Cancer Associated Fibroblasts Mediate Cancer Progression and Remodel the Tumouroid Stroma

Objective Cancer associated fibroblasts (CAFs) are highly differentiated and heterogenous cancer stromal cells that promote tumour growth, angiogenesis and matrix remodelling. Design We utilised a novel 3D in vitro model of colorectal cancer, composed of a cancer mass and surrounding stromal compartment. We compared cancer invasion with an acellular stromal surround, a ‘healthy’ or normal cellular stroma and a cancerous stroma. For the cancerous stroma we incorporated six patient-derived CAF samples to study their differential effects on cancer growth, vascular network formation, and remodelling. Results CAFs enhanced the distance and surface area of the invasive cancer mass whilst inhibiting vascular-like network formation. These processes were driven by the upregulation of hepatocyte growth factor (HFG), metallopeptidase inhibitor 1 (TIMP1) and fibulin 5 (FBLN5). Remodelling appeared to occur through the process of disruption of complex networks and was associated with the up upregulation of vascular endothelial growth factor (VEGFA) and down-regulation in vascular endothelial cadherin (VE-Cadherin). Conclusion These results support, within a biomimetic 3D, in vitro framework, the direct role of CAFs in promoting cancer invasion and that CAFs are also key components in driving vasculogenesis and angiogenesis.

macrophage polarization towards the M2 phenotype, also known as tumour activated macrophages (TAM) 4 , major orchestrators of cancer-related inflammation 9 . This process is driven mainly by interleukin-6 (IL-6), which is highly expressed by CAFs 10 . The key signature of CAFs is the overexpression of alpha smooth muscle actin (aSMA), a contractile stress fibre also expressed by myofibroblasts during wound healing 11 .
A number of "CAF markers" are used to differentiate between normal fibroblasts (NFs) and CAFs. They include fibroblast-specific protein-1 (FSP-1/S100A4), fibroblast-activating protein (FAP), platelet-derived growth factor receptor (PDGFR) and prolyl 4hydroxylase subunit alpha-1 (P4HA1), however, CAFs are a highly heterogenous population with various activation states present, which makes them difficult to be chracterised 4,7 . Initially, CAFs repress tumour growth due to gap junction formation amongst activated fibroblasts, but consequently they pave the way for extracellular matrix (ECM) remodelling and stiffening 12 . The ECM is remodelled physiologically and chemically during cancer progression due to factors expressed and released by the cancer cells and CAFs, including proteases breaking down the ECM through increased covalent cross-linking of collagen fibrils mediated by lysyl oxidase (LOX) 13,3 . This in turn increases interstitial fluid pressure within the tissue, which activates CAFs to upregulate transforming growth factor beta (TGF-b-1) 4 and matrix metallopeptidases (MMPs) thus promoting and guiding cancer cell tissue invasion 14 . Stiffness plays a major part in cancer progression and mechanotransduction and YAPdependent remodelling of the matrix is required for the creation and maintenance of CAFs 15,16 . CAFs produce and excrete a number of soluble factors which stimulate neighbouring stromal cells to excrete further tumour growth supporting soluble factors 17 . This cancer-stroma cross-talk recruits immune cells and local vasculature due to CAFs increasingly excreting vascular endothelial growth factor (VEGF) 10 . Overexpression of IL-6 by CRC cells and CAFs drives cytokinetic angiogenesis and further upregulates VEGF secretion through prostaglandin-E2 (PGE-2) mediation 14 .The recruited vascular networks promote cancer escape from the primary tumour and metastases. Colon CAFs specifically secrete growth factors, like hepatocyte growth factor (HGF), which activates mitogen-activated protein kinase (MAPK) and phosphophatidylinositol 3kinase (PI3K)/AKT pathways responsible for cell survival and invasion of the cancer 18 .
CAFs in 3D cancer models. The use of CAFs in in vitro 2D and 3D cancer models has been very limited in CRC and using patient-derived samples. CAFs cultured in collagen have increased contractility compared to NFs 13 and differentiate into cancer cells within co-culture 19 . Previous approaches have used spheroid formation, basic 2D invasion assays and microfluidic devices 20 in order to replicate the tumour stroma. These approaches are limited in their 3D representation of the tumour stroma by lacking vital components, such as vasculature and a clearly defined tumour-stroma margin through the compartmentalisation of cancer mass and stroma. By replicating the tumourstroma margin it is possible to study the interplay of different cell populations during cancer progression. Our approach to engineering a 3D in vitro colorectal cancer model incorporates patient-derived CAFs in the stromal compartment and allows us to study the patient-specific effect on vasculature formation during cancer growth and progression. This novel approach of modelling cancer-CAF interplay allows us to directly demonstrate the cellular cross-talk between the cancer and stromal cells within a stable and stiff ECM.
