3D spheroid-microvasculature-on-a-chip for tumor-endothelium mechanobiology interplay

During the final stage of cancer metastasis, tumor cells embed themselves in distant capillary beds, from where they extravasate and establish secondary tumors. Recent findings underscore the pivotal roles of blood/lymphatic flow and shear stress in this intricate tumor extravasation process. Despite the increasing evidence, there is a dearth of systematic and biomechanical methodologies that accurately mimic intricate 3D microtissue interactions within a controlled hydrodynamic microenvironment. Addressing this gap, we introduce an easy-to-operate 3D spheroid-microvasculature-on-a-chip (SMAC) model. Operating under both static and regulated flow conditions, the SMAC model facilitates the replication of the biomechanical interplay between heterogeneous tumor spheroids and endothelium in a quantitative manner. Serving as an in vitro model for metastasis mechanobiology, our model unveils the phenomena of 3D spheroid-induced endothelial compression and cell-cell junction degradation during tumor migration and expansion. Furthermore, we investigated the influence of shear stress on endothelial orientation, polarization, and tumor spheroid expansion. Collectively, our SMAC model provides a compact, cost-efficient, and adaptable platform for probing the mechanobiology of metastasis.


Introduction
During cancer progression, angiogenesis is a critical driver for early stage tumor growth, and metastasis in the late stage leads to more than 90% of cancer-related deaths in patients [1,2]. The metastatic cascade involves a multi-step process including tumor cell invasion, migration, intravasation, dissemination, extravasation, and colonization at a secondary site. These processes are often found in blood or lymphatic vessels where tumor cells are commonly subjected to shear stress. Though only around 0.01% of tumor cells that have survived from the circulatory system and colonized at the surrounding tissue, it is essential to model the process of extravasation since the role of this minor population determines the metastatic potential [3,4].
Tumor cell extravasation consists of multiple steps, including adhesion, modulation of the endothelial barrier, transendothelial migration, and crossing the vascular basement membrane [5]. The understanding of cancer metastasis in most of the studies involves individual circulating tumor cells (CTCs) that undergo epithelial-to-mesenchymal transition and escape from the primary tumor and intravasate into the blood stream [6]. During extravasation, both tumor and endothelial cells experience mechanical forces [7]. For instance, it has been shown that metastatic breast epithelial cells reduce the stiffness of the endothelium and promote epithelial cell transmigration [8]. Metastatic tumor cells have shown decreased stiffness [8,9], and their nuclei soften during transendothelial migration [10]. While the importance of CTCs has been emphasized, it is also proposed that CTC clusters, referred to as tumorderived microemboli or circulating tumor aggregates, may break off from primary tumors and lodge into distal capillaries to initiate metastatic growth. These CTC clusters have been detected within the circulation of patients with metastatic epithelial cancers [11][12][13][14][15]. More importantly, CTC clusters constitute rare but very highly metastasis-competent CTCs [16], and hematogenous cancer metastases could originate from CTC clusters at the site of primary tumor cell attachment to the vascular wall [17].
Fluid shear stress extensively affects the extravasation process. It has been reported that low fluid shear stress caused a two-fold increase in the efficiency of breast cancer cell adhesion to endothelium and promoted colonization of tumor cells after extravasating from the vascular system [18,19]. Other studies suggested that a certain range of fluid shear stress can inhibit tumor cell adhesion [20] and high shear force can lead to a malignant phenotype of tumor cells [19]. Moreover, tumor cells can be regulated by fluid shear stress to form stable CTC clusters that can facilitate tumor colonization at the secondary site [21]. Collectively, these studies demonstrate that tumor metastasis is tightly regulated and coordinated by hydrodynamic forces. The underlying effect of the bloodstream-related hemodynamic wall shear stress on tumor and their interaction with endothelium during extravasation is a fascinating question to study.
Nevertheless, recapitulating the physiological and mechanical microenvironment to study tumor extravasation and metastasis remains to be a challenge in the field. Specifically, the blood vessels where tumor invasion occurs are not accessible and feasible to investigate alongside 3D tumor tissues simultaneously. Researchers commonly used simple culture dishes with endothelial monolayer and spheroids to study the interaction [22]. Numerous studies have also used trans-well plates or porous membrane to coculture the cancer and endothelial cells to examine the transmigration [23][24][25], for example the molecularto-phenotypic features of tumors mortality and their endothelial invasion [26,27]. Recently, in vitro organon-chip platforms [28][29][30][31][32] have rapidly emerged as comprehensive humanized models that incorporate sophisticated vascular anatomies or 3D tumor microenvironment [33,34]. However, those systems still lack integration of tumor spheroids as CTC clusters and shear stress modulation built into the process.
3D multicellular tumor spheroid is a promising model that mimics the in vivo solid tumor. The tumor spheroids have potential to bridge the gap between a 2D monolayer cell culture and in vivo studies by providing similar in vivo biomechanical integrities, such as spatial interactions of cells with surrounding cells and extracellular matrix [35][36][37] in the tumor microenvironment [38]. In this regard, we employed a liquid dome model to produce more than 200 tumor spheroids in less than a day as previously reported [39]. Specifically, the surface tension created by this model, lead to the formation of the liquid dome of cancer cell suspension on the microwell array on the chip. The liquid height at different positions of the dome controls the number of cells that settle in the microwells, producing 3D spheroids with diverse sizes, critical for recapitulating the heterogeneity of human tumor spheroid.
In this study, we developed a spheroidmicrovasculature-on-a-chip (SMAC) model that incorporates 3D tumor spheroids into an endothelialized microfluidic device ( figure 1(a)). Tumor spheroids with gradient-sizes were generated using a liquid dome method. The SMAC model enabled a thorough investigation of the interaction between various tumor sizes and the endothelial layer under a diverse range of shear stresses, including: (1) tumor spheroid-induced endothelial cell compression and cell-cell junction degradation (i.e., vascular endothelial-cadherin (VE-cadherin)); (2) tumor spheroid migration and expansion in the endothelialized environment; and (3) the shear stress effects on the endothelial cell orientation and polarization as well as tumor spheroid expansion (figure 1(a)). By seeding tumor spheroids in an endothelialized microfluidic channel, our SMAC model resembles the tumor microenvironment, in particular, incorporating an endothelial layer and fluid shear stress that regulates tumor expansion and extravasation.

