Architecture-driven quantitative nanoscopy maps cytoskeleton remodeling

Significance

An imaging tool with high spatiotemporal resolution and computational sensitivity has the potential to uncover cytoskeleton remodeling which plays a key role in cell migration. Herein, we proposed an architecture-driven quantitative (ADQ) framework in spatiotemporal-angular hyperspace, based on a multimodal superresolution imaging system, to enable identification of the optimal imaging mode with well-balanced fidelity and phototoxicity and postcharacterization of microtubule remodeling. The ADQ framework relied on a metric reflecting heterogeneous intertubule alignment with improved sensitivity over previous efforts and its time-dependent features to map dynamic microtubule rearrangements. The ADQ framework successfully revealed distinct polarization patterns of microtubule remodeling from two migration modes and exhibited potential in predicting migration trajectory.

Abstract

Cytoskeleton remodeling which generates force and orchestrates signaling and trafficking to govern cell migration remains poorly understood, partly due to a lack of an investigation tool with high system flexibility, spatiotemporal resolution, and computational sensitivity. Herein, we developed a multimodal superresolution imaging system–based architecture-driven quantitative (ADQ) framework in spatiotemporal-angular hyperspace to enable both identification of the optimal imaging mode with well-balanced fidelity and phototoxicity and accurate postcharacterization of microtubule remodeling. In the ADQ framework, a pixel/voxel-wise metric reflecting heterogeneous intertubule alignment was proposed with improved sensitivity over previous efforts and further incorporated with temporal features to map dynamic microtubule rearrangements. The ADQ framework was verified by assessing microtubule remodeling in drug-induced (de)polymerization, lysosome transport, and migration. Different remodeling patterns from two migration modes were successfully revealed by the ADQ framework, with a front–rear polarization for individual directed migration and a contact site–centered polarization for cell–cell interaction-induced migration in an immune response model. Meanwhile, these migration modes were found to have consistent orientation changes, which exhibited the potential of predicting migration trajectory.

Cell migration is a highly integrated multistep process that plays a key role in development, wound healing, and immune system function (1, 2) and drives disease progression in cancer, mental retardation, atherosclerosis, and arthritis (1). Abundant knowledge in terms of cell migration has been achieved either in signaling and trafficking at the molecular level (3) or in cell movements and shape changes based on coarse-grain observations at the cellular level (4). However, it remains unclear how intracellular organelle and cytoskeleton remodel themselves independently or mutually and how these structure rearrangements affect molecular activities or large-scale behaviors in migration at the subcellular level, partly due to a lack of an investigation tool with high spatiotemporal resolution, system flexibility, and, especially, computational sensitivity to visualize and characterize remodeling activities. As a cytoskeletal component, microtubules have not been heavily studied as actin filaments regarding their roles in migration (5); however, recent works have suggested that microtubules, similar to actin filaments, can generate forces in the way of organized rearrangements to coordinate essential migration steps (6, 7). Therefore, resolving structural remodeling and dynamic evolution of microtubules may shed light on underlying physicomechanical mechanisms of migration and enable a better understanding of disease initiation and development.

As an established imaging tool, superresolution fluorescence microscopy (also known as fluorescence nanoscopy) has overcome the diffraction barrier to visualize ultrafine structures of intracellular components (8–10). However, although a wide range of superresolution variants have been proposed, all techniques excel in certain sides and may reach an impasse in others. For instance, three-dimensional (3D) structured illumination microscopy (SIM) has shown the capacity of whole-cell imaging, despite the lower acquisition speed than two-dimensional (2D) SIM and the limited axial resolution (11). We have developed multiangle interference microscopy (MAIM) that incorporates SIM with multiangle evanescent light illumination to achieve improved axial and temporal resolution for volumetric imaging, while at the cost of limited imaging depth (12). Predictably, a possible way to overcome the contradiction is multimodal imaging, which can take the advantage of information complementarity from multiple techniques or recognize the optimal imaging mode for a particular question (13, 14).

Although high-resolution microtubule images become more readily accessible, computational analysis method which is able to resolve sophisticated architecture and dynamic rearrangements of microtubules with high sensitivity is lacking to uncover underlying mechanisms that drive migration. We have previously developed a weighted vector summation algorithm which can extract voxel-wise 3D orientations of fiber-like structures (such as microtubules) with improved determination accuracy and time efficiency over existing approaches (15). Nevertheless, since orientation is a relative measure with reference to a coordinate system (16), a coordinate-independent metric is needed for a better understanding of the interfiber heterogeneous alignment of biological tissues.

In this study, we established a multimodal imaging system by integrating wide-field imaging with four SIM modes and developed an architecture-driven quantitative (ADQ) framework which enabled both identification of the optimal SIM mode for a specific application and accurate postcharacterization to resolve microtubule remodeling. A coordinate-independent metric, termed order index (OI), was proposed to uncover sophisticated microtubule architecture in a pixel/voxel-wise context and further incorporated with temporal features to map dynamic rearrangements. Besides, the direction of orientation change (DOC) from the initial to the end time point at distinct regions was extracted to investigate the relationship among local regions and its connection to migration trajectory. The ADQ framework was verified by resolving microtubule remodeling in drug-induced (de)polymerization, lysosome transport, and migration. Finally, we revealed differences and similarities in remodeling heat maps and orientation change schemes of two migration modes.

