Cell Heterogeneity Revealed by On-Chip Angiogenic Endothelial Cell Migration.

In sprouting angiogenesis, a key process involved in the development and the intravasation of tumor tissues, the growth of vessel sprouts, is determined by migration of single endothelial cells (ECs). This paper presents an on-chip assaying method to investigate the migration of individual ECs by simulating vessel sprouts with microchannels. When chemical stimulus is present, ECs were observed to migrate individually toward the source of factors instead of migrating collectively. The validity of this method is shown by inducing EC migration with glioma cell coculture and culture media doped with vascular endothelial growth factor (VEGF) 165. A positive correlation between cell displacement and VEGF 165 concentration was observed. Difference in migrating ability among cells was reflected by tracking single cells, which could reveal cell heterogeneity in susceptibility to stimulus.

Introduction

Sprouting angiogenesis, the process where new vessel branches are generated from pre-existing ones, is an important process in cancer development and metastasis for it enhances the oxygen and nutrient supply of cancer tissues and provides a passage for tumor cell intravasion. The vessel sprout generated in this process is led by an activated “tip cell” that senses chemical factors and migrates toward the gradient until it fuses with another sprout. This renders its migration decisive for vessel guiding, and its speed reflects the rate of sprout growth. A similar migrating behavior of ECs has been reported to be induced chemically by factors or matrixes or mechanically by shear stress.[1−3] The mechanism of migration involves a list of sequential events, including cell skeleton movement, forming of filopodia, and sensing of factor gradient.[1] Studies on the effects of vessel endothelial growth factor A (VEGF-A), a main angiogenic factor that induces EC migration through binding to receptor VEGFR2, have revealed that this process involves a complex set of components, including histone deacetylase 7.[4] protein kinase D1,[5] and regulator of calcineurin 1.4,[6] and the overall process is still far from being clear. ECs are already known to exhibit obvious heterogeneity in different tissue environments,[7] so we may also expect heterogeneity from microenvironment and intrinsic properties as well. As cells are known to exhibit notable heterogeneity in their response to drug stimulations,[8] considering the complex nature of the mechanism, this heterogeneity may have a major impact on cell activation and migration, which would affect the growth of vessel sprouts.

The deficits of in vivo models, including limited throughput and difficulty in manipulation and observation, urged in vitro models on the human umbilical vein endothelial cell (HUVEC) monolayer to emerge.[4,9] This led to the invention of scratch assay. It is the most common method for migration assay by now,[4,10] despite it simulates neither the geometry nor the scale of in vivo scenarios nor is capable to sustain a gradient of angiogenic factors. These deficiencies indicate a drastic difference in the microenvironment that would lead to different cell behaviors. To overcome them, microfluidic-based methods have been established, and to achieve better simulating accuracy, most are combined with 3D culturing techniques to recreate an in vivo-like tissue.[11] Such a model was first established by Zheng et al.[12] where HUVEC cells in collagen channels were reported to form tubes and invade into channel walls. Nguyen et al.[13] introduced angiogenic factor gradient into the chip, inducing directed new branches to form toward the angiogenic source. Bischel et al.[14] utilized microchannels for vessel lumen generation. Kim et al.[15] introduced glioma cell coculture into the model for a more accurate microenvironment.

Recently, microfluidic devices have been extensively utilized in the development of single-cell analysis platform.[16] Those in vitro models mentioned above have been successful in simulating the angiogenesis process and are inspirational for single-cell migration assays. However, extra effort is required as quantifying by measuring the sprout length is no longer available. In addition, cell interaction among EC tubes may conceal their individual properties. Recently, it has been reported that this single-cell mode migration can be triggered by highly degradable matrices where cell–cell junctions were broken.[17] This make culture medium-filled microchannels desirable for single cell assays: microchannel width hinders formation of cell–cell junctions, and culture media impede cell migration less than gel matrices. While these properties can achieve spatial separation of cells, they also enable the analysis of the same cell on its response to a stimulation for a prolonged time, which has been shown to be valuable for heterogeneity studies.[8,18] In our previous works, we have utilized microchannels in cell migration research studies.[19,20] Exploiting this similarity, herein, we combine previous on-chip models with microchannels and propose an on-chip angiogenesis model for cell migration.

