Development of a 3D-bioprinted tumour-on-chip with tunable hydrogel properties for monitoring colorectal cancer cell migration and invasion Authors: Adrian García, Lia Jove, María Pereira, Angelica Figueroa, Daniel Nieto Published: 2026 ABSTRACT AbstractUnderstanding the mechanisms governing cancer cell migration is essential for elucidating the early events of colorectal cancer metastasis. Here, we report the development of a light-tunable three-dimensional (3D) bioprinted tumour-on-chip platform for monitoring the migration of colorectal cancer cells under biomimetic microenvironmental conditions. The platform integrates digital light processing (DLP)-based 3D bioprinting with microfluidic technology to recreate structural and mechanical features of the colorectal tumour microenvironment. A hybrid bioink composed of gelatin methacryloyl (GelMA), Matrigel, and type I collagen was formulated to balance DLP printability with biologically relevant extracellular matrix components. Multi-compartment tumour constructs were fabricated with HCT116 colorectal cancer cells spatially confined within a central tumour region, while human umbilical vein endothelial cells (HUVECs) were incorporated into surrounding compartments under co-culture conditions. By modulating the projected light intensity, constructs with tunable properties were generated to investigate the influence of matrix mechanics and transport on cancer cell behaviour. Comparison of Day 1 and Day 4 fluorescence images revealed outward redistribution of the HCT116-positive signal beyond the initially printed tumour core, consistent with colorectal cancer cell migration. The extent of this redistribution was influenced by the light-defined matrix properties and the presence of the neighbouring HUVEC-containing co-culture compartment, highlighting the combined role of physicochemical and cellular cues in regulating tumour cell behaviour. The system also establishes a foundation for future studies of tumour invasion, endothelial barrier interactions, and therapeutic screening under controlled perfusion conditions, supporting the advancement of personalized cancer models and anti-metastatic drug development.Keywords: Microfluidic bioprinting, DLP bioprinting, Tumour-on-chip, 3D tumour models, Colorectal cancer Introduction Cancer metastasis, initiated by tumour cell migration and invasion, remains the leading cause of cancer-related mortality, accounting for about 90% of cancer deaths [1,2]. Colorectal cancer (CRC) is particularly relevant because of its high incidence and its tendency to disseminate to distant organs such as the liver, lung and peritoneum [3,4]. Although metastatic progression is determined by genetic and molecular alterations, tumour cells do not migrate or invade in isolation. Their behaviour is strongly influenced by the surrounding microenvironment, which includes extracellular matrix (ECM) composition and mechanics, stromal and endothelial cells, soluble gradients and physical confinement [5,6]. Reproducing these characteristics in vitro remains difficult, which limits the ability to study the early behaviour of tumour cells under controlled but biologically meaningful conditions. Traditionally, most studies have been conducted on two-dimensional (2D) monolayers on plastic or glass because they are simple, inexpensive and reproducible, but they do not replicate the architecture, mechanical constraints or matrix-mediated signalling to which cells in tissues are subjected [7,8]. Animal models offer a more complex physiological environment, but they are costly, subject to ethical constraints and are often limited by interspecies differences and their poor ability to predict clinical outcomes [[9], [10], [11]]. These limitations have driven the development of three-dimensional (3D) in vitro models, in which cells are embedded in materials similar to the extracellular matrix and can interact with their environment in a manner more analogous to tissues [12,13]. However, many 3D systems still lack spatial control, reproducible heterocellular organisation and compatibility with repeated imaging or integration into microfluidic devices. Organ-on-a-chip and tumour-on-chip technologies address part of this problem by enabling micro-scale control of geometry, compartmentalisation, flow control and cell-cell interactions [[14], [15], [16], [17]]. In cancer research, these systems are increasingly used to study tumour-stroma communication, vascular interactions and transport phenomena under defined experimental conditions [18,19]. However, many tumour-on-chip models still rely on manual hydrogel loading, fixed device architectures, or limited control over the initial position of each material and cell population. This makes it difficult to disentangle the contributions of matrix properties, cellular organisation, and soluble-factor signalling on tumour-cell behaviour. 3D bioprinting offers a direct route to overcoming this limitation by placing cells and biomaterials in predefined positions. Among the available bioprinting methods, digital light processing (DLP) is particularly attractive, as it uses digitally projected light patterns to photopolymerise hydrogel regions with spatial control [[20], [21], [22]]. Sequential loading and exposure can generate multicompartment constructs containing cell-laden and acellular regions. Integrating this process with a microfluidic device requires the integration of material printability, cytocompatibility, optical accessibility, soft mechanics and molecular transport into a single platform. Hydrogel-based microfluidic models have been used to study CRC migration and extravasation [[23], [24], [25]]. In parallel, Bioprinted CRC hydrogel models have also been used to investigate material-dependent tumour phenotypes [26], while multicellular GelMA-based CRC models have been developed to evaluate chemotherapeutic responses in heterocellular 3D constructs [27]. Multimaterial bioprinting has enabled cell-laden and acellular materials to be combined within the same construct [22,28], and greyscale projection has been used to introduce variation in hydrogel properties [29,30]. However, to our knowledge, these capabilities have not been integrated through direct sequential DLP bioprinting of digitally predefined multicompartment structures within a chip. In the present platform, the projected image determines the position, geometry and light exposure of each region, allowing different assay areas, bioinks and cell populations to be combined sequentially within a single chip. This enables migration assays, comparisons between matrix conditions and co-culture assays to be incorporated within the same device, supports the construction of more complex models and allows the geometry to be changed without redesigning the chip, introducing moulds, or transferring preformed constructs. Hydrogels are commonly used as bioinks because they provide hydrated environments around encapsulated cells and permit control over matrix composition and cross-linking [31]. Natural polymers, such as collagen, gelatin-derived materials and basement membrane extracts, provide biological signals that promote adhesion, survival and interaction with the matrix, whilst synthetic or chemically modified systems offer greater control over cross-linking, reproducibility and mechanical tuning [32]. Gelatin methacryloyl (GelMA) is widely used in light-based bioprinting, as it combines photochemical cross-linking capability, cell-interactive motifs and adjustable stiffness [33,34]. However, GelMA alone does not replicate the complexity of the tumour extracellular matrix. Combining GelMA with Matrigel and type I collagen introduces additional ECM components and cell-binding sites relevant to tumour cell-matrix interactions and future mechanistic studies of invasion, but also changes viscosity, cross-linking efficiency, network structure, mechanical properties and molecular transport. This trade-off between biological activity and printing control constitutes a fundamental material-related challenge in biofabricated tumour models. Matrix mechanics are particularly important in colorectal cancer, where stiffness, ECM composition and stromal remodelling influence tumour progression and cell motility [[35], [36], [37]]. Stromal and endothelial compartments can also modify tumour-cell behaviour through paracrine signalling and matrix-dependent interactions [38,39]. Therefore, a useful in vitro model must not only contain the relevant cell types but also allow the matrix properties to be adjusted without altering multiple biochemical variables simultaneously. Light-based bioprinting offers a practical strategy for this, as the applied light dose can modulate the cross-linking density, network structure, molecular transport and stiffness of the hydrogel, whilst keeping the composition of the bioink constant [29,40]. This enables crosslinking-dependent matrix conditions to be compared while