Expansion Microscopy for Imaging the Cell–Material Interface

Abstract

Surface topography on the scale of tens of nanometers to several micrometers substantially affects cell adhesion, migration, and differentiation. Recent studies using electron microscopy and super-resolution microscopy provide insight into how cells interact with surface nanotopography; however, the complex sample preparation and expensive imaging equipment required for these methods makes them not easily accessible. Expansion microscopy (ExM) is an affordable approach to image beyond the diffraction limit, but ExM cannot be readily applied to image the cell–material interface as most materials do not expand. Here, we develop a protocol that allows the use of ExM to resolve the cell–material interface with high resolution. We apply the technique to image the interface between U2OS cells and nanostructured substrates as well as the interface between primary osteoblasts with titanium dental implants. The high spatial resolution enabled by ExM reveals that although AP2 and F-actin both accumulate at curved membranes induced by vertical nanostructures, they are spatially segregated. Using ExM, we also reliably image how osteoblasts interact with roughened titanium implant surfaces below the diffraction limit; this is of great interest to understand osseointegration of the implants but has up to now been a significant technical challenge due to the irregular shape, the large volume, and the opacity of the titanium implants that have rendered them incompatible with other super-resolution techniques. We believe that our protocol will enable the use of ExM as a powerful tool for cell–material interface studies.

Graphical Abstract:

The interface between cells and materials at the nanoscale, referred to here as the nano-bio interface, is crucial for applications in biomedical research, medical implants, injectable hydrogels, and bioelectrodes.^1–4 A myriad of factors including surface chemistry, surface topography, substrate stiffness, availability of growth factors, and the presence of extracellular matrix ligands all contribute to the interaction between cells and surfaces.^5–8 Here we focus specifically on the effect of nanoscale topography, as it is well established that the physical texture of these materials can alter the cell–material interface and thus cellular behavior. For example, titanium implants with roughened surfaces on the scale of tens of nanometers to micrometers have been shown to improve osteoblast adhesion and device integration.⁹ In fact, dental implants with roughened surfaces are the current standard in dentistry today due to the significant increase in bone–implant contact and shorter healing time, compared to smooth-finish titanium implants.¹⁰ At the cellular level, nanoscale topography has been shown to affect numerous aspects of cell behavior including cell differentiation, adhesion, polarization, and migration.^11–16 Nanoscale topography has also been shown to affect cellular signaling pathways involved in membrane trafficking, nuclear mechanotransduction, cytoskeleton rearrangement, and cytokine production.^6,17–21 These downstream effects depend on the size, shape, and height of the nanostructures. How cells sense and respond to their environmental topography is of great interest in both biological research and biomedical applications.

Recent studies show that a key mechanism underlying how nanoscale topography affects cellular behavior is topography-induced local curvature of the cell membrane.^22,23 By recruiting and activating curvature-sensing proteins, these local membrane curvatures at the nano-bio interface activate intracellular processes such as clathrin-mediated endocytosis²⁴ and actin cytoskeleton remodeling.²⁵ To understand the nanobio interface, it is imperative to directly visualize how the cell membrane deforms around surface topography. It is also critical to visualize the intracellular signals occurring at the nano-bio interface. Generally, optical microscopy is the most versatile and widely used approach to visualize the cell membrane and intracellular signaling components. However, as a consequence of the diffraction limit of light, optical microscopy is limited to a spatial resolution of ~200–400 nm in the x–y plane and >800 nm in the z plane. In comparison, surface topographies that have been shown to affect cell behavior are often in the range of tens to hundreds of nanometers. Thus, the spatial resolution of optical microscopy is insufficient to resolve membrane curvature and the distribution of signaling molecules at the interface.

The current gold standard for imaging the nano-bio interface is electron microscopy.^26,27 Specifically, focused ion beam scanning electron microscopy (FIB-SEM) and transmission electron microscopy (TEM) have improved visualization of the nano-bio interface. Electron microscopy methods are able to clearly resolve membrane deformation around the surface topography but require extensive effort and are low throughput. It is also challenging to obtain chemical information, such as the signaling molecules at the interface, from electron microscopy. Super-resolution optical microscopy overcomes the optical resolution limit and has recently been applied to image the nano-bio interface to selectively visualize proteins.²⁸ However, super-resolution optical microscopy still requires complex instrumentation and extensive image reconstruction processes. It is also not compatible with nonconventional substrates such as titanium implants.

Expansion microscopy (ExM) provides super-resolution through physical sample expansion.²⁹ Encapsulation of the sample in a hydrogel allows it to expand approximately 4–5× linearly and thus effectively improves optical resolution by 4–5 times. ExM is an accessible and affordable high-resolution imaging method, and compatible with standard fluorescence microscopy. However, using ExM to study the cell–material interface is yet to be achieved because most materials do not expand with the hydrogel.

In this work, we develop a protocol that allows the use of ExM to image the nano-bio interface on any substrate. In this protocol, we preserve the nano-bio interface by imprinting the substrate topography into the hydrogel and then, the hydrogel-encapsulated cells are removed from the substrate before expansion. We show that this approach not only preserves the nano-bio interface with high fidelity but also allows us to examine the uniformity of ExM over a large area using fiducial markers designed onto the substrate surface. By comparing distances between regularly spaced nanopillars, we are able to clearly identify anisotropic cellular expansion in ExM that would otherwise be difficult to distinguish. The improved resolution of ExM reveals that even though two signaling components, AP2 and F-actin, are recruited to curved interfaces, they are actually spatially segregated. Furthermore, ExM allows us to readily visualize how primary osteoblasts interact with titanium implants: an important task for medical research that has long been a technical challenge in the field.

RESULTS AND DISCUSSION

Adapting ExM to Image the Cell–Material Interface.

Expansion microscopy cannot be readily used for imaging cell–material interfaces as most materials do not expand like hydrogels and have to be removed before expansion. For example, materials such as nanotextured silica or metal cannot be expanded with the hydrogel-encapsulated cells and proteins. To preserve the surface topography of substrate materials by ExM, we first coat the substrate with a thin layer of gelatin, usually 1–2 nm thickness,³⁰ to promote cell adhesion and then fluorescently label the gelatin via chemical dye conjugation (Figure 1a(i)). Cells are plated onto the coated substrate (Figure 1a(ii)), fixed, and immunostained, and then both the gelatin and the cellular proteins are cross-linked to the polymer backbone and encapsulated by the ExM hydrogel (Figure 1a(iii)). The surface topography is imprinted into the hydrogel surface through fluorescently labeled gelatin. Thus, when the hydrogel is removed from the substrate during sample digestion and expansion (Figure 1a(iv)), the topography of the substrate surface is preserved on the hydrogel surface (Figure 1a(v)).

