Bioprinted cancer-stromal in-vitro models in a decellularized ECM-based bioink exhibit progressive remodeling and maturation

Constant matrix remodeling and cellular heterogeneity in cancer are key contributors to its development and can profoundly alter treatment efficacy. Developing in-vitro models containing relevant features that can recapitulate these aspects of the tumor microenvironment and that are well characterized can circumvent the limitations of conventional 2D cultures and animal models. Automated fabrication methods combined with biomimetic biomaterials have provided the opportunity to create platforms that can potentially incorporate a heterogeneous population of cells in a 3D environment that allows cell–cell and cell-ECM interactions with reproducibility. This study used 3D extrusion bioprinting and a composite bioink containing a reinforced decellularized extracellular matrix (ECM) hydrogel to fabricate a head and neck cancer in-vitro model. The constituents of this model included fibroblasts and active ECM proteins to represent the stroma, along with HNSCC cells to represent the tumor component. The topographical characterization of the bioink showed a fibrous network with nanometer-sized pores. After cell encapsulation and model fabrication, we observed spheroid development and growth over time with cancer cells in the core and fibroblasts in the periphery. Our model is compatible with matrix metalloproteinase (MMP) quantification techniques and showed significant differences in the presence of MMP-9 and MMP-10 compared to the control groups. This characterized model is proposed as a tool for further translational and drug discovery applications since it provides a biomimetic scenario that allows the study of the tumor microenvironment in-vitro using nondestructive longitudinal monitoring over time.


Introduction
During tumorigenesis, neoplastic cells in solid tumors interact with cellular and non-cellular components present in the stroma, forming the tumor microenvironment (TME) [1,2]. The TME is an environment comprised of a heterogeneous population of cancer cells and the stroma, a collection of cell types recruited by the cancer cells and non-cellular components that aid in tumor development [1,3].
The TME works in synergy through two-way communication to promote cancer progression by providing nutrients and constantly remodeling the extracellular matrix (ECM) physically and biochemically through tumor-stromal interactions [4,5]. Endothelial cells, fibroblasts, and immune cells, among other cell types, are generally present in the stroma interacting with a dysregulated ECM [3,6,7]. Fibroblasts represent an abundant cell population in the TME and are known to be the main contributors to ECM changes over time [8]. They can be transformed into cancerassociated fibroblasts (CAFs), which behave in an abnormal manner [1,8,9]. Fibroblasts are known to synthesize structural proteins such as collagens and fibronectin, which modify the mechanical properties of the environment [8]. Also, fibroblasts can secrete non-structural proteins like cytokines, growth factors, and matrix metalloproteinases (MMPs) that can degrade or break mature regions of the ECM to favor cancer metastasis [8,9]. In healthy tissues, MMPs are tightly regulated with tissue inhibitors of metalloproteinases (TIMPs), and increased expression of TIMPs has also been linked to cancer progression, in particular TIMP-1, which has been proposed as a possible marker for cancer treatment [10][11][12]. The ECM is an essential actuator in tumorigenesis [6], and it has been shown to affect the sensitivity of drug treatments since it acts as a physical barrier and can sequester drugs due to affinity making it more difficult to penetrate the tumor region [13,14].
Three-dimensional in-vitro culture models have become popular as an alternative to traditional animal preclinical models to test and better understand disease [5,[15][16][17]. These models intend to recapitulate, in a controlled manner, variables that mimic the conditions found in-vivo [5,16]. The ideal 3D in-vitro model must include a 3D environment with relevant architecture and dimensions, a matrix resembling the ECM of the tissue in question, and relevant cell populations for the disease studied [17]. Several biomaterials have been proposed to mimic the ECM in-vitro [18]. Specifically, decellularized ECM (dECM) hydrogels are of great interest for their bioactive capabilities [19]. dECM hydrogels are thermally dependent tissue-derived biomaterials produced through a decellularization and solubilization process [20]. Collagens, glycosaminoglycans, laminins, cytokines, growth factors, and vesicles have been detected in dECM matrices [21][22][23][24][25]. These bioactive components allow cellular adhesion, development, and differentiation [20,21]. Mechanical properties of dECM tissues, such as stiffness, decrease after decellularization [26]. Then, the solubilization process uses enzymes to structurally break the tissue and form dECM hydrogels [27]. These processing steps significantly change the structure of the final product, making it challenging to use in conjunction with automated fabrication techniques such as extrusion bioprinting. dECM-containing blends have been proposed to keep the bioactive benefits provided by the ECM while adding reinforcements to create a material that has mechanical properties similar to the tissue of study and compatible with in-vitro culture fabrication [28][29][30].
Here, we use a bioink blend containing alginate, gelatin, and dECM derived from porcine tongue that has been previously characterized and which promoted tumor formation of encapsulated cancer cells [28]. Spheroid formation was not observed in the pure dECM [28]. We used this composite hydrogel in conjunction with head and neck cancer (HNSCC) cells and fibroblasts to fabricate and characterize a bioprinted heterogeneous in-vitro HNSCC model that is consistent and scalable. This was achieved using extrusion bioprinting since it allows the fabrication of reproducible, high-fidelity constructs with comparable cell densities in an automated and serial manner [31,32].
These models develop over more than 20 d showing cell reorganization, spheroid formation, and matrix remodeling through time. They allow their nondestructive study and quantification of collagen and small molecules such as MMPs through time. They are proposed as an in-vitro alternative to study cancer-fibroblast interactions through time. We propose this platform not only to study HNSCC but as a model that can be tailored to different neoplastic diseases.