Results and Discussion

To achieve this, a chip, as shown in Figure a, was fabricated with poly(dimethylsiloxane) (PDMS) using soft lithography (Figure S1) with a uniform height of 120 μm. The chip has three chambers: The middle holds an EC monolayer, such as HUVEC, mimicking the endothelium, and the side chambers function as the source of angiogenic factors where glioma cells such as U87 MG or angiogenic factors would be injected. The 50 μm-wide microchannels were designed to simulate the microvessels in their dimensions and, in addition, to restrict the horizontal EC movement while allowing ECs to separate vertically. The chip design is analogous with the in vivo situation, where the ECs form new vessel sprouts toward the source of angiogenic factors.

Figure 1

Design of the microfluidic device and the experiment scheme. (a) Two sets of microchannels are symmetrically placed on the endothelial cell culture chamber as control and treatment. The in vivo scenario was illustrated to show the correspondence. (b) Channels are sealed at first to allow independent cell growth in channels. They are later made open and allow factors to form a gradient, and endothelial cells would respond to the stimulus and migrate toward the glioma cell chamber. Certain cells will break away from the bulk during this process, exhibiting a single-cell migrating behavior.

When the gradient of chemical factors is generated by diffusion, the activated EC cells will migrate toward the source, that is, the U87 MG/factor chamber, through the microchannels where they are nearly not hindered as the microchannel is filled with the culture medium, as is shown in Figure b. Compared to the 3D culture or in vivo situations, intracellular connection on the 2D monolayer is much weaker, especially at low cell densities, which would allow cells to break away from the majority to display a single-cell migrating mode.

To seed the cells easily and to control the application of chemical factors on ECs, the content within the chambers needs to remain separated until a certain amount of factors was generated. To address this, the chip was made hydrophobic to prevent microchannels from opening when the solution was introduced, which was done through a long-time aging,[21] causing the polar groups (mostly silanol) on the surface generated by plasma washing to condense[22] and in turn eliminate hydrogen bonds, as is shown in Figure a. Consequently, all surfaces within the chip would require extra energy to be wetted. Because of the micrometer scale, the surface tension would prevent the liquid from entering the microchannels and separate the liquid apart (Figure b). As is shown in Figure b, 10 μg/mL fluorescein sodium was injected into the main channels to test the effectiveness of this method, confirming that no solute exchange happened at the interface.

Figure 2

Control of microchannels through hydrophobic treatment of the surface. Scale bars represent 100 μm. (a) PDMS surface becomes hydrophobic after aging eliminates surface hydroxyls. (b–d) Scheme of microchannel control. (b) Air prevents solute exchange between the main channels, shown by its bright field image with fluorescence overlay. (c) Remaining air gradually solves into the liquid phase. Image sample was obtained at 18 h. (d) Remaining air removed by 30–45 min of refrigeration, opening the microchannel.

When the liquid phase consists mostly of pure water, the amount of air within the microchannels does not decrease significantly over time. However, in the case of culture media, the organic solutes make it easier to wet and can moderately modify the surface hydrophilic. Thus, the culture medium would gradually invade into the microchannels and absorb the air in its way, causing a notable decrease in air amount over time (Figure S2). Approximately at 18 h after culture medium injection, only a small amount of air would remain, as is shown by the decreased length of air stored in the microchannels in Figure c. At this point, the remaining air can be completely removed through 30–45 min of refrigeration as air becomes more soluble at low temperatures and could be completely absorbed into the liquid phase. This would cause the microchannels to “open”, which allows the chemical factors to diffuse into the EC channel and form a gradient to stimulate EC cell migration (Figure d).

