The interface stiffness and topographic feature dictate interfacial invasiveness of cancer spheroids

During cancer metastasis, tumor cells likely navigate, in a collective manner, discrete tissue spaces comprising inherently heterogeneous extracellular matrix microstructures where interfaces may be frequently encountered. Studies have shown that cell migration modes can be determined by adaptation to mechanical/topographic cues from interfacial microenvironments. However, less attention has been paid to exploring the impact of interfacial mechnochemical attributes on invasive and metastatic behaviors of tumor aggregates. Here, we excogitated a collagen matrix-solid substrate interface platform to investigate the afore-stated interesting issue. Our data revealed that stiffer interfaces stimulated spheroid outgrowth by motivating detachment of single cells and boosting their motility and velocity. However, stronger interfacial adhesive strength between matrix and substrate led to the opposite outcomes. Besides, this interfacial parameter also affected the morphological switch between migration modes of the detached cells and their directionality. Mechanistically, myosin II-mediated cell contraction, compared to matrix metalloproteinases-driven collagen degradation, was shown to play a more crucial role in the invasive outgrowth of tumor spheroids in interfacial microenvironments. Thus, our findings highlight the importance of heterogeneous interfaces in addressing and combating cancer metastasis.


Introduction
The most life-threatening aspect of cancer lies in its metastatic progression, which takes place when tumor cells manage to escape the primary tumor and survive the transit through the lymphovascular system, and finally results in formation of secondary tumors in distant organs [1]. In epithelial cancers, the first pivotal step towards metastatic dissemination is the traversal of detached cells across two biochemically and morphologically separate but closely adjacent structures, the basement membrane and the stroma [2]. The former is a thin, dense sheet of highly cross-linked extracellular matrix (ECM) that underlies all epithelia in multicellular animals, while the latter typically consists of a significant amount of ECM produced by the embedded connective tissue cells (e.g. fibroblasts) [3]. Thus, ECM is considered a major structural component of the tumor microenvironment (TME); it plays a crucial role in the process of cancer cell invasion, acting as a physical scaffold for cell migration and imparting a spatial context for intercellular signaling events [4]. A plethora of matrix metalloproteinases (MMPs), secreted by tumor cells themselves or hijacked non-malignant cells, can enable the degradation of ECM so as to carve out local paths for deadly spread of metastatic tumor cells [5]. Currently, two fundamentally distinct patterns of invasive growth have been described: single-cell migration (i.e. mesenchymal or amoeboid movement) and collective migration (i.e. as multicellular units ranging from acini, cords, glands, sheets, to clusters) [6]. Emerging evidence suggests that the majority of solid cancers in vivo often employ the collective migration modes to navigate complex 3D environments [7]. Nevertheless, both patterns could still be detected within the same TME in clinical settings [8]. As the diversity of cancer cell invasion patterns determines the disease severity and overall patient survival [9], it is of great importance to characterize the invasive behaviors of tumor cells and explore the underlying factors.
The ECM networks of most tissues and organs are inherently heterogeneous at the microscale [10]. During cancer progression, dysregulated ECM remodeling, which is under the influence of cancer cells and tumor stroma, can not only alter the mechanical properties of the ECM, characteristically increasing stiffness of the tumor tissue, but also contribute to spatial and temporal heterogeneity of the ECM networks [11]. The heterogeneity of ECM microstructures has been considered an essential element influencing collective invasive behaviors [12]. Mechanistically, the structural heterogeneity and anisotropy of native tissues are largely attributable to the complexity of matrix physical properties, including rigidity, elasticity, density, porosity, topography (spatial arrangement and orientation) [13,14]. Besides, specific chemical constituents of the microenvironments (e.g. oxygen, nutrients, cytokines, and growth factors) also diverge greatly among different tissues [15]. The alterations in ECM structure or mechanics, in company with spatial variations of the aforestated microenvironmental factors, therefore disturb single-cell polarity, influence collective cell behaviors, induce multiple invasion patterns, and regulate the speed of metastasis. Realistically, invasive tumor cells are often likely to proceed along discrete heterogeneous anatomical structures during metastatic dissemination [16,17]. In this case, the ventral surfaces of migrating cells might be in contact with a structural interface between two different ECM microenvironments; this cell movement pattern may be considered as 'interfacial' in nature [18][19][20]. Despite the recognized significance of migration along and even across asymmetric tissue interfaces for metastatic outgrowth of many cancers, little is known about the influence of cell-ECM interactions at interfacial tissue microenvironments on patterns of cancer migration and invasion.
Mounting evidence suggests that ECM dimensionality should exert a profound influence on patterns of cancer cell migration/invasion [21][22][23]. Over the past several decades, studies of cell migration have been traditionally conducted using two-dimensional (2D) adherent culture systems where cells are grown on the surface of a substrate such as glass, plastic, and ECM gel. In an artificial 2D environment, most adherent cells undergo mesenchymal-like migration, which begins with actin polymerization-mediated protrusion of the cell membrane, followed by integrin-mediated formation of focal adhesions at the cell front that anchor the actin cytoskeleton to various external structures [24]. After that, myosin II, by sliding along the actin filaments in opposite, contracts the actin networks, generating traction forces to pull the cell along the direction of focal adhesion formation [25,26]. Integrin-mediated cell-matrix adhesions also play a central role in regulating cell migration in ECM scaffolds [27]; however, studies have reported that under confined 3D environments (e.g. interconnected porous structures), cells can switch to a much faster amoeboid-like migration phenotype [28,29]. To adopt this migration mode, cells undergo extensive shape changes, sending out bleb-like surface protrusions by which they can squeeze through the pores of the ECM and migrate via simply mechanical means [30]. Besides the differential impacts on molecular and cellular mechanisms utilized for cell migration, the dimensions of cell culture environments also dictate the effects of ECM mechanics on cell motility. For example, it has been noted that increased ECM stiffness promotes the migration speed of cells on 2D surfaces, whereas it inhibits cell motility in a 3D context [31,32]. In earlier studies, Beningo and colleagues had shown, by innovatively employing a sandwich culture system, that the migration pattern exhibited by fibroblasts sandwiched between two gel substrates was distinguished from that observed in 2D cultures, but shared some similarity with migration within 3D environments [33]. More recently, several separate research groups also availed themselves of similar sandwich strategies to investigate the role of ECM mechanochemistry (i.e. stiffness and composition) in regulating migration and invasion patterns of glioblastoma cells [18,34,35]; their findings have all conclusively pointed out that such tumor cells displayed dissimilar morphologies, migration modes, and spatial orientations depending on their mechanical and chemical tropism. Indeed, a growing number of scientists have already categorized the sandwich culture system as 2.5D since its configuration includes both dorsal and ventral planar interfaces [36]. Taken together, the above brief review of the literature highlights the value of mechanistically exploring the migratory and invasive behavior of tumor aggregates in interfacial microenvironments.
Our current knowledge of metastatic cell migration comes mainly from investigations on single cell migration, most of which are conducted in 2D culture systems and, more recently, using 3D scaffolds [37,38]. Over the past decade, the paradigm focused on single cell movements was shifting towards collective migration, which has now emerged as an important mechanistic contributor to metastatic dissemination of various epithelial cancers [39]. An increasing body of research has recognized that collectively migrating cancer cells are more aggressive and demonstrate higher resistance to chemotherapeutics as compared to individually migrating counterparts [40]. As mentioned above, metastatic cells would most likely encounter discrete and heterogeneous interfaces when they penetrate through the surrounding tissues. Nevertheless, less attention has been paid toward probing the influence of interfacial mechanochemical microenvironments on migration and invasiveness of metastatic cancers, particularly in the form of multicellular aggregates. To address this issue, we excogitated an artificial collagen matrix-solid substrate interface platform, which mimics the soft-hard tissue boundaries in vivo, to characterize the invasive behaviors of multicellular tumor spheroids (MCTSs) respectively established from different cancer cell lines. We first identified that the MDA-MB-231 spheroid, during the first 24 h of incubation, had the most powerful invasive capacity under such in vitro conditions. With respect to the mechanical and chemical properties of ECM microenvironments, we specifically addressed the influences of interface stiffness (i.e. local stiffness of the boundary layer between the collagen scaffold and its underlying substrate) and matrix-substrate interfacial adhesive strength (MSIAS) (i.e. the bonding strength between afore-mentioned contacting bodies) on invasive behavior of MCTSs. Our results showed that the increase of interface stiffness and the decrease of MSIAS could stimulate invasiveness of the spheroids, with enhanced singe-cell detachment from their lower hemispheres closely adjacent to the interface. Both mechanical attributes had the potential to regulate the morphological change and motility of the detached cells. Noteworthily, however, MSIAS also appeared to affect directedness of their movement. In the mechanistic aspect, we found that inhibition of myosin II motors rather than MMPs could significantly lessen the invasive capacity and interfacial migration of tumor spheroids located nearby the matrix-substrate interface. and sodium pyruvate (Gibco, Gaithersburg, MD, USA) supplemented with 10% heat inactivated fetal bovine serum (Gibco, Gaithersburg, MD, USA) and 1% penicillin-streptomycin 100× solution (Corning, NY, USA). All cells were grown and maintained in media in a CO 2 tissue culture incubator (5% CO 2 , 37 • C, humidified). Media were monitored daily and replaced with fresh media two to three times per week. For passage, cells at 80%-90% confluency were detached from the culture flask by treating with 0.05% ethylenediaminetetraacetic acid (Trypsin-EDTA) at 37 • C in an incubator for 3 min. Medium was then added to inhibit the enzymatic reaction of Trypsin-EDTA. Sub-culture seeding density was kept at 2-3 × 10 5 cells ml −1 per T25 flask.

