each time step. Finally, k(t) is applied with the cutoff at 30 seconds, and the local extrema of the k(t) are set as turning points shown as dots in Fig 1A. In each dot, the turning event occurs with an angle α and its magnitude is represented by color. In Fig 1B, green dots are the centroid position data (before applying the low-pass filtering) and red dots represent the turning events of the green box shown in Fig 1A. The turning events occur about 50 times for 90 minutes. Typically, a trajectory of the cell can be viewed as a sequence of “runs” and “turns” inter-turn-intervals on about 2 minutes (the average of 1.82 ± 0.69 min) similar to the value of previously published data (1.9 ± 0.2 min). And the segmented red lines (in Fig 1B), composed of two neighboring red dots, have the average angle of 17.29 ± 23.25 degrees. This value is smaller than the previously published data (48.3 ± 7.3 degrees) due to the shorter imaging acquisition time (~ 90 minutes) than before (~ longer than 24 hours).

Additionally, to clearly show the relationship between the turning direction of cell and the active zone of peripheral ruffles at the leading edge, four turning positions which have large turning angle are chosen with the corresponding time points of 12, 30, 45, and 65 minutes in Fig 1A. Their matching PCM snapshot images showing cell turning are given in Fig 1C. We have marked the previous and new directions as yellow and blue dotted lines superimposed on these images, respectively. To make more clear contrast on the active zone, the peripheral ruffles have been delineated with a saturated black color (through thresholding and binarizing) in Fig 1D and superimposed on Fig 1C. When peripheral ruffles show a bias (activated intensively in one side) as shown in Fig 1D, the cell turning follows immediately to the same direction. Hence, we could notice a close relation between the turning direction (blue dotted line in Fig 1C) and the active zone of peripheral ruffles in Fig 1D.

Periodic oscillatory behaviors of peripheral ruffle waves at the leading edge

Using QPI technique in conjunction with SAM (see Materials and methods for further details), we analyzed the 3-D morphology of peripheral ruffles and their dynamics in a time-lapse mode. Fig 2A is an example showing a color map of crawling microglia where the height (from the substrate to the top of the ruffle) is calculated based on the measured quantitative phase delay information acquired by QPI. Clearly, the cell carries long narrow strands (but in irregular shape) of ruffle waves along the leading front of lamellipodium. To obtain the height, width and length of peripheral ruffles originally segmented in 3-D, the cell image is first projected on 2-D with threshold and binarized. Then, peripheral ruffles are assumed to be segmented in rectangular shapes. From these images, the measured height, width and length of peripheral ruffles are 0.40 ± 0.13 μm, 0.39 ± 0.09 μm and 9.04 ± 2.51 μm (based on 5 cases), respectively. The height of the core area (demarcated by black dotted line) is far beyond the dynamic range of our QPI height measurement of ruffle waves, thus unable to be measured.

These ruffle waves are periodically generated and collapsed by folding back onto the cell body with a period of about 6 seconds as illustrated in Fig 2B (see Visualization 2). In addition to the temporal height modulation, the overall height is gradually increased from 0.1 to 0.5 μm within four consecutive waves. This slight increment is simply due to the fact that the cell has advanced a little (along the direction marked by a white arrow) during our observation.

The oscillatory behavior of peripheral ruffles is also apparent in the kymograph of Fig 2C taken along the center horizontal (white dotted) line inside the red-boxed area in Fig 2A. The lamellipodial boundary oscillation (guided by the yellow dotted line shown in the kymograph of Fig 2C) is in a synchrony with the ruffle height oscillation (green dotted line shown in Fig 2B). The peaks (in both Fig 2B and 2C) show up at 6, 12, and 18 seconds, approximately. The height at peaks and the average oscillating period of peripheral ruffles are 0.40 ± 0.13 μm and 6.45 ± 0.44 seconds, respectively (based on 6 different sampled area along the leading edges of 4 different cells).

