Lung Tumor Microphysiological System with 3D Endothelium to Evaluate Modulators of T-Cell Migration

Lung cancer is a leading cause of death worldwide, with only a fraction of patients responding to immunotherapy. The correlation between increased T-cell infiltration and positive patient outcomes has motivated the search for therapeutics promoting T-cell infiltration. While transwell and spheroid platforms have been employed, these models lack flow and endothelial barriers, and cannot faithfully model T-cell adhesion, extravasation, and migration through 3D tissue. Presented here is a 3D chemotaxis assay, in a lung tumor-on-chip model with 3D endothelium (LToC-Endo), to address this need. The described assay consists of a HUVEC-derived vascular tubule cultured under rocking flow, through which T-cells are added; a collag-enous stromal barrier, through which T-cells migrate; and a chemoattractant/tumor (HCC0827 or NCI-H520) compartment. Here, activated T-cells extravasate and migrate in response to gradients of rhCXCL11 and rhCXCL12. Adopting a T-cell activation protocol with a rest period enables proliferative burst prior to introducing T-cells into chips and enhances assay sensitivity. In addition, incorporating this rest recovers endothelial activation in response to rhCXCL12. As a final control, we show that blocking ICAM-1 interferes with T-cell adhesion and chemotaxis. This microphysiological system, which mimics in vivo stromal and vascular barriers, can be used to evaluate potentiation of immune chemotaxis into tumors while probing for vascular responses to potential therapeutics. Finally, we propose translational strategies by which this assay could be linked to preclinical and clinical models to support human dose prediction, personalized medicine

stroma but are not able to reach the tumor cells, or "immune desert", in which cytotoxic T-cells are absent from both the tumor nest and stroma (Hegde and Chen, 2020;Tokito et al., 2016).Numerous factors, such as lack of chemokine gradients, reduced integrin activation, increased density of tumor stroma, and abnormal vasculature, as well as immune-suppressive soluble factors and immune cells (e.g., regulatory T-cells) are hypothesized to contribute to T-cell exclusion from the tumor microenvironment (Oelkrug and Ramage, 2014;Zhang et al., 2019;van der Woude et al., 2017;Tokito et al., 2016;Chen and Mellman, 2013).Given that a high presence of cytotoxic T-cells in tumors is correlated with improved patient survival, there is a strong need to enhance T-cell migration into the 3D tumor

Introduction
Although immunotherapy has shown great promise, immune cell infiltration in many indications and/or sub-indications remains challenging, leading to mixed clinical outcomes (Oelkrug and Ramage, 2014;Zhang et al., 2019;van der Woude et al., 2017).Patients with "inflamed" tumors, in which immune cells are inhibited but in close contact with tumor cells, typically respond better to cancer immunotherapy and experience better prognoses (Hegde and Chen, 2020;Tokito et al., 2016).In contrast, patients experience poorer outcomes if their tumors are "immune excluded", in which cytotoxic T-cells have accumulated in the tumor model to investigate multiple stages of T-cell chemotaxis, including T-cell adhesion, extravasation, and migration through a 3D stromal barrier, to evaluate therapeutics that could enable T-cells to overcome these barriers and directly contact tumor cells.For maximum utility in drug discovery and development, it should be phenotypic-screening amenable, offering single cell resolution readouts without being time and data intensive to image or analyze.
Recent developments in organ-on-chip technologies have been encouraging, but many of these early models are low-throughput, made of polydimethylsiloxane (PDMS) (a hydrophobic material known to nonspecifically adsorb proteins), and contain artificial membranes (de Haan et al., 2021).The MIMETAS 3-lane 40 OrganoPlate ® is a platform containing 40 chips per plate, no PDMS, and phase guide technology that enables membrane-free material separation.Recently, this platform was used to investigate monocyte-to-endothelium adhesion, neutrophilic migration, and 3D T-cell chemotaxis in a melanoma model (de Haan et al., 2021).Building upon these models, we established a lung tumor-on-chip model with 3D endothelium ("LToC-Endo") to investigate immune cell chemotaxis in response to chemokines and antibody treatments.
Here we show that activated T-cells in the LToC-Endo model adhere, extravasate, and migrate in response to gradients of rhCXCL11 and rhCXCL12 (referred to throughout the manuscript as "CXCL11" and "CXCL12").The model can simultaneously discriminate between angiogenic (CXCL12) and non-angiogenic (CXCL11) chemotaxins, based on the observation of endothelial sprouting.Using this assay, we show functional differences between T-cells activated using different approaches and can inhibit migration by perturbing canonical endothelial receptor-T-cell receptor interactions.