Fabrication of SMAC
Using a standard soft lithography and single-mask microfabrication process [40][41][42], we created parallel microfluidic channels to model venous capillary anatomies within a few hundred microns on a polydimethylsiloxane (PDMS) chip ( figure 1(b)). To enable SMAC co-culture of endothelial cells and multiple tumor spheroids, we trialed multiple microchannel dimensions ranging from 300 × 100 µm to 1000 × 500 µm (cross section: width × height), and then optimized the design by checking the stability and reproducibility of tumor spheroids culture in the endothelialized channel after perfusion at bulk shear stress τ 0 = 1-15 dyn cm −2 . In quick conclusion, the optimal micro-channel design was set at 600 µm × 300 µm × 25 mm (width × height × length) with 6 mm and 2 mm holes punched at the inlet and outlet for static and flow culture respectively (figure 1(b), supplementary figure  1). This design allowed (i) initial tumor spheroid localization and stable attachment in shear flow condition; (ii) sufficient spaces for tumor spheroid migration and expansion during the long-term culture; (iii) capability to culture numerous tumor spheroids in the same channel (figure 1(b), supplementary figure 1).
Upon optimizing the microfluidic design, the SMAC biofunctionalization consists of four major steps (figure 1(c)): (1) Generate tumor spheroids using the liquid dome arrays; (2) seed endothelial cells and form monolayers inside the microfluidic channel; (3) seed 3D spheroids into the microvasculatureon-a-chip; and (4) apply shear stress under perfusion. In Step 1, human breast cancer cells, MCF-7, were seeded into agarose microwells that allow gradientsized spheroids formation as previously reported (figure 1(c); supplementary figure 2(a)) [39]. After overnight culture, MCF-7 formed tumor spheroids and were harvested from microwells (supplementary figure 2(a)). Immunostaining of F-actin and nucleus by 3D confocal microscopy validated the compact intercellular connection and spherical morphology (supplementary figure 2(b)). Meanwhile for Step 2, human umbilical vein endothelial cells (HUVECs; abbreviated as ECs) were seeded into the channels and successfully transformed into confluent monolayers spanning the entire inner lumen within 24 h (figure 1(d)). The VE-cadherin expression throughout the entire inner surface in our endothelialized microchannels further confirmed the existence of integral junctions, therefore the endothelial cells grossly achieved confluence and function appropriately (figure 1(d); green) [34].