We also obtain 3-D refractive index (RI) tomogram as shown in Fig 2D (see Visualization 3) using a standard tomographic algorithm [18, 20]. The tomogram, the isosurface image of RI, also well illustrates that peripheral ruffles (the region bounded by yellow dotted line in Fig 2D) are active not over the entire front area, but only in part of it. Tomogram viewing angles are -137° and 84°, azimuth and elevation, respectively. In addition, the features of peripheral ruffle waves visualized by our QPI technique are compared with those obtained with other imaging modalities such as SEM and fluorescence microscope. The SEM image of Fig 2E clearly shows surface protrusion of these peripheral ruffles (see the area bounded by yellow dotted line) as well as some internal cytoskeletal network structures. The fluorescent image of Fig 2F further confirms similar ruffle structures that are rich in F-actin fibers (see the area bounded by yellow dotted line). Taken all images and movies acquired through four different imaging modalities together, we could conclude that peripheral ruffle waves are closely associated with actin polymerization/depolymerization taking place at the moving front. They only form in the leading edge of the front and oscillate with an approximate period of 6 seconds.

Periodic dorsal waves analogous to forward migrating features in crawling cells

Our QPI data also show that, from the lamellipodium just behind the leading front, where peripheral ruffle waves exist, the travelling bands that are high in F-actin emanate once in a while and propagate toward the core of the crawling cell. We name them “dorsal waves”. Two snapshot images acquired through a PCM are shown in Fig 3A, depicting such a wave while the cell is moving (see Visualization 4). Dorsal waves are generated repeatedly while the cell body is in motion. In fact, the time series in Fig 3B well illustrates a periodic nature of the dorsal wave propagation with a period of about 2 minutes (1.91 ± 0.87 minutes). Dorsal wave ‘packets’ vary in their shapes as they are created at the leading edge and travel from different directions toward the cell centroid. Therefore, the time series of average intensity (in Fig 3B) acquired at the marked location (red bounded box in Fig 3A) fluctuates unevenly. On the other hand, by tracing the centroid position of the crawling cell (see in Ref. 5), the travelling distance is acquired, and then each frame of cell image is aligned. A typical example is shown in Fig 3C. Interestingly, the microglia move in a pulsatile fashion, alternating between a fast-run and a slow-run approximately every 2 minutes (as a period of 1.82 ± 0.96 minutes). The cell body moving speed vary and oscillate periodically with the fast-run at 10 μm/min and the slow-run at 2 μm/min. These pulsatile movements (fast-run and slow-run) of cell appears to be in a time span similar to the period of dorsal waves. Subsequently, to quantify the relationship of these two oscillatory phenomena, we calculate the cross-correlation function as the following equation:

(1) where N is the total number of samples and τ is the time delay between the two signals, the speed of cell body x_i and the intensity of observed ruffle waves at a single position y_i, 〈x〉, (or 〈y〉) and σ_x (or σ_y) are the mean and standard deviation of x_i (or y_i). As shown in Fig 3D, the correlation function has a primary peak (marked by a blue star) at around τ = 1.4 minutes, meaning the dorsal wave propagation (at the red-boxed area in Fig 3A) advances the movement of the cell body for about 1.4 minutes (the average τ is at 1.38 ± 0.14 minutes based on 4 different samples). Also, the amplitudes of the cross-correlation function do not show the significant decay as a function of time, implying the order relationship of the two waves stands not only for a single event but rather maintains with a repeating cycle. From this result, it suggests that the cell body movement seems to correlate with dorsal actin wave dynamics, or at least the migration of the cells during their “run” stage shows correlation with the dorsal waves.

As done for peripheral ruffle waves in previous section, we use several different imaging modalities to quantify these dorsal waves supported by crawling MG cells. First, we use a QPI technique in conjunction with SAM to obtain heat maps of the height of the dorsal surface shown in Fig 4A. Clearly, these two heat maps suggest the formation of a dorsal wave (indicated by yellow arrows) near the leading front and its propagation toward the core region. Height profiles of the dorsal membrane along the direction of its propagation are plotted in Fig 4B at two different time points t = 10 and 11 minutes, with red and blue lines, respectively. Two black dotted lines show smoothed curves using a low-pass filter with 6 points and the green dotted line represents the cell boundary. The propagation of the dorsal wave towards the cell core is rather clear.