Figure 1.

Adapting ExM for imaging the cell–material interface beyond the diffraction limit. (a) Schematic showing the steps of preserving the cell–material interface for the application of ExM. (b, left) Large-area bright field image of the 3D nanostructures substrate. A1 with corresponding 250 by 250 μm² grid of nanopillars is outlined in dotted white square box. A1 is outlined in solid white square box. (b, right) Zoomed-in bright field images of A1 (top) and nanopillars (bottom). (c) Angled SEM images showing nanopillars (top) and B3 structure (bottom). (“B” is 7.5 μm in width, 12.5 μm in length, and 1 μm in height; nanopillars are 100 nm in diameter, 2.5 μm pitch, and 1 μm in height). (d) Pre-ExM fluorescent imaging of C4 and zoomed-in view of area in white dashed box. White arrow points to a nanopillar, and white arrowhead points to the outline of the C structure. (e) Post-ExM fluorescent imaging of same locations as in panel d showing much higher spatial resolution. (f) Post-ExM imaging of U2OS cells expressing GFP-pDisplay cultured on a nanopillar array, using the SDS-denaturation protocol. Distortion of the nanopillar array due to incomplete cellular dissociation is obvious. (g) Heat map of triangular areas formed from adjacent pillars. Red denotes a larger than average triangular area, and blue denotes a smaller than average triangular area. Numbers on gradient legend correspond to the expansion factor. (h) Distances between neighboring nanopillars plotted in a histogram showing a broad and continuous distribution. (i, j, k) same as in panels f, g, and h, except for ExM being applied using the ProK digestion protocol, resulting in a homogeneous expansion inside and outside the cells. The two peaks in panel k correspond to the distances between adjacent nanopillars and diagonal nanopillars, respectively. Scale bars: (b) Zoom-out, 300 μm; zoom-in of A1 and nanopillars, 5 μm. (c) Nanopillars, 2 μm; B3, 5 μm. (d) Pre-ExM C4, 5 μm; pre-ExM zoom-in, 1 μm. (e) Post-ExM C4, 20 μm; post-ExM zoom-in, 4 μm. (f, i) 50 μm; zoom-in, 20 μm.

To demonstrate that the approach shown in Figure 1a can indeed preserve the substrate topography using ExM, we used photolithography to engineer quartz substrates with grids of nanopillars (100 nm diameter, 2.5 μm spacing, 1 μm height, evenly distributed in a grid area of 250 by 250 μm²), with each grid marked by a specific letter–number marker in the upper left corner of the grid (Figure 1b). These grids are laid out in a 10 by 10 array marked from A1 to J10 and are spaced with a 50 μm wide flat area between nanopillar grids (i.e., A1 is 300 μm away from A2). While the grids of nanopillars would be hard to differentiate by themselves, the letter–number markers (~12–15 μm) can be easily identified and used as fiducial markers to locate and image the same area, before and after expansion. Scanning electron microscopy (SEM) images at 30° tilted angles show vertically aligned nanopillars and letter–number markers (Figure 1c).

To apply ExM on nanotextured surfaces, we use a hydrogel with a high (29% (w/w)) concentration of sodium acrylate and acrylamide monomer components. Hydrogels with high monomer concentrations are stiffer and are more easily handled without scratching, scraping, or otherwise damaging the delicate nanostructures. Gel stiffness does not affect the ExM expansion factor as shown in previous studies.³¹ We first examined whether ExM can reliably capture the features of surface topography in the absence of cells. For this study, surface-coated gelatin was cross-linked to the polymer backbone (see Methods for details). We took pre-ExM fluorescence images of the gelatin-coated nanopillars and letter–number markers on the initial substrate before hydrogel cross-linking (Figure 1d). After the gelatin was cross-linked into the hydrogel and the hydrogel was removed from the substrate and expanded in water, we took post-ExM images of the same locations (Figure 1e). In the zoomed-in images, we can readily see that the nanopillars are well preserved in the postexpansion gels with a much higher spatial resolution and a lower fluorescence background (arrows in Figure 1d,e). The outlines of the letter–number markers are also well preserved (arrowheads in Figure 1d,e). By imaging 1.125 μm incremental z-slices from the bottom to the top of the hydrogel, we see that the substrate between pillars remains flat post-ExM and the nanopillars and letter–number markers are well preserved in the z-direction (Supporting Information (SI) Figure S1). Thus, the expanded ExM hydrogel not only comes off cleanly from the nanostructured substrate but also reliably preserves the topographic features of the substrate surface after expansion.

The regularity of nanopillar arrays serves as a functional tool for identifying anisotropic expansion. To date, ExM quantification has been carried out in two ways: either by measuring a specific intracellular feature pre- and postexpansion or measuring the size of the gel as a whole pre- and postexpansion. These methods do not allow for examination of the uniformity of the expansion over large areas or for comparing the expansion inside vs outside the cells. Nanopillars are evenly spaced and distributed over large areas, so nanopillars imprinted into the hydrogel can be used to identify uneven or anisotropic expansion in the sample. Here, we culture U2OS cells on nanopillar substrates coated with fluorescently labeled gelatin. U2OS cells were transiently transfected with a fusion plasmid (GFP-pDisplay, see Methods for details), which served as a marker for the plasma membrane. After fixing the cells and cross-linking both cellular proteins and the gelatin coating to the polymer backbone (see Methods for details), we compared two widely used ExM protocols to disrupt cellular protein–protein interactions before expansion. The first protocol incubates hydrogels in a buffer containing sodium dodecyl sulfate (SDS) at 95 °C to denature and dissociate proteins.^31,32 The second protocol incubates hydrogels in a solution containing Proteinase K (ProK) to enzymatically digest proteins at 37 °C.^29,33–35