Tumor spheroids compress endothelium and degrade cell-cell junction upon extravasation
To examine the biomechanical impacts of tumor extravasation against the endothelium, we cocultured endothelial cells and tumor spheroids statically in the SMAC model and fixed them after 6 h. We defined two regions of interest: (1) The proximal inner layer of endothelium that are in immediate contact (less than two-cell body length) with the central spheroid (figure 2(a), broken marquee); (2) The distal endothelial region ouside the proximal perimeter and not directly contacting the spheroids (figure 2(a), solid marquee). Interestingly, the proximal endothelium contacting both large (LTS; figure 2(b), x) and small (STS; figure 2(b), y) spheroids partially lost the VE-cadherin, while the distal regions in both cases (figure 2(b), z and {) displayed little difference from those intact endothelial cells in the absence of spheroids ( figure 2(b), |). Moreover, we normalized the VE-cadherin intensity profile across the orthogonal line region of interest (ROI) placed at 1/3 of the cell body distance. The ROI intensity of the endothelial cells proximal to both large (figure 2(c), x) and small (figure 2(c), y) spheroids only displayed a single peak in confirming the loss of VE-cadherin, whereas those distal to spheroids or in the absence of any spheroids displayed double peaks demonstrating the intact VE-cadherin expression in both edges (figure 2(c), z{|). These analyses confirmed that the spheroid contact edges of proximal endothelial cells no longer expressed VE-cadherin, whereas the VE-cadherin expression of distal endothelial cells remained intact. In another word, tumor spheroids degraded the endothelial junctions upon its invasion [43]. Further, the thickness of VE-cadherin was observed to be larger in the presence of tumor spheroids than nontumor microenvironment (figures 2(a) and (b)), possibly resulting from the endothelial movement and remodelling [44,45] caused by tumor invasion and migration. Intrigueingly, the proximal endothelial cells to a large spheroid were significantly compressed and elongated (figure 2(b), x) in comparison to those proximal to a small spheroid (figure 2(b), y). The large spheroid could compress the width of endothelial cell from 17 to 7 µm while the small spheroid could not (figure 2(c), x vs. y|). However, the endothelial cells distal to both spheroid subgroups showed no compression (figure 2(c), z and {)and retained their hexagonal shape comparable to those seeded without any spheroids (EC only; figure 2(c), |). To quantify the compression effects, we measured the aspect ratio (AR) of endothelial cells defined by the ratio of the long axis to the ort axis of the cell body. No significant difference was observed upon the ARs of distal endothelial cells to large spheroids (figure 2(d), AR = 1.97 ± 0.06), small spheroids (AR = 2.07 ± 0.06) and those without spheroids (AR = 1.71 ± 0.05). However, the proximal endothelial cells to large spheroids had an increased mean AR at 2.64 ± 0.13, which decreased to 2.15 ± 0.14 for proximal endothelial cells contacting small spheroids (figure 2(d)). In contrast to AR measurement, the nucleus circularity showed no significant difference among all conditions (figure 2(e)). The results suggested extravasating tumor impose compressive force on adjacent endothelial cells, and the subsequent morphological impacts depend on the spheroid sizes.   spheroids' main bodies (figures 3(a) and (b), white arrows). In contrast, both large and small spheroids retained their spherical morphologies and exhibited isotropic expansion in the bare microchannel without endothelization (figures 3(a) and (b)). We then quantified the area expansion of the tumor spheroids and the result showed that large spheroids with endothelium (0.25 ± 0.06 × 10 4 µm 2 hr −1 ) exhibited significantly lower expansion rates compared to those without endothelium (0.51 ± 0.05 × 10 4 µm 2 hr −1 ); while small spheroids had similar expansion rates with (0.11 ± 0.08 × 10 4 µm 2 hr −1 ) and without (0.096 ± 0.02 × 10 4 µm 2 hr −1 ) endothelium (figures 3(c) and (d)). The results suggested that the endothelium significantly decreased the expansion rate of large but not STS. Moreover, the expansion rate of large spheroids was significantly higher than small spheroids without the endothelium, which was not evident in the presence of endothelium (figure 3(d)). Nonetheless, the large spheroids with and without endothelium continued expanding until 12.6 and 13.6 h, respectively; while the expansions of small spheroids with and without endothelium saturated in approximately 3.3 and 11 h, respectively ( figure 3(c)). This result suggested that endothelium accelerated the saturation of tumor spheroid expansion, which was more pronounced in small than in large spheroids.