We hypothesised that invasion of cancer cells intro the stromal compartment is enhanced in the presence of CAFs as compared to normal human dermal fibroblasts (HDFs), our control used for this project. We also studied how the presence of CAFs, and the release of growth factors and cytokines, altered the formation of vascular networks and remodelled pre-existing vascular networks.

CAF isolation and propagation.
Primary human colorectal cancer associated fibroblasts were isolated from tumour tissues acquired from surgeries at the Royal Free Hospital. Patients provided informed consent for tissue donation for research, ethics code: 11/WA/0077. Fresh samples were provided by the pathology team, ensuring diagnostic margins were not compromised.
Tissue was disaggregated using a tumour dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany) and grown in Fibroblast Growth Medium 2 (Promocell, Heidelberg, Germany). For the first 72 hours (h) cells were left undisturbed, following that, media changes were done every 48 h in order to isolate the fibroblast cell population. The tissue samples were called T7, T10  and T11 for the first round of successful  samples and T6, T9 and T13 for the   second lot of successful samples  cultured. Patient-derived CAF samples  were then tested for positive vimentin  expression  and  negative  CK20  expression, to exclude colorectal  epithelial cell contamination, and CD31,  to eliminate endothelial cell contamination. CAFs aSMA and metabolic activity was assessed (see supplementary data). A range of other general fibroblast and more specific CAF gene markers were also investigated (see Figure 1B). Complex 3D models of cancer (tumouroids). All tumouroids were fabricated using monomeric Type I rattail collagen (First Link, Birmingham, UK) and the RAFT TM protocol pages 8-9 (Lonza, Slough, UK) as previously described 21 . 10X MEM (Sigma-Aldrich, Dorset, UK Sigma-Aldrich, Dorset, UK) was mixed with collagen and neutralising agent (N.A.) (17% 10 M NaOH (Sigma-Aldrich, Dorset, UK) in 1  M HEPES buffer Gibco TM through  Thermo Fisher Scientific, Loughborough, UK)) and mixed with cell suspension resulting in 80% collagen, 10% 10X MEM, 6% N.A. and 4% cells. For the artificial cancer masses (ACMs) 5x10 4 cells/ACM of either less-invasive HT29 or highlyinvasive HCT116 cells were used and 240 µL of the collagen mix was added to a 96-well plate (Corning® Costar® through Sigma-Aldrich, Dorset, UK). The gel mix was polymerised at 37˚C for 15 minutes (min), followed by plastic-compression using the 96-well RAFT TM absorbers (Lonza, Slough, UK). In order to produce 'tumouroids' 22 , the ACMs were nested into a stroma. For the stroma, collagen solution as described above was prepared, and ACMs were directly embedded into a 24-well plate (Corning® Costar® through Sigma-Aldrich, Dorset, UK) containing 1.3 mL of the non-crosslinked collagen mix. Extracellular matrix components were added to this stroma. In this case mouse laminin 23 50 µg/mL (Corning® through Sigma-Aldrich, Dorset, UK) for an acellular stroma additionally to 2.5x10 4 HDFs/ CAF samples and 10 5 HUVECs for a healthy or cancerous stroma respectively (please refer to Schematic 1 for more detail).
The tumouroids were polymerised at 37˚C for 15 min and plastic-compressed using the 24-well RAFT TM absorbers (Lonza, Slough, UK). Tumouroids were cultured for up to 21 days at 5% CO2 atmospheric pressure and 37˚C with 50% media changes every 48 h.

CAF treatment.
To study the effect of CAFs on established endothelial networks, a 1.0 mL suspension containing 2.5x10 4 CAF cells was added to the media mix at day 21 of established tumouroids. CAFs and tumouroids were subsequently left to propagate in co-culture for 7 days with continuing 48 h 50% media changes. This is additionally demonstrated in the Schematic 2 below. Investigative measurements were taken at day 21+1, day 21+3 and day 21+7 post CAF addition. ACTA2 levels were assessed after CAF additiona as an internal control (supplementary section).