Cell culture and media
All human biological samples were sourced ethically, and their research use was in accord with the terms of informed consent under an IRB/EC approved protocol.Human umbilical vein endothelial cells (HUVECs) (Lonza, product pooled from multiple donors) were cultured in complete human endothelial medium (Cell Biologics), expanded, and bio-banked in aliquots.HUVECs in all studies were used at or before passage 5. HCC0827 cells (licensed from the University of Texas Southwestern) were cultured in RPMI (Gibco) + 5% fetal bovine serum (FBS, Gibco).NCI-H520 (ATCC) cells were cultured in RPMI + 10% FBS.Primary human T-cells (AllCells, StemExpress, or BioIVT) were thawed in one of the following media solutions, as indicated in the studies described: either AIM V medium (Gibco) containing 20 IU/mL of  or RPMI + 10% FBS.For activated T-cells, 1:500 TransAct (Miltenyi), which utilizes polyclonal stimulation to activate T-cells via CD3 and CD28, was added to the medium.Activated-only T-cells were cultured for 48 h (either with or without 1:500 TransAct) prior to assay use.Activated-rested T-cells were cultured for 72 h in 1:500 TransAct, followed by a 48-h rest microenvironment and infiltration into solid tumors, thereby enhancing the effectiveness of immunotherapies (Oelkrug and Ramage, 2014;Zhang et al., 2019;van der Woude et al., 2017;Tokito et al., 2016;Chen and Mellman, 2013).Despite the clear rationale to address this aspect of the cancer-immunity cycle, there are limited potential therapeutics available to address it (Chen and Mellman, 2013).
While preclinical in vivo models have ushered in pivotal treatments in cancer immunotherapy (e.g., anti-CTLA-4 and anti-PD-(L)1), the limited translatability of preclinical models is a key challenge for the development of many immunotherapies (Hegde and Chen, 2020).Genetically engineered mouse models are considered to be the closest representation of human cancers, but mechanistic studies are challenging in whole animal models, and differences in species-specific immunology and disease progression hamper their clinical translatability (Hegde and Chen, 2020;Mestas and Hughes, 2004).Furthermore, increasing global attention on ethical issues with animal research has bolstered support for initiatives to replace, refine, and reduce animal models (Levy, 2012).
In vitro, transwell migration systems have been employed to investigate modulators of cell migration and chemotaxis.However, the effects of chemotactic triggers on migrating cells over long time windows remains challenging in these platforms due to gravity and gradient instability (Sip et al., 2014;Boyden, 1962;Zhang et al., 2016).Furthermore, these platforms are unable to recapitulate some aspects of the tumor microenvironment.Transwell membranes with rigid pores are unable to model dynamic cell extravasation through living, responsive vasculature or 3D cell migration through viscoelastic and mechanically plastic pores of extracellular matrix (Wisdom et al., 2018).Furthermore, as chemotaxis takes place along the z-axis in these assays, large confocal z-stacks, which may be time and data intensive to acquire and process, may be necessary to obtain single-cell resolution migration information.Alternatively, 3D spheroids are valuable for modeling T-cell infiltration into tumor nests (Herter et al., 2017;Rodrigues et al., 2020;Booij et al., 2019).However, optical clearing is necessary to image inside spheroids beyond 200 µm, which can only be done as an endpoint analysis.Furthermore, spheroid assays do not always model the extracellular matrix of solid tumors, even though dense stromal matrix is known to physically prevent infiltration into human lung tumors (Salmon et al., 2012).Additionally, growing evidence suggests that T-cells exhibit distinct kinds of motility dependent on both their activation state and features of their microenvironment (i.e., migration through 3D ECM versus tumor nest infiltration) (Krummel et al., 2016).For these reasons, infiltration studies with spheroids alone may not be sufficient to model the stromal constituents contributing to antitumor immunity in "immune excluded" and "immune desert" tumors.
While transwell and spheroid models can be informative and high-throughput, they lack a living endothelial barrier and vascular flow.For this reason, these platforms cannot be used to model extravasation, an early stage of T-cell chemotaxis, into tumors.There is a need for an integrated complex in vitro cells (HCC0827 or NCI-H520) were trypsinized, resuspended in endothelial medium, counted, and resuspended to density of 10 x 10 6 cells/mL.2 μL of cell suspension was then deposited into the bottom inlet port using an automatic repeater pipette.The cell suspension was regularly mixed to ensure homogenous cell seeding density.The OrganoPlate ® was placed with the lid forming a 75-degree angle against the plate stand, but with the plate rotated 180° from the previous HUVEC seeding step (i.e., top of the plate at the bottom, touching the incubator shelf), and left in this orientation for around 3 h to allow cells to attach.After, 50 μL of endothelial medium was added into the inlet of the bottom perfusion channel and it was placed back on the OrganoFlow ® rocker.
On day 0, T-cells or medium controls were seeded into the OrganoPlate ® .T-cells were harvested gently, centrifuged at 300 x g for 5 min, counted, and incubated in dye solution, either 2.5 μM CellTracker Orange CMRA (ThermoFisher) or 1:1000 NucLight Rapid Red (Sartorius), in AIM V medium.Cells were dyed at a concentration of 10 6 cells/mL, with no more than 3 x 10 6 cells per falcon tube.Tubes containing cells in dye solutions were wrapped in foil and placed in an incubator for 30 min.Halfway through the incubation period, the tubes were inverted several times to gently mix.T-cells were then centrifuged and pelleted to wash out the stain and resuspended in complete assay medium containing AIM V medium, 20 IU/mL IL-2, 5 ng/mL VEGF, and 5 ng/mL bFGF.Cells were then counted and diluted to the desired concentration in complete assay medium to deliver the number of T-cells per chip indicated in these studies in 50 μL of medium.At this stage, the top medium inlets and outlets were aspirated.50 μL of T-cell solution was added into the top medium inlet, and 50 μL complete assay medium was added into the top medium outlet.Then, the bottom medium inlet and outlets were aspirated, and replaced with 50 μL medium, each containing specified chemokine trigger or control medium solutions.For studies corresponding to Figures 3-5, a half-volume medium re-addition was implemented, in which 25 μL of additional complete assay medium was added into the top channel inlet and outlet, and 25 μL of chemokine trigger solution was added into the bottom channel inlet and outlet.For antibody blocking experiments, vehicle alone (PBS), IgG 1 antibody control (30 µg/mL, R&D Systems), ICAM-1/CD54 (10 µg/mL, R&D Systems) blocking antibody, or VCAM-1/CD106 (30 µg/mL, R&D Systems, BBA5) blocking antibody was added into the top channel inlets and outlets at the same time as chemotactic trigger addition into the bottom compartment (day 0) and also with the half medium refresh (day 2).Refer to Tables S2-S41 for additional information on equipment and reagents utilized.

Imaging and T-cell quantification
For data obtained in Figures 1-3 and Figures S1-6 1 , images were acquired using a spinning disc confocal, and migrating T-cells were quantified using a custom FIJI macro as previously described (de Haan et al., 2021).period, during which time the medium was washed out via centrifugation and replaced with RPMI + 10% FBS.Refer to Tables S2-S4 1 for additional information on cells and reagents.For Figures 2-3 and Figures S2-4 Figures 4-6 and Figures S5-8 1 , T-cells were isolated from StemExpress or BioIVT leukopaks internally at GSK. Leukopaks were received and stored at 4°C overnight (approx.16 h).First, peripheral blood mononuclear cells (PBMCs) were isolated using a custom PBMC Isolation Kit (Miltenyi), using magnetic beads to isolate out erythrocytes and granulocytes on magnetically charged cell selection columns while eluting PBMCs.T-cells were then isolated from the PBMCs using a standard Pan-T Isolation Kit (Miltenyi) using the manufacturer's protocol.T-cells were cryopreserved in CS10 (BioLife Solutions) in a rate-controlled freezer over the course of one hour and transferred to LN2 storage.Refer to Tables S3-S5 1 for additional information on cells and reagents.

T-cell chemotaxis and migration assay
Mimetas 3-lane 40 OrganoPlates ® (MIMETAS) were used for these studies.To seed the plates with collagen (day -2, indexed to T-cell addition day), 50 μL of DPBS was added into the observation port to facilitate making chip filling visible.To form the extracellular matrix barrier, rat tail collagen-1 (Cultrex) was mixed with HEPES and 37 g/L NaHCO 3 at an 8:1:1 ratio to form a 4 mg/mL collagen-1 solution.These components were mixed well > 20 times, being careful not to generate bubbles.Within 10 min, 1.8 μL gel solution was seeded into each chip using an automatic repeater pipette (Sartorius).The OrganoPlate ® was then placed in a humidified incubator (37°C, 5% CO 2 ) for 15 min to allow polymerization of the collagen-1 gel.30 μL PBS was then added into the gel inlet to hydrate the ECM layer prior to returning the plate to the incubator.To form the 3D endothelium, HUVECs were trypsinized, resuspended in endothelial medium, counted using an automated cell counter (ViCell Blu, Beckman Coulter), and resuspended to a cell seeding density of 10 x 10 6 cells/mL.PBS was removed from the gel inlets, and 2 μL of cell suspension was deposited into the top inlet port using the automatic repeater pipette.The cell suspension was regularly mixed in order to ensure homogenous cell seeding density.After, 50 μL of endothelial medium was added to the same top medium inlet in which the cells were deposited.The OrganoPlate ® was placed with the lid forming a 75-degree angle against the plate stand and left in this orientation for around 3 h to allow cells to attach.After cell attachment, 50 μL of endothelial medium was added into the top medium outlet.The plate was then placed on the Organo-Flow ® , set to an inclination of 7°C and an interval of 8 min, in a humidified incubator.
On day -1, tumor cells or empty medium were seeded into the bottom channel using a different seeding strategy.Tumor donor channel, C R (t) is the fluorescence intensity in the receiver channel at time point t, C(t) is the average concentration of the dye in the system at time point t (Soragni et al., 2023).Estimates of the permeability of the vasculature component alone were obtained by subtracting the permeability of the completely filled chips (stroma with tumor and HUVEC tubule) from the permeability of the partially filled chips (stroma with tumor, but no HUVEC tubules).Comparison to physiological measurement of tumor vasculature utilized published literature (Dewhirst and Secomb, 2017).

