Engineered colorectal cancer tissue recapitulates key attributes of a patient-derived xenograft tumor line

PDMS molds were fabricated using the SYLGARD® 184 Silicone Elastomer Kit (Dow Corning). SYLGARD® 184 Silicone Elastomer base and SYLGARD® 184 Silicone Elastomer curing agent were mixed thoroughly at a ratio of 10:1 (w/w). The mixture was then degassed using a desiccator equipped with a vacuum pump for 15 min to remove any air bubbles. The liquid PDMS was poured onto a glass, sandwiched by another glass, and cured in an oven at 70 °C for 1 h. The PDMS layer was then separated from the glasses, and 1 inch PDMS disks, each with ten 4 mm diameter wells were made using a hole-punch kit.

PEGDA was synthesized following previously established methods [36]. The conjugation of fibrinogen to PEG was also carried out according to previously described methods [58]. The characterization of PEG-Fb was performed by evaluating PEGylation efficiency and the ability of PEG-Fb to photocrosslink and form 3D hydrogels [59].

All experimental procedures were approved by the Animal Care and Use Committee. A frozen stage II CRC adenocarcinoma PDX line from a white female patient pre-treatment was kindly provided by BioConversant, LLC. To establish xenografts from frozen PDX cells, approximately 1.4 × 10⁶ tumor cells were mixed at a 1:1 ratio with Matrix High Concentration, Phenol-Red free Matrigel (Corning) and injected into the flank of a NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NOD-SCID) mouse (The Jackson Laboratory). To propagate the PDX line, the freshly excised tumor (∼0.8 g) was washed in sterile phosphate buffered saline (PBS), minced in approximately 2.5 ml Dulbecco's Modified Eagle Medium (DMEM) with 100 U ml⁻¹ penicillin and 100 µg ml⁻¹ streptomycin, and approximately 200 µl of the minced tumor suspension was injected into the flank of a NOD-SCID mouse. Experiments discussed in this study were performed using CRC-PDX tumors that were propagated in the absence of Matrigel and were passaged 3–4 times in mice.

Excised CRC-PDX tumors were dissociated according to the methods described in the supplementary methods (available online at stacks.iop.org/BF/14/045001/mmedia), and all PDX cells, including both human cancer cells and mouse stromal cells [28, 60], were isolated for subsequent in vitro culture as 2D-CRC-PDX cells and 3D-eCRC-PDX tissues.

To create the 3D-eCRC-PDX tissues, all isolated CRC PDX cells (i.e. both the adherent and non-adherent cell populations later observed in 2D culture) were mixed with a polymer precursor solution containing PEG-Fb with a final concentration of 1.5% (v/v) triethanolamine (Acros Organics), 37 mM 1-vinyl-2-pyrrolidinone (Sigma-Aldrich), and 0.1 mM Eosin Y in PBS at a density of 20 × 10⁶ cells ml⁻¹. The cell-laden precursor solution was gently pipetted into the prefabricated PDMS mold wells (10 μl well⁻¹) and photocrosslinked with visible light for 2 min. Molds were then removed, and cell culture media added. The 3D-eCRC-PDX tissues were maintained in long-term culture (29 d) in a standard incubator with a temperature of 37 °C and 5% CO₂ with media refreshes every 2–3 d.

CRC cells isolated from the CRC-PDX tumors were cultured in T-75 standard cell culture flasks using the same cell culture media as was employed for the 3D-eCRC-PDX tissues. These 2D cultured CRC PDX (2D-CRC-PDX) cells were passaged after 8 d of culture and subsequently passaged every 7 d up to a total of 29 d. Importantly, since non-adherent cell types (mostly cell aggregates) were present, the spent media containing the 2D-CRC 2D-CRC-PDX cells was also collected and centrifuged at 300 ×g. The non-adherent cell pellets were then resuspended in fresh media and recombined with the seeded adherent cells. All of the cultured cells were maintained in a standard incubator with a temperature of 37 °C and 5% CO₂, with media changes every 2–3 d.

To investigate the cell growth curve or the increase in the total number of 2D-CRC-PDX cells over time, 2 × 10⁵ cells cm⁻² were seeded into ten T-25 standard cell culture flasks, one of which was used for cell counting every 24 h (days 1–11) using 0.25% Trypsin-ethylenediaminetetraacetic acid (EDTA) (Corning) to detach the cells. The spent media containing the 2D-CRC-PDX non-adherent cells was also collected and included in cell counting. Following staining with trypan blue, the live cells were then counted using a hemocytometer. Additionally, 2 × 10⁵ cells cm⁻² were seeded onto a 24-well plate to investigate the viability of cells in 2D-CRC-PDX culture on day 1 by Live/Dead staining.

A cell viability assay was performed using the LIVE/DEAD® Viability/Cytotoxicity Kit (ThermoFisher Scientific) according to the manufacturer's protocol. To quantify the percentage of live cells, Hoechst 33342 (Calbiochem) (20 µg ml⁻¹) was used to counterstain the cell nuclei. The fluorescently labeled cells were observed, and images and Z-stacks were obtained using a Nikon Eclipse TE2000 Inverted Microscopes equipped with a Nikon A1 plus Confocal Microscope System. The confocal microscopic images of the 2D-CRC-PDX cells and Z-stacks of the 3D-eCRC-PDX tissues were analyzed using ImageJ software (version: 2.0.0-rc-68/1.52e). The percentage of viable cells was calculated by dividing the manually quantified number of viable cells (total cell nuclei count minus dead cells) by the total number of cell nuclei (total number of cells). Further details on the staining and imaging procedures are provided in the supplementary methods.

The 3D-eCRC-PDX tissues were cultured in vitro over a 29 d period and imaged every 7 d. Phase contrast images and Z-stacks of 3D-eCRC-PDX tissues were obtained using an inverted Nikon Eclipse Ti microscope equipped with an Andor Luca S camera and analyzed using ImageJ software (version: 2.0.0-rc-68/1.52e) to quantify the area, major axis, minor axis, circularity, and aspect ratio of the tumor cell colonies within 3D-eCRC-PDX tissues. For 2D-CRC-PDX cells, phase contrast images were obtained prior to cell passaging.

