3D Microfluidic model for evaluating immunotherapy efficacy by tracking dendritic cell behaviour toward tumor cells

Immunotherapy efficacy relies on the crosstalk within the tumor microenvironment between cancer and dendritic cells (DCs) resulting in the induction of a potent and effective antitumor response. DCs have the specific role of recognizing cancer cells, taking up tumor antigens (Ags) and then migrating to lymph nodes for Ag (cross)-presentation to naïve T cells. Interferon-α-conditioned DCs (IFN-DCs) exhibit marked phagocytic activity and the special ability of inducing Ag-specific T-cell response. Here, we have developed a novel microfluidic platform recreating tightly interconnected cancer and immune systems with specific 3D environmental properties, for tracking human DC behaviour toward tumor cells. By combining our microfluidic platform with advanced microscopy and a revised cell tracking analysis algorithm, it was possible to evaluate the guided efficient motion of IFN-DCs toward drug-treated cancer cells and the succeeding phagocytosis events. Overall, this platform allowed the dissection of IFN-DC-cancer cell interactions within 3D tumor spaces, with the discovery of major underlying factors such as CXCR4 involvement and underscored its potential as an innovative tool to assess the efficacy of immunotherapeutic approaches.

deacetylase inhibitor (HDACi) romidepsin (R) represents an efficacious antitumor treatment with a high potential to induce immunogenic apoptosis of colorectal cancer (CRC) cells 8 . Upon phagocytosis of dying cancer cells, DCs fulfil their primary role by processing and presenting tumor Ags to CD4 + T helper cells, while some subsets of DCs possess the capability to cross-present tumor Ags to CD8 + T cells, and thus stimulate the effector cells of the antitumor response 9 . These peculiar DC functions evoking antitumor immunity have been exploited in several DC-based therapeutic approaches. In our laboratory, we developed IFN-α-conditioned DCs (IFN-DCs) as promising candidates for therapeutic cancer vaccines 10 . These cells possess superior properties in Ag uptake and induction of both CD4 + T helper lymphocytes and CD8 + cytotoxic T cells and resemble naturally occurring DCs [11][12][13] . It is important to note that the superior functional activities of IFN-DCs, as well as the rapid acquisition of their potent migratory ability, may also depend on the expression of chemokine receptors 14 . Data from a pilot clinical study indicate that in patients with advanced melanoma, intratumoral injection of IFN-DCs after dacarbazine treatment activates antitumor immunity confirming the high capability of these cells to fulfil their functions upon Ag release in situ 15 .
Microfluidic models represent a new frontier for studying the complex interactions occurring within organ microenvironments in both physiologic and pathologic conditions 16,17 . The recent advances in both cell biology and microengineering have allowed the development of the innovative organ-on-chip approach, which is able to recapitulate in vivo biological microenvironments suitable for studying complex functions, such as cell-cell interactions and dynamic drug stimuli 18,19 . This enormous potential relies first on the recreation of complex 3D spaces characterized by both physical and biochemical cues closely mimicking the in vivo microenvironments 20 . Importantly, microfluidic platforms are able to reproduce cell confinement, a parameter imposed on cell movement in the interstitial space of tissues, which is totally absent in 2D assays. This confinement is essential for studying the behaviour of motile cells such as immune and cancer cells 21 . The coordinated integration of a microfluidic assay, advanced microscopy and computational modelling enables the observation of single events as part of the complex biological processes ultimately leading to define the physiopathological responses 22,23 . These breakthrough innovations have allowed the study of cancer-immune interactions as well as immunotherapeutic treatments using microfluidic platforms 24 . In oncology, microfluidic models have been widely used to study the metastatic potential of cancer cells 25,26 . In the past few years, our group exploited the microfluidic approach to investigate in real time the interactions between immune and cancer cells occurring during an IRF-8-deficient antitumor immune response 27,28 . This approach provides a new method to investigate these events also under therapeutic treatments 29 . However, one major challenge is the proper reconstruction of tumor and immune systems, two different microenvironments closely interconnected.
