A high-throughput microfabricated platform for rapid quantification of metastatic potential

Abstract

Assays that measure morphology, proliferation, motility, deformability, and migration are used to study the invasiveness of cancer cells. However, native invasive potential of cells may be hidden from these contextual metrics because they depend on culture conditions. We created a micropatterned chip that mimics the native environmental conditions, quantifies the invasive potential of tumor cells, and improves our understanding of the malignancy signatures. Unlike conventional assays, which rely on indirect measurements of metastatic potential, our method uses three-dimensional microchannels to measure the basal native invasiveness without chemoattractants or microfluidics. No change in cell death or proliferation is observed on our chips. Using six cancer cell lines, we show that our system is more sensitive than other motility-based assays, measures of nuclear deformability, or cell morphometrics. In addition to quantifying metastatic potential, our platform can distinguish between motility and invasiveness, help study molecular mechanisms of invasion, and screen for targeted therapeutics.

A microscopic assay can detect metastatic cells with high sensitivity and resolution.

INTRODUCTION

Metastasis, when cancer cells spread from the primary tumor site to other organs, is a major cause of mortality (1). Distinguishing a metastatic cell from a nonmetastatic one is important to improve our understanding of the underlying molecular mechanisms of cancer metastasis and discovery of future therapeutics. The process of metastasis involves multiple biophysical steps. Therefore, assays have been designed to evaluate the mechanobiological characteristics of tumor cells, such as mechanical strength of their invasive podosomes (2, 3), their ability to modify extracellular matrix (ECM) mechanics (4), their deformability (5, 6), their biomechanical adaptability (7, 8), and their migration ability (9). These in vitro assays serve as practical cost-efficient screening platforms for drug discovery while allowing control of experimental variables. In addition, they often allow direct visualization of the invasion process, which is challenging for most of the more biologically relevant in vivo models. A commonly used in vitro assay is a version of the Boyden assay system, commercially available as a Transwell permeable support (10, 11). With modifications, this assay can model three-dimensional (3D) motility, invasion, and migration (12). Although this assay is compatible with adherent and nonadherent cell types, it requires many steps. Moreover, its sensitivity depends on seeding density, and it can be difficult to evaluate whether the distribution of cells is uneven. In addition, cell morphology and signaling events are difficult to visualize with high spatiotemporal resolution.

Wound healing or scratch assays are used to study cell migration and invasion (11, 13), and, although these assays allow easy visualization of the cells throughout the experiment, they are limited in their ability to yield reproducible results (13), as they are dependent on ensuring that the control and treated groups have the same confluence and may be influenced by the release of factors by the damaged cells (11). Furthermore, scratch assays may be artifactually influenced by secondary parameters, such as cell division or cell-cell adhesion signaling events. Although both the Transwell and scratch assays have been partially optimized (and commercialized) for multiwell formats, their quantitative outputs limit widespread adaptability and scalability.

In addition to these, several other cutting-edge technologies have been used to mimic the tumor microenvironment, including spheroid (14–16) and microfluidic models (17, 18) to name a few. Microfluidic systems integrate innovative designs, such as incorporating electrical cell–substrate impedance sensing with the Boyden chamber design (17) or use a “Y”-shaped design to segregate concurrently the relative number and proliferative state of breast cancer cells (19). In addition, several assays have been developed to study various aspects of the metastatic cascade including invasion (20), migration (21), extravasation (22, 23), and angiogenesis (24, 25). These assays offer the advantage of minimal reagent consumption and mimicking a 3D in vivo phenotype; however, they are often limited by involving increased hands-on time to conduct experiments and daily exchange of growth medium. Another correlate of a cancer cell’s invasive potential is biomechanics, primarily its nuclear deformability (26). While Transwell and scratch assays are the most widely used directed migration assays, they are uninformative of the biophysics of the cells being tested. Nuclear deformability (or compressibility) is a key attribute of tumor cells to invade through the encapsulating ECM (27). Nuclei deform while navigating interstitial spaces (28) through an inelastic process (29); as such, several innovative microfluidics (27, 30) or microfabrication-based (31) assays have been proposed to measure nuclear deformability as a means of gauging metastatic potential.

These in vitro assays allow researchers to investigate various aspects of the metastatic cascade; however, they often provide an aggregate readout from many cells or require sophisticated fabrication techniques that are difficult to mass-produce. As such, there is a need for an easy, automatable, high-throughput, and quantitative assay that allows unbiased assessment of metastatic potential. Here, we present a new microchannel assay that uses a microfabricated pattern with narrow constrictions smaller than the typical nuclear diameter, requiring cells to actively deform for invasion. We observe cellular motion using live-cell microscopy on six different solid tumor lines and find that highly metastatic cells frequently pass through microchannels, whereas nonmetastatic cells cannot. On the basis of this observation, we present a simple “invasion index” that quantifies metastatic potential. The sensitivity of the invasion index is greater than that found using scratch or Transwell assays. We adapt our substrates to a 96-well plate format and assess the effectiveness of different compounds independently on invasiveness and motility. We demonstrate a robust system to rapidly quantify the metastatic potential of cancer cells and gauge their response to therapeutics.

RESULTS

Our imaging-based microchannel assay uses photolithographic patterning of SU-8–like epoxy photoresist on glass (Fig. 1A; fabrication details in Materials and Methods). The micropattern consists of 7-μm-tall structures, with large open chambers interconnected by 5-μm-wide channels (Fig. 1B). The micropatterns are treated with a 0.5% solution of Pluronic F-127 in phosphate-buffered saline (PBS) to render the epoxy nonadhesive; when cells are plated on these micropatterns, they preferentially localize to the glass and avoid the treated epoxy (Fig. 1C). This platform is easily adapted to a multiwell format, an advantage that allows multiplexing of several experiments. Here, we show (Fig. 1D) incorporation into a 96-well plate by affixing the glass to a bottomless plate, a process that can easily be used for other multiwell formats.

Fig. 1. Microfabricated platform on glass allows tracking of cells with high spatial resolution during basal invasion activity.