Formation of MCTSs
MCTSs derived from T24, A549, Huh-7, MG-63, and MDA-MB-231 cancer cell lines were generated by spontaneous aggregation of cells in nonadhesive microwells as previously described [41]. First, microwells were sterilized by incubation in 70% ethanol for at least 8 h and then cleaned by washing with 1× phosphate buffered saline (PBS; HyClone). To prevent undesired cell attachment to microwell plates, microwells were immersed with 0.2% Pluronic F127 (Sigma-Aldrich Corp., St. Louis, MO, USA) in 1× PBS for at least 40 min, followed by twice 1× PBS rinses. Then, a fixed number (i.e. 4 × 10 5 ) of cells were seeded into a 60 mm dish containing 3600 microwells, and then cultivated at the CO 2 incubator (37 • C, 5% CO 2 , and 95% humidity). Cells spatially distributed within the microwells would loosely congregate at the beginning and then further aggregate into compact MCTSs during the four-day incubation period. Afterwards, the mature spheroids were transferred from the microwells, by gently pipetting up and down with a 10 ml serological pipette, to a 50 ml conical centrifuge tube (Thermo Scientific™). The intact multicellular spheroids were first isolated by filtering the suspensions sequentially through 100 µm and 70 µm cell strainers, followed by placing the inverted 70 µm cell strainer onto another 50 ml conical tube. Eventually, suspensions containing spheroids with a diameter of approximately 70-100 µm were obtained by flushing the mesh with cell culture medium. To avoid the possible influence of size variation on the experimental outcomes, the spheroids with a diameter approximately equivalent to 70 µm would be further cautiously selected under the inverted microscope.

Polydimethylsiloxane (PDMS) substrate fabrication
PDMS, which is well-known for its excellent biocompatibility, has been widely accepted in recent years as a versatile tool for cell adhesion studies, especially with respect to the effects of microstructures on cell adhesion [42]. Sylgard 184 silicone elastomer base and Sylgard 184 silicone elastomer curing agent manufactured by Dow Corning (Midland, MI) were used to make the PDMS networks. A sheet of PDMS network (5 mm thick) was fabricated by thoroughly mixing the silicone elastomer base with the curing agent at a ratio of 10:1 (30:1, for a softer PDMS network), followed by pouring the mixture into a glassbottom mold and degassing it under vacuum. The PDMS solution was dried to cure in an oven at 70 • C for overnight, and then the cured PDMS network was cut into cylinders (6 mm in diameter) using the stainless-steel biopsy punches. Afterwards, the block of PDMS network with 6 mm wells and the autoclaved glass coverslip were bond with each other by O 2 -plasma treatment for 60 s.

PDMS substrate surface modifications
It is known that Pluronic F127 coating inhibits cell adhesion on PDMS or glass substrate since Pluronics are an amphiphilic triblock copolymer consisting of a hydrophobic central poly(propyleneoxide) (PPO) block surrounded by two hydrophilic end poly(ethyleneoxide) (PEO) blocks [43]. It is this amphiphilic behavior that contributes to the different solubilities exhibited by central and end blocks; thus, the hydrophobic central block can adsorb the hydrophobic surface through hydrophobic interactions, while the hydrophilic PEO chains can extend away from the surface, displaying a brush-like configuration, which sterically hinders protein adsorption. In this study, 0.2% (w/v) Pluronic F127 (in distilled water) (Sigma-Aldrich Corp., St. Louis, MO, USA) was used to coat the PDMS (10:1) substrate, the incubation process of which was conducted at room temperature for 30 min. Eventually, the Pluronic F127-coated PDMS membrane was rinsed with 1× PBS twice.
Poly-D-Lysine (PDL) is the most widely used cell adhesion molecule due to its excellent adhesion capability originated from its positively charged nature. It attracts cells by electrostatic interaction with the negatively charged cell membrane. Besides, PDL can also enhance adhesion of the coated substrate to negatively charged collagen scaffold [44]. In this study, we carried out PDL coating to render the PDMS substrate a positively charged interface, using the method described previously [45]. Briefly, the oxygen plasma was employed to attack the siloxane backbone that generates reactive species to promote formation of silanol groups (Si-OH) on PDMS surface and render it hydrophilic. Afterwards, the PDL surface treatment solution (1 mg ml −1 ) (Sigma-Aldrich, St. Louis, MO) was introduced to the hydrophilic PDMS substrate surfaces and incubated at 37 • C for at least 4 h. Eventually, the substrate was aspirated, washed with sterile water, and dried at 70 • C for 48 h.
Glutaraldehyde (GA), which can covalently interact with collagen I, has been employed to modify the PDMS surface using the method described previously [46]. Briefly, PDMS substrates were cleaned and dried overnight, and then underwent O 2 plasma treatment for 60 s. The treated substrates were grafted with the first layer of 10% 3-aminopropyltriethoxysilane (abbrev. APTES) (Sigma) in ethanol for 1 h in the 60 • C oven, followed by three times of ethanol rinses. The second layer of 1% GA (in deionized water) (Sigma-Aldrich) was grafted for 1 h at room temperature. Eventually, the substrates grafted with APTES and GA were washed with deionized water three times, and then incubated within deionized water overnight (at least for 20 h).

