A 3D Self-Assembly Platform Integrating Decellularized Matrix Recapitulates In Vivo Tumor Phenotypes and Heterogeneity

Abstract

Three-dimensional (3D) in vitro cell culture models are invaluable tools for investigating the tumor microenvironment (TME). However, analyzing the impact of critical stromal elements, such as extracellular matrix (ECM), remains a challenge. Here, we developed a hydrogel-free self-assembly platform to establish ECM-rich 3D “MatriSpheres” to deconvolute cancer cell-ECM interactions. Mouse and human colorectal cancer (CRC) MatriSpheres actively incorporated microgram quantities of decellularized small intestine submucosa ECM, which proteomically-mimicked CRC tumor ECM compared to traditional formulations like Matrigel. Solubilized ECM, at sub-gelation concentrations, was organized by CRC cells into intercellular stroma-like regions within 5 days, displaying morphological similarity to CRC clinical pathology. MatriSpheres featured ECM-dependent transcriptional and cytokine profiles associated with malignancy, lipid metabolism, and immunoregulation. Model benchmarking with scRNA sequencing demonstrated that MatriSpheres enhanced correlation with in vivo tumor cells over traditional ECM-poor spheroids. This facile approach enables tumor-specific tissue morphogenesis, promoting cell-ECM communication to improve fidelity for disease modeling applications.

Introduction

Three-dimensional (3D) engineered tumor models have emerged as essential tools to replicate the complexities of the tumor microenvironment (TME). The TME is a multifaceted and dynamic niche composed not only of cancer cells, but diverse stromal cell lineages. These cells predominantly manufacture and maintain a 3D tumor-specific extracellular matrix (ECM), which affects colorectal cancer (CRC) phenotype, metastatic invasiveness, and sensitivity to therapy(1–3). The ECM is composed of a diverse network of proteins and polysaccharides that have many functions. Namely, the ECM provides structural support and can directly influence cell signaling, which is mediated by cell-ECM ligand interactions(4). In the context of CRC, the ECM undergoes substantial remodeling(5), leading to alterations in its composition, organization, and mechanical properties. These alterations stimulate CRC tumor survival, proliferation, migration, and invasion, and can contribute to the development of chemoresistance(6). Thus, the ECM is a critical feature of the TME and is required to accurately mimic in vivo phenotypes that are predictive for drug discovery and precision medicine. Despite its importance, engineering an ECM-rich tumor stroma that reflects the compositional diversity of native tissues in vitro remains a challenge.

Recent technological advancements have ushered in the development of sophisticated 3D models, such as spheroids and organoids, which offer a more accurate representation of the TME, although each presents limitations(7). Spheroids offer a high-throughput and reproducible method for efficiently screening 3D cancer cell behavior; however, the long-time scale of ECM deposition and contributions from both stromal and cancer cell types are barriers to ECM development within traditional tumor spheroids. Alternatively, the most frequently applied paradigm for 3D ECM organoid culture is to embed tumor cells within hydrogels such as Matrigel and Collagen I. These models have yielded substantial insights yet are limited in several ways: (1) ECM composition is difficult to modulate and does not reflect the complexity of native tissues, and (2) hydrogel polymerization does not recapitulate the stromal organization that is characteristic of CRC and other tumors(8). Decellularized tissues are a form of ECM biomaterials that are frequently used as scaffolds to impart tissue-specific ECM composition and complexity. Optimized decellularization techniques remove cells from mammalian tissues to isolate compositionally intact ECM that is reproducible and scalable for commercial applications(9). For instance, ECM formulations such as scaffolds, particles and hydrogels have been used for tissue reconstruction in vivo and as a substrate to mimic ECM microenvironments in vitro(10–12). Therefore, decellularized tissues could be useful tools to augment tumor models, which to date have primarily involved passive ECM assembly rather than cell-driven tumor morphogenesis that occurs in vivo.

In this study, we developed tumor MatriSpheres: a high-throughput method of in vitro tumor morphogenesis using a tissue-specific small intestine submucosa (SIS) ECM and CRC cells to recapitulate the TME. We found that unlike previous methods, ECM assembly is cell-mediated and cell line specific. Furthermore, MatriSpheres recapitulate CRC transcriptional phenotypes found in vivo that are not found in traditional spheroids alone. We describe the parameters that facilitate cell-mediated ECM organization and concentration in a process that is distinct from bulk hydrogel formation, enables modular and diverse ECM composition, and is compatible with other purified ECM biomaterials such as Matrigel and Collagen I. This approach for consistently generating ECM-enhanced 3D tumor spheroids could be used to further understand the complex interplay between cancer cells and ECM and lead to the development of more effective therapies using precision medicine.

Materials and Methods

SIS ECM Decellularization and Processing

Porcine small intestines were obtained from Tissue Source LLC (Zionsville, IN) and were kept frozen at −30°C until decellularization. Intestines were decellularized as described previously(13,14). Small intestines were thawed and rinsed with MilliQ water to remove residual intestinal fluid. Tissues were cut longitudinally to create a sheet and the muscle layers were removed by mechanical delamination. Smaller strips (~6 inches wide) of scraped SIS were transferred to a flask containing MilliQ water to remove any loose residual tissue. To aid in cell removal and disinfect tissues, SIS strips were washed in a 0.1% peracetic acid and 4% ethanol solution in MilliQ water on a mechanical shaker at 300 rpm for 2 hours at ambient temperature. SIS was washed by alternating MilliQ and 1X PBS two times for 15 minutes each at 300 rpm. After washing, wet SIS ECM was transferred to 50 mL conical tubes and frozen at −80°C. Decellularized tissues were lyophilized and milled into a fine powder using a cryogenic grinder (SPEX SamplePrep, Cat. No. 6875). SIS ECM powder was enzymatically digested with pepsin from porcine gastric mucosa (Sigma, Cat. No. P6887, 1 mg/mL) and hydrochloric acid (0.01 M) at a stock ECM concentration of 10 mg/mL under constant magnetic stirring (between 700–1500 rpm) for 48 hours at room temperature. Low concentrations of SIS ECM digest were used for all CRC tumor spheroid cultures.

ECM Characterization

Histological Assessment of Decellularization:

Native and decellularized small intestine were fixed in 10% neutral-buffered formalin, embedded, sectioned and stained with hematoxylin and eosin (H&E) or DAPI (4′,6-diamidino-2-phenylindole) (Calbiochem - Cat.# 50-874-10001). Tissue sections were imaged using an inverted microscope (Zeiss Axio Observer 7) under brightfield or fluorescent light. Successful decellularization was qualitatively determined by tissue sections showing an absence of cell nuclei (H&E – purple and DAPI – blue) and intact tissue structure or ECM.

TEM of Solubilized SIS ECM:

Transmission electron microscopy (TEM) was performed on enzymatically digested decellularized SIS ECM to determine the presence of intact fibrils and nanoscale structures, such as matrix bound nanovesicles (MBVs). A 3μl drop of sample solution was briefly place on a glow-discharged 200 mesh carbon-coated copper grid. The grid is then washed three times with deionized water. The grid is then briefly stained with 0.75% uranyl formate two times and additional stain for 30 seconds. The grid was then imaged in the Hitachi electron microscope (H7650) operated at 80kv with a CCD camera.

DNA Isolation and Quantification:

To validate the removal of immunogenic materials within decellularized tissues, we quantified double-stranded DNA (dsDNA) content. First, we isolated and purified total DNA from native and decellularized small intestine using a DNeasy Blood and Tissue Kit (Qiagen, Cat. No. 69504) in accordance with the manufacturer’s instructions. Briefly, 10–25 mg of lyophilized samples were digested using proteinase K solution and lysed overnight in a water bath at 56°C. Tissue lysate was passed through a DNeasy membrane to bind DNA and remove contaminants. Purified DNA was eluted in 200 µL of AE buffer and stored at −80°C until use. A Quant-iT PicoGreen assay was used for a highly-sensitive fluorescence-based readout of dsDNA content. Native and decellularized DNA samples were serially-diluted from 1:200 to 1:1600 and 1:20 to 1:160, respectively. A dsDNA standard was used as a control and provided a reference range from 31.25–2000 ng/mL. 100 µL of diluted samples and standards were plated onto a 96-well plate and 100 µL of Quant-iT reagent was added. The samples were mixed for 5 minutes and the fluorescence intensity was determined at 480/520 nm excitation and emission wavelengths. DNA fragment size was assessed using a PowerSnap gel electrophoresis (Thermo Fisher) system. A 2% agarose E-gel with SYBR Safe was loaded with diluted samples (1:10 Decell or 1:20 Native) and run for 26 minutes before imaging via transillumination.

Sircol Assay:

ECM biomaterial solutions were prepared from stock concentrations and assayed using the Sircol kit (S1000, Biocolor Life Science Assay Kits, UK). Diluted samples and standards were mixed with 1 mL of saturated Sirius Red solution in picric acid. The dye-binding incubation step (30 minutes with gentle mechanical agitation) was performed at 4°C to prevent gelation of the Matrigel. Samples were spun at 13,000 rpm for 10 minutes. Excess dye and solution were removed and 750 µL ice-cold acid salt were added to each sample. Samples were spun at 13,000 rpm for 10 minutes. This solution was removed and 1,000 µL alkali reagent were added to each sample and the collagen-bound dye pellets were solubilized by vortexing the samples. 200 µL of each sample (triplicate) and standard (duplicate) were transferred to a 96-well plate. The absorbance was measured at 556 nm using a plate reader (SpectraMax i7, Molecular Devices).

ECM Biomaterial Digestion and LC/MS analysis

MC38 Tumor ECM MS prep:

Mass spectrometry was used to compare differences in composition between ECM biomaterials and acellular MC38 tumor ECM. Tumors were established in the right and left flanks of 6–10-week-old C57BL/6 mice (Jackson Lab; RRID:MGI:2159769) with bilateral subcutaneous injections of 500,000 MC38 cells (RRID:CVCL_B288) per injection. Ethical approval for the animal experiments was provided by the Institutional Animal Care and Use Committee at NCI Frederick (ASP No. 20–063). Tumors were harvested after two weeks (1 cm in diameter) and were frozen until thawed for future use. To prepare MC38 tumor ECM, we diced MC38 tumors into small pieces and processed with the following steps. Briefly, tissues were washed in MilliQ water then rinsed in hypotonic and hypertonic solutions. Trypsin-EDTA treatment, followed by washes in 4% sodium deoxycholate were designed to detach cells then release cytoplasmic nuclear contents. Residual DNA was broken down by a DNase treatment then 3% Triton X-100 was used to clean up any remaining cellular components. Peracetic acid and ethanol were combined to reduce pyrogenic material. The remaining tumor ECM was washed several times in alternating MilliQ water and 1X PBS washes then lyophilized for downstream analysis.

Digestion:

For decellularized SIS-ECM powdered samples, 0.54–0.83 mg was treated with 300 µL of EasyPep Lysis buffer (Thermo Fisher PN A45735, provided in EasyPep 96-well plat kit A57864) and treated with 50 µL each of reducing solution and alkylation solution provided with the EasyPep kit. The decellularized SIS-ECM samples were sonicated 3×25 sec at 15% amplitude then all samples were heated at 95°C for 10 minutes. SIS-ECM samples were treated with 50 µL of 0.2 µg/µL trypsin/LysC while the Collagen I and Matrigel, samples were treated with 10 µL of 0.2 µg/µL trypsin/LysC. Samples were incubated at 37°C overnight for 18 hrs with shaking at 1000 rpm. Added 100 µL of stop solution provided with the EasyPep kit and cleaned using the EasyPep 96-well plate and eluted with 300 µL of the elution buffer, aliquoted into 20 µL and 250 µL and dried. The 20 µL aliquot was used to assess the peptide concentration using the Pierce Quantitative Colorimetric Peptide Assay (Thermo PN 23275).

LC/MS analysis of peptides:

Each sample was resuspended to a concentration of 1 µg/µL in 0.1% FA and 1 μL was analyzed using a Dionex U3000 RSLC in front of a Orbitrap Eclipse (Thermo) equipped with a FAIMS interface and an EasySpray ion source. Solvent A consisted of 0.1%FA in water and Solvent B consisted of 0.1%FA in 80%ACN. Loading pump consisted of Solvent A and was operated at 7 μL/min for the first 6 minutes of the run then dropped to 2 μL/min when the valve was switched to bring the trap column (Acclaim^™ PepMap^™ 100 C18 HPLC Column, 3μm, 75μm I.D., 2cm, PN 164535) in-line with the analytical column EasySpray C18 HPLC Column, 2μm, 75μm I.D., 25cm, PN ES902). The gradient pump was operated at a flow rate of 300 nL/min. Each run used a linear LC gradient of 5–7%B for 1min, 7–30%B for 34 min, 30–50%B for 15min, 50–95%B for 4 min, holding at 95%B for 7 min, then re-equilibration of analytical column at 5%B for 17 min. MS acquisition employed the TopSpeed method at three FAIMS compensation voltages (−45, −60, −75) with a 1 second cycle time for each voltage and the following parameters: Spray voltage was 2200V and ion transfer temperature was 300 °C. MS1 scans were acquired in the Orbitrap with resolution of 120,000, AGC of 4e5 ions, and max injection time of 50 ms, mass range of 350–1600 m/z; MS2 scans were acquired in the Orbitrap a with resolution of 15,000, AGC of 5e4, max injection time of 22ms, HCD energy of 30%, isolation width of 1.06 Da, intensity threshold of 2.5e4 and charges 2–5 for MS2 selection. Advanced Peak Determination, Monoisotopic Precursor selection (MIPS), and EASY-IC for internal calibration were enabled and dynamic exclusion was set to a count of 1 for 15 sec.

Database search and post-processing analysis:

MS files were searched with Proteome Discoverer 2.4 (RRID:SCR_014477) using the Sequest node. Data was searched against the Mus musculus (Matrigel), Bos taurus (Collagen I), or Sus scrofa (ECM) Uniprot databases (RRID:SCR_002380). The SIS-ECM, Matrigel, and Collagen I samples were searched using a full tryptic digest. and allowed for 2 max missed cleavages with a minimum peptide length of 6 amino acids and maximum peptide length of 40 amino acids, an MS1 mass tolerance of 10 ppm, MS2 mass tolerance of 0.02 Da, fixed carbamidomethyl (+57.021) on cysteine and variable oxidation on methionine and proline (+15.995 Da). Percolator (RRID:SCR_005040) was used for FDR on all searches except the Collagen I sample in which Fixed PSM was used. NSAF values for each protein within a sample were calculated by taking the spectral counts of the protein and dividing by the number of amino acids to give the SCP/L value, which was then divided by the total PSMs of the sample to give the NSAF.

Colorectal Cancer (CRC) Cell culture

MC38 murine CRC cells (RRID:CVCL_B288) were provided by the McVicar Lab from the Cancer Innovation Laboratory in December 2020 (Supplementary Table 1). CT26 murine CRC cells (RRID:CVCL_7256) were purchased from the American Type Culture Collection (ATCC; CRL-2638) and received January 2021. HT-29 human CRC cells (RRID:CVCL_0320) were obtained from the NCI Division of Cancer Treatment and Diagnosis Tumor Repository (Frederick, MD). Prior to experimentation (during or before November 2021), all cell lines were confirmed to be mycoplasma negative by the Animal Health and Diagnostic Laboratory at Frederick National Lab. In general, cells are screened a few times per year using MycoStrip mycoplasma detection kit (InvivoGen; Cat.# rep-mys-20). Cells were grown in Roswell Park Memorial Institute (RPMI) media supplemented with L-glutamine (Thermo Fisher 11875119), 10% fetal bovine serum (Thermo Fisher 16000044), and 1% penicillin streptomycin (Thermo Fisher 15140122). Cells were maintained at 37°C with 5% CO₂ humidified air. For all experiments in this study, cells were used within 3–5 passages upon thawing.

Spheroid/MatriSphere Formation and Culture

Single cell suspensions were counted with a hemocytometer then prepared in RPMI at 2x concentration (5000 cells/well) of the final cell seeding density (2500 cells/well) and kept on ice until mixed with the ECM-containing solution. Several parameters for MatriSphere formation were screened, including basal media type (RPMI, DMEM), seeding densities (625, 1250, 2500, 5000 cells/well), and ECM concentrations (15.6, 31.3, 62.5, 125 µg/mL) to identify optimal conditions. This information has been uploaded to the MISpheroID database to enhance transparency, reproducibility and rigor(15). ECM-containing solutions were prepared at 2x (250 µg/mL: SIS ECM and Matrigel/125 µg/mL: Collagen I) the final seeding ECM concentration (125 µg/mL: SIS ECM and Matrigel/ 62.5 µg/mL: Collagen I) and kept on ice until mixed with cells. The cell and ECM solutions were mixed 1:1 and transferred to a sterile basin for cell seeding. MatriSpheres were seeded by pipetting 100 µL of the cell and ECM containing solution into the inner 60 wells of a 96-well ultra-low attachment round bottom well plate (Corning 7007). MatriSpheres were cultured at 37°C with 5% CO₂ humidified air by replenishing the media every two to three days. At day seven, 100 µL of media was removed from each well and saved for cytokine analysis by storing at −80°C or discarded. Spheroids and MatriSpheres were collected for RNA-Seq or qPCR analysis by storing in TRIzol Reagent (Thermo Fisher 15596018) or fixed by adding 100 µL 8% paraformaldehyde (PFA, Electron Microscopy Sciences 15710) to each well for a final concentration of 4% PFA and fixed at 4°C for at least 48 hours. The MatriSpheres were washed 3x with 1x PBS after fixation and kept at 4°C until histological processing. MatriSphere formation methods were validated to be compatible using high-throughput conditions, such as 384 well plate format, and bulk histologic evaluation using a 3D printed microarray mold for agarose/paraffin embedding and sectioning (Supplementary Fig. S1).

MatriSphere Diameter Analysis

MatriSphere diameters were measured on days 3, 5, and & 7 of the culture using a Celigo Image Cytometer (Nexcelom, Lawrence, MA) using the tumorsphere setting to acquire brightfield images of the samples. The largest object within each well was determined to be the spheroid for quantification and spheroid diameters were analyzed through the instrument’s software. MatriSpheres were manually traced in ImageJ (NIH, Bethesda, MD; RRID:SCR_003070) for cases where the analysis parameters could not be adjusted to capture the actual MatriSphere diameter. The equivalent diameters reported are calculated from the spheroid area and as if the spheroid diameter was an actual circle.

LIVE/DEAD Staining and Imaging

Spheroids and MatriSpheres were stained with Calcein AM and Ethidium-homodimer (Thermo L3224) at Day 7 in accordance with the manufacturer’s instructions. Stained samples were imaged with the ImageXpress under confocal microscopy at 4X magnification with an offset ~200 µm and z-stack range at depths between 100 and 300 µm. For three-dimensional MatriSphere imaging, resolution significantly diminishes after 300 µm without optical clearing. However, we were most interested in sampling both the outer edge and core of the MatriSpheres, which were imaged at the noted depths. After imaging, brightfield, FITC and TRITC images were analyzed using a custom module editor on MetaXpress software (RRID:SCR_016654). Briefly, an image mask was determined from brightfield images to identify the MatriSphere area. The fluorescent channels were used to identify Live (FITC) and Dead (TRITC) cells within the MatriSphere region.

CellTiter-Glo 3D Analysis

CRC MatriSpheres were cultured for seven days and 100 μL of media were removed from each of the wells. The plates were brought to room temperature and 100 μL of room temperature CellTiter-Glo 3D (Promega G9683) were added to each of the wells. The plates were incubated on a shaker at 151 rpm for five minutes, followed by an additional 25 minutes on the bench. Then, 100 μL was transferred from each well into a white opaque 96-well plate and the luminescence was read by a plate reader (Molecular Devices).

Live-Cell Time Lapse Imaging

After cells and ECM had been plated as described above, the plate was transferred to an ImageXpress Micro Confocal System (Molecular Devices, San Jose, CA) with environmental controls set to 37°C and 5% CO₂. Within 5 minutes of plating, images acquisition began using a 10x objective and were obtained every 15 minutes for the culture period. Imaging was suspended temporarily during the spheroid feeding. We observed differences in the time needed for spheroid formation during incubator conditions compared with spheroids that were formed during the timelapse imaging. It took longer for the spheroids to form in the imaging experiments, and we suspect this may be a result of the movement of the plate during the image acquisition.

Rheological Characterization

To determine the mechanical properties of the ECM biomaterials, we used a Discovery HR20 rheometer (TA Instruments – Schneider Lab) and a 40 mm stainless steel parallel plate geometry (Cat. No. 511400.945). ECM biomaterials were prepared at various concentrations (62.5, 125, 1000 and 4000, 5000 and 7000 µg/mL) to determine the effects of ECM concentration on gelation compared to RPMI media controls. Briefly, the rheometer temperature was set to 4°C to prevent premature gelation and ~700 µL of sample was pipetted onto the center of the Peltier plate. The geometry was slowly lowered to establish a uniform spreading, and mineral oil was added to the edge of the sample to avoid evaporation during testing. A brief temperature ramp was performed to 37°C to mimic the culture conditions followed by a time sweep.

Immunohistochemistry

Spheroid sections were deparaffinized in xylenes 3x for 3 minutes each followed by rehydration in 100% ethanol 2x, 95% ethanol, 70% ethanol, and type I water 3x. Next, sections were refixed in 10% neutral buffered formalin (Sigma; HT501128–4L) for 15 minutes. Slides were then washed in type I water and TBS-T for 3 minutes each. 10 mM sodium citrate (Sigma) buffer at pH 6 was preheated in a vegetable steamer for 20 minutes to reach 95°C. Slides were heated in the buffer for 20 minutes and then cooled at room temperature for 20 minutes. Following antigen retrieval, endogenous peroxidases were blocked with 3% H₂O₂ (Sigma; H1009–100ML) for 15 minutes. Slides were washed 2x in TBS-T. Aldehydes were quenched with glycine in TBS-T for 5 minutes at room temperature. Blocking buffer (10% BSA (Sigma; A9647-100G), 5% Goat Serum (Vector Laboratories; S-1000-20), 0.5% Tween-20 (Sigma) in TBS-T) was incubated for a minimum of 30 minutes on the sections at room temperature before primary antibody incubation. Primary antibodies (Abcam) (Supplementary Table 2) were incubated at 4°C overnight in a humidity chamber and diluted in blocking buffer at the following concentrations: E-Cadherin (ab76319; RRID:AB_2076796): 1/10000, N-Cadherin (ab76011; RRID:AB_1310479): 1/8000, Carbonic Anhydrase IX (CA9, ab243660; RRID:AB_3665691): 1/2000, and Ki-67 (ab16667; RRID:AB_302459): 1/1000. A FITC-conjugated collagen hybridizing peptide (CHP, 3Helix FLU300) was heated on a heating block for 15 minutes at 80°C for 15 minutes and then allowed to cool to room temperature. The CHP was diluted to 1 µM and incubated overnight at 4°C with the last primary antibody in each panel. After primary antibody and CHP incubations, slides were washed 3x in TBS-T for 3 minutes. Rabbit-on-Rodent Secondary HRP antibody (Biocare Medical RMR622H; RRID:AB_3665692) was added to the slides for 20 minutes at room temperature. Next, slides were washed 4x for 3 minutes each. Opal reagents (Akoya Biosciences) were added to slides: Opal 570 (1:150) for CA9 and E-Cadherin and Opal 650 (1:250) for Ki67 and N-Cadherin. For subsequent stains in a panel, slides were reheated in citrate buffer at 95°C for 20 minutes and allowed to cool 20 minutes before proceeding. Once all antibody stains were performed, 1 µg/mL DAPI was added to slides for 5 minutes at room temperature and then washed 2x with type I water. Coverslips were mounted with Dako Fluorescent Mounting Media (Agilent S302380–2) and allowed to dry. A rabbit IgG isotype (Abcam ab172730; RRID:AB_2687931) and primary deletes were performed as controls. Slides were imaged with an ImageXpress Micro Confocal Imaging System using a 20X water immersion objective. The images were analyzed in MetaXpress with multi-wavelength cell scoring.

In Vitro CRC Spheroid Bulk RNA Isolation, Sequencing, and Analysis

After 7 days of culture, CRC spheroids were transferred to 5 mL tubes in batches of 10, the excess media was removed by pipetting, 1 mL of TRIzol (Invitrogen) was added to each sample, and the samples were stored at −80°C until further processing. Samples were thawed on ice and the RNA was extracted using the RNeasy Micro Kit (QIAGEN, Germantown, MD) in accordance with the manufacturer’s instructions. RNA Samples were then pooled and sequenced on a NovaSeq 6000 S1 using Illumina Stranded mRNA Prep and paired-end sequencing. Datasets were imported to Ingenuity Pathway Analysis (IPA, Qiagen; RRID:SCR_008653) for calculating pathway activation patterns or z-scores.

qPCR Analysis of MC38 Spheroids and MatriSpheres

Total RNA was isolated at Day 7 from MC38 spheroids and MatriSpheres prepared with SIS ECM, Collagen I, or Matrigel to evaluate the effects of ECM composition on CRC cell gene expression. Purified RNA was converted into cDNA using SuperScript^™ IV VILO^™ Master Mix (Invitrogen, Cat#11756050) on a T100^™ Thermal Cycler (BioRad). Quantitative real-time polymerase chain reaction (qPCR) was performed using TaqMan^™ Gene Expression Master Mix (Applied Biosystems^™, Cat#369510) combined with 10–80 ng of cDNA template and Integrated DNA Technologies (IDT) PrimeTime Mini qPCR Assays in 20 µL reactions in triplicate, performed for 40 cycles on CFX Opus 96 (BioRad). The following genes were screened (IDT assay numbers): Reference genes – Actb (Mm.PT.39a.22214843.g), Ywhaz (Mm.PT.39a.22214831), Ppia (Mm.PT.39a.2.gs); Target genes – Nos1 (Mm.PT.58.11089624), Postn (Mm.PT.58.9884244), Ccl5 (Mm.PT.58.43548565), Serpine1 (Mm.PT.58.6413525), Mmp13 (Mm.PT.58.42286812), Crlf1 (Mm.PT.58.10244905), Tnfrsf11b (Mm.PT.58.41494681), Mvd (Mm.PT.58.14290933), Aspn (Mm.PT.56a.43192444), Igfbp3 (Mm.PT.58.6744601), Ddx41 (Mm.PT.58.10813544), Bcl6b (Mm.PT.58.13254777.gs), and Cox6a2 (Mm.PT.58.42467915.g) (Supplementary Table 3). Ct values were exported from CFX Maestro software into Microsoft Excel (RRID:SCR_016137) and normalized against the geometric mean of the 3 reference genes to obtain ΔCt for each group. ΔΔCt values were calculated using the average ΔCt of the cells alone spheroid control. Log transformed fold change (2^−ΔΔCt) values were plotted using GraphPad Prism (RRID:SCR_002798). Statistical significance was determined using the ΔCt values for each sample using a one-way ANOVA with a Tukey’s post hoc analysis to compare the differences between MatriSpheres containing various ECM biomaterials.

In vivo MC38 Tumor Establishment and Collection for scRNA Seq

Seven-month-old C57BL/6 mice were obtained from Charles River Laboratories and injected with 1× 10⁶ cells in 100 µL PBS per mouse. Ethical approval for the animal experiments was provided by the Institutional Animal Care and Use Committee at NCI Frederick (ASP No. 23–078). Tumors were carefully excised from the mice when the size reached up to 250–500 mm³, which was usually after 10–14 days. The tumor tissues were washed with cold sterile PBS. Fat and necrotic areas and blood vessels were removed from the samples. The MC38 tumor single cell suspension preparation began by preparing the enzyme mix in accordance with the protocol of the Miltenyi Tumor Dissociation Kit (mouse) and placing on ice. One mL of ready-to-use enzyme mixture was then added to sterile petri dishes for each sample and placed on ice. Subsequently, the tumor tissue was placed in the dish with the enzyme mix and cut into small pieces of 2–4 mm³ using sterile blades. Tissue pieces were then transferred into the gentleMACS C Tube containing the remaining enzyme mix. The C Tube were tightly closed and positioned upside down onto the sleeve of the gentleMACS Dissociator. The program was run in accordance with the instruction of the Miltenyi Tumor Dissociation Kit (mouse). Following completion of the program, cells were centrifuged at 300g x 10 min followed by one wash using cold PBS at 300g x 10 min. Red blood cells were removed using anti-Ter-119 microbeads (Miltenyi) according to the manufacturer’s instruction. Cells were counted and spun down at 300g x 10 min. Dead cells were removed using dead cell removal kit (Miltenyi) in accordance with the protocol. Cells were counted and the viability was determined, ensuring that the cell viability was higher than 90%. Finally, cells were cryopreserved in a solution of 90% FBS + 10% DMSO until further processing.

MC38 Tumor RNA Isolation and scRNA Sequencing

The frozen MC38 single cell suspension was thawed in 37°C water bath and promptly transferred into 5 mL of DMEM containing 10% FBS. Subsequently, the cells were centrifuged at 350 × 5 min. The cell pellet was then washed three times with PBS + 0.04% BSA at 350 × 5 min before further processing. Dead cells need to be removed if the cell viability is lower than 70%. Aiming to capture 10,000 cells, 16,000 cells per sample were used for generating scRNAseq libraries following the guidance of 10x Genomics Chromium Next GEM Single Cell 3’ Gene Expression version 3.1. Briefly, RT reagents, cell suspension, along with the gel beads, were loaded onto Chromium NextGEM chip G (separate channels for different samples). Gel Beads-in-emulsion (GEMs) generated using the 10X Genomics Chromium Controller were then transferred and incubated in a thermal cycler for GEM-RT incubation. The barcoded cDNAs generated were then purified, amplified, cleaned up. Subsequently, Bioanalyzer (Agilent Technologies) was run to determine cDNA concentration. The libraries were prepared from the cDNA according to the guidance of 3’ gene expression dual index library construction. The libraries were sequenced on NovaSeq_SP system using read lengths: 28bp (Read 1), 10bp (Index i7), 10bp (Index i5) and 90bp (Read 2) (Illumina).

Cytokine Arrays

To assess the effects of SIS ECM on CRC cell cytokine secretion, we used both mouse (111 analytes) and human (105 analytes) proteome profiler XL kits (R&D Systems, Cat No. ARY028 and ARY022B). We screened the cell culture supernatant from Day 7 spheroid cultures with and without SIS ECM. Briefly, 100 µL of media was collected from each well (n = 30) and batched by row (10 wells/row) for a total of 1 mL (n = 3 replicates). Media was frozen at −80°C and thawed on ice before use. For each sample array, 400 µL of cell culture supernatant were used to determine the presence of cytokines. Bound cytokines were detected using streptavidin-HRP conjugated antibodies and chemi reagent, which produced a chemiluminescent signal, captured on an Amersham Imager 680 digital imaging system (GE Life Sciences). Integrated intensity was determined using an ImageJ plugin (Protein Array Analyzer) and exported to Excel for data analysis.

Upstream Regulators and Target Genes Analysis

Ingenuity Pathway Analysis was used to identify conserved upstream regulators and target genes within CRC MatriSpheres compared to cells alone spheroids. IPA software predicted 6,670 combined upstream regulators associated with the addition of SIS ECM across all three cell lines. We narrowed this list of upstream regulators by 91.3% using exclusion criteria based upon |z-score| ≥ 2 and p-value < 0.001. From this list of 580 significant upstream regulators, we found 10 upstream regulators shared between each of the three CRC cells lines. We then combined the list of target genes associated with each of the upstream regulators and found 35 common target genes. The selected regulators and target genes were put back into IPA to create an interaction plot (Supplementary Fig. S2A). This was subsequently narrowed to only direct and shared interactions between upstream regulators and target genes (Supplementary Fig. S2B,C). The top 5 upstream regulators were then superimposed onto the Tumor Microenvironment Pathway in IPA to determine their role in TME signaling (Supplementary Fig. S3A,B).

Experimental Rigor

Inclusion and exclusion criteria were utilized to ensure accurate sampling from each experimental and control group. For example, spheroids and MatriSpheres that did not form were removed from the downstream analysis and indicated on quantitative plots. Determination of n-values were based upon previous experiments, and therefore, we did not conduct a power analysis. Sampling for bulk sequencing were randomized and batched to account for potential variability observed within and between experimental replicates/groups. Attrition was not applicable for this study and sex was not evaluated as a biological variable. To reduce bias and remove human influence on experimental findings, we used automated analyses from established image analysis and transcriptomics workflows.

Statistical Analysis

Statistical Analyses were performed in GraphPad Prism (v9). For data with a normal distribution and three or more groups, statistics were calculated by a one-way ANOVA followed by a Tukey’s or Sidak’s multiple comparisons test. A two-way ANOVA followed by a Dunnett’s or Sidak’s multiple comparisons test were used to calculate statistics for data with two variables and three or more groups. P values are indicated in the figure legends.

Data and Code Availability

Data and code will be made available upon reasonable request. All proteomic data has been uploaded to public databases MassIVE server (Accession#: MSV000094671). Bulk (Accession#: GSE267071) and scRNA (Accession#: GSE267714) sequencing datasets have been deposited in GEO. Code used to analyze data in this manuscript is available on the following GitHub repositories: (1) https://github.com/MelissaGall/MatriSphere (2) https://github.com/NIDAP-Community/3D-intercellular-assembly-of-decellularized-matrix-recapitulates-in-vivo-tumor-heterogeneity/tree/main.

Results

SIS ECM is a tissue-specific ECM biomaterial for modeling the CRC TME

The ECM of tissues is a complex assembly of hundreds of proteins, polysaccharides, such as glycosaminoglycans (GAGs), and matrix bound vesicles(16). We explored biologically-derived materials that preserve this natural complexity as a facile and reproducible source for harnessing ECM diversity. First, we decellularized porcine SIS as an tissue-specific ECM biomaterial to replicate the most abundant ECM components found within the CRC TME (Fig. 1A) (13). Histologic staining confirmed that cells were removed while maintaining a collagen-rich microarchitecture. Enzymatic digestion of SIS ECM created a liquid ECM dispersion for cell culture that consisted of loosely associated fibers and nanosized structures consistent with matrix-bound nanovesicles (MBVs)(16). Decellularization reduced double-stranded DNA from 4124±902.4 ng/mg in native tissue to 214.9±20.56 ng/mg in SIS ECM with substantial DNA fragmentation (DNA bands < 350 bp) (Fig. 1B,C).

Figure 1.

SIS ECM Matrisome composition reflects CRC tumor ECM diversity. A, SIS ECM processing workflow and characterization. Histological staining with H&E and Trichrome shows the removal of cells and retention of dense collagen. Transmission electron micrographs (TEM) of digested SIS ECM highlight the presence of intact fibrils and extracellular vesicles. B, PicoGreen assay indicates a significant reduction in dsDNA content and C, DNA gel electrophoresis demonstrates a substantial decrease in DNA fragment size post-decellularization. D, Sircol assay displays variance in soluble collagen content of ECM biomaterials. E, Matrisome category breakdown and F, Core Matrisome composition between ECM biomaterials compared with Mouse/Human colorectal tumors based upon relative abundance, detected by mass spectrometry (LC-MS). G, Top 10 Matrisome proteins by relative abundance. Venn diagram showing shared Matrisome proteins between ECM biomaterials and H, mouse MC38 tumor and I, human CRC tumors. J, Percentage of shared Matrisome proteins between ECM biomaterials and mouse/human CRC tumors. K, tSNE plot of in vivo cell populations within MC38 tumors (Immune, MC38, and Stromal cells) based upon clustering of scRNA Seq expression data. L, Matrisome expression based upon cell type within MC38 tumors. Plotted data are the mean ± SD. Statistics were calculated by a one-way ANOVA followed by a Tukey’s multiple comparisons test. Statistical significance where p < 0.0001 is denoted with ****.

We compared SIS ECM composition to other naturally-derived biomaterials, Collagen I and Matrigel, that are frequently utilized for in vitro tumor modeling. Although these purified biomaterials represent critical ECM components, we hypothesized that they provide only a fraction of the broad ECM diversity found within colorectal tumors. One substantial difference was the total soluble non-crosslinked fibrillar collagen content. Specifically, SIS ECM was composed of 50.9% ± 2.8% soluble collagen by mass compared to Matrigel which is 25.5% ± 2.3% (Fig. 1D).

We further characterized the proteomic composition of decellularized SIS ECM and other ECM biomaterials to determine relative similarity to in vivo mouse and human CRC tumors. We used mass spectrometry methods that emphasize sensitivity and coverage to identify abundant matrisome proteins, 1,062 known gene-encoded ECM molecules(17). SIS ECM was the most compositionally diverse material, with 145 matrisome proteins compared to 69 in Matrigel and 5 in Collagen I. SIS ECM diversity was in part due to greater representation of matrisome-associated constituents in addition to structural core matrisome proteins (Fig. 1E). The proteomic composition of SIS ECM more closely matched mouse and human CRC tumor ECM compared to other biomaterials. ECM from syngeneic MC38 mouse CRC tumors consisted of 101 total matrisome proteins, primarily of core matrisome collagens (62%) and ECM glycoproteins (26%). Many of these proteins detected in MC38 tumor ECM were found in SIS ECM, such as collagens (23), proteoglycans (10) and ECM glycoproteins (47), but were absent in Collagen I and Matrigel (Fig. 1F).

We also compared SIS ECM composition to existing mass spectrometry datasets of human healthy colon and CRC tumor (18). The most abundant matrisome proteins in SIS ECM were fibrillar collagens (COL1A1, COL1A2, COL6A1, COL6A2 and COL6A3), proteoglycans (OGN and ASPN) and ECM-affiliated protein (ANXA5), which mediates extracellular vesicle binding(19) (Fig. 1G). We confirmed that Collagen I was predominantly COL1A1 and COL1A2, but also substantial Type III Collagen (COL3A1), while Matrigel consisted of basement membrane ECM glycoproteins (LAMA1, NID1, LAMB1, and LAMC1). Out of 101 total matrisome proteins identified within in vivo MC38 tumor ECM, 76 total proteins were shared in SIS ECM. Notably, 42 of these proteins were found in SIS ECM, but were absent in the other ECM biomaterials. Matrigel had the next highest number of shared proteins, most of which were also identified in SIS ECM, except for 8 unique proteins only found in Matrigel. This suggests that SIS ECM provides greater coverage of tissue-specific constituents that are absent in traditional ECM biomaterials (Fig. 1H). A similar trend was observed when comparing the ECM biomaterials to human CRC tumors, which contained 264 total matrisome proteins. SIS ECM shared the greatest number of proteins with human CRC tumors, totaling 128 shared proteins, including 74 unique to SIS ECM among the biomaterials examined. In contrast, Matrigel shared 64 proteins in total, with 13 being unique, and Collagen I shared only 4 proteins, none of which were unique (Fig. 1I). In summary, SIS ECM contained the highest percentage of total shared matrisome proteins for both mouse (76/101: 75%) and human (128/264: 48%) CRC tumors, providing a robust and diverse ECM biomaterial for modeling the CRC tumor stroma in vitro (Fig. 1J).

Using single-cell RNA sequencing (scRNA seq), we elucidated which CRC cellular compartments were responsible for ECM production and cancer cell-ECM affinity. We defined three populations within in vivo MC38 tumors: (1) MC38 cancer cells, (2) immune cells and (3) stromal cells (Fig. 1K). Each cell compartment expressed unique ECM gene signatures in the production or maintenance of tumor ECM (Fig. 1L; Supplementary Figs. S4–S6). Stromal cells (i.e., fibroblasts, endothelial cells, pericytes) were the least numerous but expressed the highest matrisome gene expression compared to more abundant MC38 cancer cells and immune cells. This suggests that ECM biomaterial supplementation may be required to effectively model tumor cell-ECM interactions. Furthermore, MC38 cancer cell total matrisome expression was predominantly matrisome-associated genes, such as ECM Regulators (31%) and Secreted Factors (25%), rather than core matrisome (Supplementary Fig. S7A) supporting that these essential structural components have to be supplied to in vitro models. Bulk RNA sequencing data collected from CRC spheroids were used to infer matrisome expression patterns by cell line in the absence of SIS ECM (see Supplemental Results: Supplementary Figs. S7B,S8A–C).

Decellularized ECM spontaneously organizes into 3D stromal microenvironments during MatriSphere assembly

We sought to utilize the tissue mimetic diversity of SIS ECM to engineer 3D TME morphogenesis in vitro. Organoid methods, where isolated cells are seeded within a hydrogel (e.g. Matrigel), have been used to impart ECM signals in a 3D culture. However, hydrogel polymerization does not involve resident cell participation. Thus, we identified conditions to facilitate cancer cell and decellularized ECM co-assembly to build extracellular matrix-enhanced spheroids, or “MatriSpheres”. We evaluated CRC MatriSphere formation using two murine (MC38 and CT26) and one human (HT-29) cell line(s) in the presence of microgram quantities (8–125 µg/mL) of solubilized SIS ECM in ultra-low attachment plates (Fig. 2A–B; Supplementary Fig. S9A). Formation efficiency with Collagen I and Matrigel was cell line and ECM dependent. MC38 and CT26 MatriSpheres formed with the Collagen I and Matrigel, whereas HT-29 cells organized into multilayer discs with Collagen I and unconsolidated clusters with Matrigel rather than a single structure. Similar to stem cell models (12), we validated that SIS ECM particles (Supplementary Fig. S9B) could be incorporated into tumor spheroids via sedimentation rather than cell-mediated assembly in CRC cells. We determined that MatriSpheres formed similarly using several cell densities, ECM concentrations, and basal media composition (Supplementary Figs. S10–S12). However, to keep consistent across each CRC cell line, all subsequent experiments used 2,500 cells/well in 100 µL of RPMI at ECM concentrations of 125 µg/mL for SIS ECM and Matrigel and 62.5 µg/mL for Collagen I. MatriSpheres had more consistent morphology and size than other ECM types (Figs. 2B–C; Supplementary Fig. S13). Compared to the diameter of cells alone spheroids (MC38: 770 ± 30 µm, CT26: 870 ± 14 µm, HT-29: 776 ± 30 µm), MatriSpheres were larger with CT26 (919 ± 46 µm) and HT-29 (990 ± 63 µm), but smaller with MC38 (748 ± 41 µm) suggesting cell line-dependent effects in how they utilize ECM from their environment. Matrigel incorporation yielded the largest diameter structures but were variable in all 3 cell lines after 7 days, ranging from 968 ± 141 µm in HT-29 to 1,090 ± 56 µm in CT26.

Figure 2.

CRC cells assemble and organize ECM to form 3D MatriSpheres. A, Colorectal cancer cells are seeded as cells alone or with low concentration of ECM to generate spheroids in ultra-low attachment round-bottom 96-well plates for seven days of culture. B, Representative brightfield images of spheroids after 7 days of culture. Scale bars: 200 µm. C, Quantification of spheroid diameters after 7 days of culture (n ≥ 37 for groups that were quantified). D, Live-dead images of spheroids after 7 days of culture. E, Quantification of cell viability within the spheroids after 7 days of culture (n ≥ 13 for groups that were quantified) F, Representative H&E images of tumor spheroid sections demonstrates differences in micro-tissue organization. Scale bars: 200 µm. G, Picrosirius red (PSR) stained spheroids imaged with polarized light displays the presence of dense collagen fibers in SIS ECM and Collagen I groups. Scale bars: 100 µm. Statistics were calculated by a one-way ANOVA followed by a Tukey’s multiple comparisons test with significance values p < 0.05 is denoted with *, ≤ 0.01 with **, ≤ 0.001 with ***, and ≤ 0.0001 with ****.

MC38 and HT-29 MatriSpheres maintained a similar viability (74–99%) to cells alone controls (83–98%) but was lower for CT26 MatriSpheres (70–90%) (Fig. 2D–E). Other ECMs showed more pronounced decreases in variability, notably Collagen I in MC38 and CT26 cultures, and Matrigel with MC38. Similar to cells alone spheroids, we observed viable cells predominantly at the MatriSphere cortex with dead cells towards the center via live/dead staining. CellTiter-Glo quantification of metabolically active cells (Supplementary Fig. S14A) correlated with diameter except for CT26 cells alone and Matrigel. Enzymatically dissociated 3D spheroids and MatriSpheres at 7 days ranged between 13,000 to 97,000 cells per spheroid after initially seeding 2,500 cells, confirming net proliferation (Supplementary Fig. S14B). These results support that size differences between ECM groups are due to effects of cell proliferation and survival rather than compaction or space occupied by the ECM alone.

We performed detailed histologic analyses using a modified tumor spheroid microarray(20) and confirmed that SIS ECM is assembled and organized into a collagen dense stroma within MatriSpheres by 7 days across CRC cell lines (Fig. 2F). We observed particularly dense ECM regions within certain conditions, such as MC38 MatriSpheres, as shown by H&E and Masson’s Trichrome (Supplementary Fig. S15). Picrosirius red (PSR) imaging with polarized light revealed that these stroma-like regions formed birefringent collagen fibrils (Fig. 2G) that were absent in cells alone spheroid conditions. This lack of endogenous collagen production highlights the need for ECM supplementation in vitro. Concentrically-oriented large diameter collagen fibrils were abundant in both the MC38 and CT26 cell lines when adding SIS ECM or Collagen I. Collagen assembly in HT-29 MatriSpheres was enriched near the periphery in thin bundles. Matrigel assembly was observed, but it is composed of non-fibrillar collagens that are not PSR positive. These morphological patterns identified in CRC MatriSpheres can recapitulate dense collagen dense stromal regions often pronounced in tumors from patients(21).

MatriSphere formation is distinct from ECM hydrogel polymerization and traditional spheroid assembly

We found that low ECM concentrations were sufficient to spontaneously organize into collagen-dense regions within MatriSpheres, altering 3D formation kinetics (Fig. 3A–B, Supplementary Videos S1–S12). Cells alone spheroids displayed similar behavior for all three cell lines; cells sediment at the bottom of the well, aggregate within an hour of seeding, and then proceed to grow outward. However, MatriSphere formation was distinct, classified by: (1) rapid aggregation of multiple small clusters within hours of seeding, (2) coalescence via directed motion to form a single MatriSphere over the following 4–5 days and (3) outward growth. Collagen I supplementation was similar to SIS ECM except in HT-29 cells, which did not fully coalesce at 7 days despite sharing the same soluble collagen concentration. Matrigel supplementation more closely resembled cells alone spheroid assembly during MC38 formation and demonstrated rapid and localized cluster formation with CT26 and HT-29 cells within 24 hours. These results indicate that distinct 3D formation is dependent on ECM composition. Since HT-29 cells formed 3D structures more slowly than cells alone with each of the tested ECM biomaterials, we evaluated HT-29 MatriSphere assembly over longer timepoints (Supplementary Fig. S16A). MatriSpheres continued to coalesce beyond 7 days, forming a more compact structure by day 10. Likewise, Collagen I formed a single structure by 7 days at reduced ECM concentration (25 µg/mL), whereas it was inhibited at higher concentrations. HT-29 spheroids and MatriSpheres continued to proliferate over 7–21 days (Supplementary Fig. S16B–C). These results highlight that ECM composition can be tailored to facilitate MatriSphere assembly in difficult to form cell lines and retain viability in long-term cultures.

Figure 3.

Cell-mediated ECM assembly delays formation kinetics. A, Schematic overview of cell assembly and spheroid formation mechanisms for cells alone and SIS-ECM MatriSpheres. B, Representative images of formation over the first 3 days of culture during time-lapse imaging. Scale bar: 500 µm. C, Tan(δ)⁻¹ plotted as an indicator of gelation of ECM biomaterials (without cells) at working concentrations used in MatriSphere formation and traditional hydrogel formation measured by rheology (n ≥ 2 per group) D, Averaged storage (G’) and loss (G’’) moduli of ECM biomaterials at varied concentrations observed at 30 minutes after exposure to 37°C. A 2-way ANOVA with Dunnett’s multiple comparison test was used to assess significant differences (p-value < 0.05) within or between groups.

Intriguingly, MatriSphere assembly occurred well below ECM concentrations previously reported for hydrogel formation(22). Therefore, we were motivated to elucidate the differences between MatriSphere assembly from other well characterized mechanisms of gelation and to gain insights into whether tumor cells orchestrate this process. Wells with SIS ECM alone did not increase in opacity, which is characteristic of bulk ECM hydrogel formation. Further, rheometry of ECM biomaterials confirmed that there was no gelation under MatriSphere conditions in the absence of cells (125 µg/mL SIS ECM) and remained indistinguishable from media alone. Only SIS ECM concentrations above 1,000 µg/mL resulted in stable and robust gelation defined by sigmoidal increase in storage and loss moduli (G’, G’’), and Tan(δ)⁻¹ > 1 (Fig. 3C–D; Supplementary Fig. S16D). Similar rheologic behavior was found for Collagen I and Matrigel with no gelation at working concentrations. Turbidimetric gelation kinetics assays supported that spontaneous ECM assembly only occurred above concentrations used in MatriSphere cultures (Supplementary Fig. S17A–B). These results suggest that ECM biomaterial incorporation within MatriSpheres is not entirely a passive process. Rather, ECM organization and assembly is likely cell-mediated, which is an emergent phenomenon independent of ECM hydrogel formation.

Decellularized ECM promotes spatial phenotypic organization and heterogeneity within CRC MatriSpheres

Intratumoral heterogeneity arises from both cancer cell variation and environmental factors that can produce high ECM density influencing mesenchymal transition, proliferation, migration, and drug resistance(23–25). Therefore, we characterized spatial microenvironmental organization within MatriSpheres using multiplex fluorescence imaging (Fig. 4A,B; Supplementary Figs. S18–S21). ECM rich regions were defined by labeling fibrillar collagen with a collagen hybridizing peptide (CHP). The large dynamic range of the CHP stain allowed us to infer collagen density across CRC lines: MC38 high, CT26 intermediate, and HT-29 low in MatriSpheres (Supplementary Fig. S22). Low CHP signal was found in Matrigel, which has little fibrillar collagen, consistent with our PSR results. Immunofluorescent staining with cadherin-specific antibodies, E-cadherin (E-CAD) and N-cadherin (N-CAD), were used to determine cell-cell adhesion phenotypes. We found that ECM dense regions within MatriSpheres affected cadherin expression patterns. Total expression was similar except in CT26 cells, showing increased N-CAD expression at the periphery for each ECM-incorporated group compared to cells alone (Fig. 4C).

Figure 4.

CRC MatriSpheres display heterogenous phenotypes with enhanced cell-ECM interaction. A, Multiplex fluorescence staining for CHP (denatured fibrillar collagen, green), N-CAD (red), and DAPI (blue) within MatriSphere histologic sections. Scale bar: 200 µm. B, Multiplex fluorescence staining for CHP (denatured fibrillar collagen, green), Ki67 (proliferation, red), CA IX (hypoxia, yellow), and DAPI (blue) within MatriSphere histologic sections. Scale bar: 200 µm. C, Quantification of N-CAD expression in representative MatriSphere sections. Statistics were calculated by a one-way ANOVA followed by a Tukey’s multiple comparisons test. D, Quantification of Ki67 expression in representative MatriSphere sections. Statistics were calculated by a one-way ANOVA followed by a Tukey’s multiple comparisons test. E, Representative MC38 + SIS ECM MatriSphere section stained for CHP (denatured fibrillar collagen, green), E-CAD (yellow), N-CAD (red), and DAPI (blue). Cadherin expression is higher in ECM low regions than in ECM high regions. Full image scale bar: 250 µm. Inset image scale bar: 50 µm. Statistical significance where p < 0.05 is denoted with *, ≤ 0.01 with **, ≤ 0.001 with ***, and ≤ 0.0001 with ****.

All conditions expressed carbonic anhydrase 9 (CA IX) indicating hypoxia, with no differences between groups, and similar numbers of Ki67⁺ proliferating cells (Fig. 4D; Supplementary Fig. S23A). Proliferation was primarily cell line dependent (20%, 30%, and 50% for MC38, CT26, and HT-29, respectively) rather than ECM dependent during homeostatic culture conditions at 7 days. These results show that ECM organization is heterogeneous, cell line dependent, and influences dynamic cell-cell and cell-ECM interactions that determines CRC MatriSphere morphology and phenotype. Furthermore, cadherin expressing cells (either E-CAD or N-CAD) segregated into cell-dense tumor nests within MatriSpheres whereas cadherin negative cells were enriched in ECM dense regions within MC38 MatriSpheres (Fig. 4E). This suggests that these cells prefer cell-matrix interactions over cell-cell interactions when ECM is present. However, it is unknown whether this represents a phenotypic switch or preferential organization during assembly. This staining pattern was consistent in CT26 MatriSpheres, but not observed in HT-29 MatriSpheres, which may be attributed to its low ECM density (Supplementary Fig. S23B). ECM dense tumor stroma has been implicated in affecting nutrient diffusion and sequestering factors such as cytokines and growth factors(26), which could impact the ability of cells to proliferate or increase hypoxia within the spheroid. Decellularized tissues such as SIS ECM preserve this complexity and have been previously shown to maintain the ability to bind such soluble factors(27).

CRC MatriSpheres transcription profiles recapitulate in vivo tumor heterogeneity

We determined the effects of SIS ECM on CRC MatriSphere phenotype compared to cells alone spheroids using bulk RNA sequencing. MatriSpheres were transcriptionally distinct from spheroids and separated into distinct phenotypic clusters via principal component analyses (PCA) (Fig. 5A; Supplementary Figs. S24,S25). 112, 165, and 28 high-significance differentially-expressed genes (DEGs, adjusted P-val < 0.01 and FC > 2) were identified in MC38, CT26, and HT-29 MatriSpheres, respectively (Fig. 5B). Specific gene regulation patterns in MatriSpheres were cell line dependent. MC38 MatriSpheres downregulated several ECM genes (Col1a1, Col28a1, Dmp1, Vtn, Aspn) (Fig. 5C); however, increased expression of the long non-coding RNA Gm20427, which accounted for the highest fold change. CT26 MatriSpheres upregulated genes associated with CRC aggressiveness and invasion, including Tnfrsf11b(osteoprotegerin), Mmp13, and zinc finger family of proteins (Zfp704, Zfp936)(28,29). For HT-29 MatriSpheres, PLAT (plasminogen activator, tissue type) showed the largest fold change (4.73), along with upregulation of multiple APOBEC genes (APOBEC3G, APOBEC3C, APOBEC3F). APOBEC3s are known to play to a role in cancer mutagenesis and APOBEC3G is highly expressed in colon cancer(30).

Figure 5.

MatriSpheres alter CRC transcriptome and secretome capturing in vivo tumor heterogeneity. A, PCA plots clustering RNA Seq expression data of both mouse and human CRC spheroids cultured with and without SIS ECM at day 7. B, Volcano plots of differentially expressed genes across all three cells line as compared to cells alone spheroid controls. C, Top 10 genes for each CRC cell line by fold change. D, Top 10 gene set enrichment based upon GSEA. Significantly upregulated (orange) and downregulated (blue) gene sets. E, (MC38 and CT26), H, (HT-29) Heatmap displaying top cytokine fold change between cell culture supernatant from MatriSpheres compared to cells alone spheroids at Day 7. F, (MC38 and CT26), I, (HT-29). Modified correlation plot identifying similarities between conserved transcriptome and secretome signatures G, (MC38 and CT26), J, (HT-29) Chemiluminescent intensity values of correlated proteins. K, tSNE plot of cell clusters identified within in vivo MC38 tumors based upon scRNA Seq expression data. L, tSNE plot comparing bulk RNA Seq expression level data of in vitro MC38 MatriSpheres and cell alone spheroids to scRNA Seq expression of in vivo MC38 cancer cells and percentage of correlated cells. M, Top 10 upregulated and N, downregulated genes defining the high confidence correlation of in vitro MC38 tumor spheroids with in vivo cancer cells. Significant differences were defined as p-value < 0.01 for RNA Seq data.

Gene set enrichment analysis (GSEA) revealed broad phenotypic regulation in MatriSpheres. The most strongly regulated pathways in MC38 MatriSpheres (highest normalized enrichment score, NES), were negatively correlated relative to cells alone (Fig. 5D) with the largest impact in suppression of hypoxia signaling. Additionally, several metabolism-related pathways were downregulated in response to SIS ECM. Within the hypoxia gene set, there were 68 leading edge genes, including Serpine1, which blocks plasminogen activation and limits fibrin degradation (Supplementary Fig. S26A) and has been implicated in CRC TME remodeling by regulating immune cell infiltration(31). GSEA for CT26 MatriSpheres revealed an inflammatory response led by the upregulation of interferon (IFN) signaling pathways, Sp140 an activator of STAT1, as well as IFN regulatory factors (Irf9, Irf7, Irf1) and IFN-induced transmembrane proteins (Ifitm3, Ifitm1, Iftim2) (Supplementary Fig. S26B)(32–34). Similar to CT26 MatriSpheres, several enriched gene sets within HT-29 MatriSpheres are involved in inflammation, including IFN-α, IFN-γ. We also noticed an increase in the epithelial to mesenchymal transition (EMT) gene set, which is an indicator of metastatic potential, and negative regulation of cholesterol homeostasis. IFN-α response pathway was the most perturbed gene set (61 leading edge genes and NES of 3.21, (Supplementary Fig. S26C) and IFITM1 was the highest ranked gene and has been proposed as a marker of poor survival in CRC patients(35). We inferred more generalizable gene regulation across 8 shared DEGs between MC38 and CT26 MatriSpheres, including Mmp13, Aspn, and Dlx2. We posited that upon introducing SIS ECM within MatriSpheres we would perturb the Adhesome(36,37). GSEA of adhesion-related pathways from MSigDB identified 20 shared genes between MC38 and CT26 (unique to mouse) and an additional 15 genes shared by all three mouse and human lines (Supplementary Fig. S27A). Integrins, Itgb4 and Itgb5, that facilitate cell adhesion to ECM, were perturbed in response to SIS ECM, and may provide unique targets for studying cancer cell-ECM interactions within the TME (Supplementary Figs. S27B,S28–S29).

Many of the transcriptionally regulated factors in MatriSpheres corresponded with a functional CRC cell secretory phenotype (Fig. 5E). In MC38 MatriSphere supernatant, we found downregulation of both Serpin E1 (Serpine1) and Periostin (Postn), which agrees with RNA sequencing (Fig. 5F,G). For CT26, only osteoprotegerin (Tnfrsf11b) was significantly increased in MatriSpheres. In HT-29 MatriSpheres, 38 cytokines were increased (FC ≥ 2) (Fig. 5H), with three pro-inflammatory cytokines (GROα, IL-8, and IP-10), showing correlation in both secretion and RNA sequencing analyses (Fig. 5I,J). These cytokines are essential factors in the wound healing response and have been cited as pro-tumor markers leading to poor survival in CRC patients(38).

A primary aim of this study was to establish an in vitro model to recapitulate in vivo ECM conditions of the TME. We benchmarked MatriSpheres to in vivo MC38 tumor cells analyzed via scRNA seq. We defined MC38 cancer cells from other immune and stromal cell types (Fig. 5K) and found that SIS ECM MatriSpheres recapitulated a heterogenous subpopulation of in vivo MC38 cells using Scissor analysis(39), with a greater phenotypic correlation score compared to MC38 spheroids (Fig. 5L). Of 20,796 MC38 cells, 34.5% positively correlated with MatriSpheres and 18.6% for spheroids. This analysis indicates that MatriSpheres can drive unique CRC phenotypes observed within primary in vivo tumor cells. A DEG analysis between these distinct MC38 cancer cell populations in vivo showed that cells more similar to MatriSpheres upregulated histone genes (Hist2h2ac, Hist1h1a, Hist1h1b, Hist1h1e, Hist1h4d, Hist1h2ae), proto-oncogene Mybl2, and downregulated ECM protein and remodeling genes (Col3a1, Col1a1, Mmp2) (Fig. 5M,N). Aberrant histone modification and expression can disrupt DNA folding and replication dynamics(40,41) and Mybl2 overexpression has been linked to reduced CRC patient survival(42). These results suggest that in vitro phenotypes may improve modeling of specific cancer cell subpopulations.

CRC MatriSphere and spheroid phenotype depends on ECM composition

To assess the role of ECM composition within MatriSpheres, we performed qPCR across all three biomaterials. SIS ECM regulated each of the tested genes compared to cells alone in agreement with our RNA seq analysis. Comparing SIS ECM to Collagen I directly, we found differences in the magnitude of this regulation across six genes: Crlf1 (1.36-fold), Igfbp3 (1.71-fold), Mvd (1.75-fold), Nos1 (1.77-fold), Postn (1.72-fold) and Serpine1 (1.43-fold) (Fig. 6A; Supplementary Fig. S30). Since soluble collagen content was consistent between groups, these differences highlight the tissue specific composition in a decellularized tissue compared to purified ECM component. Matrigel, which lacks fibrillar collagens, demonstrated an opposite expression pattern (8 of 13 genes), such as reduced Mmp13 expression (−3.62-fold) and Aspn (−18.1-fold) versus SIS ECM. These observations show the critical importance of biomaterial composition and its effects on cancer cell phenotype within the TME supporting the use of MatriSpheres for tumor-mimetic 3D cultures.

Figure 6.

Gene expression profile of MatriSpheres altered by ECM composition. A, Gene expression analysis using qPCR evaluated by fold change (2^−(ΔΔCt)) comparing MC38 MatriSpheres with different ECM biomaterials compared to cells alone spheroids. B, Venn diagram showing the number of predicted pathways and their activity based upon RNA transcription patterns across CRC MatriSpheres. Filtered by |z-scores| ≥ 2. The table highlights the 3 significantly overlapping pathways and the corresponding z-scores across all CRC cell lines. C, Bubble plot displaying shared significant pathways. Pathways shown contained two or more CRC cell lines with |z-scores| ≥ 3. D, Flow chart of interactions between IPA predicted upstream regulators and target genes conserved between all CRC cell lines. E, Heatmap of target genes within the IPA tumor microenvironment pathway that have interactions with the 5 identified upstream regulators.

We identified pathway activation in CRC MatriSpheres utilizing Ingenuity Pathway Analysis (IPA) and found strong regulation of cancer-associated, immune-associated, and metabolism-associated pathways across cell lines. The most significant change in MC38 MatriSpheres was decreased activity in metabolism-associated pathways, specifically cholesterol biosynthesis, which agreed with the GSEA findings (Supplementary Fig. S31A) and has several potential implications within the TME (43). The top DEGs in each category showed decreases in collagen expression (Col1a1, Col28a1), immune-associated pathways (Il2rg and Rras), and metabolism-associated genes (Mvd and Dgat2) (Supplementary Fig. S31B)(44). CT26 MatriSpheres predicted immune modulation resulting from ECM interactions, activating IFN signaling and cytokine storm pathways that may indicate pro-inflammatory signatures (Supplementary Fig. S31C). Cancer-associated pathways influenced by SIS ECM in CT26 MatriSpheres were driven by an upregulation in 2’−5’ oligoadenylate synthetases 2 and 3 (Oas2, Oas3) (Supplementary Fig. S31D), which are also immunoregulatory enzymes activated by IFN signaling cascades to establish an immunosuppressive microenvironment(45). IFN-related genes (Prkd1 and Irf7) were also enriched, which interact with β-catenin to control proliferation in CRC(46) and correlates with greater immune cell-infiltrates and poor survival in CRC patients, respectively (47).

HT-29 MatriSpheres predicted activation of cytokine production, fibrosis, wound healing, and significant inhibition of cholesterol biosynthesis (Supplementary Fig. S31E). We found both PRKCG (protein kinase c gamma) and FN1 (fibronectin) upregulated within the fibrosis signaling pathways (Supplementary Fig. S31F). PRKCG has been shown to increase CRC cell migration(48) and elevated levels of fibronectin have been implicated in poor prognosis and survival in CRC patients(49,50). IL-8 signaling contained the highest fold change from immune-associated pathways with significant increases in both CXCL1 (CXC motif ligand 1) and CXCL8 (CXC motif ligand 8). CXCL1 and CXCL8 play critical roles in CRC progression and metastasis by suppressive myeloid cell recruitment(38).

We discovered three shared pathway activity patterns predicted across all CRC MatriSpheres: Superpathway of Cholesterol Biosynthesis, Role of Chondrocytes in Rheumatoid Arthritis, and Pyroptosis Signaling (Fig. 6B). Additionally, 22 other pathways were shared between at least 2 out of 3 cells lines with high activation significance (|z-scores| ≥ 3) (Fig. 6C). CT26 and HT-29 MatriSpheres showed the most similarity within the cancer and immune-associated pathways 61% (11/18) and 55% (12/22) of the total pathways shared. The most significant pathways between CT26 and HT-29 were related to IFN signaling, specifically the Role of PKR in IFN Induction and Antiviral Response ( z-score: 3.46) and IFN Signaling (z-score: 3.38).

We then identified common regulators and gene targets predicted in CRC MatriSpheres across all 3 cell lines from a list of more than 6600 upstream regulators from IPA. We narrowed these interactions to 5 conserved upstream regulators CITED2, BHLHE40, TP73, ESR2 and STAT4 that directly interacted with 4 common gene targets B2M, SERPINE1, MMP14 and SDC4 (Fig. 6D) (p-value < .001 and |z-score| > 2). We hypothesize that these regulator-target interactions provide insights of possible pathways that are dependent upon ECM interactions that are otherwise absent in traditional spheroid models. Across all 3 CRC MatriSpheres cell lines, IPA predicted CITED2 gene inactivation with strong confidence, which would indicate this regulator may have a similar signal transduction stimulated by SIS ECM. CITED2 (Cbp/P300 Interacting Transactivator with Glu/Asp Rich Carboxy-Terminal Domain) is a transcription regulator that acts as a competitive inhibitor to HIF-1α signaling and can alter activation states of macrophages(51), regulate cell proliferation(52), and mediate colorectal cell invasion(53). Lastly, we found 40 target genes that associate with at least 2 upstream regulators, out of 89 with at least one regulator interaction (Fig. 6E). These findings suggest that SIS ECM or stromal interactions in MatriSpheres influence CRC signaling pathways beyond the transcriptional level, which may lead to broad TME regulation.

Discussion

In this study, we developed MatriSpheres: a 3D culture system that incorporates decellularized ECM within spheroids through cell-directed self-assembly. This spontaneously creates a tumor-like ECM stroma that induces phenotypic changes mimicking in vivo tumor heterogeneity. This facile and reproducible approach displays characteristics of tumor morphogenesis and is rheologically and structurally distinct from traditional hydrogel formation used in organoid methods. This model system displayed a dynamic reciprocity between CRC cells and a proteomically complex intestinal ECM. Specifically, the cells organized the partially-solubilized matrix, and in turn, the integrated matrix influenced CRC MatriSphere transcriptome and secretome phenotypes. Gene set enrichment and pathway analyses showed that SIS ECM influenced matrix remodeling, immune cell signaling, cell cycle, and lipid metabolism, which each play critical roles in cancer progression. Most notably, in vitro MatriSpheres demonstrated improved correlation with in vivo CRC cells over traditional spheroids. MatriSpheres provide a unique tool to seamlessly enhance spheroid complexity as well as their applicability for disease modeling and high-throughput therapeutic screening.

A significant finding of this study was that MatriSpheres containing decellularized ECM biomaterials enhance CRC spheroid ECM diversity and morphological complexity. In vitro tumor cultures have been unable to reliably predict therapeutic responses observed in the clinic, which is hypothesized due to a lack of TME hallmarks and do not reflect tumor heterogeneity. More complex 3D in vitro models (spheroid, organoid, and 3D bioprinting) and in vivo models (patient-derived xenografts) have been used to capture the complexity of solid tumors. However, with increasing complexity comes a greater financial burden, reduced analysis throughput, and higher variability, which could limit their implementation and highlights the importance of a balance of practicality and predictiveness. Therefore, our initial objectives in MatriSphere development were to (1) bridge the gap in 3D model complexity between traditional spheroids and ECM-competent organoids, (2) facilitate intestinal tissue-specific ECM composition using decellularized tissues, and (3) maintain the ease of standard spheroid workflows.

A benefit of MatriSpheres is their modular design, which allows cancer cells to assemble a multitude of ECM types, derived from different tissue sources, to match a tumor of interest. Every tissue in the body possesses a unique ECM composition and architecture, including intestinal tissues in which CRC arises. Decellularization has allowed researchers to isolate tissue-specific ECM scaffolds that contain unique biochemical composition. The ECM can be processed into multiple forms and has been used clinically for its innate regenerative properties(54,55). Previously, spheroids made from human adipose derived stem cells or breast cancer cells were augmented with ECM particles derived from Collagen I, bone, brain, cartilage, adipose, lung, and spleen and showed discordant cell responses to varied ECM biomaterial composition(12). Another study showed that orthotopic decellularized intestinal ECM hydrogels could improve gastrointestinal organoid maturation and function suggesting that Matrigel alone may not be the most effective ECM biomaterial for tumor modeling(56). Our proteomics analysis indicated that SIS ECM substantially differs from Matrigel and Collagen I in matrisome diversity, abundance and shared homology with both mouse and human CRC tumors, supporting the use of an alternative mass producible ECM source. Another unique advantage of MatriSpheres is their ability to enable tumor cells to rapidly develop a mature ECM compartment by self-assembly rather than de novo synthesis that is otherwise absent due to the lack of specialized cell types, such as fibroblasts, which are professional ECM generators. In traditional cells alone spheroid models, we find small amounts of matrix deposited over short time scales (2–7 days) and are generally devoid of fibrillar collagens that provide structural and mechanical support for virtually all solid tumors(57). This was confirmed by single cell analysis of in vivo MC38 tumor cells which showed relatively low matrisome expression compared to stromal cells. In other words, this method provides user-defined ECM composition without the need for heterotypic cell cultures that may prevent the study of direct cancer cell responses to their changing microenvironment. Though CRC cells likely do not generate the bulk ECM stroma of tumors, recent studies have shown their contributions create a highly localized pericellular matrix that promotes drug resistant phenotypes(58). This suggests the importance of ECM in tumor modeling but requires further investigation of cell contributions toward matrix deposition in addition to the observed stroma formation from supplemented ECM biomaterials.

The primary objective of a 3D tumor culture system is to recapitulate in vivo phenotypes under controlled in vitro conditions that are not possible in simplified models. Thus, we asked whether MC38 MatriSpheres would provide a tumor-like niche that would drive similar responses to in vivo MC38 cancer cells. We found that MatriSpheres induced unique transcriptomic phenotypes that correlated with the heterogenous profiles of CRC cell populations identified from in vivo MC38 tumors. Interestingly, single cell Scissor analysis(39) revealed that in vivo MC38 tumors stratified into subpopulations that matched either MatriSpheres or cells alone spheroids. Our results suggest that a single culture condition may never fully capture in vivo tumor heterogeneity and that there is utility in examining tumor spheroids that are both ECM rich and poor. This may address a challenge in treating solid tumors where cancer cells populate heterogeneous TME niches. These characteristics have hindered clinical efficacy of targeted cancer therapies informed predominantly by genetic mutations rather than the influence of the TME(59). Additionally, our qPCR analysis showed the importance of ECM composition in directing cancer cell phenotype with major ECM components such as fibrillar collagens mediating gene expression. These findings illustrate the importance of utilizing ECM biomaterials for solid tumor modeling to mimic the diverse cancer cell phenotypes observed in vivo.

We found that the concept of dynamic reciprocity was a promising framework to describe both ECM organization within MatriSpheres and subsequent phenotypic regulation. Dynamic reciprocity, coined several decades ago by Bissell et al., describes the bidirectional relationship between cells and their extracellular environment(60). The first principle of dynamic reciprocity was at work during MatriSphere formation, as cancer cells directed assembly of the ECM dispersion into defined networks encapsulating small cellular clusters, or nests. Distinct differences in ECM organization were observed between cell lines thus demonstrating cell-specific ECM utilization. For example, both MC38 and CT26 lines showed intraspheroidal ECM uptake; however, MC38 MatriSpheres displayed regions of hyper densified collagen fibers whereas CT26 showed an even distribution of thinner collagen fibers. In contrast, HT-29 cells preferentially assembled a peri-spheroidal collagen barrier with limited ECM assembly toward the spheroid core. In addition to cell-dependent ECM assembly, the second principle of dynamic reciprocity was evident wherein each CRC line responded differently to ECM via changes to their transcriptomes and secretomes. An advantage of MatriSpheres compared to organoids is the ability to decouple the contribution of ECM in 3D cultures by comparing to traditional spheroids. We identified multiple ECM-driven changes in CRC cells, such as matrix remodeling, immune regulation, cell cycle and cholesterol metabolism. ECM-associated gene expression was particularly interesting and reinforces dynamic reciprocity. MatriSpheres typically downregulated structural ECM proteins (e.g. collagens) and enhanced ECM protease (e.g. matrix metalloprotease) expression. We found that ECM engagement affected other cancer cell-intrinsic gene expression programs that potentially shape the TME including immunomodulation of inflammatory gene signatures and cytokines. Namely, Type I and II IFN-associated genes were heavily upregulated in CT26 and HT-29 spheroids in response to SIS ECM. To date, there has been a vast amount of literature documenting the important role of IFN signaling in altering tumor homeostasis and responsiveness to immunotherapy(61,62). This inflammatory profile was supported by the enhanced cancer cell secretion of pro-inflammatory cytokines (GROα, IL-8 and IP-10) within HT-29 MatriSpheres. This would indicate that the ECM is inducing cancer cells to recruit immune cells to the tumor site, but it is unknown whether this recruitment would polarize these cells toward a pro- or anti-tumor phenotype. We observed that ECM also played a causal role in cancer cell metabolism, robustly inhibiting cholesterol metabolism pathways in each of the tested CRC cell lines, but most prominently in MC38 and HT-29 MatriSpheres. Cholesterol metabolism is a critical aspect of cellular maintenance and has become an intriguing target for cancer therapy. Although no single gene associated with this pathway was substantially altered, the collective downregulation of these genes resulted in the highest activation score among all pathways. This result predicts that the ECM within MatriSpheres significantly modifies the lipid biosynthesis machinery within cancer cells and could provide a model for studying the mechanism of ECM regulation of cancer cell metabolism. MatriSpheres highlight the influence of cancer cells on the TME as well as their plasticity and capacity for being manipulated by the ECM(63). We observed features that were consistent with matrix-bound nanovesicles (MBVs), cell secreted vesicles that have been previously shown to be sequestered within SIS ECM and may contribute to tissue specific environmental cues. MBVs contain varied cargo, such as growth factors and microRNAs and may play a role in modulating the tumor microenvironment.

As with any in vitro model, there are several limitations and questions that require further investigation. Here, we utilized healthy porcine small intestine as a tissue-specific ECM source for modeling CRC tumor stroma, but uncharacterized factors within tumor-specific ECM may be essential for generating more representative models of different tumor types (64,65). This limitation could be addressed by using matched decellularized tumor ECM that would offer a direct comparison between healthy and diseased microenvironments in vitro. As a proof of principle, we confirmed that MatriSpheres can be generated using decellularized ECM sourced from different sites and species (Supplementary Fig. S32). In each case, we noticed MatriSphere formation with unique stromal morphology, which could be important for evaluating different cancer types and models of metastasis. Although ECM is highly conserved among mammalian species, xenogeneic effects may affect these described interactions. Therefore, it would be prudent to assess differences using species-matched ECM. Transcriptional benchmarking of in vitro MatriSpheres was evaluated using a scRNA seq data set from a syngeneic MC38 tumor mouse model generated in the flank; however, this approach does not necessarily reflect the direct influence of the local microenvironment on CRC cell phenotype (66). Utilizing an orthotopic tumor model and/or a genetically engineered mouse model compared with tumor derived ECM would be valuable next step towards elucidating cancer cell phenotype compared to MatriSpheres based upon the site of cancer cell implantation or generation in vivo.

Further investigation is required to characterize the unique 3D ECM organization, composition, and quantity found within the MatriSpheres post-assembly, which may impact several factors in the TME (e.g. cell-ECM adhesion, mechanics, etc.). For example, we observed qualitative changes in ECM assembly and density within MatriSpheres that may affect intraspheroidal stiffness and mechanotransduction in cancer cell signaling, which could substantially alter cancer cell response within the TME (67). Supplemental experiments are necessary to determine the extent of cancer cell ECM deposition and remodeling at the protein level, as transcriptomic analysis suggests several critical upstream regulators that may influence TME dynamics. Although we determined that cell participation is necessary for ECM organization and assembly within MatriSphere, the mechanism remains unknown. At least 22 ECM receptor coding genes were co-expressed across the tested cell lines in different ratios and combinations, which may provide clues that govern 3D organizational complexity. Including additional cell types found in the TME, such as immune cells and fibroblasts that support tumors via paracrine and metabolic factors, would further enhance model complexity. The modular nature of MatriSpheres could accommodate their use in a heterotypic co-culture system to study the effects of the changing microenvironment on cancer cell behavior. The small ECM quantities used in MatriSphere formation makes human sourced tumor ECM a practical possibility. This highlights the potential use of this method for high-throughput drug screening and precision medicine approaches.

Significance Statement:

MatriSpheres provide a hydrogel-free 3D platform for decoupling the influence of heterogeneous extracellular matrix components on tumor biology and can broadly facilitate high-throughput drug discovery and screening applications.