By selecting random regions from phase contrast images or Z-stacks, three 3D-eCRC-PDX tissue batches with at least 50 tumor cell colonies from each batch, were analyzed and the data exported to MS Excel. The diameter represented in the section 3 is the mean geometric diameter, which was calculated using equation (1):

$$\begin{equation}{\text{Geometric diameter}} = \sqrt {{\text{major axis}} \times {\text{minor axis}}} .\end{equation} \tag{ 1 }$$

To investigate the cell subpopulations using flow cytometry, the CRC-PDX tumors, 3D-eCRC-PDX tissues, and 2D-CRC-PDX cells were dissociated to obtain single cells, and the numbers of viable cells were manually counted post-dissociation as described above. Detailed cell staining and flow cytometry protocols are provided in the supplementary methods. Briefly, to identify human-derived cells, mouse-derived cells, CRC cells, and proliferating cells the following antibodies were used, respectively: β-2-microglobulin (B2M) mouse IgG2a mAb (OriGene Technologies) conjugated to Zenon™ R-phycoerythrin (PE) mouse IgG2a (ThermoFisher Scientific), MHC class I H-2Db antibody (H2Db) mouse IgG2a mAb (OriGene Technologies) conjugated to the Zenon™ PE mouse IgG2a, cytokeratin 20 (CK20) rabbit IgG mAb (Cell Signaling Technology) conjugated to Zenon™ Alexa Fluor™ 647 rabbit IgG (ThermoFisher Scientific), and Ki-67 rabbit IgG pAb (abcam) conjugated to the Zenon™ Alexa Fluor™ 647 rabbit IgG.

To exclude dead cells and debris from the analysis, cells were labeled with Zombie Green™ dye (Zombie Green™ Fixable Viability Kit, BioLegend). To block nonspecific binding on the cell surface, cold blocking buffer (0.5% (w/v) bovine serum albumin (BSA) (Sigma-Aldrich) and 10% (v/v) fetal bovine serum (FBS) in 1X PBS) containing Human BD Fc Block™ (BD Biosciences) (12 µg ml⁻¹) and Mouse BD Fc Block™ (BD Biosciences) (5 µg ml⁻¹) was used. To fix and permeabilize the cells, 0.5 ml of cold Foxp3 Fixation/Permeabilization (eBioscience™) working solution (prepared according to manufacturer protocol) was used for each sample tube. To block nonspecific binding for intracellular staining, fluorescent-activated cell sorting (FACS) buffer (0.5% (w/v) BSA and 10% (v/v) FBS in 1X permeabilization buffer) was used. Compensation beads (ThermoFisher Scientific) were used according to the manufacturer protocol to accurately set the compensation in the event of fluorochrome overlap. To monitor for cell autofluorescence, unstained cell controls were also used. To set an accurate compensation for Zombie Green™ dye, a dead control cell population was used. A BD Accuri C6 cytometer was used to quantify labeled cells and subsequently, FlowJo software (version 10.0.7) was employed for data analysis.

In addition to investigating the percentage of cell subpopulations in culture over time, we also assessed the change in the total number of cells and the number of cells in each subpopulation. To calculate the number of cells in each subpopulation, the population percentage quantified via flow cytometry was multiplied by the total number of viable cells counted as described above. To account for obligatory cell passaging required for maintenance in the 2D-CRC-PDX culture, total numbers of 2D-CRC-PDX cells were calculated based on cell counts at each time point and the number of cells seeded at each passage.

To study the cell morphology within both 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures, the cells were labeled with Hoechst 33342, B2M mouse IgG2a mAb with goat anti-mouse IgG Alexa Fluor™ 488 (ThermoFisher Scientific), Alexa Fluor™ 568 Phalloidin (ThermoFisher Scientific), and CK20 rabbit IgG mAb conjugated to Zenon™ Alexa Fluor™ 647 rabbit IgG to observe the cell nuclei, human-derived cells, f-actin filaments, and CRC cells, respectively. The cells were imaged using a Nikon confocal microscope. Further detail is provided in the supplementary methods.

To visualize both cellular and tissue microanatomy, a portion of the CRC-PDX tumors and entire 3D-eCRC-PDX tissues were sequentially sectioned and embedded in paraffin. The paraffin-embedded sections of CRC-PDX tumors and 3D-eCRC-PDX tissues were stained using hematoxylin and eosin (H&E). The stained sections were scanned using an Aperio CS2 scanner (Leica Biosystems) at 40× magnification. The digital images were viewed using Aperio ImageScope (Ver. 12.3.2.8013). Further details on the H&E staining protocol are provided in the supplementary methods.

Total RNA was extracted from CRC-PDX tumors, 2D-CRC-PDX cells, and 3D-eCRC-PDX tissues using the RNeasy Plus Micro Kit (Qiagen). The analysis was carried out for 3–4 separately obtained tumors and the corresponding batches of 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures. Total RNA (50 ng) was used for complementary DNA (cDNA) synthesis using SuperScript™ IV First-Strand Synthesis System (Invitrogen). qPCR amplifications were performed using RT2 SYBR Green qPCR Mastermix (Qiagen) with amplification conditions: 40 cycles of 15 s of denaturation at 95 °C followed by 1 min annealing. Reactions were performed in triplicate, and data were calculated using the ΔΔCt method. Gene expression was normalized by using B2M and GAPDH as reference genes.

To study whole-tissue morphological changes of 3D-eCRC-PDX tissues over time, images were obtained from a CellScale MicroSquisher system (side view of the tissues) and through phase-contrast microscopy (top view of the tissues) and used to quantify the volume (top and side views of the tissues), mean geometric diameter (top view of the tissues using equation (1)), circularity (side view of the tissues using equation (2)), and aspect ratio (side view of the tissues using equation (3)) of the 3D-eCRC-PDX tissues using ImageJ software. A minimum of three samples per time point per batch were analyzed:

$$\begin{equation}{\text{Circularity}} = \frac{{4\pi \times {\text{area}}}}{{{\text{perimete}}{{\text{r}}^2}}}\end{equation} \tag{ 2 }$$

$$\begin{equation}{\text{Aspect ratio}} = \frac{{{\text{major axis}}}}{{{\text{minor axis}}}}.\end{equation} \tag{ 3 }$$