Results

ADQ Framework for Cytoskeletal Imaging and Architectural Mapping.

The ADQ framework was established on a multimodal imaging system (Fig. 1A) which integrated low-resolution but fast wide-field imaging with four superresolution SIM modes including total internal reflection fluorescence (TIRF)-SIM, 2D-SIM, 3D-SIM, and MAIM. Although the requirements for the incident angle, azimuth, polarization direction, and phase of incident light were different among these modes (SI Appendix, Materials and Methods), their main configurations were basically identical (SI Appendix, Fig. S1), facilitating instant switching among them.

Fig. 1.

Overview of the ADQ nanoscopy framework. (A) Optical scheme of the multimodal imaging system (Top), along with the schematic of the back focal plane of the objective corresponding to different acquisition modes (Bottom). The multicolor illumination laser is split into four channels modulated separately by scanning modules to switch the imaging mode (SI Appendix, Fig. S1). Illumination positions of the incident light are demonstrated for each mode at the back focal plane. The outer solid and inner dotted circles represent the maximum and TIRF illumination positions that incident light can reach, respectively. (B) ADQ acquisition concept. The azimuthal angle θ and the polar angle ϕ are needed to depict an orientation in 3D space. Wide-field image stack is acquired first for coarse angular preanalysis which extracts FWHM of ϕ distribution, defined as IF to determine the optimal superresolution mode. (Scale bar, 10 μm.) (C) Superresolved images are further finely analyzed to obtain quantitative orientation and alignment information of microtubule remodeling in a static or time-dependent manner, in either 2D or 3D context. WF: wide-field, SR: superresolution. (Scale bar, 3 μm.) (D) Simulated fiber stack (400 × 400 × 400 voxel in size) used to verify the OI metric in quantifying fiber alignment. The θ (Top) and ϕ (Bottom) orientation maps of two marked regions (40 × 40 × 40 voxel in size) with different levels of alignment are shown. (E) OI maps and corresponding distributions within these two regions.

To determine the most appropriate superresolution mode, the wide-field stack of the cellular cytoskeleton was first obtained and analyzed using a fast angular processing strategy to acquire in real time the distribution of the polar angle, ϕ, which reflected how microtubules stacked up in 3D space (Fig. 1B). A suitable wavelength from the three (488 nm, 561 nm, and 640 nm) was used for the acquisition of cytoskeleton image, dependent on the fluorescent labels. An imaging criterion was developed based on the full width at half maximum (FWHM) of ϕ distribution, termed inflated factor (IF), which guided the following decision of the proper imaging mode. IF was dependent on how microtubules stacked up (i.e., parallel or perpendicular to the cell surface), while not related to cell thickness or shape, nor the wavelength used for image acquisition. Enabled by 2D and 3D approximation under carefully validated situations (SI Appendix, Materials and Methods), this IF criterion guaranteed the sensitivity that can be achieved from 3D assessments whatever 2D/3D superresolution mode was triggered. According to the calculation from simulated fibers (SI Appendix, Materials and Methods), 3D-SIM was needed when IF was larger than a threshold of 12 degrees, while 2D imaging modes were appropriate at IF below 12 degrees when relatively flattened or layered cytoskeletal structure was formed.

Following the acquisition of superresolution images, a computational method was developed to extract the morphological orientation and alignment information of cytoskeletal components, and the time-dependent information was used to uncover mechanisms behind the dynamic cellular processes, such as microtubule remodeling (Fig. 1C and Movie S1). Specifically, we first extracted voxel-wise 3D orientations of fibrous structures (15) and then developed a coordinate-independent optical metric, termed OI, which directly reflected the heterogeneous alignment level of cytoskeletal components by extracting interfiber orientation relationship (SI Appendix, Materials and Methods). OI ranged between 0 and 1, with 0 corresponding to complete randomness and 1 referring to perfectly parallel alignment. To illustrate how OI worked, we quantified two fiber regions (Fig. 1D) with different levels of alignment. As evident from OI maps and corresponding distributions (Fig. 1E), OI was a direct yet informative measure of fibrous alignment. We proved that the OI algorithm was robust against the image degradation by background noise (SI Appendix, Fig. S2 A–F), and the influence of fiber density and waviness could be minimal (SI Appendix, Fig. S2 G–J). Then, we compared the OI metric with the previous order parameter (OP) metric (17), which quantified the overall alignment of a bunch of fibers without offering insights into interfiber relationship. Therefore, for the two simulated fibrous stacks with the same components but different organization (SI Appendix, Fig. S3), only the OI metric was able to recognize the difference, highlighting our methods as a sensitive tool for resolving heterogeneous fibrous organization. Furthermore, we validated that our orientation and OI algorithms were robust against different fluorescent labels and wavelengths used for excitation (SI Appendix, Fig. S4), and the influence from potential cross talk between color channels could be neglected since the architecture of microtubules and actin filaments, corresponding to different channels, was accurately resolved (SI Appendix, Fig. S5).

Verifying the ADQ Framework from Assessments of Microtubule Dynamicity.