Immunofluorescence.
Tumouroids were formalin fixed using 10% neutrally buffered formalin (Genta Medical, York, UK) for 30 min and then washed and stored in phosphate buffered saline (PBS) (Gibco TM through Thermo Fisher Scientific, Loughborough, UK). The tumouroids were permeabilised and blocked for 1 h at room temperature using a solution of 0.2% Triton X 100 and 1% bovine serum albumin (BSA) (both Sigma-Aldrich, Dorset, UK) in PBS. Primary antibody incubation was performed overnight at 4˚C followed by three 5 min wash steps with PBS. Secondary antibody incubation was carried out the next day with a 2.5 h incubation at room temperature followed by three 15 min wash steps with PBS. Antibodies were diluted in the same Triton X 100 and BSA solution and suppliers and source were: primary 1:200 anti-CK20 rabbit D9Z1Z (New England Biolabs, Herts, UK), anti-CD31 mouse JC70/A (Abcam, Cambridge, UK) anti-Vimentin mouse V9 (Santa Cruz, Texas, US) and secondary 1:1000 anti-mouse Alexa Fluor® 488 IgG H&L ab150113 and anti-rabbit DyLight® 594 ab96885 (both Abcam, Cambridge, UK). All tumouroids were counterstained with DAPI, using NucBlue TM (Invitrogen TM through Thermo Fisher, Loughborough, UK).

Measurement of invasion, endothelial networks and analysis.
All tumouroids were imaged using the Zeiss AxioObserver with ApoTome.2 and Zeiss ZEN software (Zeiss, Oberkochen, Germany). In order to measure the invasion from the original ACM into the stromal compartment and the number of endothelial structures, 4 images were taken at a 10x magnification evenly spaced out in alignment with a clock face at 12, 3, 6 and 9 o'clock on the same focal plane. This method has previously been described 21 .
All samples were assessed for distance and surface area of invasion and number of endothelial structures in the stromal compartment. The images obtained were then analysed in Fiji ImageJ software 24 .
RNA extraction, cDNa Synthesis and real-time PCR. RNA was extracted using the phase separation TRI Reagent® and chloroform method 25 (both Sigma-Aldrich, Dorset, UK). Total RNA obtained was quantified and assessed for integrity using the NanoDrop TM . Transcription into cDNA was conducted using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems TM through Fisher Scientific, Loughborough, UK) on the T100 TM Thermal Cycler (Bio-Rad, Watford, UK). Primers were designed according to the MIQE with an annealing temperature (Ta) of 60˚C, sequences and efficiencies are listed in Table 1 below, and were purchased through Eurofins Genomics (Ebersberg, Germany). Gene target amplification was conducted using iTaq TM Universal SYBR® Green Supermix on the CFX96 TM Touch System (both Bio-Rad, Watford, UK) in 10 µL reactions with 20 ng sample cDNA and primer concentration of 0.2 µM. Relative gene expression was calculated using the ∆Ct and 2 -∆∆Ct method 26  Tween 20 (TBST), both Bio-Rad, Watford, UK)), incubated with 1° antibodies for a-SMA 1A4 and loading control b-tubulin N-20 in 5% milk overnight at 4°C at dilutions 1:1000 and 1:200 respectively followed by five quick rinses and three 5 min washed with TBST. 2° antibodies IgG-HRP antigoat sc-2953 and anti-mouse sc-2314 at 1:1000 dilutions were incubated for 1 h in 3% milk (all antibodies through Santa Cruz Biotechnology, Dallas, US), followed by three 15 min washes with TBST, and developed using Pierce TM ECL Western Blotting Substrate (Thermo Fisher Scientific, Loughborough, UK). Blots were imaged using the ChemiDoc TM XRS imaging system and Image Lab TM software (Bio-Rad, Watford, UK).

Statistical analyses. All statistical analysis
was conducted using GraphPad Prism 7 software. Data was tested for normality with the Shapiro-Wilk test (n≥3) or the D'Agostino test (n≥8) and the appropriate test for statistical significance was applied depending on data parameters (t-test, Mann-Whitney, One-way ANOVA with Dunnet's Post Hoc or Kruskal-Wallis with Dunn's multiple comparisons test).