maintaining the bioink composition constant. In this study, we have developed a microfluidic platform for the in situ DLP bioprinting of multi-compartmental colorectal tumour constructs under greyscale-controlled exposure conditions. The device was designed to allow the formation of hydrogel patterns directly within the chip, whilst maintaining optical access for fluorescence imaging. A hybrid bioink comprising GelMA, Matrigel and collagen was formulated and characterised to balance cell compatibility, printability, soft mechanical properties and diffusion transport. Greyscale-controlled DLP exposure was used to vary the light dose while maintaining the same bioink composition, and the resulting matrix conditions were assessed through mechanical and physicochemical measurements. HCT116 colorectal cancer cells were placed in a defined central region, with or without a surrounding compartment containing HUVECs, to monitor how matrix properties and cellular organisation affect the early 3D redistribution of tumour cells. The contribution of this work is the integration of direct in-chip DLP bioprinting, sequential patterning of cell-containing and acellular regions, light-dose modulation and fluorescence-based monitoring of tumour cell behaviour within one platform, providing a controlled basis for future assays of migration, invasion and endothelial barrier function under perfusion. Materials and methods Fabrication of the microfluidic device All digital models were designed using Autodesk Fusion 2025 (Autodesk Inc., San Francisco, CA, USA) and exported as STL files. The final device consisted of a central culture chamber connected to two opposing microchannels acting as the inlet and outlet. The chamber measured 12 × 8 mm and was 1 mm high, with an optimised non-rectangular geometry designed to improve filling and reduce bubble trapping. The volume of the chamber was 83.5 μL. The inlet and outlet ports were designed to allow direct loading with a pipette or connection to microfluidic Luer connectors. The STL files were prepared using HALOT BOX slicing software (Creality; version 4.2.0.37) and printed on a HALOT-MAGE PRO 8K LCD printer (Creality 3D Technology Co., Ltd., Shenzhen, China) with a layer height of 50 μm. Before selecting the final material, three commercial resins were evaluated for geometric fidelity and biocompatibility: BioMed Clear (Liqcreate BV, Eindhoven, Netherlands), Standard Clear resin (Anycubic, Shenzhen, China) and Flexible UV Resin (Copymaster3D, United Kingdom). A resolution test artefact containing positive and negative features in the X, Y and Z directions was printed in each material to assess feature definition and dimensional accuracy (Supplementary Methods). After printing, the devices were washed in two successive baths of isopropanol to remove the uncured resin and were dried using compressed air. The bottom opening was sealed with a 22 × 22 mm glass coverslip by photobonding during the final UV post-curing stage to allow high-resolution imaging from below. The top was sealed with a PDMS lid to maintain optical access from above during in situ DLP bioprinting. Briefly, PDMS (10:1) was mixed, degassed, cast on 3D-printed moulds, and cured at 60 °C for 2 h; (Supplementary Methods). The PDMS lid was placed over the chip during the final UV post-curing stage, sealing the top of the device whilst the resin completed curing. The sterilisation method was tailored to the compatibility of the resin. Devices printed with Bio-Med Clear were autoclaved at 121 °C for 20 min, as this resin is formulated for biomedical applications and the manufacturer supplies it with biocompatibility documentation based on ISO 10993. Devices printed with the other resins were sterilised by exposure to UV light for 30 min prior to the cell contact experiments. Before use, the assembled chips were washed with sterile buffer to remove particles introduced during handling. Geometric characterisation Dimensional accuracy was assessed using images captured with a Euromex CMEX-10 Pro USB 3.0 microscope camera (Euromex Microscopen BV, Arnhem, Netherlands) mounted on a stereomicroscope. For each resin and for the final chip design, geometric features were measured and compared with the corresponding CAD values. These included the chamber length, chamber width, chamber height and channel diameter. For the resolution test artefact, the absolute dimensional error was calculated as the difference between the measured dimension and the nominal CAD dimension. For the resolution analysis, three independently printed artefacts were analysed per resin (n = 3). For characterisation of the final BioMed Clear chip, 12 independently printed chips were analysed (n = 12). Each feature was measured three times per sample, and the technical measurements were averaged to obtain one value per independently printed artefact or chip. Flow tolerance test The quality of the assembly was assessed by visually inspecting the glass coverslip, the PDMS lid, the chamber and the channels. First, functional leak-tightness was checked by manually filling the chips with a dyed aqueous solution and inspecting the devices for any leaks whilst handling them. Next, flow tolerance was assessed by perfusing dyed Milli-Q water through the assembled devices at increasing flow rates. Custom 3D-printed Luer connectors were fitted to the inlet and outlet ports and connected to a syringe pump system (SPM/2 ZU-I, DK Infusetek, Germany) using silicone tubing with an inner diameter of 1 mm. Flow rates were gradually increased from 0.05 to 80 ml min−1, corresponding to the maximum operating range of the pump. During perfusion, the devices were monitored for leaks, bubble formation, uneven filling, delamination and structural failure. The test was performed on three independent chips. Cytocompatibility assessment of the printed resins The cytocompatibility of the printed resins was assessed using 3D-printed inserts designed to fit into the individual wells of a 48-well plate. The inserts were printed following the same cleaning, curing and sterilisation procedures described above. HCT116 cells were seeded onto each insert at 1 × 104 cells per sample, which corresponds to the density typically used in viability assays in 96 well plates. After 24 h of culture under standard conditions, cytocompatibility was assessed using a fluorescence live/dead cell imaging kit (Sigma-Aldrich, USA). The samples were washed twice with HBSS and stained with calcein-AM, propidium iodide and Hoechst 33,342 following the manufacturer's instructions. After 60 min of incubation in the dark, images of the samples were obtained using a MATEO FL fluorescence microscope (Leica Microsystems, Wetzlar, Germany). Three images were acquired per sample, with three independent replicates per resin. Cell viability was quantified in Fiji/ImageJ as equation (1): Bioink formulation and characterisation Materials All reagents were used as received unless otherwise stated. Gelatin type A (porcine skin, CAS 9000-70-8), methacrylic anhydride (MA, CAS 760-93-0) and lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP, CAS 85073-19-4) were purchased from Sigma-Aldrich (USA). Dulbecco's Modified Eagle Medium (DMEM), fetal bovine serum (FBS), and penicillin-streptomycin (Pen/Strep) were obtained from Thermo Fisher Scientific (USA). Matrigel (growth-factor-reduced, Corning) and type I collagen (rat-tail, 8-10 mg/mL stock, Corning) were used as extracellular-matrix components. All aqueous preparations were made with sterile, 0.22 μm-filtered Milli-Q water or supplemented DMEM, as specified. GelMA synthesis Gelatin methacryloyl (GelMA) was synthesized from type A porcine gelatin and methacrylic anhydride following a modified version of the protocol described by Loessner et al. [41]. Briefly, gelatin (10% w/v) was dissolved in Milli-Q water at 50 °C under constant stirring for 40 min. Methacrylic anhydride (0.6 mL/g gelatin) was then added dropwise and allowed to react for 2 h at 50 °C. The reaction mixture was centrifuged (3500 rpm, 5 min) to remove excess anhydride and diluted 1:2 with pre-warmed Milli-Q water. The solution was dialyzed ([MWCO] = 12-14 kDa, Spectra/Por 2) against deionized water for 7 days at 40 °C with daily water changes, adjusted to pH 7.4 using 1 M NaHCO3, frozen in liquid nitrogen, and lyophilized to obtain a white porous solid. The dried polymer was stored at −20 °C until use. The degree of methacrylation was determined by 1H NMR spectroscopy based on the reduction in the lysine methylene proton signal relative to unmodified gelatin (Supplementary Methods). The synthesized GelMA exhibited a degree of methacrylation of 79.5 ± 4.4% (Supplementary Fig. S1). Metabolic activity assay To select a bioink composition compatible with the encapsulation and printing of HCT116 cells, metabolic activity was assessed using the CellTiter-Glo luminescent cell viability assay (Promega, USA). Formulations containing different