Bioink fabrication
We used A 1.5 G 5 dECMT, a combination of alginate, gelatin, and dECM derived from porcine tongue (dECMT), as the bioink for cell encapsulation. Detailed formulation methods and characterization of this blend have been previously published [28].
Our composite bioink nomenclature follows the format: A x G y dECMT, where 'x' corresponds to the w/v percentage of alginate, and 'y' corresponds to the w/v percentage of gelatin. To prepare A 1.5 G 5 dECMT, a 9% (w/v) solution of sodium alginate (FMC biopolymer) in DPBS-1X (Wisent Bio Products) was mixed overnight at RT with the neutralized dECMT to obtain a final concentration of 1.5% alginate. Next, type B gelatin from bovine skin (gel strength: 300, G2500, Sigma-Aldrich) was added to the alginate-dECMT solution at a 5% (w/v) final concentration. The neutralized dECMT has a protein concentration of f 180 ± 80 µg ml −1 [28]. The final dECMT protein concentration in the A 1.5 G 5 dECMT is 160 µg ml −1 [28]. A 1.5 G 5 dECMT was kept at −20 • C for long-term storage or at 4 • C if used within 4 weeks.

Scanning electron microscopy (SEM)
Both pure dECMT and A 1.5 G 5 dECMT were mold-casted without encapsulating cells in them. A 1.5 G 5 dECMT was crosslinked with CaCl2 (100 mM, Sigma-Aldrich) for 3 min before immersing the samples in supplemented media for 24 h at physiological conditions.
After culture, 3D printed discs were fixed with 4% PFA for 30 min, washed with PBS-1x, and gradually dehydrated up to 100% ethanol. Samples were placed in a critical point dryer (EM CPD030, Leica), and 20 × 1 min cycles of CO2 exchange were performed. Discs were attached to SEM specimen studs using carbon tape and sputter coated with an 8 nm layer of platinum (EM ACE600, Leica). Samples were stored in a desiccator at room temperature prior to SEM acquisition. SEM images were acquired using an Environmental Scanning Electron Microscope (Quanta 450, FEI).

Pore size fiber alignment and diameter
Sample porosity was quantified using ImageJ. The 'Auto threshold' tool was applied to a ROI (region of interest) using the Huang method. Then, images were binarized before analyzing the particle size using the 'Analyze Particle' tool. The output data included the area of the particles detected, which was considered the pore area. Data analysis is detailed in the statistical analysis section. Four samples were imaged per experimental group. At least four ROIs were selected per group yielding more than 400 pores measured per experimental group.
The diameter of the fibers was measured by selecting a ROI with a single fiber from the SEM images. The ROI was pre-processed using an imageprocessing algorithm developed by D'Amore et al [33] with MATLAB (The MathWorks). Briefly, the algorithm increases the contrast using image equalization, reduces the noise while preserving structure edges using three-by-three median filtering, and separates the outer fiber network from the background with a global histogram threshold using Otsu's method [34]. The pre-processed images were then analyzed using Image J by overlaying the original region of interest of the SEM image with the pre-processed image obtained in MATLAB to verify the accuracy of the fiber edges after binarization and were adjusted manually if necessary. The overlay was then removed. Finally, after proper calibration of the scale, the fiber diameter was measured by drawing a line across the fiber using the plot profile command. More than 40 fibers were measured per experimental group.
Fiber orientation and alignment were quantified by vectorizing the SEM images using Inkscape. The vectorized and original images were overlapped in Image J to verify the accuracy of the fiber edges. The images were then cropped in 2 × 2, 3 × 3, and 4 × 4, depending on the resolution, and the final images were analyzed using the Orientation J plugin. The dominant direction command in Orientation J returns the dominant fiber orientation and the alignment as a coherency value. The prevailing direction is in degrees ( • ) with values between −90 • and 90 • [35]. The coherency is a value between 0 and 1 measuring the anisotropic properties of the region of interest. A coherency of 1 indicates that the structure has a dominant direction, and a coherency of 0 indicates that the network is isotropic with no preferred direction of alignment [35]. Four samples were imaged per experimental group. From the data obtained, more than 20 ROI were measured for each group.