Flow cytometry
Cells were plated at 300,000 cells/well in 96-well U-bottom plates (Corning).Plates were spun down (300 x g, 5 min), washed 1x with 200 μL DPBS (Life Technologies), and spun down again (300 x g, 5 min) to remove supernatant.For live/dead staining, live/dead dye was resuspended as per the manufacturer's protocol and diluted 1:100 in PBS.50 μL of diluted live/dead solution was added to plate wells, mixed thoroughly, and incubated at room temperature (RT) for 15 min in the dark.Samples were washed 1x with 150 μL PBS and spun down (300 x g, 5 min) to remove supernatant.For Fc blocking and primary antibody staining, 10 μL of Fc block (Miltenyi) was added to each well and incubated for 10 min in the dark at RT.Then, 90 μL of antibody cocktail (see details in Tab.S6 and S7 1 on antibodies and reagents) prepared in FACS buffer (Becton Dickenson) was added to each well and mixed.Samples were incubated for 30 min at 4°C, wrapped in foil to protect from light.Wells were then washed 1x with 100 μL FACS buffer and 1x with 200 μL FACS buffer.The plate was then spun down (300 x g, 5 min) and the supernatant removed.For sample fixation, 100 μL Cytofix fixation buffer (Becton Dickenson) was added to the wells and incubated for 25 min at RT, wrapped in foil to protect from light.Samples were then washed 1x with 100 μL FACS buffer and 1x with 200 μL FACS buffer, spun down (300 x g, 5 min) and supernatant removed.Samples were resuspended in 250 μL FACS buffer and mixed well.Plates were stored at 4°C until they were read on the cytometer.Staining for compensation controls was conducted on the day of flow analysis as follows.One drop of UltraComp eBeads (eBiosciences) was incubated with 2 μL of the appropriate antibody for 30 min at RT protected from light.For Aqua live/dead dye compensation control, 2 drops ArC beads (Life Technologies) were incubated with 2 μL of live/dead dye for 30 min, at RT, protected from light.After incubation, beads were washed with flow buffer (500 μL), centrifuged (300 x g, 5 min), and resuspended in 400 μL of fresh flow buffer.One drop of ArC negative beads was added to the Aqua tube, and then compensation was run.Flow cytometry was conducted on the LSR Fortessa X-20 (Becton Dickinson), and data were analyzed using Flow-Jo 10.6.2.Please refer to Table S7 1 for more information on reagents.

Immunocytochemistry
Cell cultures in the MIMETAS OrganoPlate ® were fixed in 3.7% formaldehyde (Sigma) after 48 h, 72 h, or 120 h in culture and immunostained as previously described (de Haan et al., 2021).Figures 4-6 and Figures S6-8 1 , imaging was performed either on an EVOS microscope or a GE InCell 6500 high-content confocal imaging system.Two confocal z-stacks were acquired per chip (right and left sides), with the same z-stack size used across each plate.All stacks were converted into single maximum intensity projection images of equal size and used for analysis.Analyses were performed using Im-ageJ or using a custom python script.For analyzing migration distance of T-cells and the number that successfully migrate, a python script was developed which utilized the open-source scikit-image library (van der Walt et al., 2014).This analysis pipeline was run in two stages: to accurately identify the Phase-Guides™ from the brightfield image, and therefore the channel boundaries, and to identify DAPI-stained nuclei.To identify PhaseGuides™, a synthetic image that mapped out the position of the PhaseGuides™ was used as a template to convolve along the image to find the position that looked most similar to the distribution of PhaseGuides™.To increase the accuracy of this approach, the synthetic image was a 1-pixel-width image with intensity bands that are similar to a vertical cross-section of the PhaseGuides™ (as it is 1 pixel wide, this is less affected by rotation).Fast normalized cross correlation was used for template matching, and this led to a processed image with ideally a single horizontal line that had been rotated as per the rotation of the plate.Finding the maximum intensity (and therefore the highest correlation) along the x-axis enabled binarizing the image, and then edge detection was used.The original positions of Phase-Guides™ were then mapped back to this line.Separately, blob detection was used, and the distances from the blobs was measured using a signed distance function (i.e., distances are negative if they are behind the line and positive if they are in front).This meant that channels could be identified just by the sign of the distances.Once the channels had been assigned to each nucleus, it was also possible to count the number of nuclei per chamber.To assist in detecting the PhaseGuides™, illumination correction was performed retrospectively by estimating the illumination profile using a low-pass filter (using a Gaussian kernel with a large sigma) (Dey, 2019).

Barrier integrity assays
The barrier integrity of HUVEC endothelial tubes embedded within the tumor-on-chip platform was evaluated before and after the addition of T-cell compatible assay media as previously described (Trietsch et al., 2017), and the procedure is detailed within the supplementary file 1 (de Haan et al., 2021).Here, the top chip inlets and outlets were perfused with 0.5 mg/mL 20 kDa or 155 kDa FITC Dextran (Sigma).Please refer to Table S4 1 for additional information on equipment and reagents.The apparent intrinsic permeability, or P app , was quantified using the following formula: where V R is the volume in the receiver channel, V D is the volume in the donor channel, A is area of the ECM interface with the channel was comparable to, or less than, that of the no-tumor assay after only 4 hours.Altogether, these diffusion studies support that this assay models leaky tumor vasculature near a diffusion-limiting NSCLC tumor.When comparing the permeability of the vasculature barrier modeled in vitro within these tumor chips to the permeability of leaky tumor vasculature found in vivo, we estimate that the in vitro vascular barriers are roughly one order of magnitude leakier than physiologically relevant permeability estimates (Fig. S1H 1 ).
We note that the initial formation of the HUVEC endothelial tubule requires specialized media containing FBS, an animal-derived product.To make the LToC-Endo amenable to T-cell addition, we adopted an assay medium that was xeno-free once the initial tubule was successfully formed.We explored the impact of an assay medium switch on day 0 and then evaluated the corresponding platform permeability in addition to endothelial cell number and phenotype (Fig. 1B,E,F show images and data using the final assay medium selected; Fig. S1 1 shows all media evaluated).We selected AIM V medium, supplemented with 5 ng/mL rhVEGF (165 isoform) and bFGF, based on its ability to promote markers of endothelial tube stability without appreciably changing barrier diffusion properties and its being FBS-free.This enables a fully xeno-free model, which can both avoid ethical issues around animal products as well as unwanted immunogenicity caused by species mismatch.