We investigated the performance of algorithms in ADQ by first assessing microtubule remodeling induced from two agents, Taxol and Nocodazole (Fig. 2A), which were expected to alter microtubule dynamic instability through polymerization (18) and depolymerization (19), respectively. We conducted time-lapse TIRF-SIM imaging on cells with IF below 12 degrees to guarantee time-efficient acquisition for minimum disruption since the microtubule remodeling triggered by the agents was subtle and at the second level (18, 19). To extract the dynamic evolution of microtubule organization, we incorporated temporal features with OI and developed the OI variation quantity (Fig. 2B and SI Appendix, Fig. S6). Briefly, the cell image was divided into a set of local regions, and for each region, the SD of OI values throughout distinct time points was calculated to obtain OI variation, which was then pseudocolored to make the heat map. Apparently, bluer hues in representative OI variation maps were visualized from agent-treated groups (Fig. 2C), indicating a potential inhibition of microtubule dynamicity. These observations were further validated by statistical analysis (Fig. 2 D, Left). To avoid overestimation, the statistical analysis of OI variation was region based, rather than pixel based. Interestingly, although both agents led to reduced microtubule dynamics, OI assessments were able to further tell the subtle differences between them (Fig. 2 D, Left). However, such differences could hardly be recognized by the previous OP metric (Fig. 2 D, Right), highlighting the performance of our method in evaluating subtle cytoskeleton remodeling and drug efficacy.

Fig. 2.

Verifying the ADQ framework from assessments of microtubule dynamic activities. (A) Schematic showing the polymerization and depolymerization of microtubules induced by Taxol (1 μM) and Nocodazole (5 μM), respectively. (B) Schematic showing the generation of OI variation heatmap, which reflects regional microtubule remodeling as a function of time. (C) Representative OI maps (Top) and OI variation heat maps (Bottom) of U2OS cells under different treatments, acquired from TIRF-SIM. (Scale bar, 20 μm.) (D) Box plot of mean OI variation (Left) and previous OP variation (Right). n = 10 cells for control and Nocodazole groups, and n = 12 cells for Taxol group. Cross-validated classification accuracy in distinguishing different groups is marked in each plot. C: control; N: Nocodazole; T: Taxol. (E) Representative intensity images of microtubule (color-coded by depth) and lysosome (gray spots) from U2OS cells, acquired by 3D-SIM. (Scale bar, 20 μm.) (F) The representative OI map at a certain time point from 3D analysis. (Scale bar, 3 μm.) (G) Corresponding time-dependent OI variation heat map, obtained from 400 s time duration at 5 s intervals. (Scale bar, 3 μm.) (H) Relationship between the lysosome moving distance and OI variation, with correlation coefficient R and P values marked in the plot. (I) Box plot of OI variation at regions with and without lysosomes, obtained from 2D (Left) and 3D (Right) analysis. A local region is recognized as “with” as long as lysosomes pass by during the imaging session. n = 14 cells.

Then, we assessed the interaction between microtubules and lysosomes to gain insights into how microtubule transport activities affected in turn the architecture. To capture the transport throughout the cell depth, especially in a 3D context, we selected U2OS cells following the IF criterion of 3D-SIM to investigate the microtubule–lysosome interaction over 400 s time session (representative depth-coded images shown in Fig. 2E). 3D OI maps at distinct time points were generated (Fig. 2F), along with corresponding time-dependent OI variation heat map (Fig. 2G and Movie S2). Then, the relationship between OI variation and the lysosome moving distance (the total length of the moving trajectory) within each local region was assessed from 14 cells, and a significant correlation was obtained (Fig. 2H), suggesting that microtubule remodeling was interaction related and associated with lysosome traveling distances. According to the relationship, we further compared the OI variation between regions with and without lysosomes, quantified by both 2D (Fig. 2 I, Left) and 3D (Fig. 2 I, Right) analysis. We found that local microtubule regions with lysosomes traveling through exhibited a higher level of remodeling. Significant differences could be recognized by 3D analysis only, indicating that 2D imaging and analysis were not sufficiently sensitive to resolve the architecture of cells falling into the 3D IF range.

Resolving 3D Architectural Differences for Round and Polarized Cells.

We then focused on extracting microtubule alignment during the process of cell migration. To understand the microtubule architecture across a large depth scale, we started from quantitative volume imaging of fixed U2OS cells by 3D-SIM. Cells with round (Fig. 3A) and polarized (Fig. 3B) features were typically associated with a relatively static and migrating phase, respectively, and a local region along the cell periphery in both cases (indicated in Insets) was analyzed. Polarized cells generally exhibited more uniform depth-coded hues (Fig. 3B) compared with varying hues of round cells (Fig. 3A), revealing more flattened microtubule structure, especially at protrusions. The more consistent hues were also observed from both θ and ϕ maps of polarized cells. These observations were supported by a single peak in either θ (Fig. 3C) or ϕ (Fig. 3D) distributions for polarized cells, probably reflecting the microtubule bundling at a directed orientation typically visualized at the front edge (20). We found that this inclination of ϕ orientation approaching 90 degrees (i.e., parallel to the substrate plane) in polarized cells also applied to other regions besides front protrusions, indicating that the whole microtubule network tended to become more flattened than round ones. Overall, microtubule images at the protrusions from 17 polarized cells and randomly selected cell periphery regions from 15 round cells were harvested and quantitatively characterized (representative OI maps shown in Fig. 3E). Consistent with OI distributions where a peak at higher values was visualized from the polarized phase in contrast to the round phase (Fig. 3F), a significantly higher level of OI was obtained from polarized cells (Fig. 3G). Moreover, we found that round cells typically exhibited a higher level of heterogeneity in OI distribution (Fig. 3H), as represented by the SD of OI values among all the effective voxels from each image. Compared with round cells, the similarity in ϕ orientation along the depth dimension in polarized cells even led to similar accuracy level in θ determination by truly 3D analysis or simplified 2D analysis at a certain depth (Fig. 3 I and J), as validated by significantly lower error in orientation assessed from polarized cells than that from round cells (Fig. 3K).