Fluorescence imaging of GFP-pDisplay in Figure 1f shows the cell morphology after expansion using the SDS-denaturation protocol. The postexpansion images show that the cell morphologies are well preserved with no obvious ruptures.³⁶ However, fluorescence imaging of the gelatin channel in the same locations show that the regularity of nanopillars is dramatically distorted by the presence of the cells (middle row of Figure 1f). Our nanopillar array provides a frame of reference to separately quantify intracellular and extracellular expansion, as well as the anisotropic distortion at the cell boundary. We constructed a heat map made up of triangular shapes formed from adjacent nanopillars. These triangles are color-coded with their areas; i.e., a red triangle is larger than a blue triangle. A larger triangle also indicates a larger expansion factor because these triangles have the same area before expansion. From the heat map, it is clear that hydrogels inside cells are significantly underexpanded (~3.1× expansion factor) compared to the bulk hydrogel (~4.7× expansion factor) and the expansion at the cell boundary is anisotropic and severely distorted (Figure 3g and Figure S2). Quantification of the lengths of all triangle sides shows a broad and continuous distribution (Figure 1h). Using regularly spaced nanopillar arrays as a reference, the difference between intracellular and extracellular expansion can be observed. Here, nanopillar measurements demonstrate that SDS denaturation likely results in an incomplete cellular dissociation, leading to inhomogeneous expansion with underexpanded and locally distorted cells.

Figure 3.

ExM enables high-resolution imaging of the cell–nanotopography interface in 3D. (a) First column shows 100 nm diameter nanopillars before (top) and after (bottom) ExM; second column shows zoomed-in area from dashed white square of 100 nm diameter nanopillars before (top) and after (bottom) ExM (white arrow points to same nanopillar before and after ExM); third column shows 1000 nm diameter nanopillars before (top) and after (bottom) ExM. (b) Plot showing normalized pixel brightness over the length of the orange and blue line for 100 nm diameter nanopillars (second column of panel a; orange is pre-ExM and blue is post-ExM). (c) Plot showing normalized pixel brightness over the length of the orange and blue lines for 1000 nm diameter nanopillars (third column of panel a; orange is pre-ExM and blue is post-ExM).(d) Fluorescent images of B before (left) and after (right) ExM. (e) x–z projection taken though red dashed line in panel d before (top) and after (bottom) ExM. (f) Pre-ExM and (g) post-ExM images of a cell (green) grown on top of nanopillars (white). y–z projections along dotted white line are shown. Zoomed-in images are of areas inside the white squares. (h) Zoomed-in images from panels g and j with the bottom cell membrane outlined in red and the substrate topography outlined in blue. (i) Pre-ExM and (j) post-ExM images of a cell (green) grown on top of J6 (white) where the gap between J and 6 and the inside of the 6 are examples of surface indentation. Post-ExM x–z projections along the dotted white line show that the cell membrane does not adhere to the bottom of the indentations. Zoomed-in images are of areas inside the white squares. Scale bars: (a) First column, 5 μm. Second column, 2 μm. Third column, 5 μm. (d, e) Pre-ExM, 5 μm; post-ExM, 1 μm. (f, i) Left column, 5 μm; zoomed-in, 1 μm. (g, j) Left column, 20 μm; zoomed-in, 5 μm.

In contrast, using the ProK digestion protocol, we find that the hydrogel is uniformly expanded both inside and outside the cells (Figure 1i). From the gelatin images, there is no distortion of the nanopillar array inside cells or at the boundary (middle row of Figure 1i). The heat map of triangles formed by adjacent nanopillars shows a uniform coloration with no obvious difference inside or outside cells (Figure 1j). A few blanks in the heat map are due to missing nanopillars that are not properly identified by our analysis scripts. Quantification of the triangle sides shows two distinct peaks with Gaussian distributions (one peak correlates to adjacent nanopillars and the other peak correlates to diagonal nanopillars) (Figure 1k). The histogram in Figure 1k shows interpillar distance measurements from >10 cells. Interpillar distance measurements from single images from four experimental replicates all show two distinct peaks with Gaussian distributions, with each image having a standard error (σ/μ) of 3.5% or smaller (Figure S3). Thus, the ProK digestion protocol is highly replicable and results in fully dissociated cellular components and an isotropic expansion of the hydrogel. We use the ProK digestion protocol for all subsequent studies in this work.

ExM Preserves Surface Topography over Large Areas.

When surface topography of the substrate is imprinted into the hydrogel, it creates voids and islands, in addition to the bulk hydrogel. For example, the B3 letter–number pattern becomes a void in the hydrogel, i.e., the empty space left behind where 3D letter–number structures used to be (colored in blue, Figure 2a), while the two holes inside letter B get filled with hydrogel and become two islands on the hydrogel surface (colored in orange, Figure 2a). The rest of the area is bulk hydrogel (colored black). To use ExM for interface studies, it is important to determine whether voids, islands, and the bulk hydrogel expand similarly.

Figure 2.

ExM preserves surface topography over large areas. (a) Fluorescent images and outlines of B3 before (top) and after (bottom) ExM. Voids are colored blue, islands are orange, and bulk hydrogel is black. (b) Random lines generated on the voids and the islands of the letter–number pattern B3. (c) Lengths of matching line segments of voids (blue) and islands (orange) from pre- and post-ExM images. Linear fit is shown in dark blue for voids and red for islands. (d) Histogram showing the expansion factor of letter–number areas (blue voids) and the area within letter/number areas (orange islands) compared to the bulk expansion factor of the hydrogel calculated from nanopillar distances (black). (e) Fluorescent images and expansion heat maps of A2 structures before (top) and after (bottom) ExM. Heat maps show local swelling deformations with expansion factor correlating to gradient color scale. Zoomed-in view of the area inside dotted white rectangle shows deformation at a sharp corner (white arrows). (f) Histograms correlating to distances between nanopillars at specific locations A1, A4, and A7. A1 is 900 μm from A4 and 1.8 mm from A7. Top row histograms are before ExM, and bottom row histograms are after ExM. (g) Plot showing average expansion factor for each 250 by 250 μm² grid of nanopillars located at a specific letter–number location. Error bars are standard deviations. Scale bars: (a) Pre-ExM, 5 μm; post-ExM, 20 μm. (e) Pre-ExM, 5 μm; post-ExM, 20 μm.