Here, we reconstituted 3D spaces mimicking cancer and immune systems suitable to investigate the physicaland biochemical-driven interactions among these cell components. Specifically, we monitored the behaviour of IFN-DCs toward CRC cells, untreated or exposed to the innovative antitumor combined treatment with R and I (RI). We found that IFN-DCs moved through 3D immune and tumor spaces with a driven trajectory toward RI-treated cancer cells rather than untreated (NT) counterpart. Noteworthy, during dynamic migration, largely guided by the CXCR4/CCL12 axis, IFN-DCs were found to modify their motion in order to interact with drug-treated cancer cells and then to take up tumor Ags.

Results
A 3D microfluidic platform enables the evaluation of dynamic IFN-DC migration toward SW620 cancer cells exposed to combined treatment with IFN-α and romidepsin. The success of IFN-DC-based immunotherapy relies on the ability of these cells to accomplish Ag uptake by phagocytosis of dying cells in the TME and then to migrate to lymph nodes where they induce T cell-mediated antitumor responses 15, 30 . To investigate the interactions between IFN-DCs and SW620 CRC cells, we built a microfluidic platform suitable to recreate interconnected 3D spaces. To this end, we first generated primary IFN-DCs by culturing peripheral monocytes from healthy donors for 3 days with I and GM-CSF. In addition to the high expression of the co-stimulatory molecules CD80 and CD86, IFN-DCs were found to display the plasmacytoid CD123 marker as well as the type 2 myeloid DC marker BDCA3, which are involved in the cross-presentation of Ags derived from pathogens and dead cells 31,32 (Fig. 1a). Based on our previous data showing the ability of RI to promote DC recognition of cancer cells 8 , we then evaluated the phagocytic capacity of IFN-DCs toward SW620 cells pre-exposed to RI (RI SW620) for 48 h. We found that RI treatment markedly induced both early and late apoptosis in SW620 cells that in turn were phagocytosed at a higher rate by the IFN-DCs as compared to untreated cancer cells (NT SW620) (Fig. 1b,c). Therefore, in order to study the behaviour of IFN-DCs toward NT and RI SW620 we devised a novel strategy combining the design of a hydrogel-based microchannel platform with advanced microscopy as well as an unsupervised image processing algorithm. This approach allowed us to perform, on a single-cell basis, real time monitoring and analysis of DC migration as well as to evaluate the interactions occurring between IFN-DCs and SW620 cells. Figure 2a shows the microfluidic platform, containing a central immune culture compartment (immune-chamber) where floating IFN-DCs were uniformly distributed and showed non-adhesive motion, resembling the peripheral immune system. Two-side narrow chambers (tumor-chambers) were designed to establish 3D tumor microenvironments by embedding with a type I collagen matrix, where NT or drug-treated SW620 cells were seeded through two loading reservoirs, thus establishing 3D-NT and 3D-RI tumor spaces, respectively. On the basis of the already reported unique suitability of type I collagen for reproducing DC behaviour in the extracellular matrix 33 , it was used to build tumor-chambers re-creating proper biochemical and biophysical factors guiding DC migration and function. Each tumor-chamber, flanked on the external side by a parallel wide channel deputed to perfuse complete medium to the cultures, was connected to the immune-chamber via a set of short and narrow capillary migration channels (connecting-channels) (Fig. 2a,b, Supplementary  Fig. 1). This geometry allowed to analyse DCs that move under confinement performing distinct processes such as