2D cell culture and lentiviral transduction
An immortalized human head and neck cancer cell line derived from the base of the tongue (UM-SCC-38) [36] and immortalized human vocal fold fibroblasts (A8-HVFFs) [37] were used for this study. Stably transduced red fluorescent protein (RFP)expressing UM-SCC-38 (RFP-UM-SCC-38) was generated to enable long-term, non-invasive cell tracking in the 3D bioprinted cultures. Membrane-targeting monomeric RFP (Addgene plasmid # 32604) was cloned into a pHIV-blasticidin lentiviral vector. pCAG:myr-mRFP1 was a gift from Anna-Katerina Hadjantonakis of the Memorial Sloan Kettering Cancer Center [38]. Engineered lentivirus bearing the above-mentioned vector was packaged in 293 T cells and used to generate a UM-SCC-38 cell line that expresses RFP. Cells were infected with the engineered lentivirus in media supplemented with polybrene (8 µg ml −1 ) for 24 h. Cells with stable RFP expression were selected via media supplementation of selection antibiotic blasticidin (Invitrogen) at 6 µg ml −1 for two weeks. Desired RFP expression level in the final RFP-UM-SCC-38 cell line was selected via fluorescence-activated cell sorting and validated via fluorescent microscopy.
Non-transduced cells were cultured in DMEM with 10% fetal bovine serum, 1% Penicillin -Streptomycin, and 1% non-essential amino acids. Media and supplements were purchased from Wisent Bio Products. RFP-UM-SCC-38 used the same media formulation with the addition of blasticidin to maintain homogeneity and persistence of RFP transgene expression. Cells were cultured in traditional 2D conditions (37 • C and 5% Co 2 ) until they reached 85% confluency and passaged using trypsin-EDTA (Wisent Bio Products) to disrupt cell attachments.

Fabrication and culture of 3D bioprinted models
Cells were encapsulated in A 1.5 G 5 dECMT at a final concentration of 10 million cells per ml. For the co-culture experiments, a 2:1 ratio of immortalized human vocal fold fibroblasts (HVFF)s:UM-SCC38 was used. Before encapsulation, a syringe with A 1.5 G 5 dECMT was placed in a 37 • C water bath for at least 30 min to liquefy the gelatin in the composite bioink. Cells cultured in 2D conditions were detached using trypsin-EDTA (Wisent Bio Products), counted, and centrifuged. The cell pellet was resuspended in media accounting for less than 1% of the total volume, and the warm bioink was loaded into a 3 cc 3D printer-compatible cartridge. The cells were pipetted into the warm bioink and mechanically mixed until evenly distributed. The cartridge was centrifuged at 200 g for 2 min to remove air bubbles and later incubated for 15 min at room temperature to allow the gelation of the gelatin in the bioink to occur. Discs of 5 mm diameter and 500 µm height were bioprinted using a Bioscaffolder 3.1 (GeSiM). A 22 G conical tip (Nordson) was used to fabricate the discs at room temperature (24 • C) with a pressure of 45 ± 10 kPa and a printing speed of 10 ± 2 mm s −1 . After printing, samples were crosslinked with a 100 mM aqueous solution of calcium chloride for 3 min. Samples were rinsed with PBS-1X and placed in culture dishes. Cell media was changed every 3-4 d depending on the experiment. Cell-free A 1.5 G 5 dECMT discs were printed as controls.

Confocal microscopy RFP-transduced UM-SCC-38 and non-fluorescent
HVFFs were used to observe cell development over time. Experimental groups included monocultures of HVFFs, UM-SCC-38, and the co-culture 2:1, HVFF: UM-SCC-38. Prior imaging samples were immersed in Calcein-AM (Invitrogen) and Hoechst 33 342 (Invitrogen) in DPBS-1X at a final concentration of 2 µm and 18 mM, respectively, and incubated at 37 • C for 30 min, covered from the light. Samples were washed with DPBS-1X and imaged while encapsulated in a humidity chamber set to 37 • C and 5% CO 2 . Microscopy was acquired with a Nikon A1+ confocal microscope capturing the Z-stack (5 µm step size). During analysis, cells with a double positive signal were considered cancer cells, and cells only positive with Calcein-AM were considered fibroblasts. All image analysis was performed in ImageJ [39] using three discs per time point and per experimental group (n = 3).