(A) Schematic representation of the photolithographic process used to fabricate the platform. UV, ultraviolet. A negative-tone SU-8-like photoresist was used to transfer (B) a repeated array of 2500-μm² square chambers interconnected with 5-μm-wide (width) and 30-μm-long (length) channels. (C) Scanning electron microscopy (SEM) of 3D micropatterns shows cellular compliance within the patterns. (D) Schematic showing how a micropatterned substrate is integrated with standard 96-well plates allowing for controlled multiplexed experiments with different cell types or antimetastatic drug candidates. (E) Representative immunofluorescent images of triple-negative LM2-4 and MCF-7 breast cancer cells plated on micropatterns representing cellular distribution 1 day after seeding on the micropatterns. (F) Still frame high-resolution image sequence showing a triple-negative LM2-4 cell (green) squeezing through the length of a channel (purple) as imaged in 3D on a modified lattice light-sheet microscope.

To demonstrate the use of this platform as an assay to quantify the metastatic potential of cancer cells, we chose the MCF-7 and MDA-MB-231 LM2-4 breast cancer cell lines. MCF-7 is a commonly used estrogen and progesterone receptor–positive breast cancer cell line (32) that is known to be noninvasive with low metastatic potential (33, 34). In contrast, the LM2-4 cell line is a lung metastatic variant of breast cancer cells generated from the triple-negative MDA-MB-231 human breast cancer cell line (35). This line is known for its in vivo spontaneous metastatic spread (36), and, therefore, it serves as a good contrast to the nonmetastatic MCF-7 cell line. The two cell lines show distinct spreading behaviors in the micropatterns upon seeding (Fig. 1E) with increased arborization visible for the LM2-4 cells. We observe that both cell types are highly motile; however, the metastatic LM2-4 cells frequently squeeze through the narrow channels, whereas nonmetastatic MCF-7 cells do so rarely. Aberrant cell migration observed in metastatic cells is a cyclical multistep process: extension of the leading edge, adhesion to subcellular matrix, cytoplasmic contraction, and release and reprocessing of membrane receptors of the following edge (37, 38). Our system can capture all the stages of migration in high spatial resolution as a single LM2-4 cell traverses through the narrow channel of our micropattern (Fig. 1F). Still frames of volumetric image sequences, captured using a modified lattice light-sheet microscope (39, 40), show extension of pseudopodal protrusions followed by retraction of the lateral end (movie S1).

Our microchannel assay represents a subtext analogous to the in vivo process of invasion while retaining assay flexibility that allows quantitative, high-magnification optical investigation of the migratory process in real time. In our microfabricated chambers, we can use live-cell tracking to evaluate motility and invasiveness simultaneously, and, because our wells are designed to contain micropatterns only in certain areas within the multiwell chamber, we can effectively assay the physiological effects of micropatterning on cells in a paired manner (i.e., micropatterned and unpatterned controls in the same well). Using MCF-7 cells, we assessed apoptosis using immunofluorescent labeling of cleaved caspase-3 followed by high-content image analysis and found no detectable changes (fig. S1). In addition, an interrogation of parameters, such as cell proliferation and overall survival using a series of six different tumor cell lines (MCF-7, Huh7, HepG2, MDA-MB-468, LM2-4, and LCC6; fig. S2A) using 5-Ethynyl-2'-deoxyuridine (EdU) incorporation (fig. S2B) and 3-(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide (MTT) colorimetric assay (fig. S2C), showed no detectable differences between unpatterned and micropatterned cells. We confirmed that type of nuclear tracking marker also did not have any significant effect on either total traveled distance or the invasion index (fig. S3). We used the invasive LCC6 tumor cell line to evaluate the impact of different ECM coatings on motility and invasiveness; our assay revealed that LCC6 cells exhibit a similar invasion index on uncoated, fibronectin-coated, and laminin-coated microchannels (fig. S4). We likewise repeated the proliferation and survival assays on both unpatterned and micropatterned surfaces; these showed no change in survival and a similar decrease in proliferation on laminin-coated surfaces for both unpatterned and patterned surfaces (fig. S5). Similarly, morphometrics on both unpatterned (fig. S6) and micropatterned (fig. S7) surfaces showed that changes associated with the ECM coating were consistent across the two groups.

Rapid quantitative assay to assess invasiveness

To track the cells, we label the nuclei using the Hoechst 33342-derivative NucBlue. We image cells over a 24-hour period with a rate of three frames per hour on a Leica DMi8 widefield microscope equipped with an automated black-box live-cell incubator stage. We image 75% of the entire well using 20× tile scans. The centroid of each nucleus is tracked using the TrackMate plugin in Fiji (41). Schematic of the entire process is shown in fig. S8. Trajectories of MCF-7 (Fig. 2A) and LM2-4 (Fig. 2B) nuclei recorded over 24 hours show that the MCF-7 cells stay within a single chamber, while LM2-4 cells traverse into the neighboring chambers. While both cell lines were highly motile, the LM2-4 cells show significantly higher motility; by measuring the total length of each cell trajectory, we find that the median distance traveled by an LM2-4 cell is roughly 1.3 times that of an MCF-7 cell (Fig. 2C). Observing the trajectory for each cell within the micropattern allows us to examine whether differences in motility can account for the observed large difference in invasion index between these two cell lines. The higher frequency of LM2-4 traversing through the narrow channels suggests that its metastatic potential is roughly 32 times higher than that of MCF-7 cells (Fig. 2D). To account for the effect of the motility on the invasion, we compute the metastatic potential as the ratio of the calculated invasion index to average distance traveled; the invasion index is significantly higher for LM2-4 cells when normalized to the total amount of distance traveled as well (Fig. 2E). A histogram of the number of channel crossings (Fig. 2F) shows that almost none of the MCF-7 cells “invade” into the adjoining chambers, whereas roughly 40% of the LM2-4 cells invade at least once.

Fig. 2. Microchannel assay can be used to quantify metastatic potential.

(A) Mapped trajectory of the nucleus of an MCF-7 (pseudocolored blue) and (B) triple-negative LM2-4 breast cancer cell line (pseudocolored red) seeded on micropatterns. Nuclei tagged with NucBlue were imaged over 24 hours. Captured images were subsequently superimposed to show cell trajectories. Less invasive MCF-7 cells remain trapped within the 2500-μm² chambers, whereas highly metastatic LM2-4 cells invade into neighboring chambers through the narrow channels by squeezing their nuclei. (C) The mean total distance traveled by the migrating LM2-4 cells (n = 337) is significantly higher than that of MCF-7 cells (n = 252). (D) Cell invasion is quantified by computing the ratio of cellular invasions crossing in between interconnected chambers to the total number of cells visible in a viewing area. Invasion index for MCF-7 and LM2-4 breast cancer cells in the micropatterns shows nearly two orders of magnitude difference between the metastatic line compared to the nonmetastatic line. (E) When computed, the ratio of the invasion index to the distance traveled is roughly two orders of magnitude higher for the LM2-4 reiterating the high metastatic potential of this cell line. (F) A normalized histogram of the number of crossing events of the two cell types shows the percentage of cells in the viewing area that have a greater tendency to “invade” (**P < 0.001 and ****P < 0.0001).