concentrations of GelMA and Matrigel with 0.1% (w/v) LAP were prepared. Collagen screening was performed using formulations containing 5% (w/v) GelMA, 15% (v/v) Matrigel and the corresponding type I collagen concentrations. Cell-loaded precursor solutions were prepared by mixing the hydrogel precursor with HCT116 cells in suspension at 37 °C to maintain low viscosity and prevent premature gelation. Luminescence, proportional to ATP content, was measured after 24 h using an Infinite 200 PRO multimode microplate reader (Tecan, Männedorf, Switzerland). Three technical replicates were averaged within each experiment, and three independent experiments were performed (n = 3). The final formulation was selected by balancing metabolic activity, workability and printability. A targeted metabolic-activity screen was also performed with HUVECs using the six formulations shown in Fig. S6. The preparation and measurement procedures were identical to those used for HCT116 cells. Two independent experiments were performed (n = 2). Cell-free scaffold bioprinting and light calibration Acellular hydrogel scaffolds were printed to assess the effect of the light dose on the formation of the structure and the material properties. Printing was done on a customised “top-down” DLP bioprinter, equipped with a 405 nm light source and a 4K DLP projector (Wintech PRO6600; 3840 × 2160 pixels). The projection masks were generated as 8-bit greyscale images, in which the greyscale value controlled the irradiance delivered. Prior to the printing experiments, the irradiance was calibrated in the printing plane, at the same working distance used for chip printing. Measurements were carried out using a PM100D2 optical power meter equipped with an S121C photodiode power sensor (Thorlabs GmbH, Germany). Irradiance was measured from 10% to 100% of the greyscale in 10% increments, with three independent measurements per intensity. For the experimental conditions used in this study, the exposure time was set at 45 s and four greyscale values were selected: 255, 191, 128 and 64, corresponding to a projected intensity of 100%, 75%, 50% and 25%. The corresponding irradiance values were 6.776, 4.986, 3.513 and 1.898 mW cm−2, respectively. Acellular scaffolds were printed directly inside the microfluidic chips using the same bioink formulations and geometric patterns later used for cell-laden experiments. After each exposure, the chamber was washed with PBS to remove the uncured precursor solution before proceeding to characterisation. For comparison among bioink compositions, three acellular formulations were characterised: 5% (w/v) GelMA; 5% (w/v) GelMA + 15% (v/v) Matrigel; and 5% (w/v) GelMA + 15% (v/v) Matrigel + 0.2% (w/v) type I collagen. All formulations contained 0.1% (w/v) LAP and were photocrosslinked for 45 s using a greyscale value of 255. Separately, the selected GelMA-Matrigel-collagen formulation was photocrosslinked for 45 s using greyscale values of 255, 191, 128 and 64 to evaluate the effect of light dose on its mechanical and physicochemical properties. ATR-FTIR Attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR) was used to evaluate the chemical fingerprints of gelatin, GelMA and the printed bioink formulations. The spectra were obtained using a Nicolet iS10 spectrometer (Thermo Fisher Scientific, USA) fitted with a diamond ATR crystal. Measurements were taken between 4000 and 400 cm−1 with a resolution of 4 cm−1 using 32 sweeps per spectrum. For the printed samples, acellular hydrogel discs fabricated under the same exposure conditions used during bioprinting were freeze-dried for 24 h and placed directly onto the ATR crystal. Rheological characterisation The rheological behaviour of the bioink formulations was characterised using a Discovery DHR-2 hybrid rheometer (TA Instruments, Waters LLC, New Castle, DE, USA) fitted with a 25 mm parallel-plate geometry. Measurements were carried out at 37 °C under shear flow conditions, using a logarithmic sweep of the shear rate from 1 to 1000 s−1 and a fixed gap of 0.5 mm. Prior to measurement, the samples were subjected to a pre-shear at 50 s−1 for 60 s to homogenise the precursor, followed by equilibration at 0.1 s−1 for 60 s. Each formulation was measured in triplicate. Viscosity values were recorded as a function of shear rate and expressed in mPa·s. Compressive tests The mechanical properties of the printed hydrogel formulations were evaluated by unconstrained compression using a Mach-1 mechanical tester (Biomomentum Inc., Laval, QC, Canada) fitted with a uniaxial load cell with a force resolution of 0.5 mN. The cylindrical hydrogel samples, with a diameter of 4 mm and a height of 1 mm, were photocrosslinked under the same conditions used for printing and equilibrated in PBS for 24 h prior to testing. Measurements were performed in PBS at 25 °C. A preload of 3 mN was applied to ensure contact between the compression plate and the hydrogel surface. The samples were compressed at a rate of 0.1 mm/s. The apparent compressive modulus was calculated from the initial linear region of the stress-strain curve (Supplementary Fig. S2). Results are presented as mean ± standard deviation, with n = 3 per condition. Swelling The swelling behaviour of the printed hydrogels was assessed by measuring water absorption over 24 h. Immediately after printing, the acellular scaffolds were rinsed with PBS to remove any residual precursor solution and gently dried with absorbent paper to remove excess liquid from the surface. The initial weight was recorded as the printed weight (W0). The samples were then immersed in PBS and incubated at 37 °C for 24 h. Following incubation, the samples were gently dried again and weighed to obtain the swollen weight (WS). Swelling was expressed as the percentage increase in weight relative to the initial printed condition, according to equation (2): Measurements were performed in triplicate (n = 3) for each formulation. Diffusion measurement Diffusion within the printed hydrogels was quantified using rhodamine B and a modified protocol based on imaging and Fick's second law (Supplementary Methods) [42,43]. In summary, acellular hydrogels were printed inside the central chamber (Supplementary Fig. S3), leaving an unprinted channel 1 mm wide connecting the inlet and outlet. This geometry created a defined interface between the hydrogel and the liquid, from which rhodamine B could diffuse laterally into the printed hydrogel. Following photocrosslinking, the chip was rinsed with HBSS, and the unprinted channel was filled with 1.2 × 10−5 M rhodamine B in HBSS. Fluorescence images were acquired every minute using a MATEO FL fluorescence microscope. Linear profiles perpendicular to the hydrogel-liquid interface were extracted in Fiji/ImageJ at the selected time points: 0, 10, 20, 30 and 40 min. The fluorescence intensity was considered proportional to the tracer concentration and normalised with respect to the intensity at the boundary to obtain C(x,t)/C0. The apparent diffusion coefficient D was estimated by fitting the experimental profiles to the one-dimensional semi-infinite solution of Fick's second law (3): Three independently printed samples were analysed per condition (n=3). SEM The microstructure of the printed hydrogels was examined using scanning electron microscopy (SEM). The acellular scaffolds corresponding to the different bioink formulations were freeze-dried for 24 h. The dried samples were coated by sputtering with a 40 nm layer of gold using a LUXOR Gold Coater, and images were obtained using a NANOS desktop SEM (Semplor, Spain). Imaging was carried out in low-vacuum mode at an acceleration voltage of 5 kV. Representative images were obtained at magnifications of 673× and 1346× using a dwell time of 4800 ns, a line scan of 4096 and an average of 4. Computational simulations Finite element simulations were carried out in COMSOL Multiphysics 6.3 (COMSOL AB, Stockholm, Sweden) using the Transport of Diluted Species module to support the design of the static culture conditions. Two complementary models were implemented: firstly, diffusion equilibrium within the printed hydrogels; and secondly, glucose availability under cellular consumption (Supplementary Methods). The chip was represented as a 2D geometry in the image plane, including the chamber height as a parameter for volume conversion. The model reproduced the geometry of the central chamber and the arrangement of the four circular hydrogel structures used experimentally. The same geometry was used for the diffusion and glucose consumption simulations. As the experiments were conducted under static conditions, transport was modelled as diffusion-dominated, without imposed convective flow. No-flux boundary conditions were applied at the outer boundaries of the chip. For the diffusion model, the hydrogel domains were assigned apparent diffusion coefficients measured experimentally. Concentration profiles were extracted from