First, we separately quantified the expansion factor for the voids, islands, and bulk hydrogel. This was achieved by randomly generating a large number of lines (~2000 lines) on the pattern (Figure 2b; see Methods for details). With an automated image alignment, the same set of lines are projected onto the pre-expansion and the postexpansion patterns. After measuring the line segments across voids or islands, we plot the length of the matching segments before and after expansion, which shows a linear relationship with the slope being the average expansion ratio (Figure 2c). We calculated the expansion factor for each of the paired lines and plotted the corresponding histograms for all letter–number patterns (Figure 2d). We find that voids in the shape of letter–numbers expand approximately 4.53× with a standard deviation of 0.29 (Figure 2d, blue). Interestingly, we find that islands consistently expand to a slightly smaller degree, with an expansion factor of approximately 4.22× and a standard deviation of 0.42 (Figure 2d, orange). By measuring interpillar distances, we find that the overall bulk hydrogel expands to an intermediate degree of 4.47× with a standard deviation of 0.05 (Figure 2d, black).

The slight underexpansion of islands (4.22 ± 0.42×) and overexpansion of voids (4.53 ± 0.29×) compared to the expansion factor of the bulk hydrogel (4.47 ± 0.05×) are unexpected at first glance: one would expect the void areas to shrink due to the expansion of hydrogel into the empty letter–number areas. To understand this phenomenon, we performed nonrigid image registration (see Methods) by comparing the initial shape of the letter to the final shape of the letter left in the hydrogel postexpansion. This analysis showed that while the shape was preserved overall, the sharp corners of islands tended to deform more. Heat maps of local expansion factors are calculated. Areas of the heat map that are white expand less than the average, while areas of the heat map that are dark blue or or dark orange expand more than the average. We can see this effect more distinctly in heat maps of expansion factors for letter A (Figure 2e). The island in the center of the letter A (orange), distinctly shows this pattern: the gel in the corners of the triangle-shaped island are initially a sharp angle pre-ExM, but become rounded post-ExM (third column of Figure 2e). These sharp corners in islands do not expand as much as the bulk hydrogel does because the expanding hydrogel occupies only 1/6 fraction of the space surrounding the sharp point, which causes a slight underexpansion of the islands and a slight overexpansion of the voids. The clear corner distortion occurs only for sharp angles (≤60°), while the two islands inside letter B (~90°) are not distorted (Figure S4). Nevertheless, it is important to note that, despite minor distortions occurring at islands with sharp corners, the expansion ratios of voids, islands, and bulk materials are overall similar (less than 10% variation from the mean expansion). Therefore, ExM overall preserves 3D surface topography well.

In addition to examining feature expansion at the micrometer scale, we also examined the uniformity of bulk hydrogel expansion over large distances. Besides crude measurement of the whole gel before and after expansion,^37,38 previous ExM studies have not examined the uniformity of expansion over several millimeter or centimeter scales due to the lack of appropriate spatial markers. Because nanopillars are fabricated in regular arrays with precisely controlled diameter and spacing, the expansion factors can be quantified locally as well as over a large area. Here, we measure interpillar distances over a large area in the same hydrogel sample, before and after applying ExM. For example, we look at grids of nanopillars at locations marked with A1, A4, and A7, where A1 is spaced 900 μm from A4 and 1.8 mm from A7. Each grid has ~7000 nanopillars, and we used a Python script to automatically calculate all of the interpillar distances. Histograms of the interpillar distances are plotted in Figure 2f, and each histogram is fitted with a Gaussian distribution. Before expansion, the Gaussian distribution of the histogram reflects the uncertainty (~0.2 μm) in determining the center of the nanopillar from the fluorescence images. We find that the expansion is fairly isotropic as the post-ExM histograms are also well fitted by Gaussian distributions. Pre-ExM the interpillar distance histograms are centered around 2.5–2.6 μm, and post-ExM the interpillar distance histograms are centered around 11–12 μm, thus allowing us to calculate an average expansion factor for each nanopillar grid. For 30 distinct grid locations on the substrate, we find that these grid areas expand uniformly; the mean difference between the average grid expansion from the overall average hydrogel expansion is 0.08×, which is <1.9% of the average hydrogel swell (Figure 2g). This measurement is experimentally replicable (Figure S5), and over the entire substrate, we achieve a hydrogel expansion factor of 4.47 ± 0.05.

ExM Allows High-Resolution Imaging of How the Cell Membrane Interacts with Topographic Features of the Substrate in 3D.

Surface topography has been shown to significantly affect cell behavior on a substrate.^39,40 However, these topographic features are often below the resolution limit of optical microscopy,²² which makes ExM an especially valuable tool for examining how cells interact with topographic features. In the pre-expansion image under a 40× confocal microscope lens, 100 nm diameter nanopillars appear as a large bright spot with a half-width of 546 nm due to the diffraction limit of light (Figure 3a,b and Figure S6a). After ~4.47× expansion, 100 nm nanopillars now appear as a bright spot with a half-width of 505 nm, which is equivalent to 113 nm before expansion (Figure 3a,b and Figure S6a). Two overlapping spots that cannot be resolved in the pre-expansion image can be clearly identified in the postexpansion image (white arrows in Figure 3a). We also applied ExM to large 1000 nm diameter nanopillars (Figure 3a, third column). Before expansion, the fluorescent outlines marking the circumference of the 1000 nm nanopillars are thick and partially overlapping, which make the center of the nanopillar much brighter than the background. After expansion, the fluorescent ring is much thinner with a completely dark center, further demonstrating the high spatial resolution that can be attained (Figure 3c and Figure S6b). ExM not only improves spatial resolution in the x–y plane (Figure 3d) but also significantly improves spatial resolution in the z-direction, which is much worse than the x–y direction for optical microscopy. Before expansion, confocal images in the x–z direction are unable to resolve the 1 μm vertical features of the letter “B” because confocal microscopy has a resolution >800 nm in the z-direction (Figure 3e). However, after expansion, the vertical features are clearly resolved: the top surface of the letter B is fluorescently labeled while the solid body of the letter appears dark (Figure 3e).