For this cell, the dorsal waves are estimated the propagation speed of 2.5 ± 0.7 μm/min (based on 5 cases). And the height, width (FWHM) and length of dorsal waves are also obtained as 0.48 ± 0.05 μm, 7.37 ± 0.47 μm, and 46.78 ± 8.77 μm (based on 5 cases), respectively. Thus, dorsal waves are far broader than peripheral ruffle waves. A 3-D rendering of RI tomogram shown in Fig 4C, obtained by ODT using a tomographic algorithm (see Visualization 5). It also confirms the existence of a dorsal wave (encircled by a black dashed line and marked by a yellow arrow). The existence of dorsal waves is also evident in the SEM image shown in Fig 4D.

In addition, we stained a sample of MG cell to identify F-actin distribution in association with dorsal waves. As seen in the peripheral ruffle waves at the leading edge, dorsal waves are also rich in F-actin (red color) as shown in Fig 4E. It is known that actin-rich region within lamellipodia can arise as dissociated actin monomers are recruited and pulled toward cell body by myosin-based contractile force [34, 35].

Synchronized peripheral ruffles waves existing at the leading edge in turning-inhibited cells

According to the cross-correlation analysis discussed in the previous section, it is rather clear that the run stage of MG cell crawling is correlated with periodic dorsal actin waves. To understand more clearly about the relationship between cell body movements and “dorsal waves”, we try to inhibit the turning events. Unfortunately, at this moment we do not know how to manipulate only the position of “peripheral ruffles” locally around the cell perimeter without disrupting other cytoskeletal components that are involved in the cell motility, specifically in our case “runs” and “turns” of cells (in Figs 2F and 4E). The usage of any drugs or chemicals known to either block or stimulate the actin dynamics would not have a specific control for these “runs” and “turns” of cells. Thus, we decide to take the drug/chemical nonintrusive route where the micro-patterned 1-D track (defined by a 3-D PDMS groove: ridge height 4 μm, track width 20 μm; see Fig 5A) is used to make turning-inhibited cells with physical confinement. Fortunately, fully physically confined cells by the track width indeed go straight without any turns (see Visualization 6). We demonstrate the peripheral ruffles do exist in the crawling direction of the cell even when the cell can only move to one direction. Here, ‘no turn’ simply means ‘no biased localization of peripheral ruffles’, not the peripheral ruffles completely disappearing from the cell.

With the turning events being prevented, it is impossible to observe the active zone of peripheral ruffles showing any bias (either to the right or left side) of the cell. But the peripheral ruffles still exist at the leading front of the crawling cell and seem to be spatial-temporally synchronized along the active zone of peripheral ruffles. In other words, there are no travelling back and forth (as biased) of peripheral ruffles along the leading front. In Fig 5E, the kymographs are taken from left, middle, and right sides along the leading edge indicated by green lines (as L, M, and R in Fig 5A, and the yellow star marks in Fig 5A and 5E represent the same location). To calculate the cross-correlation functions of ruffling dynamics at the leading edge, we trace the edge position in kymograph using the binarized and skeletonized image analysis of ImageJ. The cross-correlation functions of peripheral ruffles at the leading edge are calculated: left-middle (red line), left-right (green dotted line), and middle-right (blue dotted line) in Fig 5F. The primary peak shows up at 0 delay time, indicating when the cells crawling forward, peripheral ruffles are formed concurrently at the leading edge positions (L, M, and R), expressed as the term “phase-coherent”.

Interestingly, there are two waves with similar period of about 2 minutes in the 1-D track: the dorsal waves (1.89 ± 0.52 min) and the periodic cell movement (1.80 ± 0.72 min). And, we quantify the order relationship of these two oscillatory series (shown in Fig 5B and 5C). In Fig 5D, the correlation function has a primary peak (marked by a blue star) at around τ = 1.2 minutes (1.19 ± 0.59 min, based on 4 cases), meaning the dorsal wave propagation (at the red-boxed area in Fig 5A) precede the movement of the cell body for about 1.2 minutes. Also, the amplitudes of the cross-correlation function do not show the significant decay as a function of time, implying the order relationship of the two waves is not considered as a single event but rather maintains with a repeating cycle. With respect to the time delay τ, there is no significant difference between the results from 1-D (turning-inhibited, τ = 1.2) and 2-D (with turning, τ = 1.3). From these results (in 1-D and 2-D), the cell body movement seems to correlate with dorsal actin wave dynamics, or at least correlated with the migration of the cells during their “run” stage.