the centre of the hydrogel structures to estimate the time required to reach equilibrium within the printed regions. For the nutrient availability model, glucose was used as a representative metabolic substrate. The initial glucose concentration was set in accordance with high-glucose DMEM. Cellular consumption was implemented as a volumetric sink term in the cell-loaded hydrogel domains using Michaelis-Menten kinetics, following previous approaches to modelling diffusion-consumption in 3D cell culture systems [44,45]. Simulations were performed for different cell densities to estimate glucose depletion over four days of static culture. Cell bioprinting Cell culture Human colorectal carcinoma HCT116 cells, sourced from ATCC, were used as the tumour component. Human umbilical vein endothelial cells (HUVEC; ATCC PCS-100-010) were used in co-culture experiments. The HCT116 cells were maintained in high-glucose DMEM, supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin. HUVECs were maintained in basal medium for vascular cells supplemented with the BBE endothelial cell growth kit. All cells were cultured at 37 °C in a humidified incubator with 5% CO2. The cells were subcultured using trypsin-EDTA and used between passages 5 and 8. Cell labelling and preparation of cell-laden bioinks Cell-loaded bioinks were prepared using the optimised formulation consisting of 5% (w/v) GelMA, 15% (v/v) Matrigel, 0.2% (w/v) type I collagen and 0.1% (w/v) LAP in culture medium. The LAP was first dissolved in the medium at 70 °C for 10 min under constant stirring. The GelMA was immersed in the LAP-containing medium and allowed to hydrate overnight at 4 °C without stirring and protected from light. The following day, the solution was incubated at 37 °C for 2 h until completely dissolved and filtered through a 0.22 μm membrane. Matrigel and neutralised type I collagen were added and mixed gently to prevent premature gelation. Prior to encapsulation, HCT116 and HUVEC cells were labelled with CellTracker probes of different spectra (Invitrogen, Thermo Fisher Scientific, USA). HCT116 cells were labelled with CellTracker Green CMFDA and HUVECs with CellTracker Orange CMTMR. Both probes were prepared at 5 μM in serum-free medium. The cells were incubated with the corresponding dye for 30 min under standard culture conditions, washed and allowed to recover for 30 min in complete medium prior to harvesting. Immediately prior to printing, the precursor solution was equilibrated to 37 °C. The corresponding cell suspension was then added dropwise under gentle mixing to minimise shear stress and promote a homogeneous cell distribution. Each bioink was prepared to a final density of 1 × 106 cells·mL−1. Therefore, the local cell density was equivalent in the HCT116 and HUVEC compartments. Printing procedure The structures were printed directly inside the microfluidic chips using the top-down DLP configuration described above. Two concentric geometries were used. In the tumour-only condition, a central core loaded with HCT116 was surrounded by a cell-free hydrogel. In the tumour-HUVEC co-culture condition, the same HCT116 core was immediately surrounded by a ring loaded with HUVECs and an outer ring of cell-free hydrogel. The central HCT116 region had a diameter of 0.5 mm. The HUVEC ring had a radial thickness of 100 μm and the outer cell-free ring had a radial thickness of 150 μm, giving a final structure diameter of 1 mm. The tumour-only controls were printed using the same general geometry, but the HUVEC-containing region was replaced by cell-free bioink. All structures were fabricated under four greyscale conditions corresponding to projected intensities of 100%, 75%, 50% and 25%, with a fixed exposure time of 45 s. Sequential multi-material printing was carried out by loading the first bioink into the chamber, projecting the corresponding mask, washing with HBSS to remove the uncross-linked precursor, and repeating the process for the next compartment. Following printing, the chips were filled with co-culture medium consisting of a 1:1 mixture of high-glucose DMEM without FBS and endothelial basal medium without the BBE endothelial growth supplement. Serum and exogenous endothelial growth factors were omitted to minimise proliferation and externally supplied growth-factor signalling during the redistribution experiment. All cellular experiments were conducted under static culture conditions. Single-plane fluorescence images were acquired daily. For analysis of HCT116 redistribution, fluorescence z-stacks were acquired on Days 1 and 4 and converted into maximum-intensity projections, as described below. Quantification of tumour-cell redistribution The redistribution of HCT116 was quantified from fluorescence images using a custom Python-based image analysis workflow, adapted from pixel-based approaches used in the quantification of 3D spheroid invasion [46]. The analysis did not track individual cells, as it was not possible to reliably resolve individual cell trajectories from the fluorescence images. Instead, redistribution was quantified at the population level using the HCT116 fluorescence signal. Fluorescence images were acquired separately for each fluorescence channel and assembled into image stacks. The channels were merged to generate the representative composite images. Quantitative analysis was performed on the HCT116 green fluorescence images. For each construct, the Day 1 image was used to define the initial tumour region. The green fluorescence channel was segmented to identify the HCT116-positive area, and a circular reference boundary was generated to enclose the initial tumour core. This boundary was then used as the reference region for subsequent time points. For the Day 4 images, the same reference boundary was re-centered on the main HCT116 core to correct for minor image positioning differences between acquisitions. Subsequently, the HCT116-positive signal located outside the initial reference boundary was segmented and quantified as redistributed area. Two main parameters were extracted: the HCT116-positive area outside the initial tumour core, expressed in μm2 and as a percentage of the initial core area; and the mean redistribution distance, calculated as the mean radial distance of HCT116-positive pixels redistributed from the initial boundary. Distances were converted from pixels to micrometers using the image calibration factor of 0.833 pixels·μm−1. The script generated results per image, overlaid images for quality control, and summary spreadsheets for statistical analysis. Three independently prepared chips were analysed for each culture configuration. Each chip contained four constructs, with one construct printed at each projected intensity of 100%, 75%, 50% and 25%. Consequently, three independent constructs were analysed for each combination of projected intensity and culture configuration (n=3). Cell viability in printed scaffolds To assess post-printing viability, independent constructs containing either unlabelled HCT116 cells or unlabelled HUVECs were printed under the four greyscale conditions and cultured for four days. These cells were not labelled with CellTracker dyes. On day 4, the constructs were stained with calcein-AM, propidium iodide and Hoechst 33,342 and imaged using the MATEO FL fluorescence microscope. Three non-overlapping fields were acquired per construct and counted manually in Fiji/ImageJ. Propidium iodide-positive nuclei were classified as dead, whereas calcein-AM-positive cells were classified as viable. The three fields were averaged to obtain one viability value per independently printed construct. Three independent constructs were analysed per cell type and greyscale condition (n = 3). Statistical analysis All statistical analyses were performed using GraphPad Prism 10.6.1 (GraphPad Software, Boston, MA, USA). Data are presented as mean ± standard deviation unless otherwise stated. HCT116 redistributed area and mean redistribution distance were compared across the four projected intensities and between tumour-only and tumour-HUVEC co-culture constructs using ordinary two-way ANOVA followed by Šídák's multiple-comparisons test. HCT116 and HUVEC viability were analysed separately using one-way ANOVA followed by Tukey's multiple-comparisons test. The four intensity conditions were included in each chip, and three independent chips were analysed (n=3). Differences were considered statistically significant at p < 0.05 and are represented as follows: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***) and p < 0.0001 (****). Comparisons without a significance symbol were not statistically significant. Results and discussion Microfluidic device characterisation The first requirement for creating the platform was to fabricate a microfluidic device that combined optical access, reliable filling and sufficient dimensional accuracy for in situ DLP bioprinting. Resin-based 3D printing offers a simple and