The mechanical stiffness of CRC-PDX tumors and 3D-eCRC-PDX tissues was evaluated using a CellScale MicroSquisher system and the associated SquisherJoy software according to the previously published protocol [50]. Briefly, the geometric core, midpoint, and periphery of CRC-PDX tumors were cut into disk-shaped slices with a diameter of approximately 3 mm and a thickness of approximately 1 mm (thus resembling the 3D-eCRC-PDX tissue geometry) using a sterile scalpel (Bard-Parker®) and a surgical punch (Sklar) (supplementary figure 1). All CRC-PDX tumor slices and 3D-eCRC-PDX tissues were subjected to parallel compression testing under physiological temperatures in a PBS bath using the MicroSquisher apparatus. The compression system was equipped with a 558.8 μm gauge cantilever beam and a square microplate. All samples were preconditioned to 30% deformation for three compressions before data were collected. The force versus displacement data were then obtained via compression and relaxation cycles at a rate of 10 µm s⁻¹ to 30% deformation. The data were then converted to stress–strain curves, linear least-squares fitting was performed on the linear portion of the curves, and the Young's moduli of the tissue samples were calculated. Two separate cell culture batches of CRC-PDX tumors and 3D-eCRC-PDX tissues were examined. For 3D-eCRC-PDX tissues, the mechanical stiffness was quantified on days 1, 8, 15, 22, and 29 post-encapsulation for a minimum of three samples. For CRC-PDX tumors, three pieces of each region were measured for each batch.

To isolate ECM enriched fractions of protein CRC-PDX tumors (obtained after 31–42 d of tumor growth) and 3D-eCRC-PDX tissues (obtained on the 22nd day of in vitro culture) were processed via Millipore Compartmental Protein Extraction Kit following the manufacturer's instruction. Of note, samples were weighed prior to processing, and buffer volume was adjusted appropriately based on the sample weight. Following sequential extraction of cytoplasmic, nuclear, membrane, and cytoskeletal proteins via the kit [61, 62], the remaining pellet was washed once with 500 µl of PBS and reconstituted in 2x Novex NuPAGE LDS sample buffer (Invitrogen) supplemented with 2x NuPAGE sample reducing agent (Invitrogen). ECM-enriched fractions were then sonicated in an ultrasonic water bath for 20 min at ambient temperature. Debris were removed by centrifugation for 20 min at 16 000 ×g, 4 °C. Proteomics workup was carried out as previously reported [63]. Briefly, the protein fractions were quantified; ∼20 µg of protein per sample was then reduced with dithiothreitol (DTT) and denatured at 70 °C for 10 min prior to loading onto 10% Bis-Tris Protein gels and separating as a short stack run. The gels were stained overnight with colloidal Coomassie for visualization purposes, each lane was cut into 6 MW fractions and equilibrated in 100 mM ammonium bicarbonate (AmBc), and each gel plug was then digested overnight with Trypsin Gold, Mass Spectrometry Grade (Promega) following manufacturer's instruction. Peptide extracts were reconstituted in 0.1% formic acid/ddH₂O at 0.1 µg µl⁻¹. Mass spectrometry was carried out, and the data were processed, searched, filtered, grouped, and quantified using normalized spectral counting as previously reported in detail [63].

All identified proteins were converted from UniProtKB AC/ID to UniProtKB using the Retrieve/ID Mapping feature in UniProt to obtain primary gene names. The primary gene names were processed using Matrisome Annotator (http://matrisome.org/) [64] to identify ECM and ECM-associated proteins and subsequently categorized into ECM glycoproteins, collagens, proteoglycans, ECM regulators, ECM-affiliated proteins and secreted factors. The categorized proteins detected in only one out of six samples (CRC-PDX tumors and 3D-eCRC-PDX tissues) were removed from the analysis. The proteins were then employed to plot Euler diagrams using the eulerr package in RStudio (version 1.2.1335) for each sample. Moreover, the categorized proteins from three samples (CRC-PDX tumors or 3D-eCRC-PDX tissues) were pooled and presented for the comparison between CRC-PDX tumors and 3D-eCRC-PDX tissues. For the pooled data, a protein was indicated as present when the protein was detected in at least two out of three samples. Further details on the proteomics analysis protocol are provided in the supplementary methods.

RNeasy Plus Micro Kits (Qiagen) were used to isolate total RNA from frozen CRC-PDX tumors (obtained after 31–42 d of tumor growth) and 3D-eCRC-PDX tissues (obtained on the 15th day of in vitro culture) following the manufacturer's protocol. Procedures for RNA-seq were performed at the Genomic Services Laboratory, HudsonAlpha Institute for Biotechnology. Extracted total RNA initial quality control (QC) quantification was done using a Qubit Fluorometer (Invitrogen), and the quality of the extracted RNA was evaluated using an Agilent 2100 Bioanalyzer. RNA with a RNA integrity score (RIN) score of 7.0 or higher was used for RNA-seq library preparation using an Illumina TruSeq Stranded RNA Kit with Ribo-Zero Gold and sequenced on NovaSeq sequencers at 100 PE.

Sequencing reads were aligned to both the human GRCh38 and mouse GRCm38 reference genomes and counted with STAR [65] using aRNApipe, a RNA-seq analysis pipeline developed at the HusdonAlpha Institute for Biotechnology [66]. Approximately 62%–65% of the reads uniquely mapped to the human GRCh38 reference genome, while approximately 12%–21% of the reads uniquely mapped to the mouse GRCm38 reference genome (supplementary table 1). Human mapped raw counts obtained from STAR were used for differential expression analysis with the R package DESeq2 [67]. False discovery rates (FDRs) p value less than 0.05 were considered significantly differentially expressed and genes with log₂ fold change greater than 1.5 were used for downstream gene set analyses.

Human mapped raw counts were used for CMS analysis with the R package CMScaller [68], with default RNA-seq settings. CMScaller uses the nearest template prediction algorithm [69] to assign each sample to a CMS [70]. In brief, CMScaller computes cosine distances for each sample and assigns it to a CMS based on the smallest distance to the CMS template. Prediction FDR p value greater than 0.05 resulted in no classification.

The R package TCGAbiolinks [71] was used to obtain HT-Seq gene counts from the Genomic Data Commons for CRC tumor and tumor-adjacent tissue as a part of the Cancer Genome Atlas Colon Adenocarcinoma (TCGA-COAD) [72] project. CRC-PDX tumor and 3D-eCRC-PDX tissue HT-Seq gene counts and TCGA counts were merged by Ensembl gene ID's and processed as described above with DESeq2. The combined sequence data were variance stabilized using the DESeq method.

GSEA was performed using the desktop GSEA program (version 4.0.3) [73] and the Molecular Signature Database (MSigDB) Hallmarks v7.0 gene sets. GSEA analysis was performed using 1000 permutations and the phenotype permutation type. Annotations of human ensemble gene ID to MSigDB.v7.0 was performed. Enrichment plots and a heatmap were presented to visualize the data. Normalized enrichment score (NES) and the FDR q-value which presents that the estimated probability that the NES represents a false positive finding were reported in the enrichment plots.