Recent studies using electron microscopy (EM) show that the cell membrane often conformally adheres to protruding or convex topographic features but does not conformally adhere to concave features.^26,41 Using ExM, we examined how the cell membrane interacts with nanopillars and letter–numbers, which are convex structures, as well as holes inside the letters or numbers, which are concave structures. Panels f and g of Figure 3 show x–y projections from confocal z-scans of a cell cultured on nanopillar arrays, before and after expansion of the same cell. Before expansion (Figure 3f), the y–z projection image along the dotted line shows poor resolution and the cell membrane appears to interact similarly in the nanopillar at the cell edge (box 1) compared to the nanopillar in the middle of the cell (box 2). However, after expansion (Figure 3g,h), it becomes clear that the cell membrane only adheres to the top of the nanopillar at the cell edge (box 1′), but wraps the entire height of the nanopillar located in the middle of the cell (box 2′). Examining many such membrane–nanopillar interfaces confirms previous EM observations that nanopillars induce tight membrane wrapping instead of puncturing into the cell.⁴²

Figure 3i shows x–y projections of a cell grown on the letter–number structure J6 before expansion. The inside of the “6” is effectively a concave surface topography. In the x–z projection image along the dotted line before expansion, the z-resolution is poor and it is impossible to discern how the membrane is interacting with the substrate surface (boxes 3 and 4). However, post-ExM (Figure 3i,h), we observe a gap between the cell membrane and the substrate surface both within the hole of the “6” (box 4′) and between the “J” and the “6” (box 3′). Confirming previous observations from EM studies, we find that the cell membrane does not tightly adhere to negatively curved areas.^26,43,44 Thus, ExM can be used as a tool to study the cell–material interface in three dimensions.

ExM Shows That AP2 and F-Actin Are Spatially Segregated at the Nano-Bio Interface.

Upon confirming that ExM preserves the nanobio-interface and images at a much higher resolution, we used ExM to visualize the distribution of two intracellular proteins, a clathrin-adaptor protein AP2 and a cytoskeletal protein F-actin. Both proteins have been shown to preferentially localize to areas of high membrane curvature at the nano-bio interface and both proteins participate in clathrin-mediated endocytosis.^24,45 But with limited spatial resolution, whether AP2 and actin are recruited together or independently to nanopillars is yet to be determined. To image AP2, endogenous protein was probed by anti-AP2 immunostaining and a secondary antibody labeled with an AF568 fluorophore that has been shown to survive the gelation and the digestion steps in ExM.³³ Cells were transfected with GFP-pDisplay to mark the plasma membrane at nanopillar locations. Fluorescence images of pre-ExM samples show that AP2 preferentially localizes to nanopillar locations (Figure 4a), agreeing with previous studies.^18,24 Yet, the limited spatial resolution prevents further information about AP2’s localization to a specific area of the nanopillar, either in x–y (Figure 4b) or in x–z planes (Figure 4c), which is highly desired in order to gain further insight. After expansion (Figure 4d), the fluorescence images show that AP2 does not uniformly distribute on nanopillars in the x–y plane (Figure 4e). Instead, AP2 often appears as dots located on the sides of nanopillars, likely indicating endocytic events that preferentially occur on curved membranes surrounding nanopillars. The distribution of AP2 along the z-direction provides even more valuable information. From the x-z image of post-ExM samples (Figure 4f), we clearly observe that AP2 tends to specifically localize to the very top of the nanopillar where there is the highest degree of cell membrane curvature, even though the cell membrane wraps around the entire height of the nanopillar. We quantified the distribution of AP2 along the height of the nanopillar and normalized against the cell membrane. Statistical measurements over many nanopillars show that AP2, normalized by the cell membrane, is consistently more likely to localize to the top 1/3 of nanopillars (Figure 4g). This result is consistent with previous EM work that shows endocytotic buds pinching off at the tops of nanopillars.^18,24 It is also consistent with a recent super-resolution imaging study of the nanobio-interface that shows AP2 accumulates more at the tops of nanopillars.²⁸ Therefore, ExM can provide valuable information on protein distributions at the nanobio-interface without sophisticated instruments.

Figure 4.

ExM shows that AP2 and F-actin are spatially segregated at the nano-bio interface. (a) Pre-ExM images of a cell grown on fluorescent gelatin-coated nanopillars (blue) with membrane shown in green and AP2 shown in red. (b) Zoomed-in x–y images of the row of nanopillars in the white box in panel a showing that AP2 preferentially localizes to nanopillars. (c) x–z slice through row of nanopillars shown in panel b (top) and zoomed-in view of single nanopillar in white dashed box (membrane, AP2, gelatin, merged (L–R)) (bottom). (d–f) Post-ExM images of panels a–c. AP2 can be seen to preferentially localize to the top of nanopillars. (g) Plot showing the fluorescence of AP2 normalized to the fluorescence of the cell membrane at the top, middle, and bottom locations on nanopillars. Error bars show mean and standard deviation. For each category (top, middle, bottom), N = 382, where N is the number of pillars (*** denotes p < 0.001). (h–m) Same as panels a–f except F-actin is stained and shown in red. F-actin can be seen to preferentially localize to the bottom of nanopillars. (n) Plot showing the fluorescence of F-actin normalized to the fluorescence of the cell membrane at the top, middle, and bottom locations on nanopillars. Error bars show mean and standard deviation. For each category (top, middle, bottom), N = 350, where N is the number of pillars (*** denotes p < 0.001). (o) Post-ExM image of co-stained AP2 (green) and F-actin (red). (p) Zoomed-in x–y images of the row of nanopillars in the white box in panel o showing that AP2 and F-actin both preferentially localize to nanopillars. (q) x–z slice through row of nanopillars shown in panel p. Merged x–z image (top), AP2 (green) and F-actin (red) only image (middle), and zoomed-in view of single nanopillar in white dashed box (AP2, F-actin, gelatin, merged, AP2 + F-actin, only (L–R)) (bottom). AP2 can be seen to preferentially localize to the top of nanopillars, while F-actin preferentially localizes to the bottom of nanopillars. (r) Schematic showing AP2 and F-actin localization relative to the plasma membrane and nanopillar substrate. Scale bars: (a, d, h, k, o) 20 μm; (b, e, i, l, p) 5 μm; (c, f, j, m, q) (top) 5 μm and (bottom) 1 μm.