The tests used for each graph are outlined within the figure legends individually. Significance was at pvalues <0.05. All data points are represented as mean with standard error mean (SEM) in graphs and values stated in text as mean with standard deviation (STDEV). In general, n=3 with 3-4 technical replicates, details described within the figure legends for each individual data set. F-values, tvalues and degrees of freedom (DOF) are stated within the figure legends for each set of statistical tests. Two-tailed tests for significance were used when appropriate.

Extraction,
propagation and characterisation of patient-derived CAF samples. Six patient-derived CAF samples (n=6) were established from tumour samples, expanded on 2D tissue culture plastic (passage ≤3) and included in the tumouroid model. The samples were of variable location and origin ( Figure 1A), but all samples were from the lower bowel, colon or rectum with 5 being of adenocarcinoma and 1 being of neuroendocrine type. Samples were of varying grade in terms of tumour margin infiltration and vascular invasion. All samples were successfully cultured in 2D monolayers and tested for a number of fibroblast markers at the gene level ( Figure 1E&F). The data showed that all six samples were positive for ACTA2, S100A4, PDGFRA, FAP, P4HA1 and IL-6. This confirms that the cells are activated fibroblasts, especially based on the high expression of S100A4, PDGFRA and IL-6 in all samples 28,29&30 . Gene expression levels were compared between the samples and their differences can be taken from Figure  1E&F. Secondly, the western blots showed that the a-SMA protein was expressed in all samples ( Figure 1B), this is a measure previously used to distinguish samples as CAFs 31 . Thirdly, vimentin staining was done in CAF tumouroids grown to confluency and the morphology was compared to normal HDFs within tumouroids ( Figure  1 C&D). It was observed that the CAF samples overall appeared to have more stress fibres and had a much less organised structure.
A healthy stroma does not upregulate cancer invasion significantly. There is an evident cross-talk between the cancer cells and surrounding stroma. In the tumouroids, extensive primitive endothelial networks are formed within the stroma whilst the fibroblast population completely overgrows the space ( Figure 2D,E&F). The effect of adding cells to the stromal compartment was measured comparing tumouroids with an acellular stroma to ones containing normal fibroblasts and a primitive vascular network within the stroma. Firstly, the number of invasive bodies increased in the presence of a cellular stroma (Figure 2A), significantly in the HT29  Figure 2B) in the HT29 tumouroids (p=0.0006) with an average distance of invasion within the acellular stroma of 264.3 µm±138.9 µm and 110.2 µm± 47.7 µm. The surface area of invasion also decreased significantly ( Figure 2C) in the presence of a cellular stroma (p<0.0001) for both the HT29 and HCT116 tumouroids. For the HT29 acellular stroma the average surface area of invasion was 321,801 µm 2 ± 182,232 µm 2 and 26,963 µm 2 ±26,871 µm 2 within a cellular stroma and within the HCT116 tumouroids it was 510,240 µm 2 ±313,463 µm 2 for acellular and 97,946 µm 2 ± 95,469 µm 2 for a cellular stroma present. This means that although the number of invasive bodies increased when incorporating a cellular stroma, the distance and surface area invaded decreased ( Figure 2J). When analysing the genes responsible for the different invasion patterns, a number of genes were significantly upregulated when going from an acellular to a cellular stroma including MMP2 for HT29 and HCT116 tumouroids (p=0.001 and p=0.0098 respectively), TIMP1 (p=<0.0001 and 0.0099 respectively) and THBS1 (p=0.0039 and p=0.0007 respectively) ( Figure 2G, H&I). Additionally, HIF-1a was upregulated ( Figure 2K) in the presence of a cellular stroma compared to an acellular stroma in HT29 tumouroids (p=<0.0001), indicating that there was more hypoxia occurring within the tumouroids. Interestingly, MACC1 was downregulated in the presence of a cellular stroma within the HT29 and HCT116 tumouroids (p=0.0041 and 0.0024 respectively) ( Figure 2L).