To test cell viability after refrigeration and to test their behavior, EC monoculture was performed with no angiogenic stimulus in which no significant cell death was observed until approximately 42 h after seeding (24 h after opening). During this process, despite the absence of stimulus, multicellular migration of ECs toward the gas–liquid interface in the microchannels was still observed, mostly in a multicellular pattern, as is shown in Figure a. One possible explanation of this behavior is that the surface in the microchannels was not occupied when it was first wetted, causing the ECs to migrate for better nutrition. The shear force generated by the flow of the culture medium when the liquid phase invaded the microchannel may also contribute to this phenomenon. Such a migrating behavior would cause a nonzero background, requiring control groups for following analyses.

Figure 3

EC migration induced by glioma cell coculture. Scale bars represent 50 μm. (a–c) Optimization of timespan. (a) Multicellular migration at 24 h after channel opening with no angiogenic stimulus. No significant cell death was observed. (b) VEGF-A generation in the coculture system. The concentration becomes relatively stable within 6–24 h after glioma cell seeding. (c) Simulation of VEGF diffusion. Within approximately 1 h, a uniform gradient can be formed within the microchannels. (d) Cell image obtained at 6 h after the opening of microchannels. ECs in angiogenic environment (left) migrate further than the control group (right), which is more conspicuous among foremost cells (marked). Fluorescence images after calcein live staining showed no observable death among migrating cells.

VEGF-A, the main angiogenic factors in this coculture system, is naturally secreted by U87 MG. It has been reported that in glioma cell bulk culture systems, the total VEGF-A concentration undergoes a dramatic increase within 72 h.[23] To apply a relatively stable migrating stimulus to the ECs, the VEGF-A concentration should not change greatly within the time of interaction. To achieve this, the enzyme-linked immunosorbent assay (ELISA) test was performed on supernatant samples obtained from the on-chip U87 MG cell culture at 6, 12, 24 h and was summarized, as is shown in Figure b. The result shows that the concentration of VEGF-A significantly rises to about 1 ng/mL within 6 h after seeding, about 10-fold compared to the result obtained from the petri dish at 24 h (0.1 ng/mL), possibly as a result from the microliter scale volume. The VEGF-A concentration remains on a steady increase from 6 to 24 h. This indicates that the latter period is fully available for testing. Considering the time for the microchannels to open and the potential risk of cell viability after 24 h, the time for cell interaction was chosen to be 18 to 24 h after seeding.

Diffusion of the chemical factors also largely affects the migration of the ECs: due to the peptide nature of these factors, it cannot reach its equilibrium instantly as smaller molecules. Take the example of VEGF165, the most abundant isoform of VEGF-A with a molecular weight of 38.2 kDa, whose diffusion coefficient can be estimated from an empirical formula[24]

Substitute T with 37 °C and η with 0.6913 mPa·s, the dynamic viscosity of water at 37 °C. This generates

Making an approximation that the factors can be instantly replenished or drained at the openings of the microchannel, the spatial distribution of concentration was determined by simulation. According to Fick’s first law, the flux is proportional to the gradient of concentration, so the proportion to original concentration (c/c0) can be used instead. The simulation result is shown in Figure c: the gradient spans over the microchannel within approximately 30 min and becomes nearly uniform in about 1 h, which remains stable. At 6 h, the final gradient does not deviate much from uniform distribution. As Fick’s first law and the empirical formula suggest with c/c0 and x remaining constant, a simulation was done with 10 μg/mL fluorescein sodium (Figure S3), which reached a consistent result with the previous simulation: diffusion of chemical factors completes rapidly and the gradient becomes mostly uniform.

To test the chip’s ability to reproduce in vivo scenario, a coculture system with glioma cells, a natural source for angiogenic factors, was established by loading U87 MG cells into a side channel. The microchannels were opened at 18 h after seeding, and the EC migration was observed at 6 h after the opening of microchannels (Figure d). To test the viability of ECs in this coculture system, at the end of observation, 10 nM calcein solution was injected into the EC channel and the chip was incubated for 30 min. As is shown in Figure d, the displacement of foremost EC cells in the coculture increases visibly compared to the control group where a nonangiogenic culture medium was injected instead. The fluorescence image shows that these migrating cells achieve good viability. This demonstrates the ability of this in vitro model to simulate EC migration cell in the glioma cell coculture.