Fig. 3.

3D microtubule architecture of cells with round or polarized morphology. (A) U2OS cells with round morphology are imaged using 3D-SIM for orientation and alignment assessment. Based on the 3D intensity image (color-coded by depth), the θ, ϕ and OI maps from Left to Right are obtained accordingly. The Inset within the intensity image demonstrates the whole cell and the representative field selected. (Scale bar, 2 μm.) (B) Depth-coded intensity image of U2OS cells with polarized morphology, along with the θ, ϕ and OI maps. (Scale bar, 2 μm.) (C and D) Distributions of θ (C) and ϕ (D) orientations for round and polarized cells. The ϕ orientation is highly centered in polarized cells at around 90 degrees, revealing that these microtubules aligned along the optical section, i.e., parallel to the substrate plane. (E) Representative OI maps from randomly selected periphery regions of round cells (Top) and protrusions of polarized cells (Bottom). (Scale bar, 2 μm.) (F) Comparison of OI distributions. (G) Box plot of OI values from 15 round and 17 polarized cells. (H) Box plot of heterogeneity in OI distribution. (I) For a marked region (shown in the second row in A) within round cells, the calculated θ distributions have obvious differences as obtained from 3D and 2D analysis (shown from the schematic). White arrows point to locations with different orientation outputs. (J) For polarized cells, the calculated θ distributions from 3D and 2D analysis are identical. (K) Comparison of the θ quantification difference between 2D and 3D analysis from round and polarized cells, calculated as the mean distribution difference in histograms at all the distinct orientations.

2D and 3D Analyses Lead to Identical Time-Dependent OI Trends for Migrating Cells.

After establishing microtubule architecture from fixed cells, we then performed live-cell imaging to obtain time-dependent remodeling patterns. Dynamics of cells were monitored to confirm migration, and then, superresolution images of microtubules were collected for a certain time session. From the representative migrating U2OS cell image (Fig. 4A), the protrusive front edge at the top right location could be recognized via the organization of microtubules in an elongated, parallel array. To compare the difference between 2D and 3D strategy in elucidating microtubule alignment, as well as in time efficiency regarding imaging and analysis, we acquired the entire cellular image by TIRF-SIM, since the calculated IF was 6.1 degrees, while performing 3D reconstruction of a local region (region 1) by MAIM across a 45 s time session. TIRF-SIM and MAIM recordings were done sequentially as TIRF-SIM was the prestep of MAIM, with their acquisition speeds determined by the number and exposure time of acquired raw images. The exposure time of both modes was 100 ms, while the number of raw images was 9 for TIRF-SIM and 29 for MAIM. Based on the MAIM images, we generated the evolution of both θ and ϕ distributions (SI Appendix, Fig. S7), where the subtle changes in θ orientation as a function of time could be visualized. Although slight changes in ϕ orientation occurred as well, it was highly centered around 90 degrees, revealing that these microtubules aligned consistently parallel to the substrate plane. We hypothesized that this specific ϕ distribution, typical in migrating cells, would diminish its influence to the determination of OI and tested this hypothesis by extracting OI of region 1 throughout the assessing time session from both 2D and 3D analyses (Fig. 4B). As expected, 2D and 3D OI followed identical trends (Fig. 4C), despite the differences in absolute values. This finding revealed that it might be sufficient to apply 2D superresolution mode in combination with 2D OI algorithm to cells, particularly at a migrating phase, since they faithfully resembled their 3D counterparts in resolving microtubule remodeling with comparable sensitivity, while in a much more time-efficient manner counting both imaging (~1 s vs. ~3 s) and analysis (Fig. 4D). These findings also supported the IF criterion for the ADQ framework.

Fig. 4.