migration, phagocytosis and cell-cell interactions, as it occurs in vivo during extravasation into peripheral tissues 34,35 . For this purpose, two experimental configurations were used: (i) a no-competition condition, where the left tumor-chambers of two distinct platforms were functionalized with collagen without tumor cells (matrix) while collagen-embedded NT or RI-treated SW620 cells were cultured in the right tumor-chambers (Fig. 2c, left panel); (ii) a competition condition, where collagen-embedded NT SW620 and RI-treated SW620 cells (hereafter NT and RI) were placed in the left and in the right tumor-chambers, respectively, of the same platform (Fig. 2c, right panel). Such design provided the opportunity to re-create a microsystem characterized by interconnected tumor and immune compartments, yet retaining specific environmental properties. This model allowed to monitor the motility of IFN-DCs (migrating IFN-DCs) through the immune-chamber, connecting-channels, and tumor-chamber, in the presence of either non-competitive signals (no-competition) or biochemical stimuli (competition) and also to evaluate the overall recruitment of IFN-DCs into the 3D tumor space (infiltrated IFN-DCs) and their interactions with SW620 cells (Fig. 2d).  showing the immune-chamber filled with PKH67 green-stained DCs, the connecting-channels and the tumor-chamber loaded with type I collagen embedded-PKH26 red-stained cancer cells. Images were acquired with an inverted Olympus IX73 microscope using a Plan Achromat 4×/0.10NA objective and processed with ImageJ software. Scale bar 100 μm. The inset photographs of the 3D projections represent: (1) a 3D rendering of a confocal image of SW620 cancer cells in the reconstructed 3D-matrix tumor space. Scale bar, 100 μm; (2) a 3D rendering of a confocal image of the connecting-channels entering the 3D-matrix tumor space showing IFN-DCs moving toward cancer cells; scale bar, 100 μm. Image stacks (100 images of 1 μm Z-step size) were acquired using a spinning disk microscope with a 20×/0.45NA objective and processed with Imaris software; Preferential IFN-DC migration toward the RI tumor environment. To explore the intrinsic propensity of IFN-DCs to migrate actively toward NT as well as RI-treated cancer cells, the microfluidic platform was used first under the no-competition condition. NT and RI tumor spaces were reconstituted into the right tumor-chamber of different microfluidic platforms with PKH26-stained cancer cells, depicted as red cells. Upon collagen polymerization, PKH67-stained IFN-DCs, visible as green-cells, were loaded and uniformly distributed into the immune-chamber. The capability of IFN-DCs to migrate toward the chambers filled with cancer cells rather than matrix alone was evaluated over a period of 24 to 48 h by wide-field imaging analysis of the entire microfluidic device, and the total number of IFN-DCs within the micro-channels and tumor-chambers was quantified. We found that, while IFN-DCs did not recognize the matrix, they actively moved from the immune-chamber toward both tumor-chambers containing NT and RI-treated SW620 cells through the connecting-channels (Fig. 3a). Consistently, the number of infiltrated IFN-DCs into the 3D-tumor spaces was significantly higher than that infiltrated into the matrix. Notably, at both time points we found that the number of IFN-DCs infiltrated into the RI space was significantly higher compared to that into the NT space (Fig. 3b), and a remarkable fold increase of infiltrated IFN-DCs, with respect to matrix, was observed in the RI space (Fig. 3c). Of interest, IFN-DCs in both cases were able to migrate upward and migrate deep within the 3D-tumor space (Fig. 3d).
