Integrative characterisation of secreted factors involved in intercellular communication between prostate epithelial or cancer cells and fibroblasts

Reciprocal interactions between prostate cancer cells and carcinoma‐associated fibroblasts (CAFs) mediate cancer development and progression; however, our understanding of the signalling pathways mediating these cellular interactions remains incomplete. To address this, we defined secretome changes upon co‐culture of prostate epithelial or cancer cells with fibroblasts that mimic bi‐directional communication in tumours. Using antibody arrays, we profiled conditioned media from mono‐ and co‐cultures of prostate fibroblasts, epithelial and cancer cells, identifying secreted proteins that are upregulated in co‐culture compared to mono‐culture. Six of these (CXCL10, CXCL16, CXCL6, FST, PDGFAA, IL‐17B) were functionally screened by siRNA knockdown in prostate cancer cell/fibroblast co‐cultures, revealing a key role for follistatin (FST), a secreted glycoprotein that binds and bioneutralises specific members of the TGF‐β superfamily, including activin A. Expression of FST by both cell types was required for the fibroblasts to enhance prostate cancer cell proliferation and migration, whereas FST knockdown in co‐culture grafts decreased tumour growth in mouse xenografts. This study highlights the complexity of prostate cancer cell–fibroblast communication, demonstrates that co‐culture secretomes cannot be predicted from individual cultures, and identifies FST as a tumour‐microenvironment‐derived secreted factor that represents a candidate therapeutic target.

Reciprocal interactions between prostate cancer cells and carcinomaassociated fibroblasts (CAFs) mediate cancer development and progression; however, our understanding of the signalling pathways mediating these cellular interactions remains incomplete. To address this, we defined secretome changes upon co-culture of prostate epithelial or cancer cells with fibroblasts that mimic bi-directional communication in tumours. Using antibody arrays, we profiled conditioned media from mono-and cocultures of prostate fibroblasts, epithelial and cancer cells, identifying secreted proteins that are upregulated in co-culture compared to monoculture. Six of these (CXCL10, CXCL16, CXCL6, FST, PDGFAA, IL-17B) were functionally screened by siRNA knockdown in prostate cancer cell/fibroblast co-cultures, revealing a key role for follistatin (FST), a secreted glycoprotein that binds and bioneutralises specific members of the TGF-b superfamily, including activin A. Expression of FST by both cell types was required for the fibroblasts to enhance prostate cancer cell proliferation and migration, whereas FST knockdown in co-culture grafts decreased tumour growth in mouse xenografts. This study highlights the complexity of prostate cancer cell-fibroblast communication, demonstrates that co-culture secretomes cannot be predicted from individual cultures, and identifies FST as a tumour-microenvironment-derived secreted factor that represents a candidate therapeutic target.