The web-based application GOrilla [74] was used to identify enriched gene molecular function ontologies between CRC-PDX tumor and 3D-eCRC-PDX tissue samples and TCGA CRC tumor-adjacent tissue. The application REViGO [75] was used to eliminate redundant gene ontology (GO) terms and visualize the enriched terms using the TreeMap function which shows a two-level hierarchy of GO terms. The following REViGO settings were used: allowed similarity, (0.5, small); and semantic similarity measure, Resnik (normalized).

To examine expression of the top 16 significantly upregulated genes between CRC-PDX tumors and 3D-eCRC-PDX tissues in the precomputed GEO datasets GSE17536, GSE14333, GSE17537, and TCGA-COAD, the PROGgeneV2 tool was used [76, 77]. All cohorts were adjusted for age, stage, and gender covariates and bifurcated based on median expression. The hazard ratios, 95% confidence intervals, and p values were reported.

The outliers were removed using the robust regression and outlier removal (ROUT) method with coefficient Q = 1%. The normality of distribution and equality of variance were then checked. T-test was performed when comparing the means of two groups. One-way analysis of variance (ANOVA) followed by Tukey post hoc test was performed when comparing the means of more than two groups. Two-way ANOVA followed by Tukey was performed for morphological analysis and mechanical stiffness. Two-way ANOVA followed by Bonferroni–Šídák post hoc test was performed when comparing groups to a control group (flow cytometry). Linear trend analysis was performed using one-way ANOVA with a post-test for linear trend. The rates of increase or decrease (slopes) were found using linear regression analysis. p value ⩽ 0.05 was considered statistically significant unless otherwise stated. GraphPad Prism 7.0 (GraphPad Software, San Diego, California USA, www.graphpad.com) was used for the above analysis. Hierarchical cluster analysis was carried out using heatmap.plus package in RStudio (version 1.2.1335).

Using an optimized tumor dissociation method, we obtained 14.6 ± 0.6 million viable cells/gram of initial tissue (n = 5 CRC-PDX tumors). The isolated cells were then encapsulated in PEG-fibrinogen (PEGylation efficiency = 90 ± 4%) to create the 3D-eCRC-PDX tissues and separately cultured as 2D-CRC-PDX cells. A schematic of the process is shown in figure 1.

Zoom In Zoom Out Reset image size

To evaluate cell viability, the encapsulated cells within the 3D-eCRC-PDX tissues were labeled with Live/Dead dye and counterstained with Hoechst 33342. Based on confocal z-stack analysis, 67 ± 4% of encapsulated cells within the 3D-eCRC-PDX tissues were viable on day 1 (figure 1(B)). To examine whether cell viability was reduced due to the tumor dissociation process, the cell encapsulation process, or a combination of both procedures, we performed viability assays on 2D-CRC-PDX cells on day 1 post-tumor dissociation. Image analysis revealed that approximately 66 ± 6% of the cells were viable (supplementary figure 2) with no significant difference between 2D-CRC-PDX cells and the cells encapsulated within the 3D-eCRC-PDX tissues. Therefore, the lower viability of tumor cells observed within the 3D-eCRC-PDX tissues on day 1 was primarily a result of the tumor dissociation process, as expected. Prior investigation of CRC-PDX tumor dissociation by Petit et al reported similar post-dissociation cell viability [78].

Long-term viability was also examined to determine the temporal impact of in vitro culture on cell survival in the 3D-eCRC-PDX tissues. The percentage of viable cells on day 8, 15, 22, and 29 was 85 ± 6%, 90 ± 2%, 94 ± 3%, and 96 ± 1%, respectively (figure 1(B)), which was significantly higher than day 1 (p ⩽ 0.0003). Furthermore, the cell viability was significantly higher on day 29, as compared to day 8 (p = 0.0149). The high viability at the later time points is a result of cell proliferation and a subsequent overall increase in the total number of viable cells throughout the long-term culture, as can be seen in the confocal images (figure 1(B)). The cells encapsulated within the 3D-eCRC-PDX tissues proliferated (quantified via flow cytometry, as reported in detail below) and altered their morphological characteristics, with some cells forming colonies and others elongating within the biomaterial matrix (figure 1(B)). Elongated cells were first observed on day 8 and increased in abundance over time, as visualized on day 29.

To assess the colony-forming ability of the tumor cells in the 3D-eCRC-PDX tissues and temporal changes in the colony size and morphology, cells within the 3D-eCRC-PDX tissues and in 2D-CRC-PDX culture were imaged over a period of 29 d. Three batches of 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures were separately prepared, each from separate tumors, and analyzed. Phase-contrast z-stacks showed that the cell colony area and diameter within the 3D-eCRC-PDX tissues ranged from approximately 70–850 µm² and 10–40 µm on day 1, respectively (figures 2(A)–(E), (K) and (L)). However, for 2D-CRC-PDX cells, the cell colony area and diameter ranged from approximately 300–1300 µm² and 20–50 µm on day 1, respectively (figures 2(F)–(J), (O) and (P)). Cell colony area and diameter within 3D-eCRC-PDX tissues increased from day 1 to day 29 for each batch based on one-way ANOVA with a post-test for linear trend (p < 0.0001); observed increases were consistent between batches with no significant difference between the slopes based on linear regression analysis. Within the 3D-eCRC-PDX tissues on day 29, the cell colony area and diameter ranged from approximately 3500–35 000 µm² and from approximately 70–220 µm, respectively (figures 2(A)–(E), (K) and (L)). This increase in the size of CRC PDX cell colonies within the 3D-eCRC-PDX tissues over time was also observed using H&E staining (figures 2(S)–(V)). For 2D-CRC-PDX cells, passaging was necessary after 8 d of in vitro culture to prevent cell death (supplementary figure 3). The 2D-CRC-PDX cell colonies increased in size from day 1 to day 8; however, following passaging the subsequent colonies did not grow as large as those quantified on day 8 (figures 2(F)–(J), (O) and (P)).