Dynamic and location-dependent microtubule remodeling in live migrating cells. (A) The representative image of microtubules from U2OS cells undergoing migration, with four regions of interest labeled. The entire microtubule images are recorded using TIRF-SIM across 45 s time scale at 9 s intervals, and 3D images of region 1 are obtained using MAIM. (Scale bar, 5 μm.) (B) Comparison of 2D and 3D methods to extract microtubule alignment. In 2D analysis, OI is calculated from θ orientation only. In 3D analysis, OI is extracted from both θ and ϕ orientations. (Scale bar, 1.5 μm.) (C) Although being different in absolute values, the time-dependent OI profiles of migrating cells extracted by 2D and 3D analysis follow identical trends. (D) Comparison of imaging and analysis time between 2D and 3D superresolution modes. (E) OI maps of four marked regions at different time points, with white arrows pointing to locations of varying hues with time. (Scale bar, 1.5 μm.) (F) The time-dependent mean OI values of these regions. The Inset shows the OI variation. (G) The schematic showing the calculation of DOC. (H) MRM is acquired by combining the OI variation heat maps (indicated by the pseudocolor) and DOC (indicated by the arrow) at each local region. The dashed magenta curve shows the ellipse fit of the cellular contour, and the dashed line corresponds to the short axis. (Scale bar, 5 μm.) (I) The relationship between the OI variation and the normalized location calculated from the front to the rear of the cell (shown in the schematic). Data are shown as means ± SD, n = 16 cells. Totally 551 effective local regions are involved in analysis. (J) Duncan’s multiple range test is used to analyze the significance of differences among normalized locations with the threshold of P set to 0.05. Characters on bars indicate significant differences. (K) Distribution of the orientation difference between DOC in each local region and mean DOC of all the effective regions within a cell, calculated from all the effective local regions of all the cells. n = 16 cells.

Front–Rear Polarization of Microtubule Remodeling Revealed in Individual Directed Migration.

To highlight the sensitivity of the ADQ framework, we first focused on short session of migration. Within the time scale below 1 min, typically few changes in cellular contours were observed (SI Appendix, Fig. S8), while dynamic alterations in microtubule architecture continuously took place, as indicated by varying hues in orientation and OI maps as a function of time at four marked regions (Fig. 4E and SI Appendix, Fig. S9). We then obtained the mean OI within these marked regions at distinct time points and found that they fluctuated at different scales (Fig. 4F), which could be characterized by the OI variation measurement. Meanwhile, the DOC from the initial to the end time point was considered as another dynamic microtubule remodeling feature at migrating phase (Fig. 4G). Based on these quantifications, we generated the microtubule remodeling map (MRM) of the entire cell, which composed of two main characteristics, including the OI variation, represented by pseudocolored heat maps, and DOC, represented by the arrows at each local region (Fig. 4H and Movie S3). Observed from this MRM, active microtubule remodeling dominated the front leading edge, as indicated by relatively redder hues compared to the rest, yet hot areas also seemed to occur at the very end of the migrating cell.

To validate these observations, we imaged 16 live migrating U2OS cells across 50 s by TIRF-SIM, generated MRM accordingly, and obtained the relationship between OI variation and normalized location (SI Appendix, Fig. S10) from the front to the rear of the cell (Fig. 4I), along with statistical analysis results from Duncan’s multiple range test (Fig. 4J). These results confirmed the observations from the representative MRM of migrating cells, where hot areas with intense microtubule remodeling mainly appeared at leading and rear sides in contrast to the interior region, while the protrusive leading edge still exhibited a significantly higher level than the rear. Interestingly and importantly, we found that DOC was identical among these local regions, and the difference between DOC in each local region and the mean DOC of all the regions was distributed with a narrow peak around 0 degrees (Fig. 4K).

Migration Trajectory Predicted by the ADQ Framework in Individual Directed Migration.

Then, we assessed microtubule remodeling in long-term migration at the hour-level time scale. Genome-edited retinal pigment epithelial (RPE) cells expressing mTagRFPt-α-tubulin formed monolayers and were scratched with a pipette tip to form a wound. These cells were treated with knockdown of vimentin intermediate filaments (VIFs), which were supposed to template microtubule organization to stabilize cell polarity during directed migration (20). Images of microtubule networks from cells at the wound edge were obtained immediately after scratching for 60 min at a rate of 1 frame per minute, and orientation and OI maps were obtained for each frame (Fig. 5A and Movie S4). As can be seen from time-dependent images, microtubule networks were less oriented toward the direction of migration than in usual cases due to VIF knockdown. Representative MRMs were obtained for three certain time periods (Fig. 5B). With a front–rear polarization in OI variation, these maps exhibited consistent patterns as the one from the short-term period within 1 min (Fig. 4H), indicating that our method was sensitive enough to uncover microtubule remodeling in migration from short-term data. It was worth mentioning that besides front protrusions, intense microtubule remodeling was also found at some extensions (pointed by red arrows in the third MRM in Fig. 5B). The relationship between OI variation and normalized location from the front to the rear of the cell (Fig. 5C) and corresponding statistical analysis results from Duncan’s multiple range test (Fig. 5D) further validated the front–rear polarization. Importantly, we found that the DOC metric, each calculated over 2 min time scale, was able to predict the migration direction (Fig. 5E), as evident from a good coincidence between the calculated DOC (yellow arrow) and the fitted migration trajectory (magenta curve), where the cell location was represented by the center of the ellipse fitted to the cell nucleus. Quantified from 80 distinct time periods along the trajectory of 4 RPE cells, the mean absolute difference and SD between DOC and migration direction (calculated as the tangent of the fitted trajectory) were only 7.8 ± 4.9 degrees.

Fig. 5.