Activated but not naïve T-cells migrate in response to chemokine gradients in the tumor chips with 3D endothelium
Next, we used the LToC-Endo model to study the effect of activation status, T-cell seeding density, tumor barrier presence, and chemoattractant type on T-cell chemotaxis.On day 0, we seeded either naïve or activated primary human T-cells into the endothelial channel of the tumor chip, along with recombinant CX-CL11 or CXCL12 in the bottom tumor channels (Fig. 2A).While naïve T-cells did not migrate into the ECM compartment (Fig. S2C 1 ), activated T-cells migrated into the ECM compartment in a seeding density-dependent manner by day 2 in response to both chemokines (Fig. 2B-D).We observed significant differences between chemokine and vehicle control chips (Fig. 2D).
We then evaluated the role of chemoattractant dose and tumor presence on T-cell presence in endothelial tubes and migration.T-cell presence in the endothelial tube did not increase with CXCL12 concentration, although it did increase with time at all doses tested (Fig. S3C 1 ).Meanwhile, all doses of CXCL12 tested, regardless of tumor presence, led to increases in T-cell chemotaxis compared to control chips, with the no-tumor version of the assay leading to greater overall T-cell migration (Fig. 2E,F; Fig. S3A,B 1 ).We especially noticed in no-tumor conditions an elevated baseline level of migration even in the absence of chemokine, compared to the with-tumor assay (Fig. 2E,F; Fig. S3A,B 1 ).One potential disease-relevant explanation for this could be soluble inhibitory factors secreted by the tumor cells.However, another explanation could be asymmetry of media consumption in the no-tumor version of the assay (i.e., endothe-Hoechst 33342 (Thermo Fischer Scientific) was used to stain nuclei.Refer to Table S4 1 for more information on primary and secondary antibodies utilized.

Statistical analysis
Statistical analyses were performed using GraphPad Prism version 8.1.2(332) for Windows, GraphPad Software, San Diego, CA, USA.Data were tested for homogeneity in standard deviations and were square root transformed if needed.Statistically significant differences between means of two or more groups were evaluated using one-way ANOVA (equal variance) or Brown-Forsythe and Welch ANOVA (Gaussian, unequal variance), with multiple comparisons corrected using Dunnett's, Tukey's, or Sidak's tests.Differences were considered significant if p < 0.05.

Tumor barrier limits chemokine diffusion throughout tumor chips with 3D endothelium
We developed a lung tumor-on-chip model with 3D endothelium ("LToC-Endo") in the MIMETAS 3-lane 40 OrganoPlate ® using three human cell types: pooled donor HUVECs, non-small cell lung carcinoma cells (HCC0827), and primary T-cells.First, we established a collagen-1 extracellular matrix barrier.Then, we seeded endothelial cells in the top channel of the OrganoPlate ® against this barrier on day -2 and cultured the chips under rocking flow (Fig. 1A).The following day, we seeded tumor cells, and by day 0, we observed that both the endothelial cells and tumor cells formed tubules in the top and bottom lanes, respectively (Fig. 1B).
To evaluate the diffusivity of both the endothelial and tumor barriers in the LToC-Endo, we performed two different assays.First, using an imaging-based barrier integrity assay, we added 20 kDa fluorescent dextran (approximately the size of chemotactic chemokines) on day 0 into the top endothelial channel, and observed dextran flow through the chips over time with fluorescence microscopy (Fig. 1C; Fig. S1 1 ).By comparing permeability coefficients throughout different chip configurations, we noticed that the 3D endothelial tube readily allowed diffusion and was comparable to no-cell chip controls (Fig. 1C-F; Fig. S1 1 ).We repeated these studies with larger fluorescent dextran (155 kDa) and found similar results (Fig. S1G 1 ).In both cases, the tumor tubule formed a more diffusion-limiting barrier within the device than did endothelial cells on their own, explaining why the combination of barriers is also significantly more diffusion-limiting than the endothelial barrier alone (Fig. 1C-F; Fig. S1G 1 ).This finding was corroborated by an ELISA-based permeability assay, in which CXCL12 chemokine was added into the bottom channel, such that it flowed into the tumor tubule or empty channel and then diffused upward through the chip (Fig. 1G,H).Media sampling over time revealed that in the no-tumor version of the assay, chemokine was detectable in the top channel after only 4 hours and increased markedly by the 48-h time point, with a gradient remaining by this time.In contrast, with a tumor tubule, we could not detect any chemokine in the top channel after 48 hours, at which point the chemokine concentration in the top 2F; Fig. S3A,B 1 ).We note that throughout the assay we did not observe T-cells entering the tumor compartment; only a limited number of T-cells in the no-tumor condition migrated all the way into the bottom compartment of the chips (Fig. 2E).

Presence and activation status of T-cells influence endothelial activation in response to CXCL12
In the LToC-Endo, we observed notable differences in endothelial tube response to CXCL12 depending on the presence of activated T-cells.While CXCL12 drives migration or angiogenic lial cells and T-cells only, and present in the top channel, with a cell-free bottom channel), which may lead to a nutrient gradient that initiates nonspecific T-cell migration even in the absence of recombinant chemokine.What is more, ELISA-based diffusion studies repeated with T-cells support that the no-tumor version of the assay is more permissive to chemokine diffusion at all doses of chemokine tested (Fig. S3D,E 1 ).Thus, more effective chemokine diffusion may also explain why T-cell response saturates at lower doses in the no-tumor assay (37.5 nM CXCL12) compared to the with-tumor version of the assay (150 nM CXCL12) (Fig. through Akt activation via atypical CXCR7 receptors, which are overexpressed only in stressed endothelial cells (Zhang et al., 2017).Under pro-angiogenic signaling, the endothelium responds by increasing endothelial wall permeability, destabilizing the vessel wall, and increasing expression of leukocyte adhesion receptors, in addition to increasing endothelial cell proliferation and migration (Romagnani et al., 2004;Distler et al., 2003;Hunt and Jurd, 1998;Strieter et al., 2005).It is possible that these CXCL12-mediated endothelial events indirectly contribute to the observed window in T-cell migration (Fig. 2B-F), in addition to sprouting of HUVECs with naïve T-cells (Fig. 3A, top row) or when T-cells were absent (Fig. 3B,C), we did not observe pervasive endothelial cell activation when introducing activated T-cells (Fig. 3A, bottom row).CXCL12 is a known driver of T-cell chemotaxis, but it is also a crucial regulator of angiogenesis.It acts by increasing VEGF-A production in endothelial cells, which then upregulates their CXCR4 expression, enhances responsiveness to CXCL12, and contributes to an amplifying angiogenic signaling loop (Salcedo et al., 1999;Staller et al., 2003;Romagnani et al., 2004).CXCL12 also promotes angiogenesis in activation time period is not the source of variation in T-cell response based on prior experience (data not shown here), an additional study would be needed to show this definitively.Under this assumption, these data suggest that implementing a T-cell culture protocol with activation followed by a rest period enables the introduction of T-cells that are more sensitive and responsive into the tumor-on-chip assay.
With respect to endothelial tube responsiveness to CXCL12 chemokine, another key difference emerged when switching from an activated-only to activated-rested T-cell culture protocol.With activated-only T-cells, a lack of response previously shown (Fig. 3A) was reproduced (Fig. 4A) in an independent experiment, using a separate donor's T-cells.In contrast, in activated-rested T-cell chips, we observed endothelial migration in response to CXCL12 by 48 h (Fig. 4C, white arrow).This endothelial response to CXCL12 with activated-rested T-cells appears to match more closely the endothelial response to CXCL12 with the naïve and no-T-cell conditions (Fig. 3).These data lead us to infer that adding rested T-cells minimizes the stress on 3D endothelial tubes caused by the addition of highly proliferative T-cells and may preserve more physiologically relevant responsiveness of the 3D endothelium to angiogenic cues.