Collagen quantification
Sircol collagen assay (S1000, Biocolor) was used to quantify collagen over time. Three-dimensional printed models were harvested every three days and frozen at −80 • C until ready to use. We used the manufacturer's protocol with some modifications. We followed the acid-pepsin protocol provided by the company to extract the soluble collagen from the samples. Due to the nature of our samples, we added 100 µl of a 55 mM trisodium citrate to decrosslink the alginate in the models, increasing the surface area for collagen extraction. Then, we performed the suggested collagen concentration protocol and the Sircol Assay. Samples were plated in a 96-well plate, and absorbance was measured at 555 nm. Experimental groups included HVFFs, UM-SCC-38, and the co-culture 2:1, HVFF: UM-SCC-38 at a final concentration of 10 million cells per ml and cell-free A 1.5 G 5 dECMT. Data were plotted using JMP Pro 16 (JMP), and the cubic spline method was used for smoothing the data with a λ of 0.033.

MMPs quantification
Samples were bioprinted, ionically crosslinked, and cultured under standard conditions (37 • C, 5% CO 2 ) for the duration of the experiment. Media was changed every three days. Three-day-old media was harvested, centrifuged at 400 g for 15 min to remove cellular debris, and the supernatant was frozen at −80 • C until ready to use. The multiplexing analysis was performed using the Luminex™ 200 system (Luminex) by Eve Technologies Corp. (Calgary, Alberta). Thirteen markers were simultaneously measured in the samples using Eve Technologies' Human MMP/TIMP 13-Plex Discovery Assay®, which consists of two separate kits; one 9 plex and one 4 plex (R&D Systems, Inc.) according to the manufacturer's protocol. The 9-plex consisted of MMP-1, MMP-2, MMP-3, MMP-7, MMP-8, MMP-9, MMP-10, MMP-12 and MMP-13. The 4-Plex consisted of TIMP-1, TIMP-2, TIMP-3, and TIMP-4. Assay sensitivities of these markers range from 0.28-253 pg ml −1 for the 13-plex. Individual analyte sensitivity values are available in the product datasheet for the 4-Plex and by building the panel on the R&D Systems Magnetic Luminex Performance product page for the 9-Plex. At least six samples per experimental group were measured.