A cancerous stroma significantly upregulates cancer invasion. CAFs were incorporated into the cancer stroma within the 3D tumouroid model containing either a HT29 or HCT116 ACM in order to investigate the effect of a cancerous stroma on cancer growth. The CAF-derived stroma caused an increase in the distance and surface area of invasion compared to HDFderived stroma ( Figure 3A,B,E&F). For the less-invasive HT29 32 Figure 3C&D), which demonstrate the increase in size of the invasive bodies in the presence of CAFs. A panel of 30 genes involved in invasiveness and angiogenesis were investigated to compare the healthy and cancerous stroma. Genes that were significantly upregulated in CAFtumouroids were HGF, ACTA2 and TIMP1 ( Figure 3G,H&I respectively). In the HT29 tumouroids, HGF was upregulated significantly in the presence of T9 (p=0.0105) and within the HCT116 tumouroids, HGF was significantly upregulated in the presence of samples T7 (p=0.0001), T10 (p=0.0071) and T11 (p=0.0255). There was a tendency for increased ACTA2 in CAF-containing HT29 tumouroids, but it was not statistically significant whilst for the HCT116 tumouroids, ACTA2 was upregulated significantly in the presence of samples T7 (p=0.0015) and T10 (p=0.0457). Finally, TIMP1 was highly overexpressed in the presence of CAF samples. In the HT29 tumouroids, the presence of samples T6 (p=0.0010), T9(p=0.0001), T10 (p=0.0026) and T13 (p=0.0001) significantly increased TIMP1 expression and in the HCT116 tumouroids, TIMP1 expression was significantly increased in the presence of samples T6 (p=0.0264), T9 (p=0.0329), T10 (p=0.0001) and T13 (p=0.0087).
The presence of CAFs within the tumouroid stroma inhibits vascular network formation. CAF containing tumouroids demonstrated an inhibition of vasculogenesis; the de novo formation of endothelial networks 33 . This was seen as a decrease in the number of elongated endothelial structures formed within the CAF stroma by day 21 of tumouroid culture. In the HT29 tumouroids ( Figure 4A) Figure 4C&D), in the CAF containing stroma the formation of complex endothelial structures was only observed around invasive bodies from the cancer mass ( Figure 4E&F). The protein levels of VE-Cadherin, a protein involved in endothelial cell end-to-end fusion 34 , showed a temporal decrease in VE-cadherin levels over the 21 day culture period. The amount of produced VE-Cadherin (ng/mL) significantly decreased in the HT29 tumouroids with sample T13 (p=0.0148) from 22.07±2.144 at day 2 to 12±1 by day 21. Within the HCT116 tumouroids, the VE-Cadherin production significantly decreased in the presence of T6 (p=0.0413) going from 25.04±2.649 at day 2 to 13.86±1.415 by day 21. The gene expression level of FBLN5, a gene which inhibits endothelial cell proliferation 35 , angiogenesis 36 and especially sprouting 37 , were measured in CAF-and HDF-tumouroids at day 21 ( Figure 4I). In the HT29 tumouroids samples T9, T10 (both p=0.0001) and T13 (p=0.0007) caused a significant upregulation in the relative gene expression while in the HCT116 tumouroids, samples T7 (p=0.0063) and T9 (p=0.0014) significantly upregulated FBLN5 compared to the HDF-tumouroids.
The disruption of pre-formed vascular networks by CAFs. In order to investigate the effect of CAFs on a developed, mature endothelial network (angiogenesis), CAF samples were added on top of tumouroids at day 21 and propagated for 7 days. Endothelial cells start off as single cells within the stroma on day 1 ( Figure 5A) and form complex, branched networks by day 21 ( Figure 4B) in the presence of HDFs within a tumouroid. After CAF addition, a disruption of the endothelial networks was observed ( Figure 5C). The disruption of vascular networks was assessed by quantifying the number and complexity of endothelial structures. The number of endothelial structures on day 21+7 decreased in the presence of all three CAF samples for both the HT29 tumouroids and HCT116 tumouroids ( Figure 5D&E) Figure 5F&G). Furthermore, FBLN5 gene expression increased significantly after CAF addition ( Figure  4H&I). In HT29 tumouroids containing samples T6 and T9 on day 21+7 (p=0.0001 and 0.0395). Within the HCT116 tumouroids, FLBN5 was upregulated significantly for T13 at day 21+1 (p=0.0001), T9 and T13 for day 21+3 (p=0.0261 and 0.0024) and T6 and T13 for day 21+7 (p=0.0300 and 0.0001). Finally, VEGFA gene levels were analysed after CAF addition ( Figure 5J&K) and a general increase was measured. Within the HT29 tumouroids sample T9 caused a significant upregulation at day 21+1 (p=0.02304) and samples T6 and T9 at day 21+7 (p=0.0233 and 0.0072). For the HCT116 tumouroids sample T13 caused a significant upregulation at day 21+1 (p=0.0372), samples T6, T9 and T13 at day 21+3 (p=0.0008, 0.0071 and 0.0019) and sample T13 for day 21+7 (p=0.0282).