Introduction
Prostate cancer affects millions of men worldwide, and is the second most common cancer in men, causing > 350 000 deaths per annum globally [1,2]. Understanding the pathophysiological mechanisms of prostate cancer progression is the key to developing novel therapeutic strategies. Strong evidence has emerged implicating the tumour microenvironment (TME) in cancer development and progression [3,4], and intercellular communication within the TME is a key regulator of the malignant behaviour of prostate cancer cells. Unique characteristics of the prostate TME, comprised of extracellular matrix (ECM) and diverse stromal cell types, distinguish it from corresponding normal tissue. Carcinoma-associated fibroblasts (CAFs) are one of the most abundant stromal components in the TME with numerous studies demonstrating their prominent roles in prostate cancer pathogenesis and progression [5][6][7][8], however the specific mechanisms of intercellular dialogue between CAFs and prostate epithelial, or prostate cancer cells, require further characterisation. These complex mechanisms can involve cell-cell contact, cell-matrix interplay or paracrine communication mediated by soluble factors [9]. Deciphering the mechanisms underpinning tumour stroma-cancer cell interactions during prostate cancer progression will enable the identification of novel biomarkers and/or treatment strategies.
The emergence of single-cell technologies has enabled the identification of different CAF subtypes within the TME [10]. Two important CAF subpopulations are termed "myCAFs" and "iCAFs". The myCAFs exhibit a matrix-producing contractile (myofibroblast) phenotype, while the iCAFs generate an immunomodulating secretome. CAF-derived soluble factors mediate autocrine and paracrine signalling, stimulate cancer cell proliferation, enhance epithelialmesenchymal transition (EMT) to facilitate metastasis, and increase neovascularisation [5,11]. Prostate cancer cells that come into contact with CAFs activate the same pro-inflammatory gene signature and become invasive through EMT [12]. For example, increased secretion of CCL2 and IL-8 from CAFs in advanced prostate cancer promotes cancer cell migration [13]. Meanwhile, IL-8 secreted by PTEN-deficient prostate cancer cells can also augment CAF-derived CCL2 and CXCL12, leading to increased proliferation and migration of the cancer cells [14]. Thus, the interaction between stromal and tumour cells results in an altered secretome, that induces pro-tumourigenic activities.
To date, various approaches have been employed to investigate the expression of secreted factors from cancer cells and CAFs, including Luminex assays, antibody (cytokine/chemokine) arrays and mass spectrometry. Of note, studies using these methods usually only focus on the secreted factors from one cell type and one culture condition (i.e. utilising monocultures) [15,16]. However, the secretome is likely to be altered in a co-culture environment, where reciprocal intercellular signalling is active. In order to address this, we have systematically profiled the secretomes of prostate epithelial and prostate cancer cells, and patient-derived normal prostatic fibroblasts (NPFs) and CAFs, and compared the secretomes from individual cell types with corresponding co-cultures. This approach revealed marked changes in the co-culture secretomes, with markedly elevated expression of several factors compared to mono-cultures. Follistatin (FST), which can bind and bioneutralise specific members of the TGF-b superfamily, including activin A [17,18], was identified as one such factor. Previous studies identified increased serum level of FST in prostate cancer patients [19], compared to that in healthy male and benign prostate hyperplasia patients. In addition, an increased serum level of FST significantly correlated with the presence of bone metastases and higher prostate specific antigen levels [19]. These findings suggested a role for FST in prostate cancer development and progression, but detailed insights into FST regulation and function in prostate cancer have been lacking. In this study, we identify FST as a key regulator of prostate epithelial/cancer cell-fibroblast intercellular communication, highlighting it as a potential target for therapeutic intervention.

Preparation of conditioned media
Cells were seeded in 10 cm culture dishes at 1.6 9 10 6 cells per dish, and 5 9 10 4 cells per well for each cell type (5 9 10 4 epithelial cells and/or 5 9 10 4 fibroblasts) in a 12 well plate for co-culture conditioned media (or Transwell conditioned media) in complete medium. Upon reaching 80-90% confluence, cells were washed with PBS and the medium replaced with 8 mL serum-free medium for 10 cm dishes, and 1 mL serumfree medium for 12 well plate. After 48 h, the conditioned media were collected, centrifuged and stored at À80°C until analysis.

Expression of fluorescence markers and luciferase
The expression vector for Luciferase/mCherry (pRV100G ofl T2A Luciferase/mCherry) was kindly provided by Prof. Paul Timpson (Garvan Institute, NSW, Australia). The expression vector pGIPZ for GFP was purchased from Thermo Fisher Scientific. Lentiviral or retroviral transduction of PC3, BPH-1 and 22RV1 cells for GFP, mCherry and Luciferase expression was performed as previously described [23] by using Lipofectamine TM 3000 reagent (Thermo Fisher Scientific). Positive cells were selected by Fluorescenceactivated cell (FAC) sorting.

siRNA transfection
Cells were transfected for 24 h with siRNA pools targeting CXCL10, CXCL6, CXCL16, FST, PDGFA or IL-17B (Dharmacon, Lafayette, CO, USA), or ON-TARGETplus (OTP) as the negative control. Sequences of siRNAs are shown in Table S2. Cells were used for further functional experiments within 48 h after transfection.

Random cell migration assay
PC3-GFP cells (1.8 9 10 4 /well) or PC3-GFP cells + WPMY-1 cells (1.8 9 10 4 + 1.8 9 10 4 /well) were seeded in a 12 well plate as mono-culture or coculture. The selected targets (CXCL10, CXCL6, CXCL16, FST, IL-17B and PDGFA) were knocked down in PC3-GFP cells and WPMY-1 cells, individually or in combination, by siRNA (or CRISPRi) before seeding. After 24 h, the medium was replaced with serum-free medium with 1 lgÁmL À1 Mitomycin C (Sigma-Aldrich). Human recombinant CCL11 (Peprotech) (8 ngÁmL À1 ) was added in PC3-GFP monoculture as the positive control for migratory promotion and a FAK inhibitor (FAKi, PF-562271, Sigma-Aldrich) was added to the co-cultures as the positive control for migratory inhibition. Human recombinant FST (hrFST) protein (R&D Systems, Minneapolis, MN, USA) at different concentrations (10, 50 and 100 ngÁmL À1 ) was used in the rescue experiments. Random cell migration was monitored using a Leica DMi8 Live Cell microscope (Wetzlar, Germany). Three fields per well were chosen under a 109 magnification objective. Time-lapse movies of the PC3-GFP cells in each mono-culture and co-culture condition were recorded over 24 h with an image acquired every 20 min. The movie files were analysed using the MtrackJ plug-in of the IMAGEJ software (National Institutes of Health, Bethesda, MD, USA). This method was repeated to validate the FST knockdown effect in 22RV1-mCherry cells and BPH-1-GFP cells.