Hydrogel topography and characterization
We previously developed and characterized a composite bioink (A 1.5 G 5 dECMT) composed of alginate, gelatin, and dECM derived from porcine tongue [28]. This blend has a comparable Young modulus to a HNSCC tumor xenograft [28]. Here, we further characterize this material by looking into the topography using SEM and comparing it with the pure dECMT hydrogel (figure 1). We are interested in determining the topographical differences between these two materials. SEM enabled the visualization of randomly interwoven fibers that have been previously observed in other dECM hydrogels and characterized as a self-assembled network of mainly collagen fibrils ( figure 1(a)) [23,27]. There is a significant difference in pore size, fiber diameter, coherence, and fiber orientation when comparing the pure dECMT hydrogel and A 1.5 G 5 dECMT (figures 1(c)-(f)). The reinforcements used to enhance the dECMT hydrogel changed the topography of the constructs making them more heterogeneous.
Pore size mean values have been shown to decrease when the concentration of dECM is increased in the sample [40]. For dECMT and A 1.5 G 5 dECMT, the average pore size is 165.9 ± 65.7 nm and 153.3 ± 84.3 nm, respectively ( figure 1(c)). Both experimental groups are skewed to the right 0.25 and 0.28, respectively. The geometric means of dECMT and A 1.5 G 5 dECMT are 151.2 nm and 126.9 nm, respectively. Examples of mean pore size for dECM hydrogels are 112 nm for small intestinal submucosa gels [23] and 152-670 nm for brain, depending on the dECM concentration [40], etc [41]. By looking at the frequency distribution; we can determine that 48% of the pores in the dECM sample are between 100-160 nm. For A 1.5 G 5 dECMT, there are two regions to highlight. 31.2% of the pores are between 1-80 nm, and 38% are between 140-220 nm. We can attribute the differences in porosity to the presence of additional components in the bioink blend group. Alginate and gelatin occupy space within the dECM fiber network, causing a decrease in pore size.
Fiber diameter quantification shows a significant difference between samples ( figure 1(d)). dECMT and A 1.5 G 5 dECMT have mean values of 70.3 ± 13 nm and 102.6 ± 44.9 nm, respectively. They present a right skewness of 0.28 and 0.2, respectively. The geometric means of dECMT and A 1.5 G 5 dECMT are 69.1 nm and 91.9 nm. For the dECMT group, 100% of the measurements are between 50-100 nm. For A 1.5 G 5 dECMT, we observe two regions in the frequency distribution. 28.6% of the measurements are between 30-50 nm, and 59.2% of the measurements are between 90-140 nm. Our dECM data resemble measurements previously reported [42,43]. The mean diameter of collagen fibrils can range between 40-80 nm in mammals, but they can measure up to 500 nm [42,43]. Specifically for dECM hydrogels, an average fiber diameter of 92-112 nm for myocardial matrices [44], 74 nm for sub intestinal submucosa matrices [23], 130-140 nm for brain matrices [40] has been measured.
Fiber orientation is shown in a −90 • -+90 • range ( figure 1(e)). For dECMT, the mean value is −22.1 ± 26.4 • with 80% of the measurements showing a preferential orientation between −10 • and −60 • . For A 1.5 G 5 dECMT, 44% of the measurements show an orientation between −30 • and −60 • . However, 36% of the population shows the opposite orientation preference, between 50 • and 80 • . These results show that dECMT has a more consistent alignment across different samples and regions when compared to A 1.5 G 5 dECMT. The bioink shows two groups with opposite orientations, suggesting that this material has a more heterogeneous alignment.
The coherence values are a representation of the isotropy of the samples. If the values tend to 1, the region of interest has a dominant orientation. However, if the values tend to 0, the image is isotropic. Coherence values are 0.15 ± 0.1 for dECMT and 0.11 ± 0.1 for A 1.5 G 5 dECMT, indicating that the bioink samples are more isotropic ( figure 1(f)). The fiber orientation data paired with the coherency values show a higher organization in dECMT samples characterized by consistent fiber alignment and higher coherency values. Overall, both matrices have a high degree of isotropy, which has also been observed in other tissues [45]. However, when comparing both samples, the decrease in organization observed in A 1.5 G 5 dECMT can be mainly attributed to the incorporation of reinforcement materials. Alginate and gelatin chains are in contact with the dECMT. Changes in fiber network formation are expected when additional materials are incorporated. However, all measurements are in the same order of magnitude, demonstrating that abrupt topographical changes in the matrix did not occur when providing mechanical stability to the dECMT.

3D co-culture fabrication and spheroid development
Three-dimensional printed constructs, after crosslinked, were immersed in cell media and cultured for 22 d. Sample immersion allows the nutrient-waste exchange to happen around the entire disc. This behavior is challenging to replicate with 3D cultures in traditional well plates since the media is in contact with the sample only from the top. Monocultures of UM-SCC-38 or HVFF encapsulated at a final concentration of 10 million cells ml −1 were fabricated as controls to observe cell development over time. UM-SCC-38 culture is characterized by forming spheroids that grow up to Day 22 ( figure 2(a)). This behavior has been previously reported [28]. HVFF culture shows a spindle morphology which has been previously reported [37] (figure 2(b)).
Co-cultures constituted by a 2:1 ratio of HVFF: UM-SCC-38 with a final concentration of 10 million cells ml −1 were fabricated with the same dimensions as the monocultures. Co-cultures start as single cells, but the formation of spheroids is observed by day 7. UM-SCC-38s aggregate in the center while HVFFs wrap the cancer cells (figure 2(c)). Representative high-magnification images are presented in figures 3(a) and (b). This cellular arrangement corresponds to what is observed in the TME in-vivo. Neoplastic cells are surrounded by stromal cells that promote cancer development, progression, and metastasis [3]. Qualitative SEM images of these experimental groups are presented in figure S1. In the co-culture group, fibroblasts present both round and spindle-like morphologies, which have been previously reported in cancer-fibroblast co-cultures encapsulated in collagen matrices [46]. Spheroid organization is consistently observed for up to 22 d. Since the observation of spheroid formation (day 7), diameter quantification is shown in figure 3(c). Figure 3(d) shows the arithmetic and geometric means. There is no significant difference in spheroid size between days 7, 10, and 13. However, spheroids significantly increase in size by day 16 and maintain a similar diameter on day 19. The diameter continues to grow by day 22. The co-culture results show a consistent organizational development through time that is not observed in the monoculture group   lower fibroblast-related MMPs are expected in the co-culture group since, at day 0, the fibroblasts corresponded to 2/3 of the total population and cancer cells to 1/3. Over time, the co-culture profile diverges from the HVFF monoculture and presents MMP levels similar to the ones observed in the UM-SCC-38 monoculture. We attribute the lower MMP levels to the presence of fibroblasts and UM-SCC-38 in this group. TIMP-1, a glycoprotein that has been associated with poor prognosis in cancer [11], presents increased levels between day 4 and day 10 in the HVFF monoculture and co-culture groups.