Dynamic microtubule remodeling in long-term individual migration. (A) Time-dependent fluorescence images of mTagRFPt-labeled microtubules from RPE monolayers are captured immediately after scratching. Representative images are shown including cell contour (row 1), microtubule intensity (row 2), orientation map (row 3), and OI map (row 4) at distinct time points. (Scale bar, 20 μm.) (B) Representative MRMs are acquired each from 2 min (with distinct time periods illustrated on Top Left of each MRM). Red arrows point to local regions with intense microtubule remodeling. (Scale bar, 5 μm.) (C) The relationship between the OI variation and the normalized location calculated from the front to the rear of the cell. Data are shown as means ± SD, n = 4 cells, with three distinct time periods randomly selected from each cell during the 60 min duration. Totally 304 effective local regions are involved in analysis. (D) The significance of differences for OI variation at distinct normalized locations is acquired by Duncan’s multiple range test and labeled on the bars, with the threshold of P set to 0.05. (E) Representative trajectory (fitted by magenta curve) of cell migration and calculated DOC (yellow arrow) at distinct time points.

Contact Site–Centered Polarization of Microtubule Remodeling Revealed in Cell–Cell Interaction-Induced Migration.

Following individual directed migration, we then mapped the kinetic changes in cell–cell interaction-induced migration. Specifically, microtubule images were harvested from bone marrow–derived dendritic cells (SI Appendix, Fig. S11) and splenic CD4 T lymphocytes of allogenic mouse strains (Fig. 6A) when they moved toward each other and established physical contact and mutual interactions required for allogenic antigen presentation, T cell priming, maturation of dendritic cells, and the subsequent immune responses (21, 22). Only microtubules of dendritic cells were clearly resolved by superresolution microscopy since they were adherent to the culture plate bottom while T cells were typically nonadherent, appearing only as a small spot in the acquired image (SI Appendix, Fig. S12), with the spot centroid designated as the contact site (labeled by the green point in Fig. 6). Owing to inherent morphological features and the interaction-induced elongation/flattening of microtubule network in dendritic cells, time-lapse TIRF-SIM images were collected, in accordance to the IF criterion. The θ orientation (Fig. 6B) and corresponding OI maps (Fig. 6C) at the initial time point of a 40 s imaging session from three representative cells are shown, along with the MRM (Fig. 6D) generated from all the discrete time points over this session (SI Appendix, Fig. S13). During this period, only slight movements of T cells were observed from corresponding trajectories (SI Appendix, Fig. S14), while highly heterogeneous microtubule remodeling took place as evident from a variety of hues among different local regions (Fig. 6D).

Fig. 6.

The MRM of dendritic cells reveals the interaction mechanism between dendritic cells and T cells mimicking immune responses. (A) The schematic showing the experimental design. (B) θ orientation maps from three representative dendritic cells. TIRF-SIM images of microtubules labeled by Alexa Fluor 647 are captured dynamically over 40 s time scale at 5 s intervals. The raw intensity image of microtubules is shown in the Inset (Top Left), and the contact site between the dendritic cell and the T cell is labeled by the green point. (Scale bar, 8 μm.) (C) OI maps of these dendritic cells. (Scale bar, 8 μm.) (D) MRM of these dendritic cells, calculated from images throughout all the time points. The arrow in each local region (9.5 × 9.5 μm in size) demonstrates the DOC of θ, and the blue arrow in the Inset shows the mean DOC of all the effective local regions. (Scale bar, 8 μm.) (E) The relationship between the OI variation and the normalized location away from the contact site. The dashed line corresponds to the linear fit of data. Data are shown as means ± SD, n = 15 cells. Totally, 497 effective local regions are involved in analysis. (F) The significance of differences for OI variation at distinct normalized locations is acquired by Duncan’s multiple range test and labeled on the bars, with the threshold of P set to 0.05. (G) The schematic of OI variation characteristics for the individual directed migration of U2OS/RPE cells and the cell–cell interaction-induced migration of dendritic cells. (H) Distribution of the orientation difference between DOC in each local region and mean DOC of all the effective regions within a cell, calculated from all the dendritic cells interacting with T cells. n = 15 dendritic cells. (I) Comparison of the variance in DOC among different local regions between noninteracting and interacting cells. There are 15 interacting cells and 12 noninteracting cells assessed. (J) Representative trajectory (fitted by magenta curve) of dendritic cell migration and calculated DOC (yellow arrow, each calculated from 40 s time duration) at distinct time points.

From OI variation heat maps (Fig. 6D), we found that local regions closer to the contact site typically exhibited more intense microtubule remodeling. To confirm this finding, we plotted the relationship between OI variation and normalized location (SI Appendix, Fig. S15) away from the contact site from 15 dendritic cells in interaction with T cells, and found a decreased trend as the distance increased (Fig. 6E), which was further statistically validated using Duncan’s multiple range test (Fig. 6F). These results indicated that the contact site–centered OI variation in cell–cell interaction-induced migration was different from the front–rear polarization of individual directed migration (Fig. 6G).