Shear stress influences the tumor-endothelialium interplay
To evaluate the impact of hydrodynamic forces on endothelium orientation and polarization with the presence of tumor spheroids, we subjected the SMAC to different levels of bulk shear stresses: static (τ 0 = 0 dyn cm −2 ), venous (τ 0 = 5 dyn cm −2 ) and arterial (τ 0 =15 dyn cm −2 ) conditions [46] using a peristaltic pump (figure 4). Shear stress was continuously applied for 40 h, providing a sufficient time window for spheroids and endothelium to respond. Taken together, these data indicated that endothelial cell orientation and polarization are highly affected by shear stresses and the tumor spheroids.
Furthermore, we evaluated the shear stress effects on the tumor spheroid expansion in the SMAC model. We reconstructed the contours of tumor spheroids derived from 3D confocal images and mapped the surface shear stress τ for both large (figures 5(a) and (b)) and small (supplementary figures 4(a) and (b)) SMAC models by the established computational fluid dynamic (CFD) analysis [40,48]. Notably, the mean shear stress of the large spheroid exposed to a bulk shear at τ 0 = 5 dyn cm −2 (τ ave = 9.7 dyn cm −2 ) was significantly lower than that at τ 0 = 15 dyn cm −2 (τ ave = 31.  left) were observed when the shear stress at 5 dyn cm −2 was introduced. We then quantified the area expansion of each individual tumor spheroids and found that the spheroid's final area was linearly proportional to the initial area at static (τ 0 = 0 dyn cm −2 ), venous (τ 0 = 5 dyn cm −2 ) and arterial (τ 0 = 15 dyn cm −2 ) conditions ( figure 5(d)). We then constructed the spheroid growth rate by normalizing with the initial area. Remarkably, the result showed that spheroid growth rate when exposed to low (2.53 ± 0.22) and high (2.45 ± 0.41) shear  stress was significantly decreased as opposed to static condition (4.32 ± 0.34) (figures 5(d) and (e)). This indicated the suppressive effect of the shear stress on tumor migration and expansion. In conclusion, venous shear stress (5 dyn cm −2 ) promoted endothelial alignment and polarization perpendicular to the flow direction, while arterial shear stress (15 dyn cm −2 ) induced parallel orientation and polarization of endothelial cells with the flow. However, the existence of tumor spheroids had a more significant disruption of endothelial orientation than polarization. We also found that both venous and arterial shear exposure downregulated the tumor migration and produced smaller cavities, which implicates less endothelial permeability and slow extravasation.