Microchannel assay is more sensitive than the current state of the art

We next compare the performance of our microchannels to two of the most commonly used methods to assess cancer cell invasiveness: Transwell and scratch assays (fig. S9). The Transwell migration assay involves seeding cells on a porous membrane and inducing migration through the 3D membrane via chemoattraction. The quantitative endpoint is the optical density of the eluted crystal violet stain that labels transmigrated tumor cells measured at the appropriate wavelength. This assay measures invasion through narrow channels and combines low cost of implementation with ease of use. However, it requires precise control over seeding density and does not independently assess cellular motility (12), and all the phenotypic single-cell information is lost because the migrated cells are all lysed at the end (11). Scratch, or wound healing, assay involves creation of a linear gap (wound) within a continuous cell monolayer and recording the rate of closure (healing); it is a measure of directed cell migration. Cellular morphology and phenotype information, unlike the Transwell assay, are not lost and may be observed in real time. Drawbacks include an inherent variability in creating the scratch and the lack of the invasion context. We examined MCF-7 and LM2-4 cells using the Transwell assay with 8-μm pore size (comparable to the 5-μm width of our channels), which showed that the two cell lines have significantly different transmigration rates (Fig. 3A); scratch assay showed that LM2-4 cells had a significantly higher rate of directed migration (Fig. 3B). When we determine the effective sensitivity of the three assays by comparing the inferred rate of invasiveness of MCF-7 and LM2-4 cells, we saw that our microchannel assay exhibits a sensitivity about 3.5 to 10 times greater than that of scratch and Transwell assay, respectively (Fig. 3C). The Transwell assay is known to be highly sensitive to the chosen geometry, i.e., pore size (42, 43). To evaluate the range of sensitivity for our assay as a function of geometric dimensions of the microchannels, we fabricated micropatterns of varying dimensions (width: 5, 7, and 9 μm; length: 20, 30, 40, and 50 μm). Our results show that our assay sensitivity was relatively unaffected by the changes of these parameters (fig. S10). Another limitation of both the scratch and Transwell assays is their sensitivity to initial plating conditions. To determine the effect of cell density on our findings, we binned the measured invasion index values as a function of starting local microchamber density. Evaluating time-course videos of LM2-4 cells over a large array of microchambers (n = 26,100 to 34,800), we noted that the relative probability of an invasion event occurrence in a microchamber was not significantly associated with the starting number of cells (fig. S11), suggesting that our assay is less sensitive to plating density.

Fig. 3. Microchannel assay is more sensitive than current state-of-the-art high-throughput platforms.

(A) Crystal violet–stained representative images of MCF-7 and LM2-4 cells migrating in the Transwell chamber migration assay, under the same conditions as those of the micropattern assay, confirm that the inferred migration of LM2-4 cells is significantly higher than that of MCF-7 cells (scale bars, 20 μm). (B) Similarly, representative bright-field microscopy images of MCF-7 and LM2-4 cells migrating in a 2D assay, where a confluent monolayer is “injured” using a 0.5-mm-wide pipette tip, under the same experimental conditions as those of the micropattern assay, show significantly faster wound closure in LM2-4 cells within a 24-hour window (scale bars, 200 μm; ****P < 0.0001). (C) The effective sensitivities of the three assays, as computed by taking the ratio of inferred invasion of the metastatic LM2-4 cell line versus that of the nonmetastatic MCF-7 cell line, show that the microchannel assay exhibits superior sensitivity.

Greater dynamic range compared to nuclear biomechanics or morphometrics

Cellular and nuclear stiffness has been used as a biophysical proxy for the metastatic potential of cells (44, 45). This approach has been extended to tissue slices (46) and various techniques developed to focus specifically on the nuclear stiffness of cancer cells on substrates mimicking physiological stiffnesses (47). Because invasion requires both nuclear deformability and contractile force generation, we chose to quantitatively compare the ability of individual tumor lines to invade through our microfabricated channels as a function of their nuclear deformability. Using atomic force microscope (AFM) elastography, we measured the nuclear apparent elastic modulus of the six selected cancer cell lines of varying metastatic potential (Fig. 4A). We noted that the nuclear deformability of the six tested cell lines varied greatly (Fig. 4B and table S1) and increased along with its invasion index (Fig. 4C and table S2). The negative correlation between the measured elastic modulus and invasion index was significant [correlation coefficient (r) = −0.87, P < 0.02; fig. S12A). This correlation was also visible in the results obtained by the Transwell assay measurements (r = −0.92, P < 0.01; fig. S12B), which showed a similar trend for the same six cancer cell lines (Fig. 4D and table S3). Correlation between the Transwell migration values and invasion index measurements was the strongest (r = 0.99, P < 10⁻⁴) and most significant (fig. S12C); however, the relationships between the two indices of invasiveness and the nuclear moduli were not statistically different (P = 0.88; table S4). While there was no difference between mean-variance relationship of the two assays, the dynamic range of the microchannel invasion assay was nearly two orders of magnitude greater than both the Transwell and nuclear elasticity measurements. Because morphology has also been associated with invasiveness, we performed morphometric analysis of unpatterned cells (fig. S13). We did not observe any strong correlative trends in nuclear spread area, cell spreading area, nuclear solidity, or cytoplasmic solidity (fig. S14). This was true for nuclear morphometric properties of cells in micropatterns as well (fig. S15), which showed similar quantitative trends (fig. S16).

Fig. 4. Decreased nuclear stiffness correlates with invasiveness, but it is not the principal factor.