We also used ExM to image F-actin, an important component in the cytoskeleton of the cell.⁴⁶ Previous studies show that highly curved membranes wrapping around nanopillars stimulate the polymerization of F-actin.²⁵ F-actin preferentially accumulates at nanopillars and may impact many cellular processes at the nanobio interface.⁴⁴ To probe F-actin, we immunostained cells expressing GFP-pDisplay with phalloidin-AF594, a fluorophore compatible with ExM.³³ Fluorescence images before expansion show F-actin localizing to nanopillars, but the low-resolution images cannot provide further details (Figure 4h–j). After expansion (Figure 4k), from the x–y images, F-actin appears to accumulate to one side of nanopillars, rather than entirely overlapping with the membrane signal (Figure 4l). From the x–z images, we discovered that F-actin tends to specifically localize toward the lower half of nanopillars, an opposite behavior to AP2 (Figure 4m). Quantitative analysis of F-actin distribution along the height of the nanopillar, normalized by the cell membrane, shows that F-actin is consistently more likely to localize to the bottom or middle of nanopillars compared to the top (Figure 4n). As expected, co-staining of AP2 and F-actin confirms that AP2 tends to localize to the top half of the nanopillar and F-actin tends to localize to the bottom half of the nanopillar (Figure 4o–q). The reasons behind the preferential AP2 accumulation to the top half of the nanopillar and the preferential F-actin localization to the bottom half of the nanopillar is yet to be fully understood. It is possible that our nanopillars have a slightly tapered top, which gives rise to sharper membrane curvature that is more preferred by endocytosis. It is also possible that proteins respond differently to the two-dimensional Gaussian curvature at the top vs the one-dimensional curvature along the side wall in the middle and bottom sections. Nevertheless, ExM clearly shows that AP2 and F-actin are spatially segregated and are most likely recruited independently to the membrane-nanopillar interface.

ExM Enables 3D Imaging of Rat Osteoblasts with Titanium Implants.

The interactions of cells with implant surface topography has been and continues to be a subject of intense interest. Currently, commercially available titanium dental implants are acid etched and/or sand blasted to create a nanotextured surface.^47–50 There are EM studies of textured implant-like surfaces only^51,52 and the interface between implants and tissues,^53,54 but very few studies using EM to image the cell–implant interface.⁵⁵ Additionally, imaging the cell–implant interface using optical microscopy remains a major challenge due to the irregular shape, the large size, and the opacity of the titanium implants. Because the ExM hydrogel is removed from the substrate before being expanded and imaged, we expect that ExM can be used to visualize cells grown on top of titanium implants using optical microscopy (Figure 5a).

Figure 5.

ExM enables 3D imaging of the rat osteoblasts with titanium implants. (a) Schematic showing ExM applied to cells grown on textured, opaque substrates. (b) Photograph of the titanium dental implant housed in a 3D-printed well for cell culture. Roughened implant screw surface is labeled in the photograph (length = 8 mm). (c) Scanning electron microscopy image of the titanium dental implant surface. (d) ExM image of gelatin coating (white) on the implant surface (top) and merged gelatin coating with osteoblast cell (green) (bottom). (e) y–z projection of the same osteoblast shown in panel d through the white dotted line (top). From left to right, zoomed-in of areas in white dashed squares of the cell membrane (green), the gelatin coating (white), merged image, and outlined interface where the cell membrane is outlined in red and the implant surface is outlined in blue (bottom). (f, g) Similar to panels d, e, showing another osteoblast–titanium implant interface. Scale bars: (b) 1 cm; (c) 20 μm; (d, f) 50 μm; (e, g) (top) 50 μm and (zoom in) 10 μm.

An important consideration for the success of bone and dental implantations is their integration with osteoblasts. To investigate osteoblast–implant interactions, we obtained commercially sourced titanium dental implants that are routinely used for human patients (Figure 5b). Our SEM images of the as-obtained titanium implant confirms that the titanium surface is rough with structural peaks and valleys on the order of hundreds of nanometers to micrometers (Figure 5c). The irregular-shaped titanium implants were housed in our custom-designed and 3D-printed case for subsequent cell culture (Figure 5b). Similar to nanotextured silica substrates, titanium implants were coated with fluorescently labeled gelatin before cell seeding. Primary osteoblasts were extracted from adult rat femur bones according to an established protocol.⁵⁶ Extracted cells were cultured and stained with an alkaline phosphatase (ALP) staining kit, thus confirming ALP activity, which is a biomarker of osteoblasts (Figure S7). Upon culturing the cells on the titanium implant for 48 h, the cells were fixed, embedded in the polymer hydrogel, and digested with ProK. During the digestion step, the gel detaches from the implant surface. Upon expansion, fluorescent gelatin shows the expected nanotextured surface of the implant (Figure 5d,f), similar to the SEM image taken of the same implant (Figure 5c).

Using ExM, we examined how osteoblasts interact with roughened titanium implants. Fluorescence imaging of gelatin (top image in Figure 5d) confirms that rough features of the titanium surface are preserved in the hydrogel. In the x–y plane, osteoblasts fully spread and adhere well to the roughened titanium surface (bottom image in Figure 5d), similar to cells on flat surfaces. The enhanced resolution in the z-direction reveals crucial information. In the y–z image shown in Figure 5e, the bottom cell membrane can be seen to attach to areas of surface protrusions (convex geometry, peaks) but not to areas of surface indentations (concave geometry, valleys). The bottom cell membrane can be micrometers away from the base of the valleys. Therefore, the cell membrane as a whole does not deform to comformally wrap around the substrate topography. Instead, the cell membrane appears to locally deform and attach to protrusions, but avoids direct adhesions to valleys. Similar behaviors are observed from multiple independent experiments of primary rat osteoblasts cultured titanium implants (Figure 5f,g), and also observed from U2OS cells cultured on titanium implants (Figure S8). These observations are consistent with the observation that U2OS cells do not adhere to the concave structures on quartz substrates shown in Figure 3.