Substrate stiffness measurement
The modulus data of the borosilicate glass, PDMS (10:1), and PDMS (30:1) substrates were collected on a Bruker Hysitron TI-980 triboindenter equipped with a Berkovich geometry diamond tip. The tip was calibrated for its size with experiments on a standard fused quartz sample. A peak indentation loads of 1000 µN and 50 were respectively used for glass and PDMS substrates. At least ten indentations were conducted for each sample. The subsequent data analysis was performed using the Hysitron TriboScan analysis software. Young's modulus of each sample was recalculated using the equation described previously [47].

Fluorescent labeling of rat-tail collagen
Rat tail collagen solutions have been widely used as polymerizable in vitro 3D ECM gels for single-cell and collective migration assays as well as spheroid formation. Such 3D hydrogels are an inexpensive, simpleto-use model system mimicking the in vivo physical features of various tissues within the body. To visualize the collagen matrix architecture under the fluorescence reflectance confocal microscopy, 3D collagen gels labeled with a fluorescence dye were generated using the method adapted from previously-described approach [48]. Briefly, an appropriate amount of stock collagen solution (Corning ® Collagen I, Rat Tail) was added into the pre-cooled culture dish, followed by addition of 10× DMEM medium and 10× reconstitution buffer and adjustment of the pH (to pH 7.0-7.4) with 1 N NaOH. Next, let the collagen polymerize at room temperature for 30 min. Once after complete polymerization, the collagen gels were added with 10 ml of 50 mM borate buffer (pH 9.0), followed by incubation for 15 min at room temperature, and then the borate buffer was aspirated from the culture dish. The Atto-488 NHS-ester dye solution (Sigma-Aldrich) was added into the culture dish to cover the collagen gel. The culture dish was wrapped with aluminum foil and let it keep rocking at 4 • C overnight for dye conjugation. Afterwards, quench the dye reaction by addition of 10 ml of 50 mM Tris buffer (pH 7.5). Eventually, the labeled collagen gels were liquefied by adding 500 µl of 500 mM acetic acid (Sigma-Aldrich), followed by dialysis with the Slide-A-Lyzer Dialysis cassette (G2) (Thermo Fisher Scientific), and then centrifuged at 20 000 × g at 4 • C for 1 h. The supernatant containing fluorescentlylabeled collagen was kept from light and stored at 4 • C.

Embedding MTCSs within the collagen matrix
The spheroid suspension was transferred to a 15 ml centrifuge tubes and spun down at 500 rpm for 3 min at room temperature. Aspirate all culture medium carefully and add about 200 µl of DMEM to resuspend spheroids. It has been reported that collagen densities in human fresh tissues are approximately 1.4-2.3 mg ml −1 [49]. Hence the final concentration of the collagen matrix was set to 2 mg ml −1 . The embedding procedure was done by mixing 200 µl of a 2.5 mg ml −1 fluorescently-labeled rat-tail collagen stock solution (pH7.4) with 50 µl of the spheroid suspension. Afterwards, the mixture with a total volume of 250 µl was added onto the substrate and then incubated at 37 • C for at least 30 min until a solid gel formed. When polymerization was finished, 120 µl warm (37 • C) culture medium was added carefully along the sidewall onto the gel to cover the collagen/spheroids drops. The PDMS substrate with collagen matrix-embedded spheroids was subjected to timelapse differential interference contrast (DIC) microscopic analysis using the IX83 fully motorized and automated inverted microscope (Olympus, Japan) equipped with a stage incubator at 37 • C for live imaging.

Myosins II and MMPs inhibition
Inhibition of actomyosin-dependent contractility was done by treating the spheroids with blebbistatin (Sigma-Aldrich), a well-known myosins II inhibitor, at a working concentration of 10 µM. To broadly inhibit MMPs, PD 166793 (S-2-4 ′ -bromobiphenil-4-sulfonylamino-3methyl-butyric acid) (Tocris Biosciences) was administered at a working concentration of 10 µM. Both inhibitors were added into the culture medium at 3 h after conducting time-lapse imaging.

Time-lapse DIC microscopy
DIC images of MCTS at matrix-substrate interfaces were automatically captured using an Olympus IX83 inverted microscope (Olympus, Tokyo, Japan), equipped with a microscope incubation chamber (setpoints: 37 • C and 5% CO 2 ) to keep MCTSs alive during the imaging process. For time-lapse imaging, the heating system had to be turned on and set to 37 • C before imaging. Next, the glass coverslip/P-DMS device loaded with the collagen-MCTS mixture was placed on the stage of the microscope. After the environmental condition (i.e. temperature, CO 2 and humidity) became stable, browse the location of the spheroids in matrix with UPLXAPO 20× Olympus objective lens. The spheroids which were fully embedded in collagen matrix and sufficiently distant from others were chosen. The spheroids of interest were positioned to the center of the field. And then, the z stack range was set to 100 µm with 7 µm for step size. The time interval was set to 10 min to result in 144 repetitions during the 24 h imaging process.

ImageJ (Fiji) software analysis
ImageJ software was used to analyze images obtained from time-lapse imaging systems and immunofluorescence imaging. The parameters to be processed were spreading area, cell path coordinates, aspect ratio, orthogonal view, and 3D viewer. As for dealing with the spread area, a series of Z-stack scans over time were loaded into ImageJ. Use the images to stack and Z project functions with standard deviation mode to combine the whole images. And then, eight-bit grayscale images and threshold function with 'Otsu' mode were applied. Next, use the Fill hole function in Binary to make the area seamless. Finally, Analyzed Particles was applied to calculate the area.