Discussion
Our findings can be summarised as follows. First, a normal healthy stroma does not upregulate cancer growth significantly in a 3D model with a defined cancer mass and defines stromal compartments. Secondly, the presence of a cancerous CAF stroma increased the distance and surface area of invasion of colorectal cancer (CRC) into the stromal compartment whilst, at the same time, inhibiting vasculogenesis. These processes were driven by the up-regulation of hepatocyte growth factor (HFG), metallopeptidase inhibitor 1 (TIMP1) and fibulin 5 (FBLN5). Next, the remodelling appeared to occur through the process of disruption of complex endothelial networks and was associated with up-regulation of vascular endothelial growth factor (VEGFA) and a down-regulation in vascular endothelial cadherin (VE-Cadherin). These results support, within a biomimetic, 3D, in vitro framework, the direct role of CAFs in promoting cancer invasion and driving both vasculogenesis and angiogenesis.
The first of our observations describes the effect of adding a healthy stroma to surround the engineered cancer mass. This 3D model, comprising a cancer mass and a stroma, permits the study of how cancer cells invade from a distinct original mass into a new healthy tissue or stroma in the context of a metastasis. Future advancements to this model would include an active immune component. The only aspect that was increased in the presence of a healthy stroma was the increase in invasive bodies, modelling the highly invasive cancers "dispersal" into a high number of small clusters to invade the tissue. This is often due to the loss of structure proteins such as cadherins and cytokeratins 38 . This underlines the necessity of a cancerous storma for a cancer to thrive. The second major finding of this novel work is the differential invasion rate and pattern of less-invasive HT29 and highly-invasive HCT116 cancer cells in 3D tumouroids in the presence of CAFs, causing an increase in the distance (up to 3-fold) and surface area of invasion (up to 10fold) over a 21-day period. CAFs proinvasive properties and their ability to enable cancer cells to metastasise has been demonstrated previously 39,40 . Logsdon et al. 41 described the importance of CAFs in a 3D pancreatic cancer model using Matrigel®-coated invasion chambers and soft-agar colony formation and although this model showed an increased in proliferation and metastasis during in vivo validation, the 3D model was not compartmentalised and did not allow for a measurement of invasion in vitro. The majority of 3D cancer models lack appropriate tensile force and stiffness associated with tumour tissue, as they commonly use soft hydrogels, which have too high a water content 42 . Our biomimetic 3D in vitro cancer model (tumouroid) has a collagen density of up to 40x higher compared to standard hydrogels and therefore mimics the in vivo stiff tumour environment more closely 22 ; an important aspect especially for CAFs 43 . Our tumouroids let us investigate invasion patterns and study the genes that may be responsible for the increase in invasion. From a personalised healthcare perspective, the prospect of culturing patient specific cells in the presence of a highly or less invasive cancer mass allows the possibility to interrogate the interaction of these two cell populations. Primary cancer cells are especially difficult to culture and therefore utilising patient specific CAFs instead creates an avenue to overcome this limitation and supplement information we can retrieve.