Our ExM observations directly refute an early hypothesis⁵⁷ that a rough titanium surface promotes better osseointegration because it presents a larger total surface area for cells to adhere to than a smooth surface does. Since osteoblasts do not conformally deform and adhere to surface topography, the total surface area is likely not the key determinant. Although surface topography potently regulates cell functions, optimizing surface topography to achieve a desired function among a high-dimensional space of ill-defined topographical parameters such as peak size, height, steepness, peak/valley ratio, feature shape, density, orientation, and symmetry remains a challenge. ExM provides a method to visualize the interface and answer fundamental questions of how cells detect their environmental topography and what key topographic features elicit cellular responses.

CONCLUSIONS

In this work, we demonstrate that ExM can be applied to visualize the cell–material interface for substrates with 3D textured surfaces. By imprinting the substrate topography into hydrogels and employing nonrigid image registration algorithms, we quantified the uniformity of the expansion over large areas and the geometry dependence of expansion isotropicity. ExM allows us to image beyond the diffraction limit of how the cell membrane interacts with nanotopography and how intracellular proteins are distributed at the interface. In particular, ExM can be used to fluorescently image samples grown on irregularly shaped metallic and opaque surfaces that typically cannot be achieved using optical and electron microscopy. Thus, ExM is a powerful tool in cell–material interface studies.

METHODS

Nanofabrication.

Nanopillar and microstructure arrays used in this work were fabricated using photolithography. Desired patterns were designed with Python. Quartz wafers (4 in.) are first cleaned with Spin Rinse Dryer (SRD) and baked. Then, hexamethyldisilazane (HDMS) is used to remove residual moisture and promote photoresist adhesion. The quartz wafers are then coated with photoresist (Shipley 3612) and then exposed to the desired pattern using ultraviolet light on the Heidelberg (MLA150). Next, wafers are baked and developed with MF-26A (Transene). A chromium mask is deposited on the wafers using an AJA e-beam evaporator and lifted off immediately with acetone and isopropanol. The quartz wafers are immersed in chromium etchant 1020 (Transene) and then incubated in 20:1 buffered oxide etch (BOE) to isotropically shrink structures to their respective nanopillar size. The quartz wafers are then diced into chips of 1 cm².

Scanning Electron Microscopy.

The fabricated nanostructures and titanium implant surfaces are characterized by SEM (FEI Nova).

Cell Culture.

U2OS cells (ATCC, cat. no. HTB-96) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin. Rat osteoblast cells were cultured in DMEM supplemented with 10% FBS, 1% penicillin–streptomycin, and 50 μg/mL ascorbic acid. Cell medium was changed every 2 days. All cells were cultured in a standard incubator at 37 °C with 5% CO₂.

Substrate Coating and Cell Plating on Nanopillar Chips.

Quartz nanopillar chips (1 cm² surface area) were plasma cleaned in a Plasma Cleaner PDC-32G and then coated with 0.1 mg/mL poly-l-lysine for 1 h, 0.5% glutaraldehyde for 15 min, and 1 mg/mL gelatin for 30 min. Substrates were washed 3× with PBS after each coating layer. To fluorescently stain gelatin, nanochips were incubated in 350 nM Atto647N-NHS ester in a sodium bicarbonate solution of pH 8 for 1 h at room temperature. Cells were then plated onto nanopillar chips. Nanopillar chips were cleaned in piranha (3 part sulfuric acid:1 part 30% H₂O₂) overnight between experiments.

Cell Plating on Implants in 3D-Printed Wells.

Commercially sold Bioconcept implants (cat. no. 013010) were purchased and used for implant experiments. Implants were soaked in ethanol for 20 min and then coated with 0.1 mg/mL poly-l-lysine, 0.5% glutaraldehyde, and 1 mg/mL gelatin. Implants were washed 3× with PBS after each coating layer. To fluorescently stain gelatin, implants were incubated in 350 nM Atto647N-NHS ester in a sodium bicarbonate solution of pH 8 for 1 h at room temperature. A 3D-printed well was designed to hold the implant in place for cell culture using Tinkercad. The 3D wells were then printed using PLA plastic filament on a Prusa i3MK 3s printer. 3D-printed wells were sterilized with ethanol for 20 min and then rinsed with ddH₂O before placing the implant into the well cavity. Cells were then plated on top of the implant. Implants were cleaned via ddH₂O soak, ultrazyme soak, and then overnight nitric acid soak between experiments. Implants were not used for more than three experiments.

Isolation of Osteoblast Cells.

Rat osteoblast cells were isolated as previously described.⁵⁶ Briefly, femora are isolated and cleaned of soft tissue, the epiphyses are removed with clippers, and the bone marrow is flushed out using a syringe with a 10 gauge needle filled with culture media. The resulting cell suspension from each femur is collected in a 75 cm² plastic culture flask. After 48 h, culture medium containing nonadherent cells is removed and replaced with fresh culture medium. After 10 days of culture, alkaline phosphatase activity was confirmed in cells using Sigma-Aldrich alkaline phosphatase staining kit (86C-1KT). Osteoblasts were then used for implant experiments.

Transfection.

U2OS cells were transfected using an Amaxa Nucleofactor II (Lonza) electroporation machine. Cells were added to a suspension of DNA in electroporation buffer (7 mM ATP, 11.7 mM MgCl₂, 86 mM KH₂PO₄, 13.7 mM NaHCO₃, 1.9 mM glucose), transferred to a 2 mm electroporation cuvette, and subjected to the manufacturer provided protocol for U2OS cells. pDisplay-GFP (Invitrogen, V66020) was transfected to show the cell membrane. pDisplay is an expression vector that targets and displays proteins of interest in the cell surface using a platelet-derived growth factor receptor (PDGFR) domain. GFP was cloned into this vector to fluorescently label the outer cell membrane surface.

Immunostaining.

After overnight culturing, cells were fixed in 4% paraformaldehyde at room temperature for 10 min and then washed once with 100 mM glycine in PBS and twice with PBS. For osteoblast samples, cells were cultured on the implants for 48 h before fixation. Cells stained with WGA were incubated at 4 °C with a 1:2500 dilution of WGA–biotin (5 mg/mL check) for 30 min prior to fixation. After fixation, WGA-stained samples were blocked for 20 min with 2% BSA and then stained with a 1:10,000 dilution of AF488–streptavidin (2 mg/mL, Invitrogen, S11223) in 2% BSA. Samples were then washed four times with 2% BSA. To immunostain AP2 or F-actin, cells were permeabilized with 0.1% Triton X-100 for 10 min, blocked with 2% BSA for 20 min, and then stained with a 1:500 dilution AP2 (1 mg/mL, abcam, ab189995) in 2% BSA or a 1:500 dilution of phalloidin–AF594 (Invitrogen, A12381) in 2% BSA for 1 h. Samples were then washed four times with 2% BSA. After AP2 staining, cells were stained with secondary antibody goat–antimouse AF568 (2 mg/mL, Invitrogen, A11004) in 2% BSA for 1 h and then washed again four times with 2% BSA.