Image acquisition and analysis
The Matlab code was used to automate analysis of all images captured from the time-lapse imaging system. Use the Image J (Fiji) code to integrate the Z-stack images and calculate the spreading area of MCTS, Matlab performed the task of automatic data loading and instructed the Image J software to process the data after loading. After image processing, all data were exported as excel files. The data regarding the spreading areas were imported to Microsoft Excel to undergo calculation of the mean and standard deviation values. And then, the data were plotted and statistically analyzed using Prism GraphPad 7.0. Use the Matlab code to plot cell tracking lines. After using the TrackMate to perform cell tracking, the files including the coordinates of the selected point were exported by ImageJ. Finally, use Matlab to plot the path of the points from the exported coordinate file.

Immunofluorescence imaging
Spheroid samples were obtained after DIC microscopic observation of interfacial MCTS migration and invasion above different substrates, and then washed with PBS, pH 7.4. Next, fixation was conducted by fixing the samples with 4% paraformaldehyde (Sigma) (in PBS) for 20 min, and then rinsed with 10× PBS three times. Afterwards, the fixed MCTS samples were permeabilized with 0.5% (v/v) Triton-X-100 (Sigma) (dissolved in blocking buffer (Thermo Fisher Scientific)) and incubated for 1 h, which was followed by three times of 1× PBS wash. MCTSs were incubated with the Rhodamine phalloidin solution (Invitrogen, diluted 1:200 in PBS) and Hoechst 33342 (Invitrogen, diluted 1:1000 in PBS), respectively, to label the actin filaments and nuclei within MCTS for 1 h at room temperature. Samples were visualized under a confocal laser scanning microscope (FV3000, Olympus, Tokyo, Japan) using 40× and 100× objective lenses, respectively. Raw images were processed using the ImageJ software (Fiji).

Statistical analysis
Statistical significance was evaluated using the oneway ANOVA followed by Tukey-Kramer multiple comparison test and represented by bar plots with error bars representing standard errors. A p-value less than 0.05 was considered statistically significant.

The hemisphere of a breast cancer spheroid in the vicinity of the matrix-substrate interface is largely responsible for its invasive outgrowth
To identify a more suitable spheroid model for exploring the impacts of interfacial mechanical attributes on single-cell and collective migration of tumor cells, we began with microscopic inspection of the invasive capacities of several MCTS types separately formed from cancer cell lines representative of different common primary tumor locations: bladder, breast, bone, lung, and liver. Considering that the differentiation stage of a primary tumor is associated with its malignancy and metastatic behavior [50], T24 (bladder), MDA-MB-231 (breast), MG-63 (bone), A549 (lung), and Huh-7 (liver) cells, with different differentiation statuses, were included in this preliminary experiment (table 1). The former three cell types T24, MDA-MB-231, and MG-63, isolated from advanced, higher-grade tumors, have been reported to be poorly to moderately differentiated and exhibit highly invasive and metastatic capacities both in vitro and in vivo [51][52][53]. A549 and Huh-7 cells, in contrast to the forgoing, were derived from well-differentiated tumors, and therefore found to be less or moderately invasive, with limited metastatic potential in vivo [54][55][56].
Each of the established spheroid models was embedded within the collagen scaffold situated above an uncoated borosilicate glass substrate. After 24 h of incubation, multiple sharp protrusions were observed to emanate from both T24 and MDA-MB-231 MCTSs. In the meanwhile, several detached single cells had already dispersed around them, with either of the spindle-and cobblestone-like morphologies (see figure S1). Comparatively speaking, the MDA-MB-231 spheroid appeared much more aggressive as it exhibited a much greater invasive area (i.e. the subtraction of the total area from the core area). Although apparent invasive outgrowth seemed not to have occurred at this timepoint, the surface of the MG-63 spheroid still displayed small bleb-like protrusions. As for A549 and Huh-7, the volume, morphology, and integrity of these two spheroids were shown to remain largely unchanged during 24 h of culture. Collectively, our pilot screening process determined that the invasive capacity of the MDA-MB-231 breast tumor spheroid stood out among others under such in vitro conditions. Apart from the 'aggressiveness' level in interfacial microenvironments, the differentiation stage of solid tumors is also an issue worth consideration since it is not only highly associated with their malignancies but also metastatic potentials [50]. To investigate whether and how interfacial mechanical properties (i.e. stiffness and MSIAS) affect the invasive and migratory behavior of metastatic tumor aggregates, MDA-MB-231 MCTSs were chosen as the model for subsequent experiments.
To have a deeper insight into invasion dynamics of MCTSs at the collagen matrix-glass substrate interface, time-lapse microscopy inspection was conducted over 24 h. As shown in figure 1(A), the MDA-MB-231 spheroid began to extend protrusions as early as the fourth hour after having been embedded within the collagen scaffold. Such protrusional outgrowth lasted until the end of the observation period, thus gradually enlarging the spreading areas of the spheroid (indicated by yellow dashed lines). Besides, single migrating cells (enclosed by red dashed lines), which escaped from the spheroid, could also be identified beyond the initial 4 h of culture. As time progressed, the number of detached single cells continued to increase ( figure 1(A)). The Z-section confocal images demonstrated that this spontaneous detachment occurred primarily in the bottom half of the spheroid (i.e. close to the substrate surface) (figures 1(B) and (C)). The detached cells were not only found scattered around the exterior of the spheroid, but also appeared to pervade the surrounding collagen matrix, by migrating along the interfacial boundary. Interestingly, we observed that a domeshaped hollow space incorporating neither cells nor collagen matrix was created beneath its lower hemisphere ( figure 1(B)). Taken together, these results suggest that the invasive outgrowth of a tumor spheroid should especially occur from its hemisphere adjacent to the matrix-substrate interface.

Interface stiffness determines the outgrowth level of breast cancer spheroids
It has been noted that substrate stiffness has an important role in spreading and proliferation of cancer cells under 2D culture conditions. To move forward along a stiffer substrate, cells typically tend to exert a larger magnitude of traction forces, which can promote remodeling of actomyosin networks and regulate maturation and turnover of focal adhesion complexes. As these two machineries coordinate cell migration and proliferation, increased substrate rigidity is considered to contribute to intensified invasiveness of adherent cancer cells [57]. The importance of augmented microenvironment stiffness on growth and invasive behavior of tumor  spheroids has also been revealed in 3D biomimetic hydrogels [58]. In our current study, we sought to determine if interface stiffness influences interfacial migration/invasion of MCTSs, and therefore employed borosilicate glass, PDMS (10:1) and PDMS (30:1), with the Young's modulus value of 69.74 ± 1.49 GPa, 5.41 ± 0.14 MPa, and 1.01 ± 0.04 MPa (figure S2), respectively, as the culture substrate of the MDA-MB-231 spheroids ( figure 2(A)). Time-lapse microscopic observations demonstrated that the cells within any of the spheroids, regardless of interface stiffness, outgrew in a time-dependent fashion ( figure 2(B)). However, the spreading area of the spheroids grown on the glass substrate was always greatest at any time point, followed by those grown on PDMS (10:1) and PDMS (30:1) sequentially (figure 2(C)), suggesting that the more rigid the substrate, the more enhanced outgrowth of the spheroid.