Discussion
In this study, we developed and characterized the SMAC model-a new microfluidic platform replicating key characteristics of tumor extravasation. This includes a size scale congruent to physiological conditions, an endothelial monolayer cultured over the entire 3D inner surface of the system, and tunable physiologically relevant hydrodynamic parameters. Our observations indicated that tumor spheroids prompted cellular compression and cell-cell junction degradation within the endothelial layer. Furthermore, we discovered that endothelial cells decelerated the spreading rate of spheroids, triggering invasive branch development. We also found that shear stress considerably hindered spheroid expansion within our model. The tumor-endothelium model affords high spatiotemporal resolution and opens avenues for promising oncology therapies development.
Previous research concentrated on the biochemical impact of tumor-endothelium communication [49,50], often underestimating the biomechanical regulation until recently. Tumor cells can secrete soluble factors triggering endothelial cell retraction, thereby facilitating tumor cell attachment and transmigration across the endothelial boundary [51,52]. Additional studies spotlighted complex interactions between tumor and endothelium, such as tumor cell miRNA regulon mediated endothelial recruitment [53] and exosome-induced endothelial barrier disruption [54]. However, the biomechanical regulation remained underestimated until recent times when the importance of biophysical factors began to gain recognition.
Former studies utilized in vitro parallel plates for examining single tumor cell interaction with endothelial cells and the In vitro mouse model for homotypic aggregation assay in microvasculature [17]. In vitro microfluidic models have been introduced to mimic the 3D structural organization of tumor microenvironment [31] or to study cancer cell extravasation in microvasculature [32], which significantly advanced the field of tumor metastasis study. However, there is currently no model available to study the CTC clusters' interaction with endothelium in a hemodynamic environment. To this end, our SMAC model incorporates a coculture of endothelial monolayer (mimicking vessel wall) and varioussized tumor spheroids (mimicking CTC clusters) in vitro to provide an experimental recapitulation of the tumor-endothelium interplay and associated mechanobiology [53,55], including endothelial compression, junction loss, polarization against shear stress, and rheological microenvironment.
In this investigation, we optimized the microfluidic channel size to facilitate stable attachment and sufficient space for tumor spheroid culture in fluid shear stress conditions. Employing a gradientdriven tumor spheroid formation (the liquid dome method) and the vascularized-microfluidic method, we achieved the production and analysis of the biomechanical interplay between the endothelium and heterogeneous tumor spheroids of diameters ranging from 30 to 200 µm.
Our findings indicate that tumor spheroids degrade the neighboring endothelial cellcell junctions, a process that is distinct from tumor cell transendothelial migration, where VEcadherin plays a significant role in intercellular mechanotransduction [56,57]. We also observed endothelial compression followed by an increased endothelial cell AR through direct contact with tumor spheroids during their migration and expansion.
Such biomechanical compression on endothelium was more pronounced in large than STS, possibly due to several factors, including the presence of higher compressive forces, the highly proliferating outer zone of the spheroids, and the increased development of solid stress and stiffness within the larger tumor spheroids [58][59][60]. Further, these factors could also regulate the migration and expansion of the tumor spheroids toward the endothelium. The SMAC model enables real-time monitoring of tumor spheroid expansion on the endothelium. Our study demonstrated that the endothelium decelerates spheroid expansion and induces tumor invasive branches. This is relevant to the recent findings that the tumor cells in the invasive branches are softer, larger, and more dynamic than those at the spheroid core [61]. The emerging concept of VE-cadherin degradation, endothelial cell compression and tumor invasion is important in the context of tumor angiogenesis and metastasis [43], and our SMAC platform provides a comprehensive understanding of the biomechanical coordination of tumor extravasation through the endothelium.
Upon incorporating shear flow into the endothelialized tumor-spheroid-on-chip, we could discern the endothelial and tumor spheroid responses under various shear stress conditions. Our findings indicate that venous shear stress (5 dyn cm −2 ) promoted endothelial alignment and polarization perpendicular to the flow direction, while arterial shear stress (15 dyn cm −2 ) induced parallel orientation and polarization of endothelial cells with the flow. However, the existence of LTS had a greater disruption of endothelial orientation than polarization. Shear stress-induced activation of mechanosensitive ion channels, such as Piezo1 [62], and clustering of mechanoreceptors, such as integrins [63,64], were highlighted in cancer metastasis. Additionally, we found that the cavity between an expanding tumor spheroid and the adjacent endothelium was significantly suppressed by shear stress as opposed to the static condition.
Despite its novel findings, our SMAC model bears a few limitations. Primarily, the culture of endothelium on a rigid surface impeded tumor cells from transmigration, hindering the full recapitulation of the extravasation process. As a result, the assessment of endothelial compression by large and STS was not entirely accurate. Additionally, without the ability to spread on the vasculature, the timeframe and speed of tumor extravasation and expansion might differ from In vitro conditions. These restrictions influence the results of our tumor expansion analysis and may not entirely replicate the actual tumor expansion process. Nevertheless, the controlled environment of our simplified model enabled us to compare LTS and STS behaviors in various microenvironments, including EC versus non-EC and shear versus static conditions. This study provides valuable insights into the tumor spheroid-endothelium interactions during extravasation. To address these limitations, future research should aim to enhance the current model or design new ones that better simulate the complexities of extravasation. These improvements will further our understanding of tumor behavior during extravasation.