MMPs analysis
MMP-9 and MMP-10 are highlighted in this study since these molecules have been linked with the progression and invasiveness of HNSCC. [47,48] MMP-9 levels are the highest in UM-SCC-38 across all timepoints. These results are expected since these cancer cells are cataloged as moderately welldifferentiated squamous cell carcinoma [49].
A significant increase of MMP-9 and MMP-10 levels is observed in both UM-SCC-38 and UM-SCC-38 /HVFF co-culture groups between day 4 and day 10 (figures S3 and S4). These timescales, when compared with the confocal images presented in figure 2, coincide with the transition from individual cells to spheroids in the co-culture group. The co-culture group has no significant changes in MMP-9 starting day 10. However, both monocultures continue changing their MMP-9 levels throughout the 22 d of culture ( figure  S4). MMP-10 levels remain stable for all groups from day 16 onwards (figure S4). Figure 5 compares the MMP levels of all experimental groups by day. No significant difference in MMP-9 levels is observed at day 4 between the coculture and the UM-SCC-38 monoculture. However, for the following time points, the UM-SCC-38 group consistently presented the highest levels of MMP-9, followed by the co-culture. The lowest levels of MMP-9 were always quantified in the HVFF group. There is a significant difference in MMP-9 for all groups starting day 10.
No significant difference in MMP-10 is observed at day 4 between the co-culture and the HVFF group. However, a decrease in MMP-10 occurs as the HVFF monoculture develops. MMP-10 is significantly higher across all time points in the coculture group compared to the monocultures. This can indicate that the heterogeneous culture promotes the production of MMP-10. On day 10, the coculture MMP-10 levels are five times higher than the monocultures. Overall, these results show significant differences in the co-culture showing the importance of recapitulating the heterogeneity in in-vitro models for a more biomimetic scenario.

Collagen quantification in 3D models over time
Sircol soluble collagen assay was performed for all groups, including the bioink in culture without cells ( figure 6). Statistical analysis shows no significant difference between experimental groups on days 1, 4, and 7. However, minor differences between the UM-SCC-38 group and the acellular bioink have been observed since day 10. The co-culture group becomes significantly different from the bioink group starting on day 16 (figure S5). Soluble collagen levels are maintained throughout the 22 d of culture in the HVFF and bioink groups. Significant changes are observed over time in the co-culture and UM-SCC-38 groups. All groups show comparable soluble collagen levels from days 1 and 7. However, a significant increase in soluble collagen is present in the UM-SCC-38 culture at days 10 and 13 compared to the other experimental groups. From day 16 onwards, the co-culture levels remain significantly higher than the cell-free bioink group (figures 6 and S5).