(A) Representative phase contrast images of the six cell lines probed by AFM elastography. (B) The six tumor cell lines [MCF-7 (breast, n = 54), Huh7 (liver, n = 56), HepG2 (liver, n = 60), MDA-MB-468 (breast, n = 51), LM2-4 (breast, n = 56), and LCC6 (breast, n = 21)] were probed around their nuclear regions via AFM indentations. Nuclear stiffness of both LM2-4 and LCC6 cell lines was significantly lower compared to all the other cell lines (see table S1). This was coupled with an increase in the metastatic potential as seen in both the (C) microchannel and (D) Transwell assays run for all six cell types. Both LM2-4 and LCC6 cell lines had significantly higher invasion indices compared to the other cell lines (see table S2). This trend was similar in the average Transwell absorbance as well (see table S3). *P < 0.01.

Separation of drug effects on motility and invasion

Most of the antimetastatic drugs identified through high-throughput screening fail in further studies (48). One of the reasons for this is the nonspecific nature of all the currently used high-throughput assays and their inability to distinguish whether a drug affects cell viability, metabolism, motility, directed migration, or invasiveness, which are all intricately linked cell biological processes. To demonstrate how our microchannel assay can be parametrized to investigate all these phenomena independently, we used a known perturbagen, the small-molecule tyrosine kinase inhibitor (TKI) BMS-754807, which is known to target insulin-like growth factor 1 (IGF-1) signaling and affect metastatic potential of LM2-4 cells in vivo (49). As previously ascertained for the baseline setting on multiple cell lines, we show that the micropatterns do not affect the inherent cell proliferation mechanism when conducting drug efficacy experiments (fig. S17). Using the scratch assay and our microchannel invasion assay, we respectively measured the directed migration and invasion index of LM2-4 cells in the presence and absence of BMS-754807. Scratch assay measurements showed that insulin receptor (IR)/IGF-1 receptor (IGF1R) inhibition by BMS-754807 reduced directed migration significantly (Fig. 5A), which correlated with a decrease in cell spreading area upon treatment with the inhibitor and increase in cell solidity (fig. S18). Median distance traveled (i.e., cumulative length of 24-hour trajectory), the invasion index, and the invasion index normalized to the total displacement were also significantly reduced in our microchannel assay (Fig. 5, B to D), in spite of a lack of significant difference in nuclear morphometrics of patterned cells (fig. S19). To validate that our unperturbed motility measurements are not biased by the microchannels, we designed an alternative “open-layout” version of our chip (Fig. 5E). Our results showed that BMS-754807 causes a similar reduction in intra-chamber motion for both systems, suggesting that the main impact of the drug was on cell motility rather than cell invasiveness (Fig. 5F).

Fig. 5. Microchannel assay can separate effects on invasiveness and motility in drug screening.

Results from representative antimetastatic drug efficacy experiments, where LM2-4 cells cultured with charcoal-stripped medium (LM2-4cs) were treated by either vehicle or a small-molecule TKI, BMS-754807, that inhibits IR/IGFR1 activity. (A) Percent wound closure in the scratch assay for LM2-4 cells versus those treated with BMS-754807. (B) TKI-treated LM2-4 cells plated on the microchannel assay showed a significant decrease in total distance traveled. (C) The invasion index of TKI-treated LM2-4 cells also showed a significant decrease compared to its untreated counterpart. (D) This trend remained when invasion index was normalized invasion to the total displacement. (E) To unambiguously separate motility from invasion, we created an open layout version of our micropatterned surfaces, where cells can migrate freely. (F) Normalized amount of intra-chamber motion (i.e., invasion index for the 5-μm layout) showed a decrease in both versions though the interaction term was not statistically significant per two-way analysis of variance (ANOVA) with Bonferroni correction. (G) Representative phase contrast images of vehicle and trichostatin A (TSA)–treated MCF-7 cells probed by AFM elastography. (H) The MCF-7 cell line at baseline (n = 22) and after TSA treatment (n = 21), when probed around the nuclear region, showed a significant decrease in stiffness. (I) The total distance traveled by MCF-7 cells treated with TSA (n = 462) exhibits a significant decrease when compared with the untreated MCF-7 cells (n = 776). (J) The corresponding invasion index shows a significant increase in the treated versus control cells, an increase that is much higher for (K) the invasion index normalized to the total traveled distance (*P < 0.01, **P < 0.001, and ****P < 0.0001).

Although our assay can separate the drug effects on motility and invasion, given the likely dual impact of most TKIs on these two parameters, we opted to pharmacologically validate our assay’s ability via another strategy. Using a first-generation histone deacetylase (HDAC) inhibitor trichostatin A (TSA) on MCF-7 cells (Fig. 5G), we were able to significantly reduce their nuclear elastic moduli (Fig. 5H). While TSA significantly reduced total traveled distance from 115 ± 121 μm to 78 ± 60 μm (Fig. 5I; P < 0.0001), it significantly increased invasion index from 6.0 ± 5.1 to 31.4 ± 7.9 (Fig 5J; P < 0.001) by increasing nuclear deformability as previously shown (26). We noted that, similar to other cell types and the BMS-754807 treatment case, the micropatterns did not have a significant impact on cell proliferation or viability (fig. S20). TSA treatment further caused a significant reduction in nuclear solidity in both unpatterned control (fig. S21) and micropatterned contexts (fig. S22), mirroring quantitative morphological observations in the more invasive cell lines. When normalized to the total distance traveled, the effect of TSA on invasiveness was even more apparent (Fig. 5K).

DISCUSSION

Metastasis is a biomechanical event that requires tumor cells to invade specialized processes through the encapsulating matrix, generate contractile forces to pull through the matrix, and sustain adequate levels of nuclear compression to squeeze through. Our results recapitulate the importance of nuclear deformability, and they mostly agree with prior comparative classification of invasiveness presented by these earlier studies. However, assays that measure any one of these parameters could potentially be missing information regarding the metastatic potential of the cell. We have engineered a system that can potentially assess all three biophysical aspects of metastasis. Because of its holistic approach, our assay shows a superior dynamic range compared to the commonly used assays, and it is more sensitive than the direct biomechanical readouts of nuclear deformability or morphology, which have been suggested as strong indicators of metastatic potential.