To further characterize the propensity of IFN-DCs to migrate preferentially toward drug-treated cancer cells, we performed experiments under the competition condition. The behaviour of IFN-DCs was evaluated over three time periods: 0-24 h, 24-48 h and 48-72 h. Wide-field imaging analysis showed that over the entire follow-up period IFN-DCs moved preferentially toward the RI-treated cancer cells, resulting in a constant higher number of infiltrating IFN-DCs into the RI tumor space as compared to that into the NT tumor space (Fig. 4a Table 1, Supplementary Fig. 4a). Software results were validated by manual ImageJ analysis ( Supplementary Fig. 4b). We found much greater migration of IFN-DCs toward RI relative to that toward NT counterpart overtime, with the tendency for the RI-induced migration to reduce gradually ( Fig. 5a,b). Importantly, IFN-DCs localized in the immune-chamber next to the RI space, where they can sense tumor-specific stimuli, moved with a higher constant speed as compared with those cells positioned in the corresponding area next to the NT space  Table 1). Of note, real time imaging revealed that IFN-DCs crowded the entrance and showed promiscuous crossing through the connecting-channels next to the RI space, in contrast the IFN-DCs moved toward the NT space by random walking (Supplementary Movie 3). Of interest, once in the RI space, IFN-DCs significantly reduced their intrinsic movement during a 0-48 h period, in contrast to the cells that migrated into the NT space that exhibited a slight speed reduction up to 24 h but increased the migratory speed over 48-72 h (0-24 h: 3.4 μm/min vs 3.0 μm/min; 24-48 h: 4.98 μm/min vs 7.7 μm/min; 48-72 h: 5.8 μm/min vs 4.9 μm/min) (Fig. 5d,e). These findings suggest that IFN-DCs are endowed with high capability to adapt their migration in response to external signals 36,37 . Moreover, migrating IFN-DCs were found to move in remarkably higher number toward RI, characterized by higher DP, relative to the migration toward NT overtime (0-24 h: 571 μm vs 480 μm, 24-48 h: 576 μm vs 474 μm, 48-72 h: 550 μm vs 441 μm) (Fig. 5f,g). Next, we characterized IFN-DC and cancer cell interactions and found that over 24-48 h single IFN-DCs exhibited a longer average interaction time with single cancer cells in RI relative to that in NT space (0-24 h: 16.1 min vs 14.9 min, 24-48 h: 11.5 min vs 8.7 min, 48-72 h: 11.6 min vs 12 min) (Fig. 5h).
The CXCR4/CCL12 axis guides IFN-DC movement toward the RI space to capture apoptotic cancer cells. DC migration within the TME and toward lymph nodes is driven by a fine-tuned regulation of chemokine and chemokine-receptor expression 38 . Indeed, the IFN-DC maturation stage and functional activities have been reported to be regulated by responsiveness to chemokines 14 . In this paper, we further investigated the role of chemokines expressed by RI-treated SW620 cells in regulating the movement of IFN-DCs. RI-treated SW620 cells were found to express higher levels of CCL2, CCL4, CCL20, CCL21 as well as CXCL1, CXCL8, CXCL9, CXCL10, CXCL12 and CXCL14 mRNA relative to their NT counterpart (Fig. 6a). Accordingly, IFN-DCs exhibited elevated expression of corresponding chemokine receptor transcripts, namely CCR1, CCR2, CCR5, CCR6 and CCR7 as well as CXCR2, CXCR3, and CXCR4 (Fig. 6b). Next, a key role of the CXCR4/CXCL12 axis in the guided movement of IFN-DCs toward the RI space was defined. In fact, after SW620 cells were treated with RI for 48 h, they produce CXCL12 in significant amounts (Fig. 6c). Moreover, IFN-DCs exposed to conditioned medium (CM) from RI-treated SW620, selectively increased CXCR4 expression on the membrane compared to that of cells exposed to CM from NT cancer cells (Fig. 6d). Therefore, we performed migration experiments in the microfluidic device in the presence of the CXCR4 inhibitor AMD3100. We found that while IFN-DC migration toward NT was not affected by CXCR4 inhibition, the IFN-DC migration toward the RI space was significantly reduced by AMD3100 already by 4 h, persisting for up to 24 h ( Fig. 6e and Supplementary Fig. 5a). The inhibition of CXCR4/CXCL12-driven IFN-DC migration by AMD3100 was confirmed in a Transwell migration assay ( Supplementary Fig. 5b). It was previously shown that DC engulfment of damaged cancer cells following their scanning of the TME is pivotal for inducing antitumor immunity 35 . Consistently, we observed that IFN-DCs were able to more efficiently phagocytose RI SW620 cells compared to NT counterpart (Fig. 7a). IFN-DCs could migrate in three directions within the RI space, confirming the add value of our model in investigating the guided movement of DCs (Supplementary Movie 4). In fact, we traced the migration of a single IFN-DC toward a single cancer cell in the RI space and thus highlighted the events of phagocytosis (Supplementary Movie 5). In this process, IFN-DC moved slowly (Fig 5d,e), extended trans-cellular dendrites to establish contact with cancer cells (Fig. 7b), and finally phagocytosed the cancer cell (Fig. 7c). We also found some tumor Ag-loaded IFN-DCs migrating back into the connecting-channels (Fig. 7d), suggesting the occurrence of a second moving phase after phagocytosis resembling DC migration in draining lymph nodes after capture of tumor antigens 39 .