The third of our observations can be interpreted in the following manner; CAFs play a key role in vascular network formation and remodelling. Whilst it is understood that CAFs play a major role in angiogenesis and recruiting vasculature towards the cancer 29 , in this study we also demonstrated that CAFs play a major role in vasculogenesis and the disruption of vascular network formation. This aspect has not been studied with the same rigour. Some studies have introduced CAFs at the same time point as HUVECs and observed end-to-end fusion of the HUVEC cells into endothelial structures 44 . Whilst this could be an observation of vasculogenesis, our model, with tissue specific parameters including biomimetic matrix density, shows no de novo formation of vascular networks in 3D in the presence of CAFs. At 25 000 CAFs per 24-well tumouroids, we observed 100% confluency of these cells in 3D by day 7, whilst HUVECs in our "normal" HDF containing cultures would not start forming complex endothelial structures until at least day 14. Our data indicates that CAFs start expressing factors that block complex vascular network formation, whilst retaining 'simple' vascular/endothelial structures. One of the factors, that was significantly increased, was VEGFA. Although VEGFA is the major player in angiogenesis and involved in recruiting mature blood vessels towards the cancer, its role in vasculogenesis is not as well understood. The major crosstalk between cancer cells and endothelial cells in our set up was growth factor driven, which was ascertained through an additional 3D set up. This set up demonstrated the chemoattractant driven movement and recruitment of endothelial structures to the cancer mass through an acellular ring placed between the cancer mass and stromal compartment 21 . Interestingly, a study looking at cardiac mouse development found a correlation between the disruption of vasculogensis and elevated VEFGA levels 45 . Within our model we observed the inhibition of vascular network formation in the presence of CAFs and, even within the HDF-tumouroids, we observed that closer to the central cancer mass, the endothelial networks became less complex. This indicates that the additional VEGFA released by the cancer cells could be inhibiting vasculogenesis close to the cancer mass and further underlines the fact that highly metastatic cancer cells and its stromal counterparts may disrupt the initial formation of highly complex endothelial networks. This was further studied by Brown et al. when looking at how the prevention of vasculogenesis but not angiogenesis prevented the recurrence of glioblastoma in mice 46 . Cancer cells are known for their high turnover and over-production of angiogenic growth factors 47 . This is in an attempt to recruit host vasculature from surrounding tissues. The unregulated and upregulated production and release of angogenic growth factors by solid tumours results in the formation of abhorrent and leaky vasculature surrounding tumours 48 . We have measured increased levels of VEGFA in our tumouroid cultures which are resulting in disrupted vasculogenesis and angiogenic remodelling to form non-complex networks. Vasculogenesis in cancer and especially in relation to the presence of CAFs is not a major focus of research as angiogenesis and remodelling of cancer is the biomimetic environment in which cancer arises. However, by gaining insights into how cancer angiogenic signalling can influence vasculogenesis may help further our understanding of tissue necrosis and vascular remodelling in cancer.
In our third observation we further showed that CAFs have the ability to disrupt pre-formed in vitro vascular network ("CAF treatment"). By day 7, post CAF addition, the endothelial networks that had previously formed were disrupted and an overall decrease in the number of endothelial structures was observed. Furthermore, we found a decrease of CDH5 levels within the tumouroids. The role CDH5 and the corresponding protein VE-Cadherin is becoming more pertinent in the study of angiogenesis as it has been specifically implicated in the local production of junctions within complex endothelial networks 34 . The literature on CAF interaction with VE-Cadherin is limited, although the role of CAFs as major sources of VEGFA production is established and understood to be mediated through HIF-1a/GPER signaling 49 . This particular gene (VEGFA) was increased after we added CAFs to our cultures and in fact this will have played an important role in the disruption (or angiogenesis) observed, however it would be crucial to study further how stromal cells cause this angiogenesis as the normalisation of these remodelled vascular networks has been a target for many antiangiogenic drugs 50 .
Our results need to be understood in the context of the following methodological limitations. We report the interactions at the interface of a cancer mass and patient-derived cancer associated fibroblasts in a 3D vascularized colorectal cancer model. A total of six patient-derived CAF samples were successfully isolated and cultured from colorectal cancer tissue samples. Primary CAF characterisation is a topic of debate in literature. Some groups have done extensive characterisation on the gene and protein level of 'CAF specific' markers 18,51,29 . In this study, we successfully demonstrated that our CAF samples expressed widely recognised markers for CAF identification. CAF populations have often been classified based on their location within the tumour margin 52 . For comparison purposes, we used human dermal fibroblasts (HDFs) as our control or "healthy" stromal cells, which are not an immortalised cell line and could also start to differentiate into CAFs while in co-cultured with cancer cells within the tumouroids. It could be argued that most primary fibroblast samples will adapt a cancerous phenotype due to being cultured on plastic or in co-culture with cancer cells 53 . For future work it would be ideal to use paired CAF and NF samples from the same patient as part of a larger comparison study. Patient-derived NFs would serve as a better control and further our investigations into what signalling is caused by CAFs. Additionally, following on from this work potential of identifying what drives CAFs and their cancer promoting properties could be pin pointed. One pathway would be to generate knockout CAFs with a deletion of HGF, TIMP1 and FBLN5; genes we found to be upregulated in the tumouroids and possibly responsible for increased invasion and vascular remodelling. Specific protagonists to VE-Cadherin such as could be used to potentially normalise the "leaky vasculature" caused by VEGFA upregulation, which are independent of one another.