From DOC (Fig. 6D), we observed a high level of consistency among distinct local regions. As expected, the distribution of orientation difference between DOC at each local region and the mean DOC of all the regions was narrowly centered around 0 degrees (Fig. 6H), similar to what we observed from individual directed migration. We then compared the variance of DOC values among local regions from 15 dendritic cells interacting with T cells and 12 noninteracting dendritic cells (Fig. 6I). A higher level of variance in DOC acquired from noninteracting cells indicated that microtubules tended to act more uniformly when migration was triggered by interaction between immune cells. To assess the relationship between DOC and migration direction, we performed relatively long-term imaging lasting 8 min when obvious movement of dendritic cells was observed. Totally seven dendritic cells interacting with T cells were involved in analysis, and we observed similar contact site–centered polarization of OI variation as the above-mentioned short-term case. Interestingly, we found that for dendritic cells, DOC (each calculated from 40 s time duration) was nearly perpendicular to migration direction, as can be seen from the representative trajectory graph (Fig. 6J). Quantified from 42 distinct time periods along the trajectory of these seven dendritic cells, the mean absolute difference and SD between DOC and migration direction were 82.5 ± 4.0 degrees.

Discussion

In this study, we have developed the ADQ framework which enables coarse preanalysis of wide-field image stack for rapid identification of the optimal superresolution imaging mode and fine postcharacterization for mapping cytoskeleton reorganization in key biological events, such as drug-induced microtubule (de)polymerization, microtubule-aided organelle transport, and cell migration. This framework is established on a multimodal imaging system integrating the wide-field imaging with four SIM modes which have different merits. To recognize the suitable imaging mode with balanced phototoxicity and resolving accuracy for whole-cell microtubule remodeling assessment, we develop the IF criterion based on the FWHM of polar angle distribution and validate this criterion by demonstrating that 3D imaging and analysis are needed at IF higher than threshold from the lysosome transport experiment (Fig. 2 E–I) and that 2D counterparts could be a suitable alternative at IF below threshold from individual directed migration (Fig. 4 A–D). Benefiting from the optimization from 3D to 2D manner in certain conditions, we are able to achieve comparable accuracy in uncovering microtubule rearrangements with a speed-up factor of 10- to 100-fold. Generally, for flattened or layered microtubule structure, the 2D imaging mode might be an optimal choice due to relatively low phototoxicity and high time efficiency. In contrast, for complicated cytoskeletal architecture as suggested by the IF criterion, 3D mode is necessary to guarantee resolving accuracy, although at the cost of higher phototoxicity and computational burden. The IF-based determination of optimal superresolution mode is currently done at the beginning of imaging while not at every time point of image acquisition, since tests from a variety of migration cases reveal that the probability of IF crossing the threshold during the imaging session is low. However, owing to the millisecond-level IF calculation, our framework enables instant determination and switching of the optimal mode, which might make sense in assessing more complicated cellular activities.

The postcharacterization of superresolution cell images basically relies on the OI metric developed here and its time-dependent features, i.e., OI variation. While OI is extracted from angular information, it itself serves as a coordinate-independent parameter and measures the level of alignment for fibrous structures, including microtubule, actin (SI Appendix, Fig. S5), and other fiber-like subcellular components. An important advantage of OI is that it enables pixel/voxel-wise quantification of structural alterations, in contrast to previous region-based characterization, such as that offered by OP (17). We verify the enhancement in sensitivity of OI over OP via both simulated fiber images (SI Appendix, Fig. S3) and microtubule (de)polymerization experiments (Fig. 2 A–D). Such pixel/voxel-level assessment makes OI suitable for heterogeneous biological events, such as microtubule remodeling. Specifically, due to fluctuations in the length of a dynamic GTP-tubulin cap at the microtubule end, microtubules do not grow steadily but instead elongate at a rate that varies in time which differs among individual tubules and provides a mechanism by which microtubules rapidly accommodate to the changing shape and advancing edge of motile cells (23, 24). Further, previous studies mainly focused on events related to microtubule ends at the margin, such as dynamic instability (25), lamellipodial protrusion (26), and microtubule retraction into the uropod (27), and different mechanisms are associated with changes in microtubule alignment at the cell margin vs. the interior, with the former originating from microtubule growth, fluctuation, and switching while the latter mainly resulting from cytoskeleton reconfiguration (23). In this context, the high sensitivity and ability to resolve inter- and intratubule differences through OI measurements at the whole-cell scale might lead to insights into influences of interior reconfiguration on cell functions, and also the margin-interior relationship that has not yet been studied comprehensively for specific events.

With these technical advances, we find that OI variation heat maps are distinct for the two migration modes under evaluation. In the individual directed migration, we observe polarized distribution of OI variation along the front–rear axis, with a higher variation level present at the protrusive front edge (Fig. 6 G, Left). This distribution pattern is closely associated with the role of microtubules in directed migration. Specifically, the four basic steps of integrin-dependent migration take place along the cellular front–rear polarity axis (28), which should be associated with polarized microtubule functions and ultimately leads to polarized distribution of spatial rearrangements to allow efficient trafficking of molecules (5). In the migration mode induced by immune cell interaction, by contrast, the level of OI variation is associated with the distance away from the contact site (Fig. 6 G, Right). The priming of T cells requires physical contact with dendritic cells, during which signals are exchanged that determine both the magnitude and the quality of the ensuing responses (22). Microtubules appear to be essential for the transport of different signaling microclusters along the membrane, thus facilitating the propagation of signals (29). It is worth mentioning that the mechanism of migration is cell type-dependent. Migration of leukocytes (including dendritic cells) does not rely on integrins in 3D environments, while driven by integrin-independent flowing and squeezing instead (30). However, when moving over 2D surfaces, such as the case in our study, migrating leukocytes use integrin-mediated adhesion (30). Therefore, the differences in OI variation patterns for the two migration modes might reveal that microtubule rearrangements are mainly mediated by the localized distribution of dynamic signaling and transport activities that microtubules participate in. Identifying the site of these activities is important in a variety of contexts, such as understanding cancer initiation at early stages (31), or offering mechanistic insights into how virus attacks cells (32).