Proliferation assay
PC3-GFP cells (5000/well) or PC3-GFP + WPMY-1 cells (5000 + 5000/well) were seeded in a 96 well plate as mono-or co-cultures for 3-5 days. The number of green fluorescent cells was recorded and analysed by IMAGEJ. In the proliferation rescue experiment, PC3 + WPMY-1 co-cultures or PC3_FST_KD + WPMY-1_FST_KD co-cultures were seeded at day 0, then three different concentrations (10 ngÁmL À1 , 50 ngÁmL À1 and 100 ngÁmL À1 ) of hrFST protein were added into the co-culture system at day 1, and culture continued for 3 days. The number of GFP positive cells was counted using IMAGEJ software.

Antibody array analysis
Cytokine analysis was performed using the Human Cytokine Array Q4000 (RayBiotech, Norcross, GA, USA) which included 200 human inflammation factors, growth factors, chemokines, cytokines and receptors. Then, a customised antibody array which included 24 prioritised proteins from the Q4000 array was analysed. Briefly, the arrays were blocked with the blocking buffer (provided by the kit) for 30 min and incubated with 1 mL of conditioned medium for 2 h at room temperature. After washing, the arrays were incubated with biotinylated antibody cocktail at room temperature for 2 h, and then with Cy3 equivalent dye-streptavidin for another 1 h at room temperature. Detection was performed according to the manufacturer's instructions.

Reverse transcription polymerase chain reaction (RT-PCR)
Total RNA was extracted using the RNeasy Ò Mini Kit (250) (QIAGEN, Hilden, Germany). Complementary DNA (cDNA) was synthesised from 1 lg of isolated total RNA by a high capacity cDNA reverse transcription kit (Thermo Fisher Scientific) in a final volume of 20 lL. The cDNA was subjected to realtime PCR with the primers (Sigma-Aldrich) as listed in Table S4.
PCR reactions were carried out in triplicate wells on ABI Ò 7500 system using power SYBR Green master mix (Thermo Fisher Scientific). The PCR reaction for each well included 5 lL FastStart Essential DNA green master (Roche, Basel, Switzerland), 0.2 lL of each primer (20 lM) and 4.6 lL cDNA and 40 cycles (95°C/15 s and 60°C/30 s) of amplification were performed.

Western blots
Standard Western blots were undertaken using RIPA lysates as previously described [24].

Animals
Male immunodeficient NOD-SCID mice (8 weeks old) purchased from Animal Resources Centre (Canning Vale, West Australia, Australia) were maintained in isolated ventilated cages under specific pathogen-free conditions in a temperature-and humidity-controlled, pellet food. Their health status was monitored daily. All procedures involving mice were conducted in accordance with the National Health and Medical Research Council (NHMRC) code for the use and care of animals for scientific purposes and the study was approved by the Monash University Animal Ethics Committee (Project ID: 21163).

Sub-renal graft experiment
A total of 40 NOD-SCID mice (8 mice per group) were used. Tissue recombinants consisting of 50 lL collagen gels that contain mixtures of cells (PC3 AE WPMY-1 cells, 10 5 AE 10 5 cells total) were sub-renally grafted under the kidney capsule of host mice. Each animal was implanted with 1-4 tissue recombinants placed on the left and/or right kidney. At the time of graft insertion surgery, a 5 mm silastic testosterone pellet was also inserted subcutaneously to help mimic the circulating testosterone levels of a human male. Mice were sacrificed after 8 weeks by cervical dislocation. During autopsy, the xenograft tumours under the kidney capsules were resected for histological analysis. Tumour size was measured with a calliper. Tumour volume was calculated using the following formula: (long length 9 short length 9 width).