Interface stiffness promotes spontaneous detachment of single cells from the spheroids and stimulates the migratory behavior of detached cells
The number of single cells detached from the MDA-MB-231 spheroids gradually increased with the elapsed time for all three substrate groups after 8 h of incubation ( figure 3(A)). Nevertheless, the levels of detachment were shown to depend on the degree of substrate rigidity: the stiffer the substrate surfaces, the greater number of the cells detached from the spheroid. These data suggested that tumor cell aggregates tend to adopt a collective mode of invasion in the softer interfacial tissue microenvironments. As shown in figures 3(B) and (C), the cells outside the spheroid displayed a heterogeneous spectrum of morphology, from a rounded, amoeboid (i.e. cobblestone-like) shape to a more elongated, mesenchymal (i.e. spindle-like) shape. Although there was no statistically significant difference in cell's aspect ratios among different substrate groups, we still noticed that cells moving along stiffer matrix-substrate interfaces tended to exhibit a more mesenchymal mode of migration (figure 3(C)). As cells displaying an elongated, mesenchymal morphology are more likely to migrate with random directionality [59], we ensuingly examined whether interface stiffness affects directedness of migration paths exhibited by the detached singular cells. By and large, this mechanical property appeared irrelevant, as inferred from tracing cell trajectories ( figure 3(D)), for regulating directional migration, because the majority of the cells exhibited an analogously persistent random walk in an anisotropic manner, irrespective of the substrate's rigidity. Nevertheless, augmented interface stiffness was found to correlate independently with a larger magnitude of horizontal cell displacement and migration velocity (figures 4(E) and (F)).

Interface stiffness plays an important role in
formation of a hollow space beneath the hemisphere of the spheroids located above the matrix-substrate interface As mentioned above, we observed a dome-shaped hollow space forming beneath the bottom half of  the breast tumor spheroid when it was situated above the collagen matrix-glass substrate interface ( figure 1(C)). Since stiffness of the matrix-substrate interface was shown to promote spheroid outgrowth and single cell detachment (figures 2(C) and 3(A)), we were curious whether such interfacial mechanical property has a certain role in formation of the identified cell-free and matrix-free microstructure. The z-section and z-stack confocal images (figure 4(A)) showcased that there existed an overtly visible dome-shaped vacant cavity (roughly 25 µm in its longitudinal length) above the matrix-glass substrate interface. By contrast, a much tinier void cavity (about 5 µm in length) was found at the bottom of the indicated tumor spheroid situated above the PDMS (10:1) substrate. As for the softer PDMS group (30:1), we did not visualize such hollow microstructure present beneath the spheroids grown above the indicated interface. Altogether, we suggest that MCTSs situated above a stiffer interface tend to demonstrate a hollow dome-like space between their bottom half and the interfacial boundary when they undergo profound invasive outgrowth ( figure 4(B)).

Interfacial adhesive strength regulates the outgrowth of breast cancer spheroids at the matrix-substrate interface
Apart from ECM stiffness, multiple lines of evidence have indicated that mechanochemistry of the microenvironments (coherence of ECM fibers, orientation of ECM components, etc) can also direct cancer cell migration [60][61][62]. Therefore, we further aimed to determine if alterations of the adhesive force between collagen matrix and PDMS substrate (i.e. MSIAS) can modulate the invasive behavior of MCTSs situated around the matrix-substrate interface. To alter its strength, three surface coating agents (i.e. GA, PDL, and Pluronic F127) were respectively utilized as the surface coating agent for the PDMS (10:1) substrate ( figure 5(A)). GA can be covalently cross-linked with collagen through the formation of the double bond (C=N, imine bond) between -CHO group of GA and -NH 2 group of collagen [63], and thereby it contributes to the strongest matrix-substrate anchoring strength among others. PDL coating provides a positively charged surface, and can moderately enhance the anchoring strength between the collagen scaffold and the PDMS substrate by means of electrostatic attraction [44]. Pluronic F127, an amphiphilic triblock copolymer consisting of a central hydrophobic PPO block flanked by two hydrophilic PEO blocks, has the ability to adsorb onto the hydrophobic surface of the PDMS substrate via hydrophobic interaction while sterically hindering protein absorption via its brush-like configuration [43]. As a result, the collagen matrix scaffold is unable to be anchored to the Pluroniccoated PDMS substrate. The ranking of the MSIASs caused by different coating agents is listed as follows: GA coating > PDL coating > Uncoating > Pluronic coating. Time-lapse microscopy demonstrated that cells within the spheroids located nearby the matrixsubstrate interface, regardless of the type of coating, could time-dependently outgrow (the white and yellow dashed lines respectively represent the contours of the areas of the MCTSs at the initial and observation time points) ( figure 5(B)). As compared to the uncoated PDMS substrate, both PDL and GA coatings reduced the spreading areas of MDA-MB-231 spheroids situated nearby the matrix-substrate interface (The white and yellow dashed lines respectively represent the contour of the areas of the MCTS at the initial and subsequent observation time points) ( figure 5(B)). Conversely, Pluronic coating seemed to give rise to a more extended spreading area, compared to the uncoated control. As shown in figure 5(C), the increase of MSIAS could lead to the decrease of the spreading area, with the most prominent difference between coated and uncoated groups at 24 h postincubation. Taken together, these results suggest that the stronger the adhesive force between the collagen matrix and PDMS substrate, the weaker the invasive capacity of the tumor spheroid at the matrixsubstrate interface.