Discussion
In cancer, stromal components, cell-cell interactions, and cell-ECM interactions play a pivotal role in tumor development and progression [50]. For this reason, there is a pressing need for clinically relevant 3D in-vitro models. We need models that can recapitulate the physical and biochemical characteristics which drive and control cancer progression [5]. These bioengineering models provide a new window to understand molecular mechanisms and find and test effective therapies. The ECM is one of the primary actuators in the stroma [5]. Hence, using ECM-containing materials to create in-vitro models is appealing for their biomimetic nature. dECM hydrogels have been successfully used in tissue engineering applications and to re-create healthy and diseased tissues in-vitro [21].
Here, we used a bioink composite containing a dECM hydrogel derived from porcine tongue (dECMT) reinforced with alginate and gelatin to fabricate a heterogeneous HNSCC model. This blend has comparable mechanical properties to in-vivo tumors and has been previously used to manufacture monocultures of HNSCC cells with high viability [28]. Additionally, this bioink blend promotes spheroid formation in HNSCC cancer cells [28]. Behavior that is not observed in pure dECMT [28].
In the topographical characterization of the pure dECMT and A 1.5 G 5 dECMT, we observed fibers previously reported as a self-assembled collagen network [27]. The dECM architecture has been reported as significant since it can allow cell-matrix interactions via integrins that are key for cell proliferation and migration [44]. Differences in pore size between our samples show that the reinforcements in A 1.5 G 5 dECMT occupy space in the matrix, making the pore size of the composite smaller. However, measurements for both groups are in the same order of magnitude and within the ranges reported for dECM hydrogels from other tissues. [23,40,41] The mean fiber diameter in dECMT is within what other studies have reported. [23,[42][43][44] However, the composite bioink has a higher mean fiber diameter suggesting that the incorporation of the rheological reinforcements is the reason for the difference. Since cells can interact with the dECMT, we expect to see changes over time as they remodel their environment. The significant changes in fiber alignment and decrease in coherency observed in A 1.5 G 5 dECMT when compared to pure dECMT demonstrate that the presence of alginate and gelatin influence the orientation and organization of the dECMT network. However, both samples can be considered isotropic, which could be beneficial for replicating the ECM heterogeneity observed in tumors. Additionally, electrostatic charges are known to be present in all constituents and can heavily influence their arrangement in a composite blend [51][52][53].
In-vivo, tumor-associated collagen signatures (TACSs) have been categorized from highly isotropic (TACS-1) to highly aligned collagen fibers (TACS-3) [54]. TACs-3 has been correlated with progression and metastasis [54]. Our materials resemble TACS-1 since the coherency values are close to zero (dECMT: 0.15 ± 0.1 A 1.5 G 5 dECMT: 0.11 ± 0.1). We hypothesize that if an increase in alignment is present in our material during cell culture, it could further indicate matrix remodeling increasing the TACS category to a more developed cancer model.
The stromal content in tumors has been shown to be a good predictor of patient prognosis in solid tumors [55]. Hence, the tumor-stroma ratio has been proposed as a valuable feature to predict patient outcome in head and neck cancer [56,57]. Patients with stroma-rich (>50%) tumors had worse disease-free survival and higher mortality when compared to patients with stroma-poor (<50%) tumors [58]. In this study, we used a 2:1 ratio of stomal cells (HVFFs):HNSCC cancer cells (UM-SCC-38) to mimic the in-vivo proportions of a stromal-rich tumor. We encapsulated UM-SCC-38 and HVFF cells in the bioink and 3D-printed monocultures or co-cultures. Cell attachment and proliferation are observed in all groups. However, architecturally relevant organoid regeneration is heavily promoted in the co-culture. It has been shown that fibroblast-cancer cell cultures have interactions that promote reciprocal activation in growth rate, ECM expression, etc [59]. Tumor spheres surrounded by fibroblasts are observed since day 7 of culture and continue to grow, reaching a mature and stable state by day 22. This organizational development can be attributed to the crosstalk between the two cell types, which is nonexistent in the monoculture groups [60]. This architecture commonly observed in cancer in-vivo indicates that this heterogeneous model has biomimetic characteristics and can provide more accurate results than traditional 2D monocultures. Studies done in fibroblasts cultured in 3D collagen matrices show diverse morphologies from highly 'activated' to 'quiescent' phenotypes [61]. Fibroblast spindle morphology in collagen co-cultures has been linked to higher invasion trajectories which indicate cell motility [46]. While round morphology, present in fibroblasts at rest, has been linked to low proliferation, low motility, and low ECM deposition [61,62]. We observed both spindle and round morphologies in fibroblasts co-cultured with cancer cells. These behaviors have been previously reported in 3D in vitro co-cultures of HNSCC and fibroblasts [46]. Understanding of fibroblast morphology in 3D environments in-vitro and their behavior during neoplasia is still in development [61,63]. We foresee our model as a tool that could be used to study the synergies between fibroblasts and the tumor milieu in a three-dimensional and biologically relevant setting.
MMP and TIMP analysis for all experimental groups shows similar MMP and TIMP signatures observed at the beginning of the culture between the co-culture and the HVFF monoculture, indicating HVFF secretions dominate the co-culture behavior at early stages. However, at the endpoint (day 22), the co-culture signature resembles the UM-SCC-38 monoculture indicating that the cancer cells dominate the co-culture towards the end. This staged cell domination can be attributed to an environment adaptation at early points of the culture to allow tumor spheroid formation to dominate after. In the early stages of cancer, fibroblasts promote tissue repair, leading to TME remodeling [64]. As cancer cells develop, CAFs later switch to tumor promoters since CAF-secreted growth factors are used by cancer cells in their survival and proliferation [64]. A specific tipping point where the shift happens is challenging to identify, but a gradual pro-tumorigenic behavior may be observed [64].
MMP-9 and MMP-10 are highlighted in this study since they are both remodeling indicators. They have been deemed responsible for promoting the invasion and metastasis of cancer cells in HNSCC [47,48]. MMP-9 and −10 levels increase by day 10, coinciding with spheroid formation observed in the confocal microscopy assay. At day 10, the MMP-10 levels in the co-culture were five times higher than in the monocultures. This behavior shows that an increase in MMP-10 can be attributed to the presence of both cells in the same culture and highlights the importance of including the stromal component in in-vitro cancer models. TIMP-1 levels also show an increase at day 10 in the fibroblast monoculture and co-culture groups. TIMP-1 has been proposed as a prognostic biomarker for multiple cancers, and it is known to be highly expressed in HNSCC [11]. This molecule is expressed by fibroblasts and tumor cells and has been shown to inhibit apoptosis and promote metastatic behavior [65]. TIMP-1 and MMPs as targets for cancer treatment is an avenue that is still in development, but further testing is required to assess their effectiveness. The morphological changes in MMP levels in the medium suggest that matrix remodeling events are taking place in the co-culture model [66,67]. The presence of metalloproteinases during cancer progression is a wellestablished phenomenon that is closely associated with ECM remodeling [66]. Therefore, the observed differences in the co-culture model against the monoculture controls could be indicative of matrix modifications driven by cancer cells and fibroblasts, which may have important implications for understanding cancer development and identifying therapeutic targets. Nevertheless, future research is necessary to investigate this hypothesis and determine the significance of these events for clinical or drug development applications. This experiment shows the capability of measuring secretomic analytes longitudinally, allowing the monitoring of the samples in a nondestructive manner which is challenging to achieve with in-vivo models.
Tumor-and stroma-derived ECM components are present in the TME [68]. Soluble collagen can be secreted by cancer cells [68]. However, HVFFs have also been reported to be capable of secreting collagen [37]. After day 16, the co-culture model shows soluble collagen levels significantly different from the cell-free bioink. An increase in collagen levels can indicate matrix remodeling and tumor progression [69]. Increase levels of collagen are associated with higher migration, invasion, and poor prognosis in oral squamous cell carcinomas [70]. Soluble collagen can also be derived from the matrix. MMP-9 can cleave mature collagen into soluble collagen as part of their ECM remodeling mechanisms [71,72]. This study highlights the compatibility of our constructs with assays such as Sircol Assay, which can reveal collagen contents through time. Additional techniques, such as second harmonic generation, could further support ECM colorimetric assays [73].
Our model has been proven to create a scenario that shows different behavior to the monoculture groups in cellular morphology, 3D organization, MMP expression, and collagen levels through time. This platform can be used to provide a more realistic representation of the complexity of cancer in-vitro and can be complemented further by including other components of the TME, such as immune or endothelial components. Using 3D bioprinting allows the fabrication of consistent models in a semi-highthroughput matter which can be used to test multiple conditions and scenarios in parallel. This technique has a micron-level resolution, and it removes the user error, ensuring models are comparable in dimensions and cell density, which is particularly challenging to achieve with manual fabrication techniques [32]. This is of high importance when using the same model for different conditions to ensure the fabrication method is not the reason for changes in sample development and results obtained. Three-dimensional bioprinting is also capable of extruding a broad range of viscosities (30 mPa s −1 ->6 × 10 7 mPa s −1 ) which is challenging for other extrusion techniques, such as liquid handlers [31]. It also allows the fabrication of constructs with high cell densities (<10 6 cells ml −1 ), which can bring us closer to cell densities found in tissues [31,74]. Here, we show it is possible to nondestructively monitor these models through time. To evaluate how the same model changes depending on stimuli or conditions without sacrificing the model is a valuable contribution since it provides a new window of opportunities to study potential targets for the disease of study. Applications for in-vitro models include the identification of relevant cancer biomarkers and testing the efficacy of new treatments. Here, we used HNSCC as the disease of study, but these constructs can be tailored to represent other neoplastic diseases. It would be interesting to create stage-related tumors with patient-derived cells to closely match more biological features and potentially use this technique for personalized medicine. Overall, using 3D in-vitro models can help us understand cancer, reduce reliance on traditional animal models and hopefully incorporate them as a standard of validation for new treatments because they recapitulate important variables to study and treat cancer.

Conclusions
We demonstrated that A 1.5 G 5 dECMT is a suitable bioink to fabricate heterogeneous HNSCC models since it allows the development of several cell types over more than two weeks of culture. We observed significant changes in morphology, and MMP expression, suggesting matrix remodeling and crosstalk between the stromal and cancer cells. This in-vitro model can be helpful for studying HNSCC's biology or new treatments due to the biomimetic components that permit the replication of crucial variables in-vivo. Furthermore, this platform has the potential to be tailored to study other applications and diseases.

Data availability statement
The data cannot be made publicly available upon publication because no suitable repository exists for hosting data in this field of study. The data that support the findings of this study are available upon reasonable request from the authors.