Kirsten D.
Endresen
a,
MinSu
Kim
a,
Matthew
Pittman
b,
Yun
Chen
b and
Francesca
Serra
*a
aJohns Hopkins University, Dept. Physics and Astronomy, Baltimore, USA. E-mail: francesca.serra@jhu.edu
bJohns Hopkins University, Dept. Mechanical Engineering, Baltimore, USA
First published on 8th March 2021
Many cell types spontaneously order like nematic liquid crystals, and, as such, they form topological defects, which influence the cell organization. While defects with topological charge ±1/2 are common in cell monolayers, defects with charge ±1, which are thought to be relevant in the formation of protrusions in living systems, are more elusive. We use topographical patterns to impose topological charge of ±1 in controlled locations in cell monolayers. We study two types of cells, 3T6 fibroblasts and EpH-4 epithelial cells, and we compare their behavior on such patterns, characterizing the degree of alignment, the cell density near the defects, and their behavior at the defect core. We observe density variation in the 3T6 monolayers near both types of defects over the same length-scale. By choosing appropriate geometrical parameters of our topographical features, we identify a new behavior of 3T6 cells near the defects with topological charge +1, leading to a change in the cells’ preferred shape. Our strategy allows a fine control of cell alignment near defects as a platform to study liquid crystalline properties of cells.
This is especially relevant near topological defects, regions where the nematic order is lost8,9 in order to minimize stresses in the ordered fluid. In LCs, one can think of defects as small isotropic regions in an ordered liquid. The disordered region is known as the defect core, and strong elastic distortions are concentrated around it. In 2D, defects are characterized by a topological charge, i.e. the angle by which LC mesogens rotate around the defect, divided by 2π.9 This quantity is additive, conserved, and determined by the topology of the LC confinement. For example, if nematic LCs are confined on the surface of a sphere, the total topological charge is +2 and it can be realized with an arbitrary number of defects whose charges add up to +2. In nematics, topological charge can be integer or semi integer, the most common defects being ±1/2 and ±1. Defects with the same topological charge may have different molecular arrangements around them, which depend on the energetic cost of various types of deformations. For example, around a topological defect with a charge of +1, LC molecules can have a radial arrangement with a large splay deformation or azimuthal arrangement with a large bend deformation, depending on which elastic distortions are energetically favorable. In common LCs, defects interact with each other strongly by elastic interactions, they are able to trap colloidal particles and small molecules,10–12 they exhibit interesting optical effects13,14 and in general they are mediators of self-assembly.15
Cell layers also form topological defects as they rearrange, and it is becoming evident that these defects have a biological role. Saw et al.16 found that near +1/2 defects the rate of apoptosis of MDCK epithelial cells is higher, due to the presence of isotropic compressive stresses. In contrast, the −1/2 defects are characterized by tensile stresses and do not trigger apoptosis. Kawaguchi et al.17 reported a strong effect of topological defects in murine neural progenitor cells moving on substrates without attaching. The density of the monolayer increases next to the +1/2 defects, eventually resulting in the formation of large cell conglomerates. Important effects of defects have been reported in bacterial systems, where the +1/2 defects drive the morphology of bacterial colonies,18 and where they are responsible for the initial formation of multi-layer structures.19
Defects with topological charge ±1/2 are commonly observed in cell cultures and have therefore been the subject of many studies. On the other hand, defects with integer charge ±1 are not typically observed in flat layers and have received less attention. However, these types of defects are still present in living systems in various forms. There are many examples of fibers forming topological defects with +1 charge, either with azimuthal or with radial alignment, such as those around the optic nerve20 and at the tumor-stromal interface.21,22 Furthermore, circular alignment is also seen in epithelial cells, forming rosettes,23 and circularly aligned clusters seem to precede tube invagination during embryogenesis.24 This behavior recalls that observed in small organisms such as hydra, where the formation of +1 defects is associated to the localization of the hydra's head and foot, thus playing an important role in determining the 3-dimensional shape.25
The control of cells through topography is relevant for the arrangement of cell orientation and subsequent tissue architecture, and thereby may be useful to guide desired tissue integration of implants in clinical applications, to restore normal physiological functions. It is well known that nano-sized topographical features such as grooves or ridges can align fibroblasts. The alignment strength depends on the width and height of the ridges. Turiv et al. recently explored the possibility to impose defects with integer charge in human dermal fibroblasts by using nano ridges caused by the swelling of liquid crystal elastomers in aqueous medium.26 They observed the spontaneous separation (unbinding) of the +1 defects into two defects with topological charge +1/2, confirming that defects with integer charge tend to be unfavorable in flat cell layers.