Interfacial adhesive strength regulates singe-cell detachment from spheroids at the interface and influences morphology, directedness, motility, and velocity of the detached cells
As shown in figures 6(A) and (B), the MDA-MB-231 spheroid nearly remained intact, with slightest single cell migration, during the whole observation period, when grown above the GA-coated PDMS substrate. However, the number of detached single cells increased with the decrease of MSIAS; PDL coating led to the greatest number of singlecell detachment. These results suggest that MCTSs should tend to remain in a collective invasion mode when situated nearby an interface with high MSIAS, whereas cells within MCTSs have a greater tendency to escape from their interior and migrate individually when they are situated above an interface with low MSIAS. Furthermore, the MSIAS was also shown to regulate the morphological change among the detached cells (figures 6(B) and (C)). The distribution of aspect ratio remarkably indicated that the great majority (or more precisely speaking, nearly all) of the migrating cells detached from spheroids above the Pluronic-coated substrate favored a roundedamoeboid phenotype. On the contrary, the enhancement of MSIAS encouraged their conversion to an elongated-mesenchymal phenotype. In light of the fact that the morphological changes of migrating cells are correlated with their migratory status [59], we then investigated if MSIAS affects the directedness, motility, and velocity of the detached cells. With regard to the migration directionality, the trajectory distribution of the detached cells (figure 6(D)) revealed that the uncoated substrate allowed the movement of cells in persistently anisotropic and random manners. Analogous migration patterns were also found in both Pluronic and PDL coating groups. Nevertheless, a small proportion of the detached cells in these two coating groups seemed to walk, without obvious directionality, closely around the spheroids, just as those migrating on the GA-coated substrate. It was riveting to discover that displacement of the cells migrating above the Pluronic-coated substrate is approximately comparable to that of the uncoated substrate group, but apparently greater than that of the cells above the PDL-and GA-coated substrates (figure 6(E)). Furthermore, Pluronic coating significantly accelerated migration speed of the detached cells, with the highest mean velocity value among uncoated control and experimental groups (figure 6(F)). It is also worth to mention that mean velocities of the cells migrating above the PDL-and GA-coated substrates were statistically tantamount to that of the control group, even though both coatings appeared to interfere with cell detachment (figure 6(A)) and lessen migratory displacement (figure 6(E)). Altogether, these results might suggest the less the interfacial adhesive strength between anatomically heterogenous microstructures, The distributions of aspect ratios (C), migratory trajectories (D), mean squared displacement (E), and mean velocity exhibited by the cells detached from the above indicated spheroids. N = 30 cells for each stiffness group. Statistical analyses were performed using the one way-ANOVA followed by Tukey-Kramer multiple comparison test. * (P < 0.05), * * (P < 0.01), * * * (P < 0.005), and * * * * (P < 0.001) denote significant differences between substrate groups. the greater invasive potential of the cancer clusters when encountering such an interface.

Decrease of interfacial adhesive strength promotes invasiveness of spheroids at the matrix-substrate interface and contributes to formation of a cell-free and matrix-free space beneath their lower hemisphere
As stated above, an unoccupied space (with neither cells nor collagen matrix) was brought into existence beneath the lower hemisphere of spheroids undergoing invasive outgrowth, when situated in the vicinity of stiffer matrix-substrate interfaces ( figure 4). Next, we tried to address if the degree of MSIAS modulates the formation of this dome-shaped cavity.
As shown in the Z-section and Z-stack images ( figure 7(A)), the spheroids located above the GAcoated substrate, with the greatest MSIAS among the control and coating groups, not merely exhibited obviously attenuated invasiveness, but were also devoid of such microstructures. The decrease of MSIAS, as compared to substrate stiffness, seemed to conversely facilitate their formation. In addition, these results also suggest that surface coatings moderately to severely diminish the total displacement of migrating cells. Taken together, we hypothetically consider that increment of interfacial adhesive strength is able to compromise the invasive capacity of MCTSs located nearby the matrix-substrate interface ( figure 7(B)).

Myosin II motors, as compared to MMPs, play a more important role in regulating interfacial migration and invasion of MCTSs
It has been reported that cancer cells can dynamically undergo a concerted switch between proteasedriven and actomyosin-dependent motility modes in response to alterations in the local microenvironment [64], which relies on coordinate involvement of MMPs and myosin II motor proteins. At the end of this study, we were interested in mechanistically exploring the modes of motility that cancer cells use to foster single-cell detachment from MCSTs, thus promoting their invasiveness at a matrix-substrate interface. To this purpose, the MDA-MB-231 spheroids situated above the uncoated matrix-PDMS (10:1) interface were administered respectively with inhibitors against myosin II (Blebbistatin, 10 µM) and MMPs (PD 166793, 10 µM) after 3 h of incubation. Regarding the untreated control spheroids, there had already been a profound increase in their outgrowth areas when they were incubated for 16 h (figures 8(A), 9(A) and (B)). At the same time point, it appeared that myosin II inhibition achieved an anti-outgrowth effect on MCTSs in such interfacial microenvironments, while being more efficacious with the increase of time (figures 8(A), 9(A) and (B)). These results suggest that actomyosin-dependent cell motility should play a prominent role in regulating interfacial invasiveness of tumor aggregates. Not as relevant as actomyosin-driven contraction, we still revealed that MMPs-mediated collagen fiber degradation might contribute to initiation of spheroid outgrowth, because temperately suppressed invasiveness was also made possible by the selected MMPs inhibitor (figures 8(A), 9(A) and (B)). As visualized in figure 8(B), the Z-section and Z-stack confocal microscopic images also confirmed that treatments with either Blebbistatin or PD 166793 could exert an inhibitory action that attenuated interfacial spreading of the breast tumor spheroids. Moreover, the size of the dome-like hollow space was shown to reduce as a consequence of MMPs inhibition, and much more considerably due to myosin II repression ( figure 8(B)).
Moreover, even under weaker MSIAS circumstances (i.e. being situated above PDL-coated, uncoated, and Pluronic-coated PDMS substrates), treatment with Blebbistatin was also capable of more significantly abrogate the eventual outgrowth of the spheroids in the interfacial microenvironments, as compared to PD 166793 (figures 9(A) and (B)). Altogether, our results suggest that myosin II motors should play a crucial role in regulating the invasive behavior of tumor aggregates at a heterogeneous interface. In spite of this fact, the value of MMPs' participation in single-cell detachment and early protrusional outgrowth from these tumor aggregates should not be overlooked.