rapid route to customised microfluidic devices, but the choice of material is not straightforward. In cell-contacting applications, the resin must provide sufficient detail fidelity for the fabrication of channels, whilst maintaining acceptable cytocompatibility following washing, post-curing and sterilisation [47]. For this reason, three commercial resins were evaluated using the same resolution test artefact. The resolution artefact was designed to compare positive and negative features in both the XY and Z directions (Fig. S4A and B). Representative printed artefacts showed that all resins could reproduce the general geometry, but the accuracy of the features depended largely on the type of resin, the orientation of the features, and whether the structure was a protrusion or a cavity (Fig. S4C). Quantification of absolute dimensional error showed that positive features were generally reproduced with greater accuracy than negative ones, whilst small holes and cavities became progressively less reliable as the nominal feature size decreased (Fig. S4d–g). Copymaster resin performed best for some negative features in the Z-direction and was the only material that consistently preserved the smallest cavities in that orientation (Fig. S4G). However, all three materials reproduced the 600 μm feature range with acceptable fidelity, which was the size chosen for the inlet and outlet channels used in the final chip design. As the final device was not designed around the smallest printable detail, the choice of resin was not based solely on resolution. Instead, the aim was to identify a material that offered an appropriate balance between geometric fidelity and cellular compatibility. Cytocompatibility assays performed on 3D-printed inserts (Fig. S5a–c) showed a clear difference between the three materials after post-processing and sterilisation (Fig. S5d and e): BioMed Clear supported 84.74 ± 5.38% viability, Anycubic Standard Clear 61.33 ± 7.25%, and Copymaster Resin only 12.15 ± 7.77% after direct cell contact. Therefore, despite not being the best-performing resin in the resolution artefact, BioMed Clear was selected for fabrication of the final devices, as it provided the best overall compromise between sufficient geometric fidelity and markedly superior cytocompatibility. The final chip was then fabricated using BioMed Clear and its geometry was characterised. The dimensions of the chamber and the channel corresponded closely with the CAD model (Fig. 1a and b). The measured values were 8.21 ± 0.10 mm for the chamber width, 12.16 ± 0.13 mm for the chamber length, 0.62 ± 0.05 mm for the channel diameter and 1.01 ± 0.11 mm for the chamber height (Fig. 1c). These small deviations from the nominal dimensions confirmed that the resin and the selected printing parameters were suitable for the final geometry. The device also retained the intended chamber height of approximately 1 mm, which allowed for the formation of confined 3D hydrogel constructs rather than monolayer cultures. Geometric characterisation of the final microfluidic chip. (a) CAD rendering of the final chip design indicating the main measured features: (1) chamber width, (2) chamber length, (3) channel diameter, and (4) chamber height. (b) Corresponding printed features used for dimensional analysis. Scale bar: 1 mm. (c) Comparison between CAD and measured dimensions. Quantitative data in (c) are presented as mean ± SD from 12 independently printed chips (n = 12). Three measurements per feature were averaged to obtain one value per chip. Another advantage of the design related to optical imaging is that resin-based microfluidic devices can introduce autofluorescence, background signal or optical artefacts that interfere with fluorescence microscopy [48,49]. To minimise this limitation, the central culture region was designed so that imaging was performed through the glass coverslip rather than through a printed resin surface. This configuration reduced optical interference from the printed material and maintained a clear imaging path for monitoring fluorescently labelled cells within the hydrogel. The assembled devices were also evaluated using a flow tolerance test. A dyed aqueous solution was circulated through the chips at increasing flow rates, ranging from 0.05 to 80 ml min−1, whilst monitoring for leakage, bubble formation, uneven filling and structural failure (Supplementary Vid. 1). The chips withstood the entire test range without any detectable leaks or delamination, indicating that the glass-PDMS-resin assembly was mechanically stable. However, uniform filling became increasingly difficult to maintain at high flow rates, particularly above approximately 5 ml min−1, where bubble formation and irregular fluid fronts were observed more frequently. This distinction is important because bubble formation in microfluidic devices can alter flow profiles, modify local transport and compromise image quality [50]. Therefore, the device was mechanically robust at flow rates far higher than those required for routine culture, but practical operation should be maintained at lower flow rates to preserve uniform filling and bubble-free operation. After validating the chip's geometry and sealing, the next step was to establish the in situ printing workflow (Fig. 2). The device was positioned into the printer, allowing the patterns projected at 405 nm to reach the hydrogel precursor from above, whilst it remained possible to acquire fluorescence images from below. Sequential loading, exposure and washing steps enabled the independent photopolymerisation of different bioinks within the same chamber (Fig. 2a and b). This process allowed the direct fabrication of concentric structures with multiple compartments within the microfluidic device (Fig. 2c–e). A proof-of-concept print using three visually distinguishable materials confirmed that the workflow could generate spatially defined heterogeneous structures within the chamber, with a central region, an intermediate ring and an outer ring clearly separated after printing (Fig. 2e). This confirmed that the chip was not only compatible with passive loading and flow handling, but also with controlled multi-material DLP patterning. DLP bioprinting workflow on a chip for creating sequential patterns. (a) A top-down DLP printer with a rotor that positions multiple microfluidic chips during the printing process. (b) The sequential bioprinting process, including loading each bioink, projecting the corresponding pattern, washing away the uncross-linked precursor and repeating the cycle for the next bioink. (c) A schematic representation of the final concentric structure, which is composed of three hydrogel regions that have been printed using different bioinks. (d) An illustration showing the spatial arrangement of the multi-material structure printed within the microfluidic chip. (e) A structure printed with three different materials within the chip showing the central region, intermediate ring, and outer ring. Scale bar: 2 mm. Bioink characterisation Once the chip and printing workflow had been established, the next step was to formulate a bioink compatible with cell encapsulation, DLP printing and 3D culture. GelMA was selected as the structural component, as it provides photocrosslinkable groups, cell-interactive motifs and tunable mechanical properties [[51], [52], [53]]. However, GelMA alone does not replicate the biochemical complexity of the tumour extracellular matrix. Therefore, Matrigel and type I collagen were incorporated to provide signals related to the basement membrane and fibrillar ECM. The main challenge of the formulation was to enhance biological activity without compromising printability, as increasing the amount of non-photocrosslinkable ECM components can reduce control over crosslinking, increase heterogeneity and disrupt transport. The first step was to select the composition, because matrix composition and mechanics strongly regulate tumour cell survival, proliferation and invasion in 3D [36,54]. The bioink was first selected based on cytocompatibility and only then tuned for mechanical range and transport. The formulation was evaluated using the metabolic activity of HCT116 cells. Luminescence values were normalised against the control to compare the relative effect of GelMA and Matrigel concentrations on cellular metabolic activity. The screening revealed a clear composition-dependent response (Fig. 3a). Increasing the GelMA concentration reduced metabolic activity (Fig. 3b), whilst increasing the Matrigel content enhanced the response of HCT116 cells within the range analysed (Fig. 3c). Under the 5% GelMA condition, the addition of 15% Matrigel increased metabolic activity compared to the control, supporting the use of Matrigel as a biologically active component. The subsequent addition of a low concentration of type I collagen further improved the response in the GelMA-Matrigel system (Fig. 3d). The initial