A different strategy for alignment relies on the use of larger features, such that the alignment does not rely solely on the interaction between the cells and the substrate but also on the spontaneous nematic alignment of the cells. Here, we use micron-sized topography to investigate the possibility to induce stable defects with topological charge +1 and −1 (Fig. 1a). The spacing of the ridges is on the scale of several cell sizes, so that the mechanism of cell alignment on the ridges is likely entropic.27 In this paper, we focus mainly on fibroblasts and epithelial cells, chosen as epitome of two different cell types: fibroblasts interact strongly with the substrate and are able to assume very anisotropic shapes; in contrast, the EpH-4 epithelial cells have strong cell–cell junctions and are quite isotropic in shape. Our patterns allow us to investigate situations where cells monolayers experience the frustration of undesired topological structures, thereby gathering information on the LC behavior of the cell layers.
Here we need to add a consideration on this system. It is known that ±1/2 topological defects form spontaneously and proliferate in active nematic systems. An increase in cell motility or myosin activity lead to an increase in the density of the defects which are spontaneously generated. We have verified that the number of defects is not altered by influencing the myosin activity. Blebbistatin is a myosin II inhibitor which decreases force generation; calyculin is a phosphatase inhibitor which increases force generation. By treating the cells on our pattern with these chemicals we have not detected a change in the number of defects. In fact, in all cases the cell monolayers form the minimal number of defects set by their boundary conditions. In virtue of this observation, in the present work we focus on the static configurations of cells.
The mechanism for cells’ alignment on ridges is a debated topic and several mechanisms concur to it. In our system, the cells closest to the ridges sense the edges of the features, or the curvature of the walls, and that this alignment is propagated to the other cells via nematic order. In addition to the varying spacing between ridges, we utilize different heights for the ridges, from 0.5 to 3 μm. As suggested by Bade et al.,28 cells need to rearrange their cytoskeleton in order to climb on higher features. Therefore we expect a change in alignment with the height of the ridges. The alignment of fibroblasts increases significantly from the 0.5 μm- to the 1.5 μm-tall ridges, but does not improve significantly from 1.5 to 3 μm ridges, as shown in ESI,† Fig. S1. We therefore decided to utilize 1.5 μm-tall ridges for most of our measurements.
While the quality of the alignment of fibroblasts does not depend on the spacing between ridges, it shows a strong dependence on the average density of the cell monolayer, as shown in Fig. 2. We study the cases of monolayers ranging from 600 cells per mm2 corresponding to 60% confluency, to 2000 cells per mm2 corresponding to a tightly packed monolayer well above confluency. We estimate a 100% confluency at around 1000 cells per mm2. In Fig. 2, we quantify the degree of alignment by the root-mean-square deviation of the angle α, which measures the difference between the cell alignment and the perfect alignment around a defect. We can see that increases for densely packed monolayers, and the behavior indicates a significant decrease in cell alignment above a density threshold around 1400 cells per mm2, both for +1 and −1 defects. This is significantly larger than the density at which the monolayer becomes confluent, but it coincides with the density value at which the cells become less elongated. Fig. 2b and c show cells near a +1 defect below and after the threshold density value, and a visible change in their aspect ratio and alignment.