Conclusions
Our innovative SMAC model offers a practical avenue for investigating extravasation mechanobiology within various contexts, paving the way for future mechanomedicine and drug discovery. It is an effective tool to explore the biomechanical coordination of tumor extravasation through the endothelium, and other key factors involved in the metastatic process. In essence, the SMAC model emerges as a versatile and promising platform for future mechanobiological studies of metastasis.

Microfluidic chip fabrication
The microfluidic devices were designed and optimized for dimension enabling high-throughput production. Of note, we screened the best design sizes of the microfluidic channel from 300 × 100 µm to 1000 × 500 µm (width × height). Though a large channel with 1000 × 500 µm caused no significant issue of co-culture, we further reduced the size of the channel for miniaturization, saving culture medium and cell usage. However, we found that a small channel with 300 × 100 µm was not able to accommodate tumor spheroids in neither static nor dynamic culture due to their inability to settle down on the endothelialized surface, which is caused by higher shear stress from narrow space (supplementary figure 5). In result, we optimized the microfluidic channel to 600 µm × 300 µm × 25 mm (width × height × length) to allow stable attachment, sufficient spaces for tumor spheroids culture in fluid stress.
After being designed in AutoCAD@ software (version 2015; AutoDesk, San Rafael, CA, USA), the microchannel master was fabricated using dry photoresists [40]. Firstly, a 6-inch silicon wafer was cleaned and baked for 10 min at 200 • C, followed by hexamethyldisilazane vapor priming for 30 s at 120 • C. A layer of 200 µm constant height dry photoresist film (DJ micro laminate SUEX ® ) was then laminated on the silicon wafer at 65 • C. Then a 100 µm dry photoresist film was then laminated similarly to the previous photoresist, followed by patterning using a direct lithography writer (Heidelberg MLA-100, 4500 mJ cm −2 ) to create a channel with constant height at 300 µm. Upon lithography completion, the wafer was baked for 5 min at 90 • C, followed by PGMEA rinsing until excess photoresist was completely removed. To prevent permanent PDMS adhesion, the patterned wafer was treated with silane in a vacuum for 2 h and became the master mold.
To fabricate the microfluidic device, PDMS (Sylgard ® 184 by Dow Corning) was mixed with the curing agent at a 10:1 ratio (w/w). Then the PDMS was poured on the master mold and heated in the oven at 60 • C for 4 h. After it was solidified, the cured PDMS was peeled off from the silicon master and cut into pieces. Then the inlets and outlets of the PDMS chips were punched with Ø6 mm and Ø2 mm biopsy punchers (World Precision Instruments) for static and flow experiments, respectively. Lastly, the PDMS chip was permanently bonded to 1 mm glass slides by plasma treatment for 3 min.

Endothelization of microfluidic chips
HUVECs were obtained from Thermo Fisher Scientific and cultured with EGM-2 medium (EGM™-2 BulletKit™, Lonza). Once reached 80%-90% confluency, HUVECs (passages 3-7) were washed with phosphate-buffered saline (PBS, ThermoFisher) and detached by trypsin/EDTA solution (ThermoFisher). After centrifuge, HUVECs were resuspended in EGM-2 medium at a seeding density of ∼5 × 10 6 cells ml −1 . Prior to HUVECs seeding, the microfluidic chip was sterilized with 80% ethanol for 20 min and washed thrice with PBS. Then the entire channels were coated with 100 µg ml −1 human plasma fibronectin (Thermo Fisher), incubating in a 4 • C fridge overnight. The channels were then rinsed with PBS twice, and then 5 µl of prepared HUVECs suspension was injected into each microchannel. The microfluidic chip was then immediately flipped upside down to allow HUVECs attachment to the channel's upper surface for 20 min. Then the chip was flipped again to allow HUVECs to attach to the bottom surface for 20 min. After that, the EGM-2 medium was added to the reservoirs to culture HUVECs statically overnight, and the endothelialized microfluidic chip was completed.