composition screening was performed with HCT116 cells to establish the working formulation. The selected conditions were subsequently evaluated using HUVECs to confirm that the endothelial cells showed a comparable overall response and retained metabolic activity in the selected bioink (Fig. S6). HUVECs showed a similar overall pattern: The highest mean response was obtained with the Matrigel-collagen formulation without GelMA, whereas the complete GelMA-Matrigel-collagen formulation maintained a comparable HUVEC metabolic response. Based on this screening, 5% GelMA, 15% Matrigel and 0.2% type I collagen were selected as the working formulation, as this composition provided the best balance between cellular response and practical printability. Metabolic-activity screening of the bioink composition using HCT116 cells. (a) HCT116 metabolic activity in formulations containing fixed GelMA concentrations of 15% (left), 10% (centre) or 5% (right) combined with increasing Matrigel (MG) concentrations. In each graph, the bars represent, from left to right, the medium-only control, the corresponding GelMA concentration without MG, and the same GelMA concentration supplemented with 1%, 5%, 10% or 15% MG. (b) Effect of GelMA concentration without MG on HCT116 metabolic activity. (c) Effect of MG concentration without GelMA on HCT116 metabolic activity. (d) Targeted evaluation of the selected formulations. The conditions shown, from left to right, are the medium-only control, 5% GelMA, 15% MG, 5% GelMA + 15% MG, 15% MG + 0.2% type I collagen (Col I), and 5% GelMA + 15% MG + 0.2% Col I. GelMA, Col I and LAP concentrations are expressed as % (w/v), whereas MG concentrations are expressed as % (v/v). All bioink formulations contained 0.1% (w/v) LAP. Luminescence was measured after 24 h and normalised to the corresponding medium-only control. Data are presented as mean ± SEM from three independent experiments (n = 3). Statistical analysis was performed separately for each graph using one-way ANOVA followed by Bonferroni's multiple-comparisons test (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). The ATR-FTIR technique was used to compare the chemical fingerprints of the individual and combined protein formulations. The spectra of the GelMA formulations, as well as those containing Matrigel and collagen, showed the expected broad amide regions, with no new, clearly separated bands observed following the incorporation of Matrigel or collagen (Fig. 4a). This result was predictable, as all three components are protein-rich materials and therefore share overlapping amide signatures. Consequently, ATR-FTIR proved useful for confirming the presence of functional groups of protein origin, but not for resolving each component of the ECM independently once combined in the final bioink. Physicochemical characterisation of the selected hydrogel formulations. Color code used throughout the figure: 5% GelMA (red), 5% GelMA + 15% Matrigel (blue), and 5% GelMA + 15% Matrigel + 0.2% collagen I (black). (a) ATR-FTIR spectra of the bioink formulations. (b) Viscosity of the three hydrogel formulations. (c) Compressive modulus of the printed hydrogels for the three hydrogel formulations. (d) Swelling percentage of the corresponding printed hydrogels. (e) Apparent Rhodamine B diffusion coefficient measured for the three formulations. (f) SEM images of freeze-dried hydrogels. Scale bar: 100 μm. (g) COMSOL simulation of diffusion equilibration within the printed hydrogel region using the experimentally measured apparent diffusivities of the three formulations. (h) COMSOL simulation of glucose concentration over time under cellular consumption at different seeding densities. Data in (b-e) are presented as mean ± SD from three independent samples per formulation (n = 3). Data in (c-e) were analysed by ordinary one-way ANOVA followed by Tukey's multiple-comparisons test. Rheological characterisation showed that all formulations remained within a low viscosity range suitable for chip loading and handling (Fig. 4b). Viscosity decreased as the shear rate increased in all cases, indicating shear-thinning behaviour. GelMA alone exhibited the lowest viscosity, whilst the addition of Matrigel moderately increased viscosity. The incorporation of collagen did not produce a clear increase in mean viscosity compared to the GelMA-Matrigel formulation, but it increased variability in the low-shear region. This suggests that the formulation containing collagen became more heterogeneous, likely due to early collagen self-assembly and local fibrillar organisation. Nevertheless, the viscosities remained in the order of mPa·s, which is compatible with filling and manipulation in microfluidic environments [55]. Changes in composition were also reflected in the mechanical properties of the printed hydrogels. The apparent compressive modulus decreased from 4.24 ± 0.34 kPa for 5% GelMA to 3.32 ± 0.59 kPa after the addition of Matrigel, and to 2.11 ± 0.47 kPa after the addition of both Matrigel and collagen I (Fig. 4c). This decrease is consistent with the dilution of the photocrosslinkable GelMA network by ECM components that are not covalently incorporated into the same photocrosslinked network under these conditions. This was not interpreted as a limitation, as the objective was not to maximise stiffness, but rather to obtain a soft matrix within the range relevant to colorectal tissue and soft tumour/stromal microenvironments. Human colon tissues exhibit heterogeneous mechanical characteristics, with normal epithelium generally being softer than tumour-associated regions, and the progression of colorectal cancer is strongly influenced by the stiffness and remodelling of the ECM, commonly reported from the Pa region up to a few kPa depending on matrix composition and measurement method [37,56]. Furthermore, Matrigel itself forms very soft gels, with elastic moduli ranging from ∼10 to 300 Pa depending on protein concentration and measurement method [57]. The final ink therefore brought the system closer to that target range than GelMA alone. Swelling measurements pointed in the same direction (Fig. 4d). The percentage of swelling increased from 38.81 ± 9.59% in GelMA alone to 56.13 ± 15.29% after adding Matrigel and remained high at 59.74 ± 25.00% after the incorporation of collagen. Although these differences were not statistically significant, the overall tendency suggests a less compact network. This becomes especially relevant when considering transport. In 3D in vitro models, hydrogels must allow the exchange of nutrients and oxygen while also permitting the removal of metabolic waste, and these processes are strongly influenced by network porosity and mesh density [58]. Rhodamine B diffusion followed the same trend as swelling (Fig. 4e): the apparent diffusivity increased from 2.17 ± 0.33 × 10−6 cm2 s−1 in GelMA to 4.21 ± 0.34 × 10−6 cm2 s−1 in GelMA + Matrigel and 4.95 ± 0.47 × 10−6 cm2 s−1 in GelMA + Matrigel + collagen. These values are in the same order of magnitude as those reported for small fluorescent tracers in soft hydrogels. For comparison, Shkilnyy et al. measured Rhodamine B diffusion in fibrin around 3.43 ± 0.25 × 10−6 cm2 s−1 [42], while Dogan et al. reported that Rhodamine B transport in GelMA decreases as the network becomes denser [43]. The SEM images provided a complementary structural comparison between the formulations (Fig. S7). GelMA exhibited the most compact freeze-dried microstructure, whilst GelMA + Matrigel and GelMA + Matrigel + collagen displayed a more open and heterogeneous architecture (Fig. 4f and S7). These images are useful for comparing relative differences between samples prepared under the same drying conditions but should not be interpreted as direct measurements of native hydrated pore size. Lyophilisation can substantially distort the hydrogel architecture, and SEM-derived pore structures may differ from the hydrated network observed under aqueous conditions [59,60]. Within this limitation, the SEM images were consistent with the mechanical, swelling and diffusion data: the incorporation of Matrigel and collagen produced a softer, more hydrated and more permeable matrix. Finally, COMSOL simulations were used to assess whether the measured transport properties were consistent with static culture over the timescale of the biological assays. In static hydrogels, transport is dominated by diffusion, and Fick's law provides the simplest way to describe how concentration gradients relax with time [61,62]. The first model simulated diffusion equilibrium within the printed hydrogel domains using experimentally measured rhodamine B diffusivities (Fig. 4g). GelMA alone showed the slowest equilibrium in the centre of the structure, whilst formulations containing Matrigel and collagen reached equilibrium more rapidly, consistent with their higher apparent diffusivity. The second model included glucose diffusion coupled with cellular consumption under