We perform the same experiment with EpH-4 epithelial cells (Fig. 3). If we use the same patterns used for the fibroblasts (r = 60 μm, Fig. 3a and b, or larger), the alignment observed in the cells is present, but much weaker. We then utilize a pattern with ridges which are more finely spaced, with 30 μm spacing, and indeed that gives a much better alignment to the epithelial cells, comparable to that of fibroblasts, as can be seen from the scatter plots in Fig. 3c and d and from the angle distribution in Fig. 3e. We notice, however, that the tighter confinement means that only 1–2 cells can fit in between adjacent ridges, indicating that cells can align only if each cell is in direct contact with the topographical features. We verify that the height of the ridges chosen for fibroblasts is also suitable for epithelial cells. In epithelial cells we again observe that the alignment improves with taller ridges, but ridges above 2 μm prevent the cells from going over them, as shown in ESI,† Fig. S2. Therefore we also favored the use of the 1.5 μm ridges for epithelial cells. Unlike the fibroblasts, the density of the monolayers does not affect the quality of alignment of epithelial cells, which do not significantly alter their aspect ratio as they increase in density. It should be noted, however, that once they reach confluency the epithelial cells stop dividing and die, unlike fibroblasts, where we are allowed to explore a higher range of density above confluency.
Cell density, however, plays a role in determining the relative increase/decrease of density near the +1/−1 defects for fibroblasts. The inhomogeneity in the cell density is maximum when the cell alignment is the highest. Fig. 4a and b shows the density variation for monolayers at various densities. The behavior is non-monotonic. At low density, the monolayer shows more homogeneity in density. When the monolayer reaches confluency the inhomogeneity is maximum, with defects densely packed near +1 defects and loosely packed near −1 defects. The comparison with Fig. 2 suggest that once the aspect ratio of cells starts to change, and the alignment quality decreases, the monolayer becomes more homogeneous and the difference between +1 and −1 defects is reduced, although never entirely suppressed. On the other hand, the spacing between ridges does not significantly affect the density variation, as suggested by Fig. 4c and d, not even where high-density monolayers are created on ridges with larger spacing. Preliminary data also suggest that seeding density may influence the packing, and cells planted initially at density above 20 × 104 cells per mL cannot achieve an ordered packing near the defects. In our work, we have used cells at low seeding density to ensure that they had time to align as they reach confluency.
The case is different for epithelial cells, which only show a slight increase of density near both defects of positive and negative charge (Fig. 4e and f). Also in this case the difference is more marked for the finely spaced pattern. Consistently, the work of Saw et al.16 showed an increase of apoptosis near +1/2 defects not correlated with an increase in cell density of epithelial cells. For EpH-4 cells, we could not observe any difference in the density distribution as a function of cell density.
Moreover, this behavior does not change by altering the distance between the patterned defects, if the defects are more than 100 μm apart. We probe this by reducing the number of topographical cues. For this, we use a pattern with only one or two ridges per circle (ESI,† Fig. S3). If the radius of the circular ridge is smaller than about 40 μm, fibroblasts ignore the topographical cues and grow all over the ridges, showing random +1/2 and −1/2 defects (ESI,† Fig. S3a). If the radii are between 100 and 200 μm or larger, the results are like those shown in Fig. 6. In that case, we observe the two possible configurations for the defect within the inner ring (one +1 or two +1/2 defects) and the splitting of the −1 defects into two defects (ESI,† Fig. S3b). If the separation between ridges is larger, fibroblast monolayers will form additional topological defects with charge +1/2 and −1/2 (ESI,† Fig. S3c).
Fibroblasts near defects with charge −1 do not pack densely near it, and consequently we observe fewer cells, especially for the more finely spaced ridges. However, the negative defects split into two −1/2 defects in the vast majority of cases (Fig. 6d and e), for all ridge spacing. A more isotropic packing of cells in the center, with elongated cells around it that follow the hyperbolic alignment, is observed in relatively few cases (Fig. 6f) and we call this configuration a −1 defect. There are two reasons for the prevalence of −1/2 defects: first, fibroblasts are less compressed near negative defects compared to positive defects, thereby they adopt the default fibroblast morphology of an elongated spindle; secondly, the region of the negative defect has a four-fold symmetry. As fibroblasts are elongated, they can easily align along the two alternative axes of symmetry of the defect, and this provides a natural way to “split” the defect into two −1/2 defects.