Zoom In Zoom Out Reset image size

Cell growth and morphology within the 3D-eCRC-PDX tissues were consistent between batches. No significant differences in 3D-eCRC-PDX tissue mean cell colony area were observed between batches at any time point; the only significant difference between batches in mean colony diameter occurred on day 8 between batches 2 and 3 (p = 0.0017) (figures 2(K) and (L)). Furthermore, the number of viable cells within each 3D-eCRC-PDX tissue was counted after dissociation for flow cytometry analysis. The total number of viable cells increased linearly from day 1 to day 29 based on one-way ANOVA with a post-test for linear trend (p < 0.0001); the data showed that this increase did not differ significantly between batches based on linear regression analysis (supplementary figure 4). In contrast, for 2D-CRC-PDX cells significant differences were observed between batches, including differences in both cell colony area and diameter on day 8 and day 29 (p = 0.0121 and p = 0.0187, respectively) (figures 2(O) and (P)). In 2D-CRC-PDX cell culture, the total number of viable cells for all three batches was found to increase from day 1 to day 29 based on one-way ANOVA with a post-test for linear trend (p < 0.0001); however, in contrast with 3D-eCRC-PDX tissues, the slopes differed significantly for the 2D-CRC-PDX cells (p = 0.0274).

Circularity and aspect ratio of the cells and cell colonies, which are indicators of invasiveness [79–81], were also analyzed. Within the 3D-eCRC-PDX tissues, circularity of the cell colonies decreased and aspect ratio increased from day 1 to day 29 based on one-way ANOVA with a post-test for linear trend (p < 0.0001) (figures 2(M) and (N)). Circularity ranged from 0.97 to 1.00 on day 1 and from 0.55 to 0.97 on day 29; the aspect ratio ranged from 1.00 to 1.80 on day 1 and from 1.00 to 3.15 on day 29. However, for 2D-CRC-PDX cells circularity increased and aspect ratio decreased from day 1 to day 29 (one-way ANOVA, post-test for linear trend, p < 0.0001) (figures 2(Q) and (R)). Circularity ranged from 0.27 to 0.97 on day 1 and from 0.72 to 0.98 on day 29, while the aspect ratio ranged from 1.00 to 2.58 on day 8 and between 1.00 and 2.00 on day 29. Significant differences in cell colony circularity and aspect ratio were not consistently observed between batches (figures 2(M), (N), (Q) and (R)).

Using flow cytometry, we quantified both the human and mouse cell subpopulations from the CRC-PDX tumor, 3D-eCRC-PDX tissues, and 2D-CRC-PDX cultures; PDX models contain mouse stromal components in addition to the human malignant cells [28, 60] and maintaining both subpopulations is important for recapitulation of the tumor microenvironment. Human cells were identified using the cell marker β-2 microglobulin (B2M); the cells were also labeled with a mouse cell marker, H2Db, to confirm that those cells which were negative for B2M were the mouse cell subpopulation (supplementary figure 5). For dissociated CRC-PDX tumor cells, 71 ± 3% of the cell population was positive for the B2M human marker (three separately dissociated tumors, figure 3(A)). Importantly, as the cells grew within the 3D-eCRC-PDX tissues, the human cell subpopulation (B2M⁺) percentage remained constant during long-term culture (through day 22); there was an average percentage of 68 ± 6% B2M⁺ cells (days 8, 15, 22) with no significant difference between the 3D-eCRC-PDX tissues and dissociated CRC-PDX tumors (n = 3 separately prepared batches). On day 29, the percentage of B2M⁺ cells (53 ± 12%) was significantly lower than that on days 1 and 8 (71 ± 3% and 74 ± 7%, respectively, p < 0.003, figure 3(A)). In sharp contrast to the 3D-eCRC-PDX tissues, the percentage of B2M⁺ cells in 2D-CRC-PDX cultures had already decreased by over 80% after 8 d of culture (from 71 ± 3% to 13 ± 11%, n = 3 separate batches, p < 0.0001) and then remained constant over 29 d. It is important to note that whereas the 3D-eCRC-PDX tissues were maintained in continuous culture for 29 d, it was necessary to passage the 2D-CRC-PDX cells every 7 d starting on day 8 to avoid cell confluence; data presented are prior to each passage for 2D-CRC-PDX cells.

Zoom In Zoom Out Reset image size

In both the 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures, the total number of viable cells, human cells, and mouse cells increased over time (figures 3(B) and (C)). However, which cell subpopulations contributed to this increase differed drastically. In the 3D-eCRC-PDX tissues, both human and mouse cell subpopulations increased in number, maintaining similar ratios to the those of the original CRC-PDX tumor over time; no significant difference was observed between the rates of increase for the 3D-eCRC-PDX tissue B2M⁺ and B2M⁻ subpopulations (based on linear regression analysis, p = 0.4695, figure 3(B)). In sharp contrast, the mouse cells were the major contributor to the increase in total cell number in 2D-CRC-PDX cultures; although numbers of human cells increased over time, the rate was significantly lower (0.23 ± 0.04 fold, p < 0.0001) than that for the mouse cell subpopulations.

To further investigate temporal cell subpopulation changes, Ki-67 was employed as a proliferative marker in tandem with B2M to identify human proliferative cells. Although only 23 ± 4% of all cells harvested from the CRC-PDX tumor were Ki-67 positive (supplementary figure 6(A)), 81 ± 5% of these Ki-67⁺ cells were human cells (B2M⁺) (figure 3(D)). Remarkably, the 3D-eCRC-PDX tissues maintained this human proliferative cell subpopulation over 29 d of culture with no significant changes (85 ± 7%, 80 ± 5%, 82 ± 5%, and 73 ± 9% on day 8, day 15, day 22, and day 29, respectively), whereas in 2D-CRC-PDX cultures, a significant decrease (p = 0.0113) in the percentage of human proliferative cells was observed after 8 d of culture (from 81 ± 5% to 22 ± 26%) with the percentage remaining low thereafter.