Assay timeline extension is facilitated by alternate T-cell activation protocol
Given that an activated-rested T-cell protocol allowed us to circumvent proliferative burst in-chip, preserve 3D endothelium responsiveness to activation, and mitigate live cell dye dilution, we hypothesized that we could extend the assay timeline.Repeating the assay with activated-rested T-cells, we compared day 2 and day 5 numbers of T-cells migrating and number of T-cells within the endothelial tubes.While the activated-only T-cell version of the assay results in a decline in T-cell chemotaxis after day 2 (Fig. 2C) and complete endothelial dissolution by day 5 (Fig. S6 1 ), the activated-rested T-cell version of the assay shows higher levels of T-cell migration and residence within the endothelial tubules, as well as more intact endothelium, by day 5 (Fig. 5 for images and chemotaxis data, Fig. S6 1 for endothelial tube brightfield images).While by day 2 we observe comparable levels of T-cells within endothelial tubules between control and chemokine conditions, by day 5 we observe significantly fewer in the CXCL12 condition.This may reflect that although higher numbers of T-cells are present within the endothelial channel over time in all conditions, a significant number have migrated due to extravasation in the CXCL12 condition (Fig. 5D).
For further assay characterization, we focused on CXCL11, which was determined to be the clearer positive control given the confounding endothelial sprouting introduced by CXCL12.Using CXCL11, we confirmed that with this new T-cell activation strategy and extended timeline, T-cell presence within the endothelial compartment and migration still scales with T-cell seeding density (Fig. S7 1 ), as observed in prior studies (Fig. 2C).We also observed that the absence of a HUVEC tubule diminished migrating T-cells down to nearly zero, significantly below even basal infiltration levels in the presence of HUVEC tubules, but no chemokine (Fig. S7 1 ).These data suggest that although the direct effect of CXCL12 driving T-cell chemotaxis.In contrast, CXCL11 is an angiostatic chemokine, known to counterbalance the vascular changes described above (Romagnani et al., 2004).Therefore, as expected, we did not see angiogenic sprouting in response to this chemokine in the assay (Fig. 3A,B).
The reduction in migration and angiogenic sprouting responsiveness to CXCL12 suggests that the 3D endothelium in the LToC-Endo may be under stress caused by the addition of activated T-cells.Images of the 3D endothelium 3 days after activated T-cell addition show large holes that are suggestive of endothelial stress (Fig. 3A).Abundant T-cell proliferation is suspected to play a role, as both in-chip and off-chip T-cells exhibit an expected, post-activation proliferative burst (Fig. S4 1 ), leading to a higher effective number of T-cells than initially seeded.As a consequence, the rapidly proliferating T-cells may not only be contributing to endothelial stress at later time points, but also diluting live cell dye, all of which may contribute to the plateau or the decline in migrated T-cells after 2 days, which was seen both here (Fig. 2C) and in a previous study (de Haan et al., 2021).

Alternate T-cell activation protocol impacts T-cell phenotype and enhances functional response
We hypothesized that the introduction of a rest period, mimicking the time lag between T-cell activation and homing to a tumor site in vivo (Chen and Mellman, 2013), would allow us to overcome the proliferative burst prior to seeding activated T-cells.Additionally, we switched to a live nuclear dye, which we expected would be stable over longer culture periods.We performed the assay side-by-side with activated or activated-rested T-cells.As expected, the activated-rested T-cells, which undergo proliferative burst during the 2-day rest, increase 3-4-fold in number prior to seeding, compared to the activated-only T-cells (Fig. S4A 1 ).Surprisingly, in spite of the lack of in-chip proliferative burst, more activated-rested T-cells were present in the endothelium and migrated in greater numbers in response to CXCL11 (~4x more on average) and CXCL12 (~2x more on average) compared to activated-only T-cells (Fig. 4A-D).To better understand these changes, we profiled T-cells prepared using both approaches for expression of CXCR3 and CXCR4, the cognate receptors for CXCL11 and CXCL12.We saw that the introduction of a rest period following a T-cell activation enhanced CX-CR3 and CXCR4 expression in all T-cell subsets compared to activation-only T-cells, increasing the overall proportion of double-positive (CXCR3 + CXCR4 + ) T-cells from ~30-50% to ~85% (Fig. S5B 1 ; Tab.S1 1 ).Furthermore, we observed that the rest period led to more central memory (CD45RO -CCR7 + ) and effector memory (CD45RO -CCR7 -) T-cell phenotypes, indicating a more durably activated state (Mahnke et al., 2013) (Fig. S5C 1 ; Tab.S1 1 ).Our incorporation of an additional control in these studies allowed us to attribute changes in T-cell phenotype to differences in activation regimen, rather than culture medium, as this was also changed (Fig. S5B,C 1 ; Tab.S1 1 ).We note that due to logistical constraints of performing this study, the activated-only T-cells were activated for 72 hours while the activated-rested T-cells were activated for 48 hours, prior to resting both populations for 48 hours.While we have made the assumption that this difference T-cell culture protocol and a long-lasting, live nuclear dye can enable a sufficient assay timeline extension to observe significant differences between vehicle and chemokine conditions for a variety of experimental conditions: with and without tumor barriers; HCC087 and NCI-H520 tumor cells; high and low T-cell and tumor cell seeding densities; and multiple T-cell donors (Fig. 5; Fig. S7, S8 1 ).