Tumor spheroid preparation and seeding into microwells
The gradient-sized multicellular tumor spheroids were prepared using the reported method [39]. Briefly, MCF-7 cells were obtained from Australian Cell Bank and cultured with RPMI1640 medium supplemented with 10% fetal bovine serum (Sigma-Aldrich) and 100 U ml −1 penicillin/streptomycin. When the cells were up to 70%-80% confluent, MCF-7 cells were washed with PBS and detached by trypsin/EDTA solution. After centrifuge, MCF-7 cells were resuspended in RPMI1640 medium at a seeding density of 5 × 10 6 cells ml −1 . Then, 100ul of MCF-7 cell suspension was added on top of the 1.8% agarose chip (containing 217 microwells) in a 6-well plate and let settle for 5 min. Afterward, 2 ml of RPMI1640 medium was added to the 6-well plate, and the MCF-7 was cultured in the agarose chip for 24 h. Due to the difference in cell suspension volume, the number of MCF-7 cells in each microwell varies, with more cells in the middle and fewer cells towards the side of the chip. Lower concentration would lead to smaller spheroids. The MCF-7 spheroids are then retrieved from the agarose chip to a microtube for usage later.

Co-culture of tumor cells and HUVECs on the chip under static and medium flow conditions
After 24 h, the MCF-7 spheroids formed and were harvested into a microtube. Once the spheroids were settled down at the bottom of the microtube, remove the supernatant and add 1 ml of the EGM-2 medium. When the spheroids are settled again, inject 10 µl of the spheroid suspension into the endothelialized channel and allow 6 h for the spheroids to stay firmly attached. For static culture, the culture medium was then added to the reservoirs and changed once a day. For dynamic culture, the tubings were connected with the microfluidic chip and the peristaltic pump (Harvard, P70) which was set at 375 µl min −1 and 1.12 ml min −1 for 5 dyn cm −2 and 15 dyn cm −2 respectively. The microfluidic chip was then cultured for 40 h continuously.

Immunostaining and confocal microscopy
The SMAC was fixed with 4% paraformaldehyde (company), then thoroughly washed with PBS and permeabilized with 1% Triton X-100 (Roche) at room temperature for 1 h. Then the chip was blocked with 5% bovine serum albumin (Sigma-Aldrich). Subsequently, the SMAC were incubated with anti-VE-Cadherin antibody (Invitrogen) conjugated with Alexa Fluor™ 488 (1:100) for endothelial cells and anti-EpCAM antibody (Cell Signaling Technology) conjugated with Alexa Fluor™ 555 (1:100) for MCF-7 tumor spheroids for 1 h at 37 • C. After washing with PBS, the chip was stained with Hoechst 33 342 (Abcam) or Phalloidin (Abcam) for nuclei and Factin for 20 min at room temperature, respectively. After washing, the microfluidic device was imaged using an Olympus FV3000RS laser scanning confocal microscope and operated by FLUOVIEW software.