Discussion
Given the importance of ECM in establishing and maintaining tissue architecture as well as participating in regulation of many biological processes, accumulating evidence shows that dysregulation of ECM dynamics leads to development of diseases such as cancer [14,65,66]. To accommodate physiological and pathological events, the ECM networks per se must constantly undergo dynamic alterations in the composition and microstructure via synthesis, degradation, reassembly, and chemical modification of their molecular constituents. On one hand, during tumorigenesis, ECM breakdown facilitates release of ECM-sequestered growth factors and cytokines, thus provoking tumorigenesis and cancer-associated angiogenesis. On the other hand, the diversity in relative abundance of ECM proteins and their crosslinking densities contributes to differential degrees of tissue rigidity. Stiffened ECM within the TME has been considered a mechanical hallmark of cancer [67]. Indeed, during the process of cancer spreading, metastatic tumor cells are very likely to navigate discrete tissue spaces composed of inherently dissimilar ECM microstructures [16,17], the heterogeneity between which may possibly create structural interfaces. Several lines of evidence have suggested that the mechanical heterogeneity within simulated tissue interfaces act as a trigger for the switch among different single-cell migration modes [12,[18][19][20]. Nevertheless, relatively less attention has been paid to investigating how different mechanochemical characteristics of interfacial microenvironments affect invasive behaviors of collectively migrating cancer cells.
In this research, we excogitated a facile in vitro collagen matrix-solid substrate interface platform simulating in vivo interfacial migratory and invasive behaviors of cancer cell clusters occurring at the softhard tissue boundaries. To identify a robust spheroid model that exhibits formidably invasive capacity in such man-made circumstances, we started with a screening process by characterizing the interfacial invasiveness of tumor spheroids formed from five cell types. Of the chosen candidates, both T24 and MDA-MB-231 tumor cells are reported to be poorly differentiated and highly invasive, with mesenchymal traits [51,52]. Being highly invasive, MG-63 cells are not as poorly differentiated as the forging T24 and MDA-MB-231 cells [53]. Contrariwise, A549 and Huh-7 cells were isolated from well-differentiated tumors, with a more epithelial-like phenotype and limited metastatic potential [54][55][56]. The results of this pilot screening revealed that MDA-MB-231 spheroids displayed the most pronounced increase in interfacial protrusional outgrowth during a 24 h incubation period. However, we did not observe any morphological signs of invasion in either A549 or Huh-7 spheroid groups within the same duration, which suggests that tumor aggregates consisting of epithelial-like cells should be relatively less aggressive than those with a mesenchymal-like phenotype even when they are metastasized to soft-hard tissue interfaces.
We illuminated that the stiffness and topographic feature (i.e. levels of MSIAS) of the interface had important roles in dictating invasiveness of the MDA-MB-231 spheroid model. Increased interface stiffness could enhance invasiveness of the tumor aggregates, whereas MSIAS had the opposite effect. In another word, the stronger the adhesive strength between matrix and substrate, the less invasive the spheroids situated at their interface. By contrast, MSIAS might also dictate the morphology, directedness and motility of detached single cells. At the molecular level, we showed that myosin II-mediated contractile forces were largely responsible for overall invasiveness of the spheroids at the matrix-substrate interface; nevertheless, our data also suggested that MMPs play a certain role in single-cell detachment, especially during the initial hours of spheroid invasion.
Malignant tumor masses tend to stiffen as they grow, which is particularly attributable to increased numbers of cancer cells, stromal cells, and deposition of ECM constituents [68]. Evidence has emerged that ECM stiffening alone can enable malignant transformation of various types of organs and tissues [69]. The current mechanistic knowledge indicates that deposition and remodeling of ECM can gradually cause stiffening of the tumor stroma, and the degree of its stiffness is particularly dependent on the density and alignment of the ECM fibrous proteins, collagens and elastin [70,71]. It has been reported that activation of several key signaling pathways such as TGFβ, IGF/IGF1R and PI3/Akt within cancer and/or stromal cells can regulate synthesis or lysyl oxidase-mediated crosslinking of the two proteins [72]. Moreover, a growing body of evidence also reveals that mechanic cues arising from tissue stiffening are able to not merely regulate metabolic reprogramming in cancer cells, but also stimulate processes associated with epithelial-mesenchymal transition [73,74]. Even so, there also exist contradictory in vitro findings regarding the separate effects of ECM rigidity on cell migration speed under 2D vs 3D conditions [31,32].
In the current study, we characterized the invasive patterns of MDA-MB-231 breast cancer spheroids embedded within soft collagen scaffolds, which were situated above substrates of differential stiffness: glass, PDMS (10:1), and PDMS (30:1) (ranked in descending order of Young's modulus). These designs constructed a so-called 2.5D artificial microenvironment simulating the soft-hard tissue boundaries in vivo. In the first place, the invasive outgrowth of the tumor spheroids appeared to occur considerably from their lower hemisphere closely adjacent to the matrix-substrate interface (figure 1). Such interface-mediated tropism of spheroid outgrowth presumably ought to be independent of gravitational force since a similar phenomenon had been observed in T24 spheroids by flipping the similar culture device upside down (i.e. in the direction parallel to the gravity vector) (data not shown). Yet it has been stated that microgravity could modulate proliferation and metastasis of cancer cells, making them less aggressive [75,76]. Moreover, we found that substrates with higher Young's modulus, which contribute to higher interface stiffness, could promote interfacial detachment of single cells and increase their migration velocity (figures 2 and 3). Increased interface stiffness also appeared to encourage the morphological switch of the detached cells, with a growing trend towards elongated-mesenchymal phenotype. Intriguingly, in some instances, the spheroids laid above such boundary might exhibit a dome-like vacant space beneath their lower halves, which also appeared to be influenced by the stiffness of the substrate (figure 4). However, although the rigidity of the interfaces could affect the afore-described interfacial invasion and migration of the spheroids, our results suggest that the existence of an interface provides a 'necessary and sufficient' condition for initiating invasive outgrowth and single-cell detachment as well as directing the migration of detached cells towards itself, regardless of its elastic modulus. In a more scientific sense, we coined the term 'diepafitaxis' (the prefix 'diepafi' comes from the Greek word for 'interface' , while the suffix 'taxis' is for 'orientation in response to an external stimulus') to describe this unique interfacemediated tropistic behavior.
It is well-established that regulated cycles of polarization, cytoplasmic protrusion, and adhesion formation and detachment, which are guided by microenvironmental signals, are involved in the process of cell migration [77]. Although much of the earlier work emphasized the role of chemical gradients, it has gradually become evident that physical features of the substrate have equally important contributions toward guiding cell migration. Apart from substrate stiffness, the effects of topographic cues have also been demonstrated in an assortment of contexts. For example, cells migrate preferentially along grooves on a substrate, the phenomenon of which has been referred to as contact guidance [78]. Other surface topography, such as pillars, has also been shown to affect the morphology and migration of cells grown on a substrate [79]. A considerable amount of research has utilized micro-and nano-patterned rigid substrates, such as PDMS, poly(methylmethacrylate), and other silicon materials, to specifically examine the effects of tightly controlled topographical properties on various cellular behaviors [80].
In this study, we determined the effects of adhesive strength between collagen matrix and PDMS substrate, by taking advantage of polymer coating technology that allows modifications of substrate surface topography and chemistry, on interfacial migration and invasiveness of the breast cancer spheroids. As described in Results, three types of surface coating agents, PDL, Pluronic F127, and GA, were respectively utilized to cover the PDMS (10:1) substrate. GA, which can be covalently cross-linked with collagen [63], contributes to the strongest matrix-substrate anchoring strength among others. A PDL coating can slightly enhance the anchoring strength between collagen matrix and PDMS substrate by means of electrostatic attraction [44]. Pluronic F127 possesses the ability to adsorb the hydrophobic surface of the PDMS substrate via hydrophobic interaction while sterically hindering absorption of collagen fiber via its brush-like configuration [43]. Therefore, the GA-coated substrate produced the highest MSIAS, followed in descending order by the PDL-coated, uncoated, and Pluronic-coated substrates (figure 5). Our results suggest that increase of MSIAS should attenuate invasive capacity of the tumor spheroids located at the matrix-substrate interface (figures 5 and 7). Moreover, the degree of MSIAS, as compared to that of substrate stiffness, exhibit a more prominent effect in regulating the shape, motility, and directedness of the single cells detached from the spheroids ( figure 6). It seems that MSIAS also plays a role in dictating the spatial dimensions of the collagen-free microstructure identified beneath the spheroid's lower hemisphere (figure 7). Our findings, taken together, underscore the importance of interfacial microenvironments in cancer cell metastasis and implicate alterations in interface stiffness and anchoring strength between heterogenous structures in regulating interfacial invasiveness of tumor spheroids.
MMPs are a family of zinc-dependent endopeptidases that have the ability to degrade almost all ECM components and thus play a vital role in tissue remodeling. The proteolytic activity of many MMPs drives different aspects of cancer progression, including uncontrolled tumor growth, local invasion and metastasis, and the formation of suitable metastatic niches [81]. In addition to disrupting matrix barriers, MMPs can facilitate the release of matrikines (i.e. ECM fragments possessing cytokine-like functions), which have also been implicated in regulating multiple facets of cancer development and progression [81,82]. Myosins comprise a large superfamily of motor proteins that are capable of hydrolyzing adenosine triphosphate (ATP) and interacting with actin to generate force and motion. Myosin II motors are the best studied among all myosins, and virtually all kinetic models of myosins moving along actin filaments are established based on this 'conventional' class. Despite the recognized importance of participating in skeletal, cardiac, and smooth muscle contraction, myosin II motors are also found to be present in all non-muscle eukaryotic cells and responsible for fundamental processes that require cellular reshaping and movements, such as cell adhesion, cell migration, and cell division. Several lines of evidence have suggested that myosin II has an indispensable role in governing the diverse cell migration patterns during cancer metastasis [83,84]. Our results suggest that myosin II-dependent cell contraction, as compared to MMPs-driven ECM degradation, should have a more prominent role in regulating migratory and invasive behaviors of cancer clusters at a heterogeneous interface. Nevertheless, we should not overlook the involvement of MMPs in facilitating single-cell detachment and early protrusion outgrowth of tumor aggregates (figures 8 and 9).
In general, cancer cells utilize filopodia and filopodia-like protrusions to support directed singlecell migration and invasion at distant metastatic sites [85]. It has been reported that the function of filopodia is the sensing of the ECM; filopodia may even have the ability to recognize topographical features in the underlying substrate [86]. In our matrix-substrate interface, the cells within the lower hemisphere of tumor spheroids might make use of 'filopodia' to sense the topographic cues from the interfacial surface, and therefore their maximum length might determine a threshold distance, beyond which the invasive outgrowth of the tumor spheroids might not be able to take place, even when the situated above the interfacial surface. As for MDA-MB-231 cells, it has been demonstrated that they can develop very long flexible filopodia when cultured in 3D collagen matrix, the length of which is about 5-40 µm [87]. In our study, the distance between the interfacial surface and the middle of the spheroid, on average, was about 25-30 µm, the threshold distance attributed to the length of cell's filopodium might account for why invasive outgrowth always occurred from the lower hemisphere of the spheroids situated above the matrix-substrate interface.
Mechanistically, the current data suggest that the formation of dome-shaped void (cell-free and collagen-free) space should be majorly attributable to actomyosin-based contractile forces (figure 9), as the inhibition of myosin II appeared to totally abrogate the formation of this microstructure. On this basis, when the cells are being detached from the lower hemisphere and then moving towards the interfacial boundary (i.e. the above-mentioned diepafitaxtic migration), the contractile forces, generated by actomyosin networks in these migrating cells, would pull against and eventually tear off the surrounding collagen matrix. From a different aspect, the contractile forces might also be transmitted to the top of the spheroids and then cause 'bending' of its upper hemisphere, which also contributes to the formation of such hollow microstructure. However, this assumption has yet to be proven with further evidence. Furthermore, in connection with all of the above investigations, it might be presumably suggested that an increase in the longitudinal length of such dome-like void space can serve as indicators of enhanced interfacial invasiveness, single-cell detachment, and the mechanical forces experienced by cells inside the spheroids. At the molecular level, although this microstructure contains neither cells nor collagen matrix, the issue regarding whether (and which of the secreted molecules) are present in this space is still unclear and has yet to be investigated in the future.
By and large, we developed an in vitro interface platform to explore the migratory and invasive behavior of MCTSs at the soft-hard tissue boundaries in vivo. For most soft tissues and organs, the Young's modulus is shown to range from 0.1 KPa to 1 MPa, depending on their location and function [88], while hard tissues, such as cortical bone, have the Young's modulus between 10 and 30 GPa [89]. To simulate interfaces between tissues exhibiting differential stiffness, collagen matrix hydrogels, which are widely used for soft-tissue regeneration, were placed above solid substrates made of either glass or PDMS, both of which have been broadly used as substrate materials of cell culture microchips. Indeed, they are mechanically much stiffer than most soft tissues, or even harder than the cortical bone. Apart from this, chemical composition and structure of these two substrate materials are largely dissimilar from those of animal tissues, which would be the potential confounders in this research that potentially led to observations distorted from the real-world phenomena. Moreover, either of these substrate materials is impenetrable to the cells, and therefore insights as to whether and how tumor aggregates penetrate through the soft-hard tissue interfaces have yet to be obtained by using different in vitro interface platforms, such as a hydrogelhydrogel model system with stiffness gradients.