Discussion
In this paper, we exploited a novel microfluidic platform recreating interconnected 3D immune and tumor systems, to study the behaviour of IFN-DCs toward cancer cells upon treatment with two complementary antitumor agents, the epigenetic drug romidepsin and the immunomodulator IFN-α 8 . We identified the processes undertaken by IFN-DCs for approaching dying cancer cells for Ag uptake. In particular, the following events were monitored and evaluated in real time: (i) IFN-DC scanning of environmental signals and adjustment of migratory capacity toward RI-treated cancer cells; (ii) modulation of movement parameters by IFN-DCs in performing guided migration in a confined space; (iii) establishment of IFN-DC-cancer cell interactions leading to Ag uptake.
DC-based immunotherapy is a promising tool for cancer treatment currently under intense investigation for its unique potential to initiate and direct Ag-specific T cell immune responses 40 . Furthermore, improved clinical protocols suggest that the combination of DC-based therapies with other agents may enhance the potency of the antitumor effects 41 . In this landscape, IFN-DCs are promising candidates to be combined with agents that promote cancer cell death and favour Ag uptake 42 . Here, we tested this idea and found that a high rate of CRC cell apoptosis following RI treatment resulted in a significant increase of phagocytosis by IFN-DCs. The choice to treat cells with RI in combination relies on the complementary antitumor properties of these drugs. While the direct and immune-mediated antitumor properties of IFN have recently renewed interest in the use of this cytokine in new therapeutic protocols 43 , HDACi have been reported to induce both cancer cell death and immunomodulatory effects 44 . Accordingly, we found that the RI combination was able to exert direct anti-tumor effects and to drive DC functions against cancer cells. Importantly, although these events represent pivotal markers of the efficacy of immunotherapy their detection remains a true challenge 45 . Therefore, in the attempt to overcome this limitation, we employed a microfluidic approach for dissecting the intimate dialogue between IFN-DCs and RI-treated SW620 cells. In the past few years, the rapid technological advances have led to the development of new platforms for monitoring in real time the events linked to cancer development 46 . In order to examine the dialogue between cancer and immune cells, we previously employed a simplified microfluidic device. Spleen cells from mice knocked out for IRF-8, compartmentalized across a network of microchannels, approached cancer cells only by random movements, whereas single immunocompetent splenocytes accomplished drifted random migration leading to a highly coordinated response 27,28 . Importantly, these data agreed with in vivo functional results 47 . Herein, the new design of the microfluidic platform allowed building of immune-tumor 3D spaces similar to the physiological environment characterized by simultaneous compartmentalization as well as by interconnections of the tumor and immune systems. Both the physical device architecture and the type of 3D matrix, that regulate the phenotype and functions of cells 35,48 , were taken into consideration for reproducing spaces mimicking in vivo environments where DCs and cancer cells migrate and interact. First, immune-chamber dimensions recreated a spatial confinement, mimicking vessels and tissue interstitial areas, in which DCs, under non-specific adhesive conditions, were able to tune guided migration upon sensing factors released by tumor cells 49 . In addition, the geometry of the connecting-channels allowed DCs to migrate actively through very narrow constrictions to the same extent as when they cross endothelial barriers 50 . On the other hand, the microarchitecture, porosity and stiffness of type I collagen conferred tumor spaces with the ability to release biochemical signals from dying cells to guide and adapt DC movements 51 . Therefore, our microfluidic platform was able to recapitulate physical and biochemical environmental cues driving the interactions between DCs and tumor cells, closely