Gelation, Digestion, and Expansion.

After all immunostaining steps were finished, cells were incubated overnight at room temperature in a 1:100 dilution of AcX in PBS. After washing with PBS, cells were then incubated for 15 min in the gelation solution (19% (w/w) SA, 10% (w/w) AA, 0.1% (w/w) BIS) at room temperature. Then, nanochips were flipped cell side down onto a 70 μL drop of the gelation solution supplemented with 0.5% TEMED and 0.5% APS on parafilm and incubated at 37 °C for 1 h. APS must be added last to the gelation solution. After gelation, the nanochip with the hydrogel still attached was incubated in a 1:100 dilution of Proteinase K (New England BioLabs) in digestion buffer (50 mM Tris HCl (pH 8), 1 mM EDTA, 0.5% Triton X-100, 1 M NaCl). After 5 h of incubation at 37 °C, the hydrogel has come off of the nanochip and has expanded ~1.5×. Hydrogels were then soaked 2× in ddH₂O for 30 min and then incubated overnight in ddH₂O at 4 °C. For implant samples, the gelling solution was placed directly into the 3D-printed well on top of the implant. After gelation, the implant with the hydrogel still attached was removed from the 3D-printed well, incubated in the digestion buffer in a separate well, and then expanded in ddH₂O as written above. Specific chemical information is shown in the expansion microscopy chemical list below.

Denaturation.

For samples that are denatured instead of digested, all steps up to gelation are carried out the same as written above. Denaturation was carried out by soaking samples in denaturation buffer (200 mM SDS, 200 mM NaCl, 50 mM Tris, and adjusted to pH = 9) for 15 min at room temperature until gels came off of the nanopillar substrates. Gels were then placed into fresh denaturation buffer and incubated at 95 °C for 2 h. After denaturation, samples were expanded in ddH₂O as written above.

Fluorescent Imaging and Gel Mounting.

Fluorescence images were acquired using a Nikon A1plus confocal microscope. Images were taken using a 40× water immersion lens. For pre-ExM expansion images, nanochips were placed pillar side down into glass-bottom Petri dishes. For post-ExM expansion images, excess water on ExM hydrogels was carefully removed using a Kim wipe. Then, ExM hydrogels were mounted onto PLL-coated glass coverslips to prevent sliding. After mounting, hydrogels were covered with a few drops of water to prevent shrinkage during imaging.

AP2 and F-Actin Analysis.

The fluorescent gelatin channel of images was used to generate arrays of square masks centered on nanopillars, using MATLAB code written for this study. The z-slices containing the top, middle, and bottoms of nanopillars were identified from the z-stack. The regions of interest (ROIs) were identified for each area (top, middle, or bottom), and then the ROIs were then applied to each pillar in the membrane and protein channels (AP2 or F-actin) and the mean intensities in each ROI were measured. For each measurement, the protein intensity (AP2 or F-actin) was normalized by the membrane intensity. The normalized intensity of the protein at the top, middle, and bottom for each nanopillar was reported. The error bars represent mean ± standard deviation. P-values were calculated using a paired two-tailed t test.

Nanopillar Analysis.

In order to segment nanopillars from the image background an interactive predictive annotation software, ilastik (v.1.3.0), was used.⁵⁸ The software is free to download and tries to predict the category (pillar or background) of each pixel in the image by learning from user-input pen strokes which codify the category by the pen’s color. Interpillar distance: the analysis pipeline was implemented in Python. The pillars were identified as single entities with the regionprops function of the scikit-image package. The centerpoints of the pillars were then used to construct a Delaunay (scipy package) triangulation. The triangulation yields each nanopillar’s direct and diagonal neighbor and the distance between them. Therefore, histograms of the interpillar distance in regular grids will contain two peaks: one at the direct neighbor distance and one at the diagonal distance. The area of each triangle can be computed by the triangle formula and thus the triangles are color-coded based on their size. Overlaying this coloration on the original image allows for visually inspecting irregular constrictions or expansions in the gel.

Letter Deformation Analysis.

Comparing the deformation of letters between the initial and expanded state is much more involved than for the nanopillars since image registration is needed. The initial position of the corner of a letter needs to be identified in the expanded state in order to quantify local dislocations, constrictions, expansions of the letters. To that end we used large deformation diffeomorphic metric mapping (LDDMM) through the open-source Linux-bound framework Deformetrica.⁵⁹ In brief, LDDMM assumes that deformations of shapes are spatially smooth and relatively slowly changing fields. The technique initializes a vector field that obeys a deformation differential equation. The differential equation is solved when the initial shape is successfully deformed, following the parametrized vector displacements, into the final shape.

To that end, letters were carefully hand-segmented in the initial and expanded images. The contours of letters were collected with Python’s matplotlib package. This allowed one to run LDDMM from the initial letter’s contour and try to fit it onto the expanded contour. The LDDMM framework was run via its command line operation through an IPython notebook in order to keep the pipeline concise. The resulting displacement vector field finds the smooth deformation field that takes the initial contour onto the expanded contour. Then, the initial letters were meshed with triangles via the Pygalmesh package. The generated triangular mesh is denser where most precision is needed. The triangular mesh is then deformed into its corresponding expanded state by passing it to the deformation field fitted by Deformetrica’s LDDMM. This procedure allows for comparing the same regions between the initial and expanded letters. Because a given triangle in the mesh represents the same region before and after expansion, the ratios of the triangles’ areas can be computed to extract a local swelling factor.

Data availability:

The analysis codes for producing panels g, h, j, and k of Figure ¹ and panels b–f of Figure ² can be found at https://github.com/bcuilab/ExpansionMicroscopy/.