The case is different for epithelial cells. The alignment of cells is not greatly perturbed in the inner regions of the defects. For +1 defects, in the 30 μm patterns only few cells can fit in the inner circle, and they are always arranged isotropically, in the default cuboidal morphology of epithelial cells. We do not see the tendency we saw in fibroblasts, i.e. the formation of two +1/2 defects near the outer ridges. As we increase the size of the inner circle to a radius of 60 μm, epithelial cells are still isotropic (Fig. 6g). However, we observe either a slight symmetry breaking along an axis in the inner circle or a rosette-like structure, with cells arranged in roughly concentric rings and with central cells meeting at a single vertex, as shown in Fig. 6g. Increasing the size of the inner circle to 100 μm radius, the behavior remains similar to the 60 μm case.
The data presented in our study provide further insight on the nature of the topological defects in cell systems. The search for the so-called core of a topological defect is very challenging in thermotropic LCs, mostly studied with theoretical models and computer simulations.32,33 In typical thermotropic LCs, this size is a few tens of nanometers, corresponding to about 10 times the molecular size. Defect core is usually defined as an isotropic region within the ordered fluid, even though studies in chromonics and other lyotropic LCs, where the core size is larger, have revealed a more complex alignment even inside the core of the defect.34,35
In our experiments, fibroblasts 3T6 have one more degree of freedom: due to the weak cell–cell adhesion strength, they can easily shift their morphology from elongated spindle to isotropic cuboid, no longer acting as mesogens in the core of the defect. This behavior leads us to identify a length-scale associated with this truly isotropic core, whose linear size is around 150–200 μm, corresponding to about 10 cells, in analogy with other LC systems. We can therefore start characterizing the core size as this length-scale where the transition takes place. While the unique shape-changing behavior of the fibroblasts allowed us to identify this length scale, these features were not observed for EpH-4 epithelial cells, most likely because the combination of their smaller aspect ratio and their strong cell–cell adhesion reduce the cells' ability to adjust density and shift between different shapes.
Our studies are a characterization of the behavior of cells near defects of topological charge ±1, which are hard to see in monolayers without confinement, but are observable in living systems near protrusions or along aligned fibers. There is increasing evidence not only that topological defects affect cell behavior, but also that the effects are dependent on the topological charge. We envision our work raises the awarenes of the complex nature of cells in forming defects and inspires more detailed studies and hypotheses development to understand the liquid crystal behavior in cells as sophisticated interplay between biological elements of the cells and the physical conditions imposed externally.
The orientation of nuclei was used as a proxy for the cell orientation. In order to quantify the orientation of the nuclei, we used ImageJ to first create a binary mask which showed contained only the cell nuclei, minimizing background features as much as possible. The watershed method was used to separate nuclei which appeared too close to each other to be distinguished in the image. The Fit Ellipse function in ImageJ was used to determine the center locations, major and minor axis lengths, major axis orientations, and areas of each nucleus. Cutoffs for the number of pixels for ellipses were chosen to have a sufficient signal level to background noise.
However, counting cell nuclei with fluorescence leads to under-estimating the number of cells in the very high density region as shown in ESI,† Fig. S5. For this reason, we have utilized a different method to analyze the density distribution near +1 defects for fibroblasts, where instead of counting nuclei we integrate the fluorescent intensity over concentric rings at various distances from the defect. At low density, however, cell counting is a more accurate method, as the fluorescence signal in those regions is lower and tends to underestimate the inhomogeneities. All this is discussed in Fig. S5 (ESI†).
The angle from the center of the defect was determined based on the location of the center of the defect defined from the phase contrast image and the location of the center of the nucleus given by the ellipse fitting. We verified the accuracy of this method at low cell density comparing it with an analysis of cell shapes from FogBank as explained in ESI,† Fig. S6.
Analysis of YAP localization was performed as in ref. 29. Briefly, the nucleus of each cell was segmented using the DAPI channel, and the average intensity of YAP was measured inside the nucleus and just outside the nucleus. The nuclear YAP score was then obtained by taking the ratio of nuclear to cytoplasmic YAP mean intensity.