resembling in vivo conditions in a time-and spatial-dependent manner. Complementary events, undetectable on 2D flat surfaces used in traditional assays 49 , such as IFN-DC migration and interaction with single cancer cell, were linked in real time using high-resolution time-lapse imaging. Moreover, different parameters characterizing IFN-DC behaviour toward NT and RI-treated cancer cells were evaluated with an automated data analysis system using a revised image analysis algorithm thus overcoming for the first time one of the most important challenges of microfluidic studies, i.e. time-consuming manual data analysis 52 . Specifically, IFN-DCs when sensing stimuli exclusively from either NT or RI-treated SW620 cells displayed a propensity to be recruited into the tumor mass but not in the competing matrix space. In this scenario, IFN-DC migration toward the RI space was significantly higher than that of cells recruited toward NT space. However, the elevated attractant potential of the molecules released by RI-treated dying cells toward IFN-DCs was clearly evident under competition conditions with the massive recruitment of these cells into RI tumor space relative to the NT counterpart. In this case, migrating IFN-DCs adjusted their migratory properties with respect to their functional activity 49    Scientific RepoRts | 7: 1093 | DOI:10.1038/s41598-017-01013-x they acquired higher pre-channel speed generating crowding in entrance at connecting-channels where they acquired an "amoeboid-like" locomotion mode, with dendritic protrusions at the cell front and contractions at the cell rear, propelling the nucleus through narrow spaces thus allowing persistent cell migration 53 . Once in the RI space, IFN-DCs significantly reduced their velocity, in agreement with Ag internalization along with a concomitant decrease in migration upon maturation stimuli 54 . IFN-DCs migrating toward the RI tumor space had longer displacements with high DP, in contrast to the migration toward the NT space. In line with the evidence that chemokines may drive DC behaviour 38 , biochemical signals released in the RI tumor space drove IFN-DC migration. Enhanced transcript expression of several chemokines by RI-treated tumor cells was observed along with high levels of the corresponding chemokine receptors on the IFN-DCs. Such chemokines drive DC recruitment into inflamed sites 55 . Of note, elevated CXCL12 production by RI-treated dying cells agreed with the increased expression of CXCR4 on IFN-DCs, a receptor known to guide mature DC motility 56 , thus contributing to IFN-DC migration toward RI tumor space. The key role of the CXCR4/CCL12 axis in driving IFN-DC chemotaxis toward the RI space was further confirmed by the functional inhibition with the specific CXCR4 inhibitor. This finding is consistent with the report showing that CXCR4 as well as the cooperation between both inducible CXCR3 ligands and constitutive CXCL12 regulate DC trafficking to peripheral sites 57 . On the contrary, the high CCR7 expression on IFN-DCs did not associate with the induction of CCL21 transcripts on cancer cells, suggesting a minor role of the CCR7/CCL21 axis for the movement and function of IFN-DCs under these conditions.
The further analysis of DC-cancer cell interactions revealed a high competence of IFN-DCs to take up Ags by exploring the RI tumor space. In fact, the time of interaction between one IFN-DC and one SW620 cell was higher in the RI space than that in the NT space over 0-48 h period, whereas it decreased along with the augmented migration velocity during the subsequent 48-72 h period. These data support the superior competence of IFN-DCs in patrolling drug-treated space with alternating migration phases characterized by pausing times for endocytosis 12 . IFN-DCs also established communications with cancer cells throughout trans-cellular dendrites when the velocity was reduced, suggesting the capability of these cells to modulate multiple functions. Finally, we also report that Ag-loaded IFN-DCs may make a directional choice toward the connecting-channels pointing out the potential of our platform for studying DC and T lymphocyte interactions, after Ag uptake 58 .