For both 2D-CRC-PDX cultures and 3D-CRC-PDX tissues, approximately 20%–30% of all cells (human and mouse) were Ki-67⁺ over the 29 d of culture (supplementary figure 6(A)). The lower human cell proliferation in 2D-CRC-PDX cultures was offset by higher mouse cell proliferation; therefore, less than 25% of 2D-CRC-PDX cells were B2M positive after a week of in vitro culture (figure 3(A)). Moreover, total proliferative (Ki-67⁺) and non-proliferative (Ki-67⁻) cell numbers increased over time for both the 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures; however, the major contributors to these increases differed. For 3D-eCRC-PDX tissues the rates of increase in human proliferative (B2M⁺ Ki67⁺) and mouse proliferative (B2M⁻ Ki-67⁺) and human non-proliferative (B2M⁺ Ki-67⁻) and mouse non-proliferative (B2M⁻ Ki-67⁻) cell numbers were similar (no significant difference), maintaining ratios equivalent to the original tumor. In sharp contrast, for 2D-CRC-PDX cultures, mouse cells were the major contributor to both the total proliferative and non-proliferative cell number increases, whereas human proliferative and non-proliferative cell numbers increased at significantly lower rates (0.53 ± 0.07 and 0.12 ± 0.01 fold, respectively, p ⩽ 0.05) (supplementary figures 6(B)–(E)). Taken together, these results indicate that the maintenance of human CRC PDX cells and equivalent rates of increase in human and mouse cell subpopulations in the 3D-eCRC-PDX tissues throughout long-term culture was a result of maintenance of these proliferative cell subpopulations, whereas proliferative subpopulation changes in 2D-CRC-PDX cultures resulted in domination by the mouse cells over time.

To examine temporal changes in cancer cell subpopulations in the CRC-PDX tumor, 3D-eCRC-PDX tissues, and 2D-CRC-PDX cultures, CK20 was used to examine a CRC positive (CK20⁺) cell subpopulation. Even though CK20 is a CRC marker, not all CRC cells express the CK20 protein; the CK20 negative cell subpopulation may include other CRC cells [82–86]. As expected, the CK20⁺ cells were a subpopulation of the B2M⁺ cells while all mouse cells were negative for B2M as well as CK20 (supplementary figure 5). The number of cells in both B2M⁺ CK20⁺ and B2M⁺ CK20⁻ subpopulations within the 3D-eCRC-PDX tissues and 2D-CRC-PDX cultures increased over time (supplementary figures 6(H) and (I)). For the dissociated CRC-PDX tumors, 32 ± 2% of cells were positive for CK20 (figure 3(E)). Importantly, 3D-eCRC-PDX tissues were able to maintain a subpopulation of 32 ± 2% CK20⁺ cells over 29 d of culture, whereas a significant decrease (p = 0.0001) in the percentage of CK20⁺ cells was observed by day 8 in 2D-CRC-PDX cultures in relation to the CRC-PDX tumor (from 32 ± 2% to 6 ± 6%). This trend was also observed for the dual-labeled B2M⁺ and CK20⁻ cell subpopulation (supplementary figure 6(G)).

Overall, we found that the percentage of key cell subpopulations within the 3D-eCRC-PDX tissues were maintained over time with no significant changes through day 22 of culture as compared to the CRC-PDX tumors. Conversely, 2D-CRC-PDX cells exhibited significant changes within the initial days of culture (figure 3(F) and supplementary figure 7). Moreover, cluster analysis of these cell subpopulations revealed that the 3D-eCRC-PDX tissues from all time points clustered in a group with the CRC-PDX tumors, with day 8 of the 3D-eCRC-PDX tissues clustering closest (figure 3(G) and supplementary figure 8). In contrast, the 2D-CRC-PDX cells clustered separately from both the 3D-eCRC-PDX tissues and CRC-PDX tumors.

Morphological study of the distinct cell types supported the results found using flow cytometry. To visualize the cells within the 3D-eCRC-PDX tissues and in 2D-CRC-PDX cultures, immunostaining for nuclei, B2M, F-actin, and CK20 was carried out using Hoechst 33342, anti-B2M (human marker), phalloidin, and anti-CK20, respectively. Two distinct subpopulations were noted—a subpopulation of colony-forming cells and another of elongated cells. All of the colony-forming cells were positive for B2M and a portion was positive for CK20, whereas none of the elongated cells were positive for either B2M or CK20. Therefore, we can infer that the elongated cell subpopulation was comprised of mouse stromal cells (figure 3(H) and supplementary figure 9) which is consistent with observations that human stromal cells are replaced by mouse-derived cells in cancer xenografts [28, 60]. Consistent with these findings, labeled 2D-CRC-PDX cells also demonstrated that colony-forming cells were positive for CK20 and B2M, whereas the elongated cells were not positive for either maker (supplementary figure 10).

To examine gene expression of colon cancer markers of the epithelial to mesenchymal transition (EMT), stem cells, and angiogenesis, we performed RT-qPCR on total RNA isolated from CRC-PDX tumors, 3D-eCRC-PDX tissues, and 2D-CRC-PDX cells. We compared the expression of the colon cancer markers in the 3D-eCRC-PDX tissues and 2D-CRC-PDX cells to that in the CRC-PDX tumors. The expression of the EMT markers SNAI1 and TWIST1, which play a role in tumor metastasis, were 8.15 ± 0.71 and 4.91 ± 0.66 fold higher, respectively, in the 2D-CRC-PDX cells as compared to the CRC-PDX tumors (p = 0.001 and p = 0.005, respectively) (figure 4(A)). However, no significant differences in SNAI1 and TWIST1 were found between the 3D-eCRC-PDX tissues and CRC-PDX tumors. The expression of the EMT marker TGFβ1 in both the 2D-CRC-PDX cells and 3D-eCRC-PDX tissues significantly increased by 3.85 ± 0.16 and 4.82 ± 0.51 fold, respectively, (p = 0.009 and p = 0.003, respectively). Next, we examined the expression of the stem cell marker genes ALDH1A1, GPX2, LGR5, and OLFM4. We observed a reduction in expression of ALDH1A1 (0.49 ± 0.7 fold), LGR5 (0.43 ± 0.11 fold) and OLFM4 (0.18 ± 0.14 fold) in the 2D-CRC-PDX cells (p = 0.016, p = 0.022, and p = 0.001, respectively) (figure 4(B)). However, in the 3D-eCRC-PDX tissues, ALDH1A1 expression increased by 1.61 ± 0.17 fold (p = 0.006) and OLFM4 expression decreased by 0.14 ± 0.03 fold (p = 0.001). No significant difference in the expression of GPX2 was observed in the 2D-CRC-PDX cells and the 3D-eCRC-PDX tissues. Lastly, we examined the expression of the angiogenesis marker genes KDR and VEGFA (figure 4(C)). We observed that KDR expression increased by 11.42 ± 3.95 fold in 2D-CRC-PDX cells (p = 0.028), while the VEGFA expression also increased but did not reach statistical significance (figure 4(C)). In contrast, the expression of KDR and VEGFA was similar between the 3D-eCRC-PDX tissues and the CRC-PDX tumors. Taken together, our investigation into the expression of EMT, stem cell, and angiogenesis markers suggests that gene expression is more similar between the 3D-eCRC-PDX tissues and the CRC-PDX tumors, whereas significant differences were observed between 2D-CRC-PDX cells and the CRC-PDX tumors.