AR analysis
To quantify the influence of the tumor spheroid on the endothelium, we calculated the AR of endothelial cell body and the circularity of the nuclei. The ARwas defined as the ratio of the long axis to the short axis automatically calculated by IMARIS (Bitplane AG, 9.0.1, Oxford Instruments). The circularity of a cell was defined by the ratio of the cell surface area of a sphere (with the same volume as the given particle) to the cell surface area of the particle (Wadell 1932), which was also automatically calculated by IMARIS.
The histogram graphs were plotted in figures 2(d) and (e).
The auto-detection of cells performed by IMARIS is in need for manual editing to have a better detection efficiency (correct detection of cell nuclei, shapes, and area, etc.). We employed two different image processing methods for 2D and 3D confocal image. For 2D images, the original images containing VEcadherin, nuclei, and F-actin signals were first transported to the format of TIF through IMARIS, and then imported into Adobe Photoshop CC 2019. Due to the compromised image resolution, the defects of VE-cadherin signals at boundary of the endothelial cells have brought difficulty in detecting the whole cell body. Therefore, we manually edited the boundary of the cells according to the F-actin signals. To avoid false detection by IMARIS, we manually added a fake nucleus at the empty space where endothelial cells were not present which were then filtered out by selecting the regions of interest.

Orientation and polarization analysis
We quantified the endothelial cell orientation by calculating the angle between the flow direction and the 'cell vector' determined by the long axis of each endothelial cell. The orientation of the endothelial cells (∼200 cells for one sample, n = 3) was quantified and divided into 18 subgroups from 0 • to 180 • and plotted into the hemi-rose graph as shown in To quantify the polarization of the endothelial cells, we defined the 'polarization angle' as the angle between the vector of flow direction and the vector formed by the nucleus and Golgi body. The polarization of the endothelial cells (∼200 cells for one sample, n = 3) was quantified and divided into 18 subgroups from 0 • to 360 • and plotted into the wind rose graph as shown in figures 5(e) and (f). Similarly, to plot the histogram graph presented in figures 5(g) and (h), we categorized the endothelial cells into three types: parallel (135 • -225 • ), vertical (45 • -135 • , and 225 • -315 • ), antiparallel (0 • -45 • and 315 • -360 • ) to the flow direction. The angle calculation for orientation and polarization was automatically generated from the detection of IMARIS and a handmade script in EXCEL (Microsoft Office 2016) which was validated by ImageJ (National Institutes of Health, USA, 1.53 K). The graphs were plotted using Origin (OriginLab, 2019b).

CFD analysis
The CFD simulation was performed over a virtual microfluidic channel (length x 0 = 1000 µm, width y 0 = 600 µm, and height z 0 = 300 µm) utilizing ANSYS ® Fluent 2020 R1 software (version 20.1; Canonsburg, PA, USA) as previously described [48,65]. The tumors were reconstructed from the 3D confocal scan using Imaris software (Bitplane) [40] and placed in the middle of the channel. The tumor and the channel were then meshed into 1 µm grids for an iterative solving process. The fluidic dynamic properties of the culture medium was set at a density of 1000 kg m −3 and a viscosity of 0.72 Pa s. The whole interstitial fluid was assumed laminar and incompressible, and the channel was assumed to be rigid. In the solving process, the standard secondorder scheme and the second-order upwind scheme were selected with the COUPLED algorithm utilizing pressure-velocity coupling [66]. The bulk shear stress σ 0 at the micro-channel inlet was set at 5 and 15 dyn cm −2 . After solving the Navier-Stokes equation for the whole interstitial space, the simulated shear stress contours and fluid streamlines around the tumor were constructed and exported as the results.

Live cell imaging and analysis
For live cell imaging, before seeding the tumor spheroids, the endothelialized channel was stained with anti-CD31 antibody conjugated with Alexa Fluor™ 488 (AbCam, 1:100), and the tumor spheroids were stained with CellMask™ Orange (ThermoFisher, 1:1000) for half an hour at 37 • C. After washing with PBS, the stained tumor spheroids were seeded into the endothelialized chip for live cell imaging using an Olympus FLUOVIEW FV3000 confocal laser scanning microscope adapted with Tokai Hit Incubator for 18 h with 20 min intervals. To track the area expansion of tumor spheroid, the bottom layer of the tumor was extracted and manually selected using the polygon selection tool in ImageJ. The tumor spreading area was then measured in ImageJ using the measurement tool.

Statistical analysis
All graphical data are presented by GraphPad Prism 9.0. Statistical differences between each group were tested by a one-way analysis of variance (ANOVA) test. A p-value below 0.05 was accepted as significant.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).