In summary, the microfluidic model developed in this study establishes the proof of concept for real time tracking of immune-tumor cell interactions and validates the efficacy of combination strategies in potentiating DC-based immunotherapy. Beyond this unquestionable advantage, this platform offers a valid and manageable alternative to studies in animals for evaluating the efficacy of novel immunotherapeutic treatments while its further implementation may lead to the development of suitable assays for pre-clinical testing.

Methods
Cell Culture. IFN-DCs were generated using a standard protocol previously described 59  Fabrication of the microfluidic devices. 3D co-culture experiments were performed in an ad hoc-designed cell-culture microfluidic device fabricated at the CNR-IFN facility according to already published protocols 60 . PDMS replica and microscope glass slides were plasma oxidized (Oxford Plasma Lab 80 plus, RF Power: 20 W, Flux: 60 sccm, Pressure: 800 mtorr, Time: 30 s) and bonded together at 90 °C for 4 h on a hotplate. Selective hydrophilic activation of chamber walls by O 2 plasma treatment was needed to guarantee the correct gel matrix positioning and the proper filling of connecting-channels before biological experiments. Microfluidic co-culture experiments were performed under static conditions, with passive diffusion of medium, oxygen and all molecular components.
Cell loading and culture in the microfluidic platform. Devices were sterilized under UV light in a laminar flow hood for 15 min before loading cells. Rat-tail type I collagen (Corning, MA, USA) was prepared at 3 mg/ ml, according to the manufacturer's instructions, and the pH was adjusted to 7 with 1 N NaOH. Collagen was kept for 30 min on ice before adding 25 × 10 4 PKH-26 labelled-NT or RI-treated SW620 cells. Then, 3.5 μl of cell/ collagen mixtures were inoculated into the tumor-chambers. In some experiments, one of the tumor-chambers was filled with only collagen, without tumor cells. Gels were allowed to polymerize at 37 °C for 30 min in a tissue culture incubator (5% CO 2 ). Devices were filled with complete medium and stabilized at 37 °C for 30 min. Then, 5 × 10 5 PKH-67 labelled-IFN-DCs were inoculated, through the loading reservoir, into the immune-chamber, after removing the medium. All reservoirs were then filled with 100 μl medium. Migration experiments in the presence of CXCR4 inhibitor were performed by adding 20 µM AMD3100 (Sigma-Aldrich) to the IFN-DCs before loading into the devices. For time-lapse experiments with the Juli microscope, IFN-DCs and SW620 cells were loaded into the devices in the absence of fluorescent staining.
Microscopy. Wide field live cell imaging was carried out with an inverted Olympus IX73 microscope equipped with a Lumencor Spectra X LED illumination using an Evolve Delta CCD camera (Photometrics) and MetaMorph V7.8.0 software (Molecular Devices). The Oko-lab incubation system was employed to control temperature, humidity and CO 2 during acquisition. Fluorescence and brightfield images of the entire device were collected at 24, 48 and 72 h both with a Plan Achromat 4×/0.10NA and 10×/0.25NA objectives (Olympus), and processed to quantify DC migration by automatic cell counting using ImageJ software (NIH). 3D movies were made by acquiring wide field image stacks (100 images of 1 μm Z-step size) every 5 min for each wavelength using a UPLANFL long working distance 20×/0.45NA objective (Olympus). Images were processed with Imaris software v.8.1.2 (Bitplane). For 3D reconstruction confocal images were acquired using a X-light Nipkow spinning-disk head (Crest Optics). Images of 1 μm Z-step size were acquired for each wavelength (n = 100) and processed with Imaris software. Image stacks were edited by Adobe Photoshop CC Software (Adobe Systems) preserving the original parameters of image acquisition. To analyse CXCL12 expression, SW620 cells were treated with or without RI for 48 h. Cells were then fixed with 4% PFA for 15 min at 4 °C and stained with anti-CXCL12 mAb (R&D, clone 79018). Images were taken with a FV1000 confocal microscope (Olympus, Tokyo, Japan) using a (Olympus) planapo objective 60× oil A.N. 1,42. Excitation light was obtained by a Laser Dapi 408 nm for DAPI, and an Argon Ion Laser (488 nm) for FITC (Alexa 488). DAPI emission was recorded from 415 to 485 nm and FITC emission was recorded from 495 to 550 nm.