different seeding densities (Fig. 4h). Glucose was selected because it is a central nutrient for cell metabolism in 3D culture, and its depletion can become limiting in static constructs as cell number increases [63]. Cellular uptake was described with a Michaelis-Menten term, which is commonly used for nutrient consumption because it accounts for the reduction in uptake rate as substrate concentration falls [45]. As expected, low cell densities maintained greater glucose availability over four days, whilst higher densities led to faster depletion within the hydrogel domains. Within this simplified modelling framework, 1 × 106 cells·mL−1 was a conservative working density for subsequent cell-printing experiments. These simulations were not intended to reproduce all aspects of tumour metabolism, but rather to verify that the geometry of the selected structure, the hydrogel formulation and the cell density were compatible with short-term static culture. Bioprinting Once the composition of the bioink had been established, the next objective was to determine whether its properties could be modulated without reformulating the material. This is important because altering the concentration of GelMA, Matrigel or collagen would affect not only the stiffness, but also the ligand density, viscosity, diffusion and printing behaviour. One of the main advantages of DLP bioprinting is the ability to modify local cross-linking density through greyscale projection, while maintaining the same bioink and the same geometry [30,64,65]. Firstly, a greyscale calibration pattern was projected to measure the irradiance received on the printing plane at intensities ranging from 10% to 100% (Fig. 5a). The relationship between the projected greyscale level and the measured irradiance was approximately linear across the entire range analysed (Fig. S8). Next, four printing conditions were selected for the biological experiments: 100%, 75%, 50% and 25% projected intensity (Fig. 5b and c). These conditions yielded irradiances of 6.776, 4.986, 3.513 and 1.898 mW cm−2, respectively (Fig. 5d). When using the final GelMA-Matrigel-collagen bioink, the decrease in projected intensity reduced the apparent compressive modulus from 2.11 ± 0.47 kPa at 100% to 1.58 ± 0.32 kPa at 75%, 0.84 ± 0.06 kPa at 50% and 0.43 ± 0.10 kPa at 25% (Fig. 5e). Swelling values were increasing across the same conditions (Fig. 5f), but no significant differences in swelling were detected, although the 25% condition showed the highest mean value. The apparent Rhodamine B diffusion coefficient increased from 4.95 ± 0.47 × 10−6 cm2 s−1 at 100% to 5.08 ± 0.30 × 10−6 cm2 s−1, 5.47 ± 0.38 × 10−6 cm2 s−1 and 6.21 ± 0.21 × 10−6 cm2 s−1 at 75%, 50% and 25%, respectively (Fig. 5g). Diffusion at 25% greyscale was significantly higher than at 100% and 75%. This confirmed that the same bioink could be printed across a range of apparent compressive moduli by changing the light dose without altering its biochemical composition. However, the accompanying changes in apparent Rhodamine B diffusivity demonstrate that projected intensity modifies multiple crosslinking-dependent properties. The lower stiffness values approached very soft conditions similar to those of the ECM, whilst the higher values remained within the low kPa range associated with soft tumour and stromal environments in colorectal cancer models and human colon tissue measurements [[59], [60], [61]]. Therefore, this optical tuning strategy allowed for mechanical modulation whilst preserving the same ECM composition, enabling comparison of light-defined matrix conditions without introducing composition-dependent differences. Optical tuning of hydrogel properties by greyscale-controlled DLP exposure. (a) Projected greyscale calibration pattern used to measure irradiance over the full range of greyscale levels, from 10% to 100% in 10% increments. (b) Representation of the projected greyscale mask used to modulate light dose within the same printing layer. Brighter masks correspond to higher projected light intensity. (c) Summary of the greyscale levels, delivered light doses, apparent compressive moduli, swelling percentages and apparent Rhodamine B diffusion coefficients. (d) Light dose delivered at each selected greyscale level. Light dose was calculated from the irradiance measured at the printing plane and the fixed exposure time. (e) Apparent compressive modulus of the final bioink printed under each greyscale condition. (f) Swelling percentage of the corresponding printed hydrogels. (g) Apparent Rhodamine B diffusion coefficient under each greyscale condition. (h) Projected masks used for sequential printing of the multicompartment constructs. The first mask defines the central HCT116 region, the second the HUVECs intermediate ring, and the third the outer region. Data in (e-g) are presented as mean ± SD from three independent experiments. Each independent experiment included three separately printed replicate samples per greyscale condition (n = 3). Statistical analysis was performed using ordinary one-way ANOVA followed by Tukey's multiple-comparisons test (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). Scale bars in (h): 1 mm. The same greyscale strategy was then applied to multicompartmental structures loaded with cells. Hydrogel mechanics and transport are closely linked to cell behaviour in 3D. GelMA hydrogels are highly hydrated and permeable to small molecules, yet the rate of diffusion strongly depends on their crosslink density, porosity, and degree of methacrylation, as well as on the molecular size of each solute [58]. In living tissues, oxygen diffusion is only effective over short distances, cells are rarely more than 100-200 μm from a capillary, so thicker, non-perfused constructs can develop hypoxic or nutrient-depleted cores over time [66]. These gradients do not only limit cell survival, they can also influence how cells spread and reorganize within a matrix. In 3D co-culture systems, cancer cells have been shown to migrate toward endothelial or stromal cells in response to chemotactic signals and other soluble factors secreted by the surrounding healthy tissue [[63], [64], [65],67]. The available literature suggests that, in a co-cultured GelMA bioprinted construct, tumour cells will tend to migrate towards regions populated by healthy cells due to these biochemical signal gradients and favourable conditions. Consequently, it is expected that HCT116 cells will gradually move closer to healthy cell aggregates over time, provided there is sufficient attractant stimulus and adequate porosity in the hydrogel to facilitate their movement. Projection masks were designed to print a central core of HCT116 cells, a surrounding ring containing HUVECs, and an outer ring of acellular hydrogel (Fig. 5h). The central tumour region had a diameter of 0.5 mm, the HUVEC ring had a radial thickness of 100 μm, and the outer acellular ring completed a final structure diameter of 1 mm. Tumour-only controls were printed using the same geometry but replacing the ring regions with acellular hydrogel. This design allowed for the comparison of two variables: matrix properties, controlled by greyscale exposure, and cellular organisation, controlled by the presence or absence of the endothelial-cell compartment. The aim of this experiment was not to present a fully established model of migration or invasion, but to test whether the printed platform could monitor the early redistribution of tumour cells from a defined initial geometry. In three-dimensional hydrogels, cell movement and spread are influenced by matrix mechanics, porosity, confinement, and soluble or matrix-associated signals [68]. Therefore, it was expected that the redistribution of HCT116 would depend both on the mechanical state of the hydrogel and on the presence of the neighbouring compartment containing HUVECs. Immediately after printing, fluorescence images confirmed that the intended spatial organisation had been achieved. HCT116 cells were initially confined to the central printed region, whilst HUVECs were located in the surrounding compartment under co-culture conditions (Fig. 6a). This Day 1 configuration served as a baseline for quantifying subsequent changes in the distribution of HCT116 cells. After four days, the signal from the HCT116 cells was no longer confined to the initial tumour core, and fluorescent signal was detected in the surrounding hydrogel, indicating a time-dependent redistribution of the tumour cell population (Fig. 6b and f). 