T-cell chemotaxis in tumor-on-chip requires ICAM-1
Finally, we evaluated the ability of this tumor-on-chip microphysiological system to recapitulate mechanisms of T-cell extravasation.Migration into the tumor microenvironment requires chemokine-induced polarization of T-cells and attachment to the these in vitro HUVEC tubules are leakier than their physiological counterparts, they still play a critical role in T-cell migration, likely by providing endothelial receptors as anchors for T-cell arrest in the rocking platform, in order to initiate transmigration.
Finally, to demonstrate versatility of the LToC-Endo assay, we repeated the T-cell activated-rested protocol using an additional non-small cell lung carcinoma cell line (NCI-H520), using two different T-cell donors (Fig. S8 1 ).We did observe variability between T-cell donors throughout these studies (see Fig. 2F vs. Fig.4B for Donor 1 to Donor 2 comparison, 150 nM CXCL12, activated only; see green boxes in Fig. S8 1 for Donor 2 to Donor 3 comparison, 300 nM CXCL11, activated-rested).However, our studies support that the adoption of an activated-rested after T-cell activation and selecting a long lasting, live nuclear dye.Similar to in vivo, activated T-cells in the LToC-Endo extravasate and migrate in response to chemotactic gradients, and the living endothelial barrier responds to pro-angiogenic cues through sprouting.We have also shown the dependence of T-cell migration on the presence of non-small cell lung carcinoma cells and on ICAM-1 endothelial receptors.Although the portion of T-cells migrating through ECM appears small compared to the total number of T-cells deposited into each chip (15,000 for the majority of studies performed here), we note that the observation portion of the endothelial tubule only depicts a fraction (~7%) of the space, by volume, into which T-cell solution is deposited within the upper chamber of these microfluidic chips.We refer to the diagram shown in Figure 1A to emphasize that the inlet and outlet ports, in addition to the diagonal lanes through which fluids wick in order to reach the observation channel, are not included in imaging or computational analysis.Therefore, if 15,000 T-cells are deposited into the inlet port of the microfluidic chip, and we assume ideal microfluidic wicking and uniform diffusion of the cell solution across the upper chamber of the chip, we would expect only about 1000 T-cells to be present in the observation channel.Indeed, we find that in the control conditions depicted in Figure 6D, at the conclusion of a 5-day experiment, we observe around 1000 T-cells present in the observation portion of the endothelial tubule in both the with-tumor and without-tumor conditions.Furthermore, within this fraction of cells in the observation window of the 3D endothelial tubule, our data (using specific and nonspecific adhesion blocking molecules) suggest that up to half of the T-cells can make sufficient contact with the endothelial cells lining the tubule to achieve arrest on this 3D cell layer.Of the several hundred T-cells that may achieve arrest, it is reasonable to expect up to a few hundred of them to successfully extravasate through the endothelial barrier and migrate through the extracellular matrix barrier.
While animal models typically recapitulate immune cold tumors, the LToC-Endo and described chemotaxis assay can also recapitulate features of immune-excluded tumors (i.e., angiogenesis, immune 3D migration into stroma) (Hegde and Chen, 2020).Given differences in chemokines present and antigen-presenting functions of endothelial cells between human and animal models (Hegde and Chen, 2020), this assay has potential to serve as a valuable tool for probing humanized tumor-immune-endothelial multicellular interactions.Additionally, this in vitro assay simultaneously offers the ability to observe compound efficacy (i.e., T-cell adhesion and migration) with safety (i.e., drug induced vascular injury, exacerbated angiogenesis in the tumor microenvironment (TME)), bringing safety information into the discovery research pipeline earlier.
Similar to what has been shown for an angiogenesis assay using this platform (van Duinen et al., 2020), the next step will be to evaluate the reproducibility and robustness in the LToC-Endo.Establishing a positive control with a clinically meaningful 2-5-fold window, yet without angiogenic side-effects, would be ideal based on prognostic differences between immune phenotypes in cancer tumors (Pagès et al., 2005) and in alignment with robust assay design (Iversen et al., 2006).Further work is need-endothelium through VCAM-1/ICAM integrin activity (Melder et al., 1995;Riegler et al., 2019).Therefore, we repeated this assay using CXCL11 as the chemotactic stimulus and added blocking antibodies against endothelial receptors VCAM-1 and ICAM-1 or isotype controls at the same time as adding chemotactic triggers.
By day 5 of T-cell incorporation into the platform, we observed that ICAM-1 blocking antibody treatment significantly reduced the number of T-cells migrating into the extracellular matrix in response to CXCL11 down to control (i.e., no-chemokine) levels, while VCAM-1 blocking antibody does not (Fig. 6A,B).The median migration distance of T-cells in chips with ICAM-1 blocking antibody is significantly reduced compared to those with isotype control (Fig. 6C).These trends hold in with-tumor and without-tumor versions of the assay (Fig. 6A-C).We observe that the addition of IgG control antibody significantly impacts the number of T-cells adhering to the endothelium, even in the absence of chemokine (Fig. 6D).In the presence of CXCL11, and in the with-tumor assay condition, we observe a significantly lower number of T-cells adhering to the endothelium using ICAM-1 blocking antibody.Altogether, these data suggest that blocking ICAM-1 is sufficient to block chemokine-induced T-cell adherence, extravasation, and chemotaxis.
It is unclear why VCAM-1 blocking did not result in decreased adhesion and chemotaxis.In preclinical animal models, VCAM-1 density and tumor perfusion are predictive of T-cell migration and treatment response to adoptively transferred and endogenous T-cells (Riegler et al., 2019).However, blocking VCAM-1 was only marginally effective at blocking T-cell adhesion to endothelial cells in vivo.In contrast, combined blocking of CD49d/integrin-α4 (a VCAM-1 binding partner), and CD18/ integrin β2 (an ICAM binding partner) offered substantially improved blocking, with this cocktail shown to prevent T-cell mediated tumor rejection (Riegler et al., 2019).