Cell tracking analysis. IFN-DC migration toward and within the 3D tumor spaces was recorded using time-lapse imaging on a Juli Smart microscope placed in a CO 2 incubator at 37 °C for 0-72 h. The unsupervised image analysis algorithm used for 2D cell tracking in a 3D environment is articulated into three main steps: (i) recognition of confined environment of interest in the microfluidic platform (background elimination); (ii) recognition of cells (segmentation); and (iii) construction of connections (linking) 61 . See Supplementary Methods for details. Once spaces and cells were identified in the immune-chamber, IFN-DC motion was analysed by evaluating the following parameters 62 : (i) number of migrating IFN-DCs performing active migration from the immune-chamber to the tumor-chamber; (ii) speed of IFN-DCs in the immune-chamber (Pre-channels speed); (iii) speed of IFN-DCs in the tumor-chambers (Post-channels speed); (iv) IFN-DCs displacement, which is the covered distance from the starting point in the immune-chamber to the ending position in the tumor-chamber; and (v) DP of IFN-DC migration, which is calculated by the ratio of the Euclidean distance between the starting and the ending position of each path and the total path length. Directional persistency values close to 1 indicate a strong directional path, while values close to 0 indicate dominant Brownian motion. Speed was calculated by the ratio of total distance travelled and the duration of migration.

Analysis of phagocytosis by Confocal Microscopy.
Phagocytosis of SW620 apoptotic cells was evaluated qualitatively by confocal microscopy by placing 3D microfluidic devices on 0.17 μm coverslips and by using immunostaining procedures adapted to previously described protocols applied to 3D cultures 63 . Briefly, after a 48 h cell culture in the device, chambers and channels were washed with PBS. Then, all loading chambers were filled with 50 μl of 4% paraformaldehyde for cell fixation. After washing, specific labelling of the IFN-DCs was performed at 4 °C for 15 min, by adding 5 μl of anti-human HLA-DR-FITC Ab to the upper loading chambers. High-resolution images (800 × 800 pixel, 8 μs/pixel) were acquired with an IX83 FV1200 laser-scanning confocal microscope equipped with FluoView 4.2 software using a UPLANSAPO 20×/0.75NA objective (all from Olympus) 64 . Image stacks were processed with Imaris software for orthogonal view and 3D rendering.
Transwell migration assay. 5 × 10 5 IFN-DCs were resuspended in RPMI-1640 containing 0.1% bovine serum albumin (BSA) and seeded in the upper chambers of 8-μm porous polycarbonate membranes of 24-well plates (Corning, NY), in the presence or absence of 10 μM AMD3100 (Sigma-Aldrich). The lower chambers were filled with supernatants (CM) from NT or RI-treated SW620 cells. After a 4 h incubation at 37 °C, IFN-DCs qPCR using dedicated function-tested RT-qPCR assays. Applied RT-qPCR assays for selected human gene targets were obtained as Real-Time ready Custom Panel plate for 96 reactions (Roche Diagnostics) in a dried-down format. For qPCR assays an extensive list of selected human genes is reported in Supplementary  Table 3. Each well contained specific primers and a Universal ProbeLibrary Probe. The mean of expression of three different housekeeping genes (HPRT1, GAPDH andPGK1) was used to normalize the data. The qPCR assays were run in a Roche LightCycler ® 480 following the manufacturer's instructions. qPCR assays were run in at least 3 independent assays. Statistical analyses. Each experiment was repeated at least three times, yielding comparable results.