Statistical analysis was done in Prism using Ordinary one-way ANOVA with Tukey's test for multiple comparisons. For each defect, a minimum of 4 samples and 6 samples were analyzed for 3T6 and EpH4, respectively.
To enhance actomyosin activity, cells were treated with Calyculin-A (Cell Signaling Technology, Cat. No. 9902). We diluted the Calyculin-A in DMEM to a 50 nM concentration, and added the mixture to confluent cells on the topographic patterns after sonicating for 15 minutes. Before Calyculin treatment, cells were stained with Hoechst 33258 dye at a 1:
2000 dilution and incubated for 15 minutes, after which the samples were washed with PBS. The cells were then incubated with the Calyculin mixture for 30 minutes, after which it was removed and the samples were washed three times with 5 mL PBS. The cells were imaged at 37 °C in CO2 independent media (same mixture used for blebbistatin treatment).
The uncertainty reported in the observed variations from the average density was reported as the standard error of the mean (SEM). We performed t-tests using Matlab to test the null hypothesis: H0: the average density of cells within 200 μm of a +1/−1 topological defect is the same as the average density of all the cells within 600 μm of the +1/−1 defect. H1: the average density is greater/lower within 200 μm of a +1/−1 defect than the average density of cells within 600 μm of the +1/−1 defect.
For Fig. 4a, n = 5 samples were observed in each reported density range. For 800–1000 cells per mm2, a p-value of 0.02 was obtained for observing a density increase within 200 μm of the center; for 1200–1400 cells per mm2, p-value = 0.0015; for 1400–1600 cells per mm2, p-value = 0.02; for 1600–1800 cells per mm2, p-value = 0.002.
For Fig. 4b: for 600–800 cells per mm2, n = 4 samples were observed, with a p-value of 0.015 for the observing a density decrease within 200 μm of the center; for 800–1000 cells per mm2, n = 3 and p-value = 0.04; for 1000–1200 cells per mm2, n = 4 and p-value = 0.002; for 1200–1400 cells per mm2, n = 8 and p-value < 0.0001; for 1400–1600 cells per mm2, n = 10 and p-value < 0.0001; for 1600–1800 cells per mm2, n = 2 and p-value = 0.18, meaning that we cannot reject H0 only for this high density range.
For Fig. 4c: on the r = 90 μm patterns, n = 7 and p-value = 0.046; on the r = 120 μm patterns, n = 5 and p-value = 0.25. Therefore we cannot reject H0 on the pattern alone. However, a p-value of 0.018 is obtained for both pattern sizes combined, which we determined to be appropriate given that the fibroblasts obtained similar alignment on both pattern types.
For Fig. 4d: on the r = 90 μm patterns, n = 5 and p-value = 0.003; on the r = 120 μm patterns, n = 5 and p-values = 0.013; combined, a p-value < 0.0001 was obtained.
For Fig. 4e: on the r = 30 μm patterns, n − 4 samples were observed, and a density increase within 200 μm of the center of the +1 defect was observed, with a p-value = 0.026; on the r = 60 μm pattern, n = 5 and p-value = 0.97. For Fig. 4f: on the r = 30 μm patterns, n = 3 and p-value = 0.004; on the r = 60 μm patterns, n = 5 and p-value = 0.18. Therefore, we cannot reject H0 for the r = 60 μm patterns for EpH-4 cells.
The uncertainty in the defect prevalence is the uncertainty in a binomial distribution, . To determine the statistical significance between the binomial distribution for two different feature sizes, the following formula was used to obtain the test statistic, which was then used for a two-tailed t-test:
For the +1 defect prevalence shown in Fig. 6d, at r = 30 μm, 13 samples were observed; at r = 60 μm, 18 samples were observed; at r = 90 μm, 17 samples were observed; at r = 100 μm, 93 samples were observed; at r = 120 μm, 60 samples were observed.
For the −1 defect prevalence shown in Fig. 6d, at r = 60 μm, 18 samples were observed; at r = 90 μm, 7 samples were observed; at r = 120 μm, 7 samples were observed.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/d1sm00100k |
This journal is © The Royal Society of Chemistry 2021 |