Conclusion and future outlook
In conclusion, we developed a microfluidic lung tumor-on-chip assay with a 3D endothelium (LToC-Endo) perfused with rocking flow to evaluate modulators of T-cell extravasation and migration through 3D extracellular matrix in a non-small cell lung carcinoma (NSCLC) context.Due to the orientation of the platform, T-cell chemotaxis takes place across the x-y plane.This orientation readily facilitates snapshots of T-cell chemotaxis profiles across the stromal matrix, making the assay amenable to phenotypic screening and migration time point analysis.To more wholistically capture T-cell migration strategies, future studies could incorporate tumor spheroids into the bottom compartment of the platform, enabling distinct modes of migration, i.e., directed chemotaxis through 3D ECM in addition to Brownian "hunt and kill" tumor nest infiltration, to be modeled within the same platform (Krummel et al., 2016).
In alignment with a need for future work highlighted previously (de Haan et al., 2021), we extended the assay timeline and improved the assay window by introducing a rest period could replace this animal-derived biomaterial.Alternatively, human material or in situ, tumor-generated stroma may more accurately recapitulate the donor-specific, heterogeneous niche encountered by T-cells.However, we note that the variability of these more complex materials will likely be higher, the sourcing more challenging, and the physicochemical/microfluidic properties of these materials will need to be evaluated.While lung tumors are not generally considered to be amongst the most fibrotic tumors, future adaptation of this system toward the most dense, fibrotic tumor types (i.e., breast and pancreatic) may require much higher density extracellular matrix than what was utilized in this study (Piersma et al., 2020).Regarding stromal ed to validate the translatability of the assay by using standard of care molecules and comparing outcomes to clinical responses (Baran, 2022).Moreover, there is a need to identify the T-cell subtypes that potential therapeutics successfully induce to migrate; in this case, enhancement of CD8 + cytotoxic T-cells would be desirable.
There are many opportunities for adding increased physiological relevance and complexity into this platform, starting with the extracellular matrix.For this initial study, rat-tail collagen, an animal product, was utilized due to its availability, well-characterized physicochemical properties for reliable chip-filling, and batch-to-batch consistency.In the future, synthetic biomaterials timepoint readouts between complex in vitro and in vivo models of immune migration.Noninvasive imaging techniques can detect and monitor anatomical, functional, metabolic, or molecular-level changes within the body of animals with minimal pain, distress, or premature termination (Beckmannand and Ledermann, 2017), and can do so in a temporal and spatial manner.For example, migration and infiltration of specific T-cell populations (e.g., CD8 + ) can be tracked into specific organs, tumors, or tumor-draining lymph nodes over time within a single animal (Rashidian et al., 2017).In this way, noninvasive imaging can enable comprehensive, longitudinal immune response datasets to be derived from fewer animals, thereby increasing the statistical power of the data gathered by reducing experimental variation (Weissleder et al., 2016;Rashidian et al., 2019;Alsaid et al., 2023;Tavaré et al., 2014).This is in contrast to traditional methods requiring animals to be sacrificed at given time points, i.e., using histology and flow cytometry (Zitvogel et al., 2016).Instead of relying exclusively on these informative yet endpoint-requirement techniques, which now include scRNAseq (Rashidian et al., 2019), they could instead be employed as needed to verify or supplement noninvasive longitudinal imaging.Ideally, these noninvasive in vivo approaches would translate to evolving clinical imaging techniques, which are expected to gather similar longitudinal immune migration data, monitor therapeutic response in individual patients, and enable precision oncologic medicine (Weissleder et al., 2016;Rashidian et al., 2017Rashidian et al., , 2019)).
With this translational strategy in mind, an imaging-based, humanized, immune 3D migration complex in vitro model such as the LToC-Endo could be well suited to establish an in vitro/in vivo correlation in the future.Longitudinal, imaging-based T-cell migration datasets, gathered per chip, per animal, and per patient, could then be used to calibrate in silico models, enable better in vivo response prediction, refine the selection of candidates to progress into animal studies, and ultimately provide better medicines to patients.
Mathematical modeling reveals how the density of initial tumor and its distance to parent vessels alter the growth trend of vascular tumors.Microcirculation 27, e12584.doi:10.1111/micc.12584Alsaid, H., Cheng, S. H., Bi, M. et al. (2023).Immuno-PET monitoring of CD8 + T cell infiltration post ICOS agonist antibody treatment alone and in combination with PD-1 blocking antibody using a 89 Zr anti-CD8 + mouse minibody in EMT6 syngeneic tumor mouse.Mol Imaging Biol 25, 528-540.doi:10.1007/s11307-022-01781-7 Baran, S. W., Brown, P. C., Baudy, A. R. et al. (2022).Perspectives on the evaluation and adoption of complex in vitro models in drug development: Workshop with the FDA and the pharmaceutical industry (IQ MPS affiliate).ALTEX 39, 297-314. doi:10.14573/altex.2112203 Beckmann, N. and Ledermann, B. (2017).Noninvasive small rodent imaging: Significance for the 3R principles.In F. Kiessling, thickness, it is expected that the distance between a tumor and parent blood vessel not only varies between patients, but also throughout the course of cancer progression.Mathematical modelling suggests that the distance between a tumor and its parent blood vessel (between 500-2500 µm, as seen in animal models) impacts the growth trends of vascular tumors by changing the spatiotemporal dynamics of angiogenesis (Akbarpour Ghazani et al., 2020).Future studies using this or a similar microfluidic platform could evaluate the role of these distances in vitro as a means to validate this or similar mathematical models and inform treatment strategies for patient-specific tumors.Additional future directions for the LToC-Endo involve incorporating the stromal cells that perpetuate immune suppression in the TME, such as cancer-associated fibroblasts, myeloid-derived suppressor cells, and tumor-associated macrophages (Mariathasan et al., 2018;Joyce and Fearon, 2015;Bremnes et al., 2011).It will also be important to evaluate how other immune cell types (i.e., T-regulatory cells, natural killer cells, and B-cells (van der Woude et al., 2017)) migrate into the TME in response to chemotactic cues and compounds, and to include a tumor cell killing component into the assay.After additional characterization and validation, our hope is that the LToC-Endo complex in vitro model can serve as a valuable tool to study multicellular and cell-extracellular matrix mechanisms of immune suppression, and screen for drug candidates that target these processes to improve patient responses to immunotherapies.
We note that the initial culture of the HUVEC endothelial tubule in this assay is not animal-free, as it requires media supplemented with FBS.Recent studies suggest that human platelet lysate could serve as an animal-free replacement for FBS in endothelial cell culture (Peters et al., 2022).This should be additionally investigated in a 3D microfluidic platform such as the one described here, where functional assessment (i.e., barrier integrity) could be evaluated in addition to features such as morphology, proliferation, and apoptosis.Furthermore, given the estimate that the tumor vasculature modeled here is about one order of magnitude leakier than tumor vasculature in vivo (Fig. S1H 1 and Dewhirst and Secomb, 2017), and that these measurements are somewhat noisy, it would be beneficial to adopt new developments in media optimization and barrier permeability measurement techniques (Ehlers et al., 2023), and consider automated methods of filling these multi-component tumor chips, to both reduce the variability and improve the resolution of these measurements while reducing the permeability of the in vitro vasculature to more closely mimic vasculature in vivo.
For this complex in vitro model to support the replacement, refinement, and reduction of animal immuno-oncology models, whether classical syngeneic (i.e., MC38, 4T1) or "humanized" mouse tumor models (Zitvogel et al., 2016), further validation studies are needed which incorporate additional human and animal tumor cell lines and primary, single-donor, tissue-specific endothelial cells (compared to pooled donor HUVECs), which would enable consideration of HLA-matching in selection of T-cells.As these studies should be guided by a translational strategy, we propose that noninvasive imaging techniques serve as a translational link to align imaging-based pharmacodynamic (PD) 1 , T-cells were obtained directly from AllCells and shipped to the MIMETAS research facility.For

Fig. 1 :
Fig. 1: T-cell chemotaxis assay development, medium evaluation, and barrier analysis in the LToC-Endo model (A) Experimental setup and timeline of platform seeding with collagen-1, endothelial cells, and tumor cells.(B) Representative images of the platform seeded in monoculture and coculture configurations, with endothelial medium and triculture assay medium, on day 0. The transition from endothelial medium to triculture assay medium occurs on day 0 to mimic T-cell seeding at that time point.Refer to Fig. S1 1 for information on the different assay media formulations considered.(C,E) Barrier integrity assay fluorescence images, where white shows 20 kDa FITC dextran presence.(D,F) Permeability coefficient measurements for different configurations of the assay, for day 0 and day 2, respectively.Data show measurements per chip for n = 2, or n > 2 for those used for statistical testing, within-experiment, technical replicate chips per condition.Bars indicate means, and error bars indicate SD.Data was square root transformed prior to statistical testing to account for unequal SD.Outcomes are indicated for statistical tests comparing barrier diffusivity among the conditions tested (One-way ANOVA, ***, p < 0.001; ****, p < 0.0001).(G,H) ELISA data of CXCL12 concentration in the bottom and top channels 48 h after 150 nM CXCL12 was first introduced into the bottom channels of the chips.Data show measurements per chip for n = 2 chips per condition.The bars represent means, and error bars indicate SD.The scale bars in (B), (C) and (E) are 100 µm.Please refer to Fig. S1G 1 for permeability measurements utilizing 155 kDa FITC dextran and other data obtained within the same study as those in panels (B) through (F).