3D bioprinted model for monitoring tumour-cell redistribution over time. (a) Fluorescence image immediately after printing, showing the initial spatial organisation of HCT116 cells (green, CMFDA) in the central core and HUVECs (orange, CMTMR) in the surrounding compartment. (b) Fluorescence image after 4 days in culture, illustrating the outward redistribution of HCT116 cells from the initially printed region. (c) Quantification of the HCT116 redistributed area outside the initial tumour core at day 4. (d) Mean redistribution distance of HCT116 cells relative to the initial printed core. (e) Live/Dead viability analysis of the printed cells at day 4 under the different greyscale conditions. (f) Fluorescence and brightfield images of the printed constructs at day 1 and day 4 for the different greyscale conditions and culture configurations. Scale bar: 200 μm. Data in (c,d) are presented as mean ± SD from three independently printed chips per culture configuration (n = 3). Each chip contained one construct at each projected intensity. Data were analysed by ordinary two-way ANOVA followed by Šídák's multiple-comparisons test. Data in (e) are presented as mean ± SD from three independently printed constructs per cell type and projected intensity (n = 3). HCT116 and HUVEC viability were analysed separately by one-way ANOVA followed by Tukey's multiple-comparisons test (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). Quantification was performed using a pixel-based image analysis strategy adapted from approaches developed for objective analysis of 3D spheroid invasion, where area- and distance-based metrics are used to describe cell spreading at the population level rather than tracking individual cells [46]. In the present system, the HCT116-positive area outside the initial printed core was quantified as the redistributed area, and the mean redistribution distance was calculated relative to the initial boundary (Fig. 6c and d and S9). A qualitative comparison of the images from day four revealed a signal for HCT116 that was limited to the outer regions in constructs containing tumour cells only, whereas constructs containing HUVECs showed more extensive regions positive for HCT116, particularly at intermediate projected intensities. These observations were then quantified using redistributed area and mean radial redistribution distance as complementary metrics at the population level. Quantitative analysis showed that HCT116 redistribution was associated with both projected intensity and the presence of an adjacent HUVEC-containing compartment. In the HCT116-positive redistributed area, two-way ANOVA revealed a significant interaction between projected intensity and culture configuration (p = 0.0076), alongside significant main effects of projected intensity (p = 0.0011) and culture configuration (p < 0.0001). In tumour-only constructs, the redistributed area remained low across all greyscale conditions, indicating that changing greyscale exposure alone produced only limited outward expansion under the current static culture conditions. In contrast, constructs containing the HUVEC compartment showed a marked increase in the redistributed area at intermediate light doses, with significant differences between the tumour-only and HUVEC-containing constructs at 50%, 75% and 100% projected intensity. The strongest response was observed at 75%, whilst the 100% condition showed a pronounced reduction despite corresponding to the highest apparent compressive modulus. A similar pattern was observed in the mean redistribution distance. The presence of the adjacent HUVEC-containing compartment significantly increased the overall mean redistribution distance (culture-configuration main effect, p < 0.0001), from 48.03 μm in tumour-only constructs to 101.2 μm in HUVEC-containing co-cultures. The highest mean distance was observed in the HUVEC-containing co-culture at 75% projected intensity, and this condition was significantly greater than its tumour-only counterpart. However, the interaction between projected intensity and culture configuration was not significant for this metric (p = 0.2551), indicating that the presence of the HUVEC-containing compartment was associated with a strong overall increase in radial HCT116 redistribution, whereas the influence of projected intensity was more clearly captured by the redistributed-area measurement. It is important to note that the differences in the redistribution of HCT116 were not due to a significant loss of viability caused by the printing conditions. Day-4 Live/Dead analysis showed high viability in both independently printed HCT116-and HUVEC-laden constructs across the four projected intensities, with no significant intensity-dependent differences for either cell type (Fig. 6e and Fig. S10–S11). This indicates that the selected light doses were compatible with short-term culture and that the redistribution patterns observed were more likely associated with the properties of the matrix and the multi-compartmental organisation than with cytotoxicity. Conclusions In this study, we have developed a 3D bioprinted tumour-on-chip platform with adjustable properties for the spatial organisation and short-term monitoring of the redistribution of colorectal cancer cells. The system integrates a custom resin-printed microfluidic chip, in situ DLP bioprinting, a bioink of GelMA, Matrigel and collagen and greyscale-controlled modulation of hydrogel mechanics, all within a single experimental workflow. The selection of resins identified BioMed Clear as the most suitable material for device fabrication, as it offers an appropriate balance between geometric fidelity, chip handling and cytocompatibility following post-processing. The final device reproduced the intended dimensions of the chambers and channels, allowed optical access for fluorescence imaging, and supported sequential multi-compartment printing directly within the chip. The selected bioink formulation, comprising 5% GelMA, 15% Matrigel and 0.2% collagen I, provided a soft, permeable matrix with improved diffusive transport compared to GelMA alone. It is important to note that greyscale-controlled DLP exposure allowed the apparent compressive modulus of the same bioink to be adjusted from 0.43 to 2.11 kPa without altering the chemical composition of the matrix. This combination of material selection, optical tuning and design enabled the construction of simplified three-dimensional tumour models directly on the chip. The platform's main strength lies in its ability to define and rearrange different cellular and material configurations precisely within a single geometry by locally controlling the light dose and the resulting hydrogel properties and combining different bioinks according to the experimental question. This provides a readily reconfigurable system for controlled comparison of matrix and co-culture conditions within a single microdevice. The aim is to provide a versatile, controlled tool for studying how the architecture of the microenvironment influences tumour behaviour in a simpler way. The presence of the HUVEC-containing compartment was associated with greater HCT116 redistribution, supporting the relevance of tumour-endothelial interactions in future versions of the model. Future work should focus on extending the model towards dynamic perfusion, longer culture times, endothelial maturation, permeability measurements and molecular readouts of matrix remodelling and paracrine signalling. Sequential printing could also support the future incorporation of drug-loaded hydrogel compartments or nanocarriers, consistent with emerging hydrogel-based approaches for CRC drug delivery [69]. These developments would enable the platform to progress from a redistribution assay towards more mechanistic studies of migration, invasion, endothelial interaction and therapeutic response under better-defined tumour-microenvironmental conditions. CRediT authorship contribution statement Adrian García: Writing – original draft, Investigation, Data curation. Lia Jove: Investigation. María Pereira: Investigation. Angelica Figueroa: Writing – review & editing, Conceptualization. Daniel Nieto: Writing – review & editing, Supervision, Investigation, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Daniel Nieto reports financial support was provided by University of A Coruña. Daniel Nieto reports a relationship with University of A Coruña that includes: employment. Daniel Nieto has patent No pending to No. N/A If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements DN thanks to the ERC Consolidator grant (101125172) Holographic Optical Tweezers Bioprinting (HOTB): Towards precise manipulation of cells for multi-scale vascularized tissues/organ printing. Daniel Nieto is supported by the Opportunius Programme (Xunta de Galicia) since 2024. Thanks to Healthbiolux for technical support with the DLP bioprinter development. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.mtbio.2026.103646. Supplementary data The following are the Supplementary data to this article: Multimedia component 1 Multimedia component 2 Data availability No data was used for the research described in the article. References Associated Data Supplementary Materials Multimedia component 1 Multimedia component 2 Data Availability Statement No data was used for the research described in the article.