To further examine the migration process and to find out the impact of angiogenic factor concentration on cell migration, an assay was conducted with the culture medium dosed with 0.5, 1, 2 ng/mL VEGF165, a common isoform of VEGF-A. The cells were observed at an interval of 3 h. Cell migration tests were performed at different factor concentrations, where the migrating cells was identified and tracked according to their location and morphology (Figure S4). Tracing results of migrating foremost cells are shown in Figure S5 and summarized in Figure a, where a significant increase was observed at higher VEGF165 concentrations, which establishes a positive correlation between EC migration and the factor gradient. This result reassures the validity of the model.

Figure 4

EC migration stimulated by VEGF165 gradient. (a) Summary of migrating distance over 0–3 and 3–6 h time span. EC migration increases significantly at higher VEGF concentrations. (b) Tracking result of cells in a microchannel, showing cell difference in their migration under angiogenic stimulus. Three cells and their displacement over time were marked with color to visualize the difference in migrating ability. Scale bars represent 50 μm. (*) p < 0.05, (**) p < 0.01, and (***) p < 0.001.

Apart from the overall increase of migration under an angiogenic stimulus, certain cells display a distinct displacement as is marked in Figure d and the extreme data in Figure a, a phenomenon reminiscent of in vivo tip cell migration. In these cases, the variation of response among individual cells is possibly magnified by analogy to the in vivo tip-cell selection process, where the activated cell suppresses the response of nearby cells through intercellular communication: Binding of VEGF on VEGFR2 upregulates the Notch ligand Delta-like 4 (DLL4). This will activate the NOTCH1 receptor of nearby cells, promoting their VEGFR1 expression, which competes with the binding to VEGFR2.[25] If an activated cell outcompetes nearby ones in this signaling pathway, it would exhibit a more distinct migration.

Obvious individual difference can be observed among cells, which can be magnified by high VEGF concentrations, which in turn suggests heterogeneity on migration at the cell level. Interestingly, the foremost cell is not necessarily the one with highest displacement in a certain microchannel, as is shown by 3–6 h in the last example of Figure b. Such a result eliminates the possibility that the difference is caused by the factor concentration alone as the concentration is always higher for foremost cells. This suggests that the difference in cell activation is related to their intrinsic properties, which implies that the ECs exhibit heterogeneity on their susceptibility to angiogenic factors.

Conclusions

In summary, we developed an assaying method of single EC migration on an on-chip angiogenesis model. Its validity was proven by reproduction of the in vivo scenario with glioma cell coculture. It was shown that the angiogenic effect exerted by the glioma cell causes single cell migration, as opposed to migration with adjacent cells. By further testing its reliability by stimulating the EC cells with a series of VEGF165-dosed culture medium, we confirmed its reliability by showing that the migrating distance of cells is positively related to the concentration of factors.

Furthermore, the heterogeneity of EC cells can be revealed by its migrating pattern. The difference in susceptibility, which has not been shown in the previous studies, is shown here as assays of individual cells can be more easily done within microstructures. The effect of this heterogeneity is possibly amplified by intercellular communication described before, resulting in a distinctive migration pattern among certain cells. As the chip design in this work is similar to previous angiogenesis studies, results obtained from this method would be comparable with earlier findings on which future works may eventually unveil the effect of EC heterogeneity on tip cell selection and migration.

Experimental Section

SU-8 2050 epoxy resist (MicroChem) was spinned at 1500 rpm onto a round silicon wafer with 7.62 cm diameter, generating a uniform photoresist layer of about 120 μm. The wafer was baked at 65 °C for 2 min, 95 °C for 2 min, and then 65 °C for 2 min. The photoresist layer is exposed to ultraviolet light through the mask in Figure S1b for 2 min. The wafer was again baked at 65 °C for 2 min, 95 °C for 2 min, and then 65 °C for 2 min and was soaked into the developer solvent while being stirred until the extra resist are removed. The surface of the resist is further silylated through gas phase deposition in a vacuum dryer with 10 μL of trichloro(1H,1H,2H,2H-perfluorooctyl)silane for 1 h.