Fig. 2 :
Fig. 2: T-cell seeding density, chemokine type and dose, and tumor barrier presence regulate activated T-cell migration in the LToC-Endo model (A) Experimental setup and timeline with platform seeding of extracellular matrix, endothelial cells, tumor cells, and activated T-cells.(B) Representative phase contrast and fluorescence images of T-cell migration through the collagen barrier of the tumor-on-chip, in response to chemokines CXCL11, CXCL12, and vehicle controls.Images depict data using a T-cell seeding density of 15,000 cells/chip.(C) Number of migrated T-cells, by seeding density and over time, in CXCL11 and CXCL12 and vehicle controls, with data points indicating means of n = 3 chips and error bars indicating SD. 48-hour time point data, from black boxes in (C), are highlighted in (D), with bars indicating means of n = 3 within-experiment technical replicate chips per condition and error bars indicating SD.Results shown for Welch's t-tests to accommodate unequal variances (**, p < 0.01; ***, p < 0.001).(E) Representative fluorescent images of T-cell migration, for chips by dose of CXCL12 chemokine, with and without tumor barriers, at the day 2 time point for two independent experiments evaluating dose-response to CXCL12 chemokine.(F) Number of migrated T-cells by CXCL12 dose, with and without tumor barriers, at day 1 and day 2 time points.Day 1 data for tumor and no tumor conditions overlapping.Markers indicate means of n = 6-7 chips per condition, and error bars indicate SEM.Significant differences between CXCL12 dosages and respective vehicle controls are shown (Brown-Forsythe and Welch ANOVA tests, corrected for multiple comparisons, *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001).Scale bars in (B) and (E) are 100 µm.Panels (B) through (D) show data from a single experiment, whereas panels (E) and (F) show representative images and data pooled from two other independent, biological repeat experiments, performed in the same lab, on different days (please refer to Fig. S3A, B 1 for data shown separately).Images and data in green boxes (B, D, E, F) show the same condition amongst the first study and combined second and third studies.Please refer to Fig. S2 1 and Fig. S4 1 for data obtained within the same experiment as the data shown in panels (B) through (D); refer to Fig. S3 1 for additional data obtained within the same experiments as shown in (E) and (F).These two studies were conducted using Donor 1 T-cells.

Fig. 3 :
Fig. 3: Migration and sprouting of 3D endothelium in response to CXCL12 in the LToC-Endo model (A) Hoechst and CD31 staining of the indicated conditions, for naïve and activated T-cells, low and high T-cell seeding density, and CXCL11 and CXCL12 chemokines, on day 3.With no T-cells in the chips, (B) CD31 staining depicting 3D endothelium response to control, CXCL12, or CXCL11 conditions after 3 days in culture and (C) brightfield images showing endothelial response to CXCL12 or media control, with and without tumor cells, after 3 days in culture.In (A) through (C), middle channel width is 350 µm as indicated by the vertical bars.Panel (A) was obtained in the same experiment as shown in Fig. 2B-D and Fig. S2 1 ; panels (B) and (C) were obtained in the same experiment as shown in Fig. 2E,F and Fig. S3 1 .These two studies were conducted with Donor 1 T-cells.

Fig
Fig. 4: Activated-rested T-cell protocol enhances T-cell adhesion and chemotaxis, and restores CXCL12-driven endothelial activation, in the LToC-Endo Representative brightfield and fluorescent images of T-cells (15k per chip) within the endothelial tubule and migration into the ECM compartment in response to the chemokine and dose indicated, at day 2, for (A) activated-only T-cells (AIMV) and (C) activated-rested T-cells (RPMI).White arrow indicates evidence of endothelial sprouting.Refer to Fig. S5 1 for flow cytometry data from this experiment that includes an additional activated-only (RPMI) condition.Scale bars in (A) and (C) are 100 µm.In (B) and (D), quantification of migrated T-cells and T-cells within the endothelial tube for both T-cell preparation protocols, respectively.Markers indicate mean T-cell numbers per chip (n = 4 same-donor technical replicates per condition), bars indicate mean T-cell numbers per condition, and error bars indicate SD.Statistical testing was performed on square root transformed data to satisfy criteria of equal SDs.Significant differences between chemokines and respective vehicle-alone controls are shown (one-way ANOVA corrected for multiple comparisons, *, p < 0.05; ***, p < 0.0001).This study was conducted with Donor 2 T-cells.

Fig
Fig. 5: Activated-rested T-cells enable an extended assay endpoint (A,B) Representative brightfield (BF) and fluorescent images of T-cell migration in response to the indicated chemokines and doses, on days 2 and 5, for assay with 15,000 T-cells seeded and with tumor barrier.Scale bars are 100 µm.(C) Quantifications of migrated T-cells and (D) T-cells within the endothelial tubes on days 2 and 5.In (C), statistical testing was performed on square root transformed data to satisfy criteria of equal SD.In (C) and (D), significant differences between chemokines and respective vehicle controls are shown (one-way ANOVA corrected for multiple comparisons, *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001).Please refer to Fig. S6 1 for bright field images of endothelial tubules obtained from this experiment.This study was conducted with Donor 2 T-cells.

Fig. 6 :
Fig. 6: T-cell extravasation and chemotaxis in response to CXCL11 are dependent on ICAM-1 endothelial receptor in the tumoron-chip platform (A) Representative fluorescent images of T-cell migration in response to chemokine or vehicle control, with additional treatment as indicated with blocking antibody or IgG control.Images show day 5 assay data both without and with tumor barrier.Scale bar is 100 µm.Day 5 quantifications of (B) mean migrated T-cell number, (C) median migrated distance, and (D) mean T-cell number within the endothelial tubes.In (B-D), markers indicate metrics per chip (n = 4 within-experiment, technical replicates per condition), whereas bars indicate means per condition, across all technical replicates, and error bars indicate SD.Significant differences between chemokines and respective vehicle-alone controls, with or without antibody treatments, are shown (one-way ANOVA corrected for multiple comparisons.*, p < 0.05; **, p < 0.01; ***, p < 0.001; **** p < 0.0001).In (B), statistical testing was performed on square root transformed data to satisfy criteria of equal SDs.This study was conducted with Donor 2 T-cells.