Several novel specialized assays have been recently proposed to probe cancer cell biomechanics in relationship to invasiveness (2–4, 6); among these, nuclear deformability assays that use elastomeric microchannels exhibit a robust ability to classify metastatic potential with great diagnostic potential (19). Despite these exciting developments though, most of the commonly used high-throughput assays provide endpoints for aggregate invasive activity (e.g., Transwell assay measures bulk migration of seeded cells and scratch assay measures average motility). Such generalized metrics may lose the granularity that the metastatic process is known to exhibit, which is critical because metastasis is an inefficient process (50, 51) with studies estimating that only 0.01% of circulating tumor cells form metastatic foci (50, 52). In addition, the generalized results of the scratch assay can be substantially affected by peripherally related metrics, such as cell-cell adhesion and proliferation. Further, in vivo tumors are heterogeneous, comprising subpopulations of cells (53). The result of this heterogeneity is a diverse collection of cells with different genetic and molecular signatures resulting in a nonuniform response to therapies (54). A spatial assay that is capable of phenotyping single-cell tumor response while simultaneously visualizing their individual invasive behavior can enable improved therapeutics and basic science discovery while identifying whether a tumor contains cells that could metastasize.

The role of cellular motility in the context of metastasis has been studied in detail (9). While increased cellular motility is one of the hallmarks of a metastatic tumor (55), it is not the sole factor that determines invasive potential. The ability of a cell to form invasive peripheral structures, such as invadopodia (56); to respond to growth factors (57); or to elastically deform its nucleus through the lamellar shell of the originating organ (28) are all important in establishing its metastatic potential. As such, an assay that can distinguish the independent therapeutic impact of drugs on these distinct metrics could be a powerful way to decode mechanisms of action. We note that our micropatterns have only minimal impact on overall motility of cells (fig. S23).

Motility is a fundamental biological process that is necessary for the function of numerous organ systems, whereas invasion is a hallmark of cancer and an aberrant biological process that does not have any functional representation in healthy tissues. One of the unique attributes of our assay is its ability to distinguish drug effects on cellular motility from cell invasiveness that better correlates with dangerous metastatic potential. We present two case studies, one in the context of IR/IGF1R signaling and another using heterochromatin modification. Mitogenic role of IGF-1 in breast carcinoma cells (58) and the therapeutic potential of targeting the receptor binding domains with small molecules have been discussed previously (59, 60). BMS-754807 was also shown to be an effective IR/IGF1R targeting kinase inhibitor (60). This small molecule functions by inhibiting the catalytic domain of the IGF1R while blocking the activity of both IGF1R and IR in in vitro assays (49). While our conclusions may not apply to other tumor lines, we see that BMS-754807 acts primarily through its effect on reducing cellular motility, not necessarily its impact on the invasive machinery of the highly metastatic LM2-4 cell line. One may consider that, pragmatically, a reduction of invasion events is desirable regardless of the mechanism. However, it should be noted that metastatic tumor cells are known to have numerous unique molecular mechanisms associated with invasion (e.g., invadopodia), which, if targeted independently, may lead to reduction in metastasis with lower chance of adverse events. In another example, we show that HDAC inhibitor TSA can reduce nuclear elasticity leading to a 500% increase in invasiveness while paradoxically reducing motility by 33%. Our results may help explain the seemingly contradictory reports of increased matrix invasion (61) and reduced motility (62) observed in TSA-treated cells.

We have shown a proof-of-principle application of our system and its ability to dissect differential drug effects using two different solid tumor line and drug combinations. However, our system is versatile enough to be deployed in large-array formats, which would allow unbiased high-throughput screening of compound libraries. Using this system, one could identify compound families that preferentially target only a specific segment of the molecular machinery that is responsible for metastasis. In addition, while it would require further study and optimization, our system could potentially be deployed as a translational tool to screen for in vivo metastatic potential by determining the percentage of invasive cells within a dissected primary tumor. When we compare the invasion index rankings of the cell lines we tested with some of the recent nuclear deformability studies using elastomeric microchannels, we saw that the conclusions on the metastatic potential of different cells lines were very similar (19). We note that the surface chemistry and material composition of elastomic systems are quite different from our epoxy-based platform, which presents a relatively inert interface. Nevertheless, to ensure that material properties of our epoxy did not affect the invasion index, we repeated our LM2-4 measurements using both SU-8-2005 and mr-DWL5 photoresists, which showed that the surfaces had no effect on the invasion index (fig. S24).

Our study is not without limitations. It inherently lacks the context of the native tumor microenvironment, including tissue-like substrate elasticity or a degradable 3D matrix. Particularly, the inert and stiff nature of microchannels in their current form limits the study of several biomechanical aspects of metastasis. However, lacking in vivo contextual complexity makes it possible to study subcellular metastatic events as well as heterotypic interactions using high-power optical microscopy in real time. While most of these studies were performed without any specific surface coating or directionally diffusible cues, we show that treatment of different ECM proteins can be applied to our system (fig. S4). The invasion process, however, may involve a number of other microenvironmental cues, such as heterotypic interactions that guide cytoskeletal dynamics and collective cell behavior. While some of our analyses suggest that cell-cell interactions may influence invasion index, these require further study. While we noted that our system is less sensitive than other assays to both initial geometric selection and cell seeding density, it is possible that the invasiveness of other cancer cell types (that we have not yet probed) may require optimization. Future iterations of our system may involve optimization of geometric characteristics on primary tumor lines, which may help “tune” the system response to cell-type– or tumor-specific contexts.

We note that our system allows unparalleled access to visualizing various morphological details that are difficult to capture in other assay forms. An example of this is the formation and cycling of podosomes on the leading edge of the LM2-4 cell as it extends through the microchannel as visualized in movie S1. While we visualized a line that stably expresses an F-actin biosensor, LifeAct, such high spatiotemporal resolution could allow the study of other cell signaling events through different biosensors. Combined with newly developed high-speed superresolution techniques, our platform could allow the discovery of new molecular mechanisms responsible for invasiveness.

MATERIALS AND METHODS

Micropattern fabrication and assembly

We used #1.5 glass (Electron Microscopy Sciences, 63793-01) as a substrate on which to fabricate micropatterns because glass with this thickness is compatible with high-fidelity optical imaging on most imaging platforms. The glass was cleaned thoroughly by submerging in a bath of acetone followed by isopropyl alcohol for 15 min each. To eliminate any adsorbed moisture, the samples were dried on a hot plate at 150°C for 30 min. Following this, the samples were cleaned in a combination of oxygen and argon plasma for 10 min at a power of 100 W. For large surface areas, we performed direct write photolithography using a Heidelberg DWL 66+ Laser Writer. This uses the negative-tone photoresist mr-DWL5, which we spun to achieve a thickness of at least 7 μm. The spin coat was followed by a 2-min postexposure bake at 95°C and a 10-min annealing at room temperature to reduce any inherent stress in the resist. The pattern was transferred onto the surface with a direct-write maskless system. After direct-write exposure, the sample went through a postexposure bake at 95°C for 3 min. It was then developed in an SU-8 Developer followed by a wash in isopropyl alcohol and dried with N₂. The sample was hard-baked at 150°C for an hour to further ensure robust adhesion of the patterns to the glass.