Zoom In Zoom Out Reset image size

Locational heterogeneity within the 3D-eCRC-PDX tissues was observed through immunostaining and confocal imaging. Elongated cells were primarily located on the outer layer (periphery) of the 3D-eCRC-PDX tissues, while the colony-forming cells were more prevalent within the 3D-eCRC-PDX tissues (figure 5(A)). Similar locational heterogeneity of cell morphology was also observed in the H&E-stained 3D-eCRC-PDX tissues (figure 5(C)). Furthermore, H&E staining illustrated that the colony size within the 3D-eCRC-PDX tissues increased over time and displayed similar cellular morphology to the CRC-PDX tumor on days 22 and 29 (figures 5(C) and 2(S)–(V)). As the density of elongated cells increased on the periphery of the 3D-eCRC-PDX tissues, the 3D-eCRC-PDX tissues remodeled themselves from a disk shape on days 1 and 8 to a near-spherical tissue by day 29 (figures 5(A) and (B)).

Zoom In Zoom Out Reset image size

To examine the contraction of the 3D-eCRC-PDX tissues, which plays an important role in ECM remodeling, mechanical stiffness and cell phenotype, the size and dimensions of the 3D-eCRC-PDX tissues were investigated over 29 d in two separate batches. The 3D-eCRC-PDX tissue volume and diameter decreased significantly from day 1 to day 29 (figure 5(B)). The volume of acellular control hydrogels increased (8.8 ± 0.2%, p < 0.0001) from day 1 to day 8 and remained constant after day 8 over 29 d of incubation at 37 °C and 5% CO₂ (supplementary figure 11); this change in volume was a result of an increase in the acellular tissue thickness, whereas diameter remained constant. Thus, the 3D-eCRC-PDX tissue contraction and matrix remodeling were cell mediated. Furthermore, tissue circularity and tissue aspect ratio were examined over time to quantify changes in tissue morphology. As the volume and diameter decreased, the 3D-eCRC-PDX tissue circularity increased and aspect ratio of the lateral surface (side view) decreased from day 1 to day 29, which was consistent with the change from the initial disk-like shape to a spherical shape; importantly, this result was consistent between batches (figure 5(B)).

To assess tissue mechanical stiffness, two CRC-PDX tumors and two paired batches of 3D-eCRC-PDX tissues were subjected to parallel plate compression testing. CRC-PDX tumor stiffness was found to vary significantly depending on the geometric region of the tumor from which the sample was taken. An increasing trend (p < 000.1) in stiffness was observed from the core to the periphery of the tumor for both tumors (figure 6(A)). The core region of the CRC-PDX tumor was the least stiff (average of ∼0.5 kPa), while the periphery region had the highest stiffness (average of ∼2.8 kPa), with no significant difference between tumor 1 and tumor 2 for a given region.

Zoom In Zoom Out Reset image size

The average stiffness of the 3D-eCRC-PDX tissues increased temporally from day 1 to day 29, where a stiffness of ∼0.4 kPa was quantified on day 1 and ∼2.8 kPa on day 29, with no significant difference between batch 1 and batch 2 for a given time point (figure 6(A)). Most notably, the range of stiffness of the 3D-eCRC-PDX tissues was similar to that of the CRC-PDX tumors, while the time-dependent stiffness of the 3D-eCRC-PDX tissues mimicked the region-dependent stiffness of the CRC-PDX tumor. To determine if the PEG-Fb biomaterial contributed to the increase in stiffness, acellular hydrogels were created and maintained in similar conditions to the cell-laden tissues. In contrast to the mechanical stiffness of 3D-eCRC-PDX tissues, we found that the stiffness of acellular hydrogels decreased from day 1 to day 8, primarily due to swelling, and then remained constant from day 8 to day 29 (supplementary figure 12). Therefore, the observed temporal increase in 3D-eCRC-PDX tissue stiffness is consistent with a cell-mediated mechanism.

To compare the ECM compositions of CRC-PDX tumors (obtained after 31–42 d of tumor growth) and 3D-eCRC-PDX tissues (obtained on the 22nd day of in vitro culture), proteomic analysis using ECM protein enrichment and mass spectroscopy characterization was performed on three separate batches of paired CRC-PDX tumors and 3D-eCRC-PDX tissues. Between all samples, 872 unique proteins were identified, 74 of which were identified as matrisome [64]. Correlation in ECM proteins observed in CRC-PDX tumors and paired 3D-eCRC-PDX tissues was first examined on a batch-by-batch basis. Approximately 61% of identified ECM proteins (38 out of 62) were observed in both the batch 1 CRC-PDX tumor and its paired 3D-eCRC-PDX tissues, while approximately 10% were identified only in the CRC-PDX tumor and 29% identified only in the 3D-eCRC-PDX tissues (supplementary figure 13(A)). Similarly, in batch 2 approximately 54% of identified proteins (38 out of 70) were observed in both the CRC-PDX tumor and the 3D-eCRC-PDX tissues, whereas in batch 3, approximately 47% of proteins were found in both CRC-PDX tumors and 3D-eCRC-PDX tissues (supplementary figures 13(B) and (C)). Overall, when comparing the pooled ECM proteins identified, there were 39 proteins (approximately 57%) observed in both the CRC-PDX tumors and the 3D-eCRC-PDX tissues, whereas 11 proteins (approximately 16%) were identified only in the CRC-PDX tumors and 19 proteins (approximately 27%) were identified only in the 3D-eCRC-PDX tissues (figure 6(B)). In addition, we evaluated the similarity between CRC-PDX tumors; approximately 60% of ECM proteins within CRC-PDX tumors were the same for all three batches, 28% of the ECM proteins were present in at least two batches, and 12% of the ECM proteins were specific to each batch (supplementary figure 14(A)). Similarly, within 3D-eCRC-PDX tissues, 60% of ECM proteins were found to be the same among all batches, 25% were present in at least two batches, and 15% were batch-specific (supplementary figure 14(B)).

Therefore, batch-to-batch consistency of the ECM proteins from the 3D-eCRC-PDX tissues was similar to that of the CRC-PDX tumors.