Poly(dimethylsiloxane) (PDMS) base and curing agent (both Sylgard 184, Dow Corning) was mixed at approximately 10:1 (w/w), adequately stirred, dried in vacuum for 30 min to remove bubbles and solved air, filled onto the surface of the template, and is baked for 2 h. After baking, the PDMS layer is disintegrated from the template. Desired shapes are cut out and holes are punched. The PDMS moieties were later cleaned under a plasma cleaner for 2 min and fixed onto the glass, forming the desired channels and microchannels. The chip was then incubated at 65 °C for at least 36 h to make the surface hydrophobic.

Fluorescein sodium powder (Wako) was added into deionized water, producing a solution of 10 μg/mL. The solution was injected into the side channels, and then water was injected into the middle channel. Since the liquid will not advance in the microchannels, pressure was added to the surface of chip to force the microchannels to open, which would cause undesired mixing near the microchannels, resulting in a nonzero background. Pictures were taken at 15 min and 1 h (equivalent to ∼70 min and ∼5 h, respectively) upon the opening of microchannels, where the fluorescence pictures are shown in Figure S3.

Human umbilical vein endothelial cells (HUVECs) and U87 MG cells (both from Cancer Hospital, Chinese Academy of Medical Sciences) were respectively chosen as the sample of ECs and glioma cells. These cells were cultured using the minimum essential medium (MEM) (Corning) with 10% fetal bovine serum (FBS) (Sijiqing) and 1% Penicillin−Streptomycin (Corning) upon acquisition and was made into a suspension of about 106 cells/mL before injection.

ELISA kit was obtained from Abbkine, lot number ABB-KET6033-48T. The ELISA test was mostly performed according to the manual, except the supernatant obtained from U87 MG dishes was not diluted due to the low concentration. Ten microliters of U87 MG cells was injected into each side channel of the chip. Twenty microliters of sample was aspirated from each channel to ensure complete collection of VEGF, respectively, at 6, 12, and 24 h after seeding. The samples were centrifuged at 1000 g (3,800 rpm) for 20 min. Supernatant was separated and stored at −20 °C until use.

After the heating described above, 0.1 mg/mL poly-l-lysine (PLL) was injected into the main channels and was removed after 15 min to coat the chip surface, which improves cell adhesion. This is required as the previous heating destroyed the hydroxyl groups located on the surface, thus causing great difficulty for cells to adhere. With the complete removal of PLL solution, either by pipetting or replacing it with the culture medium, HUVEC cells were seeded into the middle channel. After 45 min of incubation at 37 °C, the U87 MG cell suspension was injected into the side channels. Liquid seals were added after another 45 min. The chips were later refrigerated at 4 °C for 30 min to allow the microchannels to open at exactly 18 h after seeding of U87 MG. Induction of significant EC migration would spontaneously initiate within the following 6 h by the angiogenic factors secreted by U87 MG and can be easily observed through tracing of the cells during this period using a reverted microscope. At the end of the observation, 10 nM calcein-AM (Dojindo) was injected into the channels and the chip was incubated for 30 min, which will specifically generate fluorescence within live cells. Cell viability was checked through the fluorescence signal produced by the cells, which could be used to determine the validity of the migration data.

The VEGF165 (PeproTech) powder was diluted to 10 μg/mL using phosphate-buffered saline (PBS) (Corning) and was further made into final concentrations of 500, 1000, and 2000 pg/mL with MEM. PLL modification and HUVEC seeding were the same as described before. The operating and observing procedures were almost identical to the cell coculture, except the U87 MG cell suspension was replaced with the respective solution of VEGF165: At 1 h after seeding, the contents on one side was replaced by injecting and aspirating 10 μL of medium. The liquid seal is broken between the side channel and the main channel to prevent undesired mixing or diffusion, so the concentration of VEGF165 within the channel was kept constant through the supply from the liquid seal. The following steps are identical to the U87 MG-induction test to replicate the result during the coculture. Images were taken at an interval of 3 h, and cells were identified by morphology and position, as is shown in Figure S4. The displacements of foremost migrating cells at these concentrations are summarized in Figure S5.