Statistical analysis
GRAPHPAD PRISM 9.0 (GraphPad Software, San Diego, CA, USA) was used for statistical calculations. For all comparisons between two groups, t-tests were performed and P value and standard deviation of the mean (SD) were reported. For all comparisons among more than two groups, one-way ANOVA was performed and P values and SD were reported. Results from all in vitro assays are representatives of at least three independent biological replicate experiments unless otherwise specified.

Overview
In order to characterise intercellular communication between prostate epithelial/cancer cells and normal or cancer-associated fibroblasts (NPFs and CAFs, respectively), we employed an integrative approach involving three steps: (a) characterisation of mono-and coculture secretomes via antibody arrays; (b) functional interrogation of high priority candidates using a siRNA screen; and (c) a detailed characterisation of a lead candidate using in vitro and in vivo pre-clinical models. The overall approach is summarised in Fig. 1A.

Profiling of conditioned media using a custom antibody array
Benign prostate epithelial cells (BPH-1) or prostate cancer cells (PC3) were co-cultured with CAFs or NPFs from 2 pairs of primary patient-matched fibroblasts (NPF/CAF 128R, NPF/CAF 332R; Table S1) derived from radical prostatectomy specimens [21,22]. The coculture conditioned medium, together with corresponding mono-culture conditioned media, was collected and then assayed using antibody arrays that covered 200 human inflammation factors, growth factors, chemokines, cytokines and receptors (Fig. 1B, Table S5).

Secretome characterisation for prostate cell mono-cultures
In total, 142 proteins were identified in the analysis of mono-culture conditioned media (Fig. S1A). The secretion pattern of prostate epithelial/cancer cells was distinct from that of prostate fibroblasts in a number of ways. For example, DKK-1, ErbB3 (soluble receptor, cleaved extracellular domain), Lipocalin-2 and CEACAM-1 were expressed at higher levels in conditioned media from BPH-1 and PC3 cells compared to the primary fibroblasts. On the other hand, CCL11 (Eotaxin), HGF, TIMP-1 and MCP-1 (CCL2) were secreted more by the fibroblasts compared to prostate epithelial/cancer cells (Fig. S1B).
Different secretion patterns were also observed in the conditioned media between prostate epithelial and cancer cells (Fig. S1C). Secreted factors such as MIF, IL-6, CD40L, FST, SDF-1a and IL-8 were expressed at higher levels in conditioned medium from PC3 cancer cells. In contrast, some factors were present at a higher level in BPH-1 conditioned medium, including activin A, PAI-1 and IGFBP-3.

Comparison of mono-and co-culture secretomes
The antibody arrays were also used to characterise how conditioned media from co-cultures of BPH-1/ PC3 with either NPFs or CAFs differed from the corresponding mono-cultures. Unsupervised clustering of secretome patterns from the different cultures resulted in segregation of co-cultures and mono-cultures (Fig. S2).
While heat maps are a conventional approach to present different '-omics' data, in order to resolve the complex datasets to the greatest extent, an X-plot approach was developed (Fig. S3). Here, each dot in the X-plot represents a secreted factor, whose xcoordinate value indicates expression value in monoculture conditioned media and y-coordinate value indicates expression value in co-culture conditioned media. comparison of BPH-1 + NPF 128R co-culture with BPH-1 mono-culture and quadrant II shows BPH-1 + NPF 128R co-culture versus NPF 128R monoculture. The red dots represent factors with an increase of expression in co-culture, as they are closer to the yaxis; the blue dots denote the secreted factors decreasing in co-culture, as they are closer to the x-axis. Accordingly, the comparison between BPH-1 + CAF 128R co-culture and mono-culture is plotted in quadrant III and IV in the similar way (Fig. S3).