We further categorized the ECM proteins into ECM glycoproteins, collagens, proteoglycans, ECM regulators, ECM-affiliated proteins, and secreted factors. The number of ECM proteins per each subtype (e.g. ECM glycoproteins) found in the CRC-PDX tumors and 3D-eCRC-PDX tissues are presented in figure 6(B) and supplementary figures 14(A), (B). Identified ECM proteins for all CRC-PDX tumors and 3D-eCRC-PDX tissues from all batches are also reported by subtype in a heatmap (figure 6(C) and supplementary figure 15). We found that ECM glycoproteins, collagens, proteoglycans, and ECM-affiliated proteins exhibited the most ECM proteins shared between CRC-PDX tumors and 3D-eCRC-PDX tissues. ECM glycoproteins and collagens showed the highest overlap among all subtypes, whereas ECM regulators and secreted factors contained the least common proteins. In general, a higher total number of ECM proteins were identified in the 3D-eCRC-PDX tissues, thereby resulting in a higher number of proteins that were specific to the 3D-eCRC-PDX tissues.

Transcriptome analysis using RNA-seq was performed to undertake a molecular characterization of the CRC-PDX tumors and the 3D-eCRC-PDX tissues. To assess the CRC molecular subtype [70] we classified the CRC-PDX tumors and the 3D-eCRC-PDX tissues together with 478 primary CRC tumors from the TCGA-COAD cohort [87]. Our classification of TCGA-COAD cohort tumors resulted in the majority of tumors to be classified as CMS 2 and 4 which is consistent with previous classifications of CRC tumors [70, 88] (figure 7(A)). We observed that both the CRC-PDX tumors and the 3D-eCRC-PDX tissues classified as CMS1. Thus, the process of engineering our CRC PDX tissue did not alter the CRC molecular subtype.

Zoom In Zoom Out Reset image size

To assess cancer-specific differentially expressed genes (DEGs) and gene ontology functions in the CRC-PDX tumors and the 3D-eCRC-PDX tissues, we determined DEGs in the CRC-PDX tumors and the 3D-eCRC-PDX tissues compared to tumor-adjacent normal colon tissue in the TGCA-COAD patient cohort. Approximately 66% of the DEGs were overlapping when the samples were compared (figure 7(B)). ECM structural component (GO:000521) was the top significantly enriched gene ontology function term observed in both the CRC-PDX tumors (5.65 × 10⁻¹⁵ FDR p value, 5.05 enrichment) and 3D-eCRC-PDX tissues (4.12 × 10⁻¹² FDR p value, 3.41 enrichment) (figure 7(C)). Further, the top ten significantly (FDR p value < 0.05) enriched gene ontology function terms observed in the CRC-PDX tumors were also significantly enriched in the 3D-eCRC-PDX tissues. Because there is overlap between gene ontology terms, a REViGO analysis [75] was performed to remove redundant terms from the significantly enriched gene ontology function terms identified in the CRC-PDX tumors and the 3D-eCRC-PDX tissues. As shown in the TreeMaps in supplementary figure 16, enrichment in receptor binding, signaling receptor activity, ECM structural constituents were observed in both the CRC-PDX tumors and the 3D-eCRC-PDX tissues.

To gain cancer-related biological insight into gene expression in the CRC-PDX tumors and 3D-eCRC-PDX tissues, we performed GSEA using the Hallmark gene sets [89] from the Molecular Signatures Database (Broad Institute) [73]. The most enriched gene set (Myc targets v1, NES 3.10, q-value < 0.001) in the CRC-PDX tumors (figure 7(D), upper left panel) was also the most enriched (NES 2.50, q-value < 0.001) in the 3D-eCRC-PDX tissues (figure 7(D), middle left panel). We also observed that eight of the top ten Hallmark gene sets (Myc_Targets v1, E2F_Targets, Myc_Targets v2, DNA_Repair, Unfolded_Protein_Response, P53_Pathway, and MTORC1) significantly enriched in the CRC-PDX tumors were also significantly enriched in the 3D-eCRC-PDX tissues (supplementary figure 17, left and middle panels). To assess whether the Hallmark gene sets are also enriched in patient tumors, we next examined whether CMS1 classified primary tumors in the TCGA-COAD cohort were enriched for the ten significantly enriched Hallmark gene sets in the CRC-PDX tumors and the 3D-eCRC-PDX tissues. The top enriched gene set in the CRC-PDX tumors and the 3D-eCRC-PDX tissues, Myc_Targets v1 gene set, was also significantly enriched (NES 1.83, q-value < 0.017) in the CMS1 patient CRC tumors (figure 7(D), lower left panel). Indeed, nine of the top ten Hallmark gene sets significantly enriched in the CRC-PDX tumors were also significantly enriched in CMS1 patient CRC tumors (supplementary figure 17, right panel); only the P53_Pathway gene set was not significantly (q-value 0.174) enriched in the CMS1 patient CRC tumors (supplementary figure 17, right panel). To visualize the enrichment of the Myc targets v1 gene set in the CRC-PDX tumors, 3D-eCRC-PDX tissues, CMS1 patient CRC tumors, and tumor-adjacent normal colon tissue in the TGCA-COAD patient cohort, a heatmap was generated using the GSEA software. As shown in figure 7(D), right panel, tumor-adjacent normal colon tissue differentially expresses genes in the Myc targets v1 gene set compared to CRC-PDX tumors, 3D-eCRC-PDX tissues, and CMS1 patient CRC tumors.

To determine transcriptomic differences between the CRC-PDX tumors and the 3D-eCRC-PDX tissues, a differential expression analysis was performed. We observed 2193 DEGs (adj p value < 0.05, baseMean > 10, Log₂-fold > 1.5). Gene ontology was performed to assess potential function differences between the CRC-PDX tumors and the 3D-eCRC-PDX tissues. The significantly (FDR p value < 0.05) regulated gene ontology function terms enriched in the 3D-eCRC-PDX tissues were cell cycle processes (figure 7(E)). To examine the prognostic relevance of the top 16 upregulated DEGs in the 3D-eCRC-PDX tissues compared to the CRC-PDX tumors on overall survival and relapse free survival in CRC patients, we generated Kaplan–Meier plots [76]. In the CRC cohort GSE17536, there was not a significant difference in overall survival (p = 0.064) and relapse free survival (p = 0.201) in CRC patients with high expression of the 16 DEGs in the 3D-eCRC-PDX tissues (figure 7(F)). We also did not observe significant differences in overall survival (p = 0.246) in the CRC cohort GSE17537 and the TGCA-COAD cohort (p = 0.697) and with relapse free survival (p = 0.112) in the CRC cohort GSE14333 (supplementary figure 18).