Conclusions
To our knowledge, the present study is the first attempt to investigate the interfacial cell migration and invasion behavior of the MCTSs, in which both cell-substrate adhesion and matrix degradation play important roles. Here, we illustrate the roles of substrate stiffness and MSIAS in regulating the switch from collective migration to single-cell migration (i.e. single-cell detachment from MCTSs) at heterogeneous interfaces. The stiffer interface the spheroid contacts, the greater number the single cells are detached. In addition, substrate stiffness also contributes to the formation of a dome-shaped void (i.e. collagen matrix-free) space beneath the lower hemisphere of the spheroid that is adjacent to the interface. On the contrary, the increase of MSIAS appears to negatively influence the above consequences. Mechanistically, we demonstrate that myosin II motors serve as a key molecular player involved in the interfacial single-cell detachment process. Taken together, our findings highlight the significance of discrete heterogeneous anatomical tissue interfaces in addressing and countering cancer metastasis, since interfacial microenvironments foster invasiveness of aggregated cancer cells. Besides, this study provides a deeper insight into how mechanochemical properties affect cell migration and invasion at the interface. Finally, our data also suggest that myosin II motors might act as a potential therapeutic target to alleviate metastatic spread and outgrowth of tumor cells at tissue interfaces.

Data and materials availability
All data needed to evaluate the conclusions in the paper are present in the paper and/or the supplementary materials. Additional data related to this paper may be requested from the authors.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://doi.org/10.17632/4v625rp3cj.1