We further verify the polarized microtubule remodeling in the individual directed migration mode at different time scales, i.e., as short as minute-level scale when few moving displacements take place (Fig. 4), and as long as hour-level scale with varying migrating direction and cell morphology (Fig. 5). The front–rear polarization of the directed migration mode, observed from the short-term case, is confirmed from the long-term investigation, emphasizing the sensitivity of our OI-based method in uncovering microtubule reorganization and telling the difference among local regions even with cellular contours remaining almost unchanged. Therefore, to minimize photobleaching and photodamage, the rate of imaging and corresponding OI analysis can be optimized, switching between a slow rate while detecting the onset of specific events and a fast rate during their progression, since a common goal of fluorescence microscopy is to collect data on specific biological events (33).

Despite differences in OI variation patterns, these two migration modes exhibit a high consistency in the feature of orientation change, with microtubules at the migrating phase found to have a similar DOC among different local regions. This similarity may facilitate force generation by microtubules to maintain the polarized microtubule network for a persistent migration (5, 6). Interestingly, for wounded monolayers of RPE cells with VIF knockdown, we find that the calculated DOC is highly consistent with migration direction at distinct time points throughout the 1 h assessing period (Fig. 5E), with the average error below 10 degrees. However, in the case of dendritic cell migration, we find that DOC is almost perpendicular to the migration direction throughout the 8-min assessing period (Fig. 6J). A possible insight into this inconsistency is that for RPE cells the VIF knockdown causes microtubule networks to align at a direction deviating from, or even vertical to the migration direction (20), which is different from the usual case. Although future studies are needed to test more treatments, cell types, and migration modes, it is encouraging from our preliminary study that the DOC measurement might be potentially an indicator of or associated with cell fate. From these insights, researchers could manipulate cells into taking one path instead of another, a common goal in biomedical research, and move toward a more quantitative version of single-cell biology (34).

In this study, we mainly focus on cell cultures in vitro, while we expect that the ADQ framework proposed here would potentially guide the quantitative and adaptable SIM imaging toward more complex 3D models, such as 3D-engineered spheroids or organoids. Pilot studies have been done which employed SIM in such 3D models (35, 36). However, the imaging depths of current studies are too limited (typical below 40 μm) to encompass entire 3D models, although techniques have been developed aiming at mitigating the influence of high background and optical aberrations in denser sample regions and at larger depths by combining SIM with reversible fluorescent probe (35, 36), selective plane excitation (36), and adaptive optics (37). Other possible solutions for improving imaging depth further could involve integrating SIM with optical clearing (38), expansion microscopy (39), and adaptive optical lattice light-sheet microscopy (40). As a preliminary study, the range of samples assessed is small. It is also worth mentioning that the IF criterion is currently applied to the whole cell since we quantify differences in microtubule remodeling among local regions of a cell; however, this criterion can also be applied to a specific region, e.g., perinuclear space, which might suggest a different SIM mode for this specific purpose. Moreover, future efforts are needed to tell the differences resolved from 2D-SIM and TIRF-SIM and figure out the necessity and cellular events that can benefit from recognizing one mode from the two. Nevertheless, the ADQ framework developed in this study opens a window for the growing field of smart microscopy techniques (41, 42) and serves as a promising tool for assessments of key events, especially cell migration. Together, our application-oriented research elucidates the significance of an optimal imaging strategy and a computational view for easily overlooked cytoskeleton rearrangements to enlighten biological discoveries.

Materials and Methods

A detailed account of materials, imaging system construction, experimental protocols for all studies, and all additional data are provided in SI Appendix. For each experiment, detailed information regarding cells, fluorescent labels, and imaging modalities is summarized in SI Appendix, Table S1. Imaging was performed on a home-built multimodel system integrating wide-field imaging with four SIM modes, with spatial resolutions of these modes shown in SI Appendix, Table S2. The OI metric was extracted on a pixel/voxel-wise basis, with details of computational algorithms shown in SI Appendix. All animal experiments were conducted in accordance with the institutional guidelines from the Animal Research and Ethics Boards of Zhejiang University.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Comparison of microtubule intensity images and order index maps acquired by wide-field imaging and 3D-SIM.

Movie S2.

Time-dependent order index and order index variation heat maps for microtubules interacting with lysosomes.

Movie S3.

Illustration of order index variation heat map for individual directed migration.

Movie S4.

Representative orientation (left) and order index (right) maps for migration of retinal pigment epithelial cells over 60 min.

Data, Materials, and Software Availability

Custom MATLAB source code for calculation of OI data have been deposited in Github (https://github.com/liuzhiyi07/Order-index) (43). All other data are included in the manuscript and/or supporting information.