The secretome of prostate epithelial cell/fibroblast co-cultures
A total of 116 secreted factors were detected and included in the analysis. The secretomes of BPH-1 cocultured with NPF/CAF 128R and NPF/CAF 332R are visualised in Fig. 2 and Fig. S4, respectively. Of note, the levels of some proteins were increased in coculture compared to all mono-cultures, such as CXCL10, IL-6, LYVE-1 and GDF-15 ( Fig. 2 and Fig. S4A). CXCL10, a known chemoattractant that exhibits anti-tumour activity [25,26], was particularly prominent, with its expression up-regulated in the cocultures with either NPFs or CAFs by up to three orders of magnitude (Fig. S4B). Another example with high expression is GDF15 (growth differentiation factor 15), a member of the transforming growth factor beta superfamily involved in regulating apoptosis, cell growth and carcinogenesis [27] (Fig. S4C).
In addition to proteins that showed increased levels in the co-culture compared to all mono-cultures, some secreted factors were expressed at different levels in cocultures with NPFs versus CAFs, including IGFBP-3, IGFBP-4, DR6, SCF and activin A in BPH-1 cocultures with NPF/CAF 128R (Fig. S5A) and NPF/ CAF 332R (Fig. S5B). Death receptor 6 (DR6, a cleaved extracellular domain in this antibody array), also known as tumour necrosis factor receptor superfamily member 21(TNFRSF21), is a cell surface receptor of the tumour necrosis factor receptor superfamily which activates the JNK and NF-jB pathway [28]. DR6 was expressed at a relatively high level in BPH-1 and low levels in fibroblast mono-culture conditioned media respectively, but its level was markedly higher in BPH-1 + CAFs co-culture conditioned medium versus BPH-1 + NPFs co-culture conditioned medium (Fig. S5C).

The secretome of prostate cancer cell/fibroblast co-cultures
Ninety-six proteins were detected and included in the analysis of the conditioned media from co-culture versus mono-culture. Compared to BPH-1, a lower number of proteins decreased expression in co-culture and most of the changes manifested in proteins that increased in expression (Fig. 3A and Fig. S6). This was particularly evident in 128R co-culture, which revealed only one protein [LIGHT, also known as tumour necrosis factor superfamily member 14 (TNFSF14)] that markedly decreased in expression compared to PC3 mono-culture (Fig. 3A). Similar to the BPH-1 co-culture, some proteins were up-regulated in the co-cultures compared to all corresponding mono-cultures, including DAN, FST, EpCAM, VEGF and CXCL16 (Fig. 3A and Fig. S6). For example, FST was expressed at low levels in all mono-cultures, but it showed an increase in the co-cultures of between one and two orders of magnitude (Fig. 3B).
Other secreted factors exhibited contrasting changes in expression when PC3 cells were co-cultured with either NPFs or CAFs, for example PDGF-AA, TGFa and CXCL4 (Fig. S7A and S7B). Transforming growth factor alpha (TGFa) is a ligand for the EGF receptor and has been associated with many types of cancer [29,30]. NPFs secreted a high level of TGFa, while the CAFs and PC3s did not (Fig. S7C). However, after co-culture, the PC3 + CAF co-cultures expressed a substantial level of TGFa, much higher than CAF and PC3 mono-cultures, but still lower than PC3 + NPF co-cultures.
Overall, the antibody array analysis revealed distinct secretion patterns where proteins were markedly increased in co-cultures versus corresponding monocultures (e.g. CXCL10, IL-6 in BPH-1 co-culture and CXCL16, DKK-1, FST in PC3 co-culture). In addition, it identified differential effects of CAF versus NPF co-culture (e.g. on DR6). Importantly, these data highlight how intercellular communication between the cell types affects the co-culture secretome so that levels of particular secreted factors in the co-culture cannot be predicted based on the mono-cultures.