To retrofit the micropatterned sample onto a standard 24- or 96-well format, we began with a bottomless black Greiner multiwell plate (VWR, 89085-338/655000-06). The rear surface was cleaned with 70% ethanol to remove any debris or particulate matter that would prevent flat surface contact. Biocompatible double-sided tape (ARcare 90106NB) was laser cut with a Thunder Laser Nova 24 laser cutter to the desired specification and mounted on the bottom surface and smoothened down with a blunt flat slab of acrylic, ensuring no air bubbles remain, avoiding uneven flatness of the mounted glass. Once the second layer of protective covering was removed from the double-sided tape, the micropatterned glass was carefully lowered onto the tape with even pressure applied on the same slab ensuring a robust seal between tape and glass.

Cancer cell lines

MCF-7 (breast; ATCC, HTB-22), HepG2 (liver; ATCC, HB-8065), and MDA-MB-468 (468, breast; ATCC, HTB-132) cells were purchased from the American Type Culture Collection (ATCC), VA, USA. Huh7 (liver; XenoTech, JCRB0403) cells were purchased from XenoTech, KS, USA. LM2-4 cells (63) were generously supplied by P. Reddy. LCC6 cells (64) were generously supplied by D. Yee.

Cell seeding on micropatterns

We prepared the surfaces following a previously established protocol (65). Briefly, the multiwell plate was sterilized under ultraviolet-C light for 30 min. Following this, each well was filled with sterile 0.5% solution of Pluronic F-127 (Sigma-Aldrich, P2443) in 1× PBS for 1 hour followed by thorough serial washes with 1× PBS to ensure no residual surfactant. This treatment renders the pattern structures more hydrophobic, allowing the subsequently seeded cells to adhere preferentially to the glass. After the last wash with PBS, cell culture medium suitable to the respective cell to be seeded replaced the PBS in the wells. Following this, the wells were seeded with at a density of 20,000 cells per well of a 24-well plate or 3000 cells per well of a 96-well plate.

Cell motility and invasion

Media in the wells were supplemented with 1:10,000 Hoechst 33342–based NucBlue dye (Thermo Fisher Scientific, R37605), which allowed tracing of cellular motion without any other endogenous fluorophores. After a 30-min incubation, the plate was transferred to a Leica DMi8 widefield microscope equipped with a temperature-controlled black-box incubator stage supplemented with 5% CO₂ for live-cell imaging.

After a preliminary scan for adjustment of exposure time and laser power to the minimum value that allows high signal-to-noise ratio, a region of interest (ROI) to fill 75% of the well area was selected. Continuous focus adjustments were prescribed, and the autofocus control feature of the microscope was enabled. Time-lapse imaging was performed with a frame rate of three frames per hour for 24 hours. Movies were exported to be analyzed after a mosaic merge was performed on the tiled ROI.

To compute the invasion index, an ROI was identified, and the total number of nuclei in the viewing area was counted. The total number of invasion events is counted; the definition of invasion is the act of the nuclei traversing through the width of the pattern. The invasion index is computed using the ratio of these numbers

$$\text{Invasion index} (\%)=\frac{\text{Total number of invasion events in} \text{ROI}}{\text{Number of cells in} \text{ROI}}\times 100$$

Generating biosensor LM2-4 cell line

Replication incompetent lentivectors were produced via polyethylenimine (Polysciences Inc., Warrington, PA)–mediated co-transfection of 293T cells with vvPW and pPAX2 and pMD.G2 packaging plasmids (Addgene, Cambridge, MA) followed by concentration by centrifugation. Transduced cells were flow sorted 48 hours after transduction to establish stable polyclonal populations expressing low and intermediate LifeAct levels.

TSA treatment

For experiments that required controlled concurrent measurements, MCF-7 cells were cultured in parallel on micropatterned and unpatterned substrates. Culture medium with TSA at a final concentration of 200 ng/ml (Sigma-Aldrich, T1952) was made. Cells were incubated in this solution for at least 2 hours before the experiment was started. TSA-treated cells were evaluated for proliferation, viability, nuclear stiffness, invasion index, and Transwell migration using the protocols described in this study.

Evaluation of surface coating effects

ECM proteins laminin and fibronectin were used to determine whether the impact of different matrix molecules could be assessed on our microchannel assay. Stock solution (1 μg/μl) of fibronectin (Gibco, 33016015) was made. This was diluted in 25 ml of sterile water for a final concentration of 10 μg/ml. For the microchannel assay, the surfaces were treated as detailed in the section on cell seeding with microchannels. Following the treatment with a sterile 0.5% solution of Pluronic F-127 and its subsequent wash, the solution (10 μg/ml) replaced the last wash and was incubated for 2 hours at 37°C. Following the incubation, the fibronectin containing solution was replaced by cell culture medium, not allowing the surface to dry followed by seeding the cells. For unpatterned surfaces, the solution (10 μg/ml) was placed on the prepared sterile glass surface followed by a 2-hour incubation at 37°C. The fibronectin solution was then replaced by cell culture medium followed by seeding the experiment relevant cells.

Recombinant human laminin-511 E8 fragment protein (iMatrix-511; Nippi, 892012) was the second ECM protein used to interrogate the effect of surface coating. A dilution of the stock solution in PBS was used at a final concentration of 0.5 μg/cm². The diluted laminin was then placed on micropatterned substrate and allowed to incubate at 37°C for 1 hour, making sure that the micropatterns were not allowed to dry at any point of the incubation. Following surface coating, laminin solution was replaced with cell culture medium, and cells were immediately seeded.

Lattice light-sheet microscopy

LifeAct biosensor transduced LM2-4 cells were plated on micropatterned 25-mm-diameter #1.5 coverslips. Lattice light-sheet microscopy (LLSM) was performed on a modified lattice light-sheet microscope (39, 40) housed at the Advanced Imaging Center at Janelia Research Campus. A Thorlabs 0.6 numerical aperture (NA) water dipping lens (TL20X-MPL) with a 5.5-mm working distance was used for light-sheet excitation, and a Zeiss 1.0 NA water dipping objective (421452-9800) with a 2.2-mm working distance was used to collect emitted fluorescence (system magnification of ×63). The emission light was reflected off of a 561-nm long-pass filter (Semrock, Di03-R561-t3-32x40) and subsequently filtered by a 520-nm band-pass filter (Semrock, FF01-520/35-25) and a 561-nm notch filter (Semrock, NF03-561E-25) before being detected by a Hamamatsu Orca Flash 4.0 sCMOS camera. A square lattice pattern (inner NA, 0.34; outer NA, 0.4; envelope, 3; crop, 10) was used for generating the lattice light sheet. Volumes were acquired in the “X Stage” motion modality, wherein the sample stage is scanned laterally at an angle of 32.45° relative to the optical axis of the detection objective. Excitation was performed with a 488-nm laser line with an exposure time of 50 ms (15.1 s per volume, 3.3 s pause between volumes, 41 time points). The voxel size was 0.108 μm by 0.108 μm by 0.215 μm for a total image volume of 55.3 μm by 140.4 μm by 64.7 μm. Immediately after completion of the experiment, autofluorescence from the micropattern was imaged using 405-nm laser excitation (100-ms exposure time, one time point, all other imaging parameters are as described above). A system correction for the excitation and detection paths was performed as previously described (40). Imaging was performed in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS; Invitrogen Life Technologies, Grand Island, NY, USA) and penicillin (100 U/ml) and streptomycin (100 μg/ml; Mediatech, Manassas, VA, USA) at 37°C with 5% CO₂; conditions were maintained using an Okolab CO₂-O₂ Unit-BL [0-20; 1-95]. Images from the lattice light sheet were deskewed and subsequently deconvolved processing with 10 iterations of Richardson-Lucy deconvolution using experimentally measured point spread functions for each excitation wavelength (https://github.com/aicjanelia/LLSM).

Imaris volume rendering was used to generate volumetric representations of these cells, before which, cell volumes were background subtracted in ImageJ [National Institutes of Health (NIH), Bethesda, MD, USA]. Cell volumes were extracted from the subsequently reconstructed volume in cubic micrometers. For 3D visualization, manual segmentation via surfaces tool, rendering, and animation were also performed in Imaris.

MTT assay

Using MTT on a custom-made 96-well plate containing both micropatterned and unpatterned wells, we compared the metabolic activity of all the cells used in this study simultaneously and uniformly. The cells were plated in the wells and allowed to stay in culture for 48 hours. For 100 μl of medium per well, 10 μl of a MTT (5 mg/ml) stock solution dissolved in 1× PBS is added 4 hours before the end of the incubation. The plate was incubated in a CO₂ incubator at 37°C for 4 hours. The labeling medium was removed, and 100 μl of dimethyl sulfoxide (DMSO) was added to each well. The plate was shaken gently to dissolve the formed crystals. The absorbance was then recorded at 550 nm in a plate reader.

Cell division (EdU) assay

The Click-iT EdU cell proliferation kit for imaging from Thermo Fisher Scientific was used to quantify cell proliferation. EdU (10 μM) was added to the chamber containing the cells whose proliferation was to be quantified. Following a 2-hour incubation, the EdU-containing medium was removed and washed once with a 1× solution of PBS. This was followed by a 12-min incubation in 4% paraformaldehyde (PFA). This was followed by a triple wash with a solution of 3% bovine serum albumin (BSA) and 20-min permeabilization in 0.5% Triton X-100, respectively. Once the samples had been permeabilized, they were once again double washed in a 3% solution of BSA at which point they were ready for staining. The wash solution was removed, and the Click-iT reaction cocktail was added to the well. The sample was incubated in the cocktail protected from light for 30 min at room temperature. In this case, the Alexa Fluor 647 Azide was used to stain EdU-positive cells. Following this, the reaction cocktail was removed, the sample was washed once with 1× PBS, and the nuclei were subsequently stained with Hoechst 33342 (1:10,000; Thermo Fisher Scientific, 62249) for 10 min.

High-content image acquisition to quantify the EdU-positive nuclei was carried out on an InCell Analyzer 2200 (GE Healthcare) using a 20× air objective. The sample was then imaged using the 4′,6-diamidino-2-phenylindole (DAPI) and Cy5 filter sets. The total number of Cy5- and DAPI-positive nuclei was calculated, and the relative ratio of EdU-positive cells was calculated as a ratio of the two values.

Apoptosis assay

MCF-7 cells were cultured concurrently on micropatterned and unpatterned substrates. One group of cells under the micropatterned and unpatterned condition was treated with cycloheximide (CHX; 50 μg/ml) for 16 hours, while the others were left in CHX-free medium. The cells were all fixed at the same time point with 4% PFA. These were all subsequently stained with the anti–cleaved caspase-3 antibody (Cell Signaling Technology, D175) (1:100) and counterstained with Hoechst 33342 (1:10,000). The coverslips were then mounted on slides with ProLong Diamond (Thermo Fisher Scientific, P36965) and imaged. Apoptotic cells showed up clearly in the cleaved caspase-3 stain. The percentage of cell death was calculated as a percentage of the total number of cells in the viewing area calculated from the nuclear stain.

Confocal microscopy

Laser scanning confocal microscopy was carried out using a Zeiss LSM 880 with the pinhole at 1 Airy unit. High-resolution representative images were acquired at 2048 × 2048 line resolution without any binning or cropping using a 0.8 NA Zeiss 40× water objective at room temperature. Laser power, gain settings, magnification, zoom, pixel size, and slice thickness (for z-stacks) were held constant across all samples used. For visualization and quantification of cell and nuclear morphology, cells were fixed with 4% PFA for 15 min, washed three times with PBS, and then permeabilized in 0.5% Triton X-100 for 20 min. Following three more washes with PBS, cells were incubated with Hoechst 33342 (1:10,000) and Alexa Fluor 647 Phalloidin (1:400; Thermo Fisher Scientific, A22287) for 10 to 15 min to stain for nuclei and F-actin respectively.

Atomic force microscopy

Nuclear AFM elastography measurements and analyses were carried out as previously described (66) using an Asylum MFP-3D-BIO AFM coupled with an Olympus IX-80 inverted spinning disk confocal microscope. Briefly, cells were plated onto collagen type I–coated 50-mm short-profile plastic dishes (BD Biosciences) and cultured for at least 24 hours. Using gold-coated silicon nitride blunt pyramidal AFM tip of either 0.03 or 0.09 N/m (Asylum, TR400PB), which was calibrated using the thermal noise method, nuclear regions of cells were probed with 4 × 4 homogeneous array of 5-μm-deep indentations (at 10 μm/s) with a uniform trigger threshold of 40 nm at 37°C. Contact point and depth-dependent pointwise apparent elastic modulus were computed from each indentation curve as previously described (67). Initial contact points were estimated by minimizing the sum of squared errors through piecewise fitting of a linear regime before contact and a quadratic regime after contact. Subsequently, contact points were manually inspected and corrected by an expert AFM user using a custom MATLAB graphical user interface, which is publicly available on our GitHub page (https://github.com/AzelogluLab/Atomic_Force_Microscopy). To minimize batch effects, several criteria were followed (68): No more than 10 to 12 cells were probed at one given time; the assay time was limited to 30 min; the entire experiment was repeated three times independently, whereby all cell lines were tested at each given time and the testing order of cell lines was randomly modified.

Scanning electron microscopy and sample preparation

After completion of the live imaging assay, samples were fixed in glutaraldehyde for at least 48 hours. Samples were subsequently treated with osmium tetroxide for 1 hour. Following this, samples were dried with progressive washing in 50 and 70% ethanol for 5 min at each step finishing with a 10-min incubation in 100% ethanol. Before critical point drying, samples were washed in a two-step format in 200 proof ethanol. Samples were then critical point dried on a BAL-TEC critical point dryer. Last, the samples were coated with a 20 nm layer of gold-palladium before scanning electron microscopy (SEM) imaging. SEM images were captured on a Zeiss SEM at 5 kV.

Scratch (wound healing) assay

MCF-7 and MDA-MB-231 LM2-4 cells were cultured in DMEM supplemented with 10% FBS (Invitrogen Life Technologies, Grand Island, NY, USA) and penicillin (100 U/ml) and streptomycin (100 μg/ml; Mediatech, Manassas, VA, USA). Cells were grown at 37°C in a 5% CO₂ humidified incubator. When cells were subconfluent, they were detached using trypsin (Corning, Corning, NY, USA), and plated in six-well plates. Before plating the cells, the bottom of each well of the six-well plates was marked with a horizontal line, using a blue marker. MCF-7 cells were plated in three wells at a 1:3 ratio, and MDA-MB-231 LM2-4 cells were plated in six wells at a 1:10 ratio. When the cells reached confluence, the medium was removed, and a scratch was made in the cells perpendicular to the blue marker line, using a sterile 200-μl pipette tip. The cells were washed once with PBS, and the medium was replaced with DMEM supplemented with charcoal-stripped FBS and penicillin/streptomycin. MDA-MB-231 LM2-4 cells were immediately treated with 5 μM of the IR/IGF1R TKI, BMS-754807, or 0.0005% DMSO control in triplicate. MCF-7 cells were treated with 0.0005% DMSO in triplicate. Images were then immediately taken (time, 0 hours) of the wound at 4× objective, using an inverted microscope and cellSens software (Olympus, Center Valley, PA, USA). The blue horizontal line was used as a landmark for each image, with one picture of the wound taken immediately above the blue line and one immediately below the blue line for each well. The wound was imaged intermittently until the cells covered the wound in one group. Wound closure was quantified using ImageJ.

Transwell migration assay

A 24-well plate with 8-μm pore polycarbonate membrane inserts was used (Corning Inc., Corning, NY). DMEM (600 μl) supplemented with 10% FBS and 1% penicillin/streptomycin was added to the lower chamber to equilibrate the membrane for at least 1 hour before cell plating. A total of 100,000 MCF-7, MDA-MB-231 LM2-4, MDA-MB-468, Huh7, HepG2, and LCC6 (MDA-MB-435) cells per transwell were resuspended in DMEM, 0.1% BSA (Sigma-Aldrich, St. Louis, MO, USA), and 1% penicillin/streptomycin, and cells were allowed to migrate for 16 hours. For the TKI study, 600 μl of DMEM supplemented with 10% charcoal-stripped FBS and 1% penicillin/streptomycin was added to the lower chamber. A total of 100,000 MCF-7 or MDA-MB-231 LM2-4 cells per Transwell were resuspended in DMEM, 0.1% BSA (Sigma-Aldrich, St. Louis, MO, USA), and 1% penicillin/streptomycin with 5 μM BMS-754807 or 0.0005% DMSO. A total of 100,000 cells were seeded in 100 μl of DMEM, 0.1%. BSA, and 1% penicillin/streptomycin with appropriate treatment in the insert in triplicate. Cells were allowed to migrate for 16 hours.

After migration, medium was removed from the wells, and non-migrated cells were removed from inside the insert using a cotton swab. The migrated cells were fixed with 70% ethanol and stained with 0.5% crystal violet solution, as previously described (10). The migrated cells were photographed in four quadrants of each insert using an inverted microscope at 4× objective. Following imaging, the crystal violet dye was eluted in 33% acetic acid in distilled water. Eluate (100 μl) from each well was transferred to a 96-well plate, and the optical density was measured at 595 nm.

Statistical methods

For high-content analyses of morphometrics and basal total traveled distance measurements, nonparametric Wilcoxon (two groups) or Kruskal-Wallis one-way analysis of variance (ANOVA) followed by post hoc Tukey multiple comparisons tests (more than two groups) was used, with a P value of 0.001 considered to be statistically significant. For all other experiments, significance was achieved at a P value of 0.05. Statistical comparisons for MTT viability, EdU cell division, and apoptosis assays were performed with unpaired t tests between unpatterned and micropatterned groups. To evaluate cellular stiffness with AFM elastography, one-way repeated measures ANOVA with post hoc Tukey test was used.

All imaging experiments and high-content assays were performed with at least three well replicates each time. Every experiment was repeated at least three times using biological replicates.

Supplementary Materials

This PDF file includes:

Supplementary Text

Figs. S1 to S24

Tables S1 to S4

Legend for movie S1

Other Supplementary Material for this manuscript includes the following:

Movie S1