Functional evaluation of secreted factors via a custom siRNA screen
Next, we set out to functionally characterise specific factors in the co-culture setting. Due to the limited lifespan of the primary fibroblasts and difficulties in manipulating them by transfection, we decided to utilise WPMY-1, an immortalised prostate stromal cell line, in combination with PC3s in functional screens. In order to implement this model, it was first necessary to undertake a secondary antibody array screen to confirm that changes observed in co-cultures of PC3s with NPFs or CAFs were observed with WPMY-1 cells. This was undertaken using a customised antibody array that detected 24 secreted proteins (Table S6), selected from the primary screen on the basis of novelty; biological function; known relationship to cancer; and tractability, including availability of validated reagents. Upon comparing PC3/WPMY-1 co-cultures with the corresponding mono-cultures, many proteins exhibited similar trends to the primary screen, including CXCL1, CCL2, CCL5, IL-8, CXCL10 and FST ( Fig. S8A-J, Table S6). Then, we excluded proteins based on (a) a different trend in the WPMY-1 co-culture system compared to the primary screen, despite their significant biological roles in tumourigenesis and progression (BDNF [31], IL-6sR [32], TGFa [33] and IL-6) (Table S6), or (b) lack of novelty, reflecting well-characterised roles in cancer, such as CXCL1, CCL2, CCL5, IL-6 and IL-8. This led to the selection of six proteins (CXCL10, CXCL16, CXCL6, FST, PDGFAA and IL-17B), all upregulated in co-culture conditioned media compared to corresponding mono-cultures (Fig. 4A, Fig. S8E-J, Table S6).
Then, the six candidate targets were subjected to a functional screen based on random cell migration. Briefly, targets were individually knocked down by corresponding siRNA in both PC3-GFP and WPMY-1 cells, then the cells were co-cultured and PC3-GFP migration characteristics assayed in terms of accumulative migrated distance (the total distance that one single cell travels within a defined time; Fig. 4B,C). Human recombinant CCL11 protein, a known pro-migratory cytokine [34], was added to PC3-GFP mono-culture as the positive control for migratory enhancement and the selective FAK inhibitor (FAKi) PF-562271 [35] was added to the co-culture as the positive control for migratory inhibition (Fig. 4B,D, Fig. S9A,B). After 24 h random migration, co-culture with WPMY-1 cells significantly increased the accumulative migrated distance of PC3-GFP cells compared to PC3-GFP culture alone. The motility of PC3-GFP cells was markedly impaired when the FAKi was added to the co-culture system. In addition, their migration in mono-culture was significantly increased by human recombinant CCL11 protein. When each of the six targets were individually knocked down in both cell types in the coculture, it was found that FST knockdown ( Fig. 4E and Fig. S9C) significantly inhibited PC3-GFP cell migration in co-culture, whereas this effect was not seen with the other targets ( Fig. 4B-D, Fig. S9A,B). These data demonstrate that expression of FST by PC3 cells and/or WPMY-1 fibroblasts is required for WPMY-1 cells to enhance PC3 cell migration. Scatter plots with log10 transformed axes, where up-regulated cytokine/chemokines (red dots) were selected using a stringent cut-off value of FC > 10 (protein level in co-culture/corresponding mono-culture), and decreased cytokine/chemokine (blue dots) were selected by a cutoff value of > 2 FC (protein level in mono-culture/corresponding co-culture). Green background: BPH-1 cytokine/chemokine secretion, yellow background: fibroblast cytokine/chemokine secretion. The primary screen was undertaken with one biological replicate. FC, fold change; CAF, cancer-associated fibroblast; NPF, non-malignant prostate fibroblast.

Characterisation of FST as a key regulator of intercellular communication between prostate cancer cells and co-cultured fibroblasts
To characterise which cell type in the co-culture system was producing the FST responsible for the promigratory effect, FST was knocked down in either PC3 or WPMY-1 cells or both cell types, and random cell motility of the PC3-GFP cells assayed in co-cultures. The motility of the PC3-GFP cells was only significantly inhibited when FST was knocked down in both cell types (Fig. 4F, Fig. S9D,E), not in one cell type only. This indicates that FST is produced by both PC3 and WMPY-1 cells, so that blocking its function in promoting cell migration requires its knock down in both cell types. In order to characterise the requirements for FST production in terms of cell-cell proximity, we analysed conditioned media from mono-cultures, direct coculture, and Transwell co-culture of PC3 and WPMY-1 cells by western blot analysis (Fig. 5A). Strikingly, a marked increase in FST expression was only observed when the cells were directly co-cultured, indicating that the two cell types must be in close proximity to promote enhanced FST production (Fig. 5B,C).
Determination of the role of FST in regulating longer term biological endpoints, such as proliferation, as well as tumour growth in vivo, necessitated generation of stable FST knockdown cell lines by CRISPRi. The knockdown efficiency in both PC3 and WPMY-1 cells was validated by western blot. Two out of five sgRNAs exhibited strong knockdown efficiency (Fig. S10A-C). The cells were then subjected to random cell migration assays, where consistent with the data using siRNA (Fig. 4F), knockdown in both cell types in the co-culture system was required to significantly reduce the accumulative moving distance of the PC3-GFP cells in co-culture (Fig. S11A-C). The same result was observed for cell displacement (maximum distance from start point during the migration; to validate the role of FST and confirm that the data did not reflect off-target effects. Addition of human recombinant FST (hrFST) rescued the decreased accumulative moving distance and cell displacement of these cells in the double knockdown co-cultures (Fig. S12A-D). We then determined the role of FST in regulating cell proliferation. Co-culture of PC3-GFP cells with WPMY-1 fibroblasts significantly increased the proliferation of the former (Fig. 6). However, similar to the cell migration result, this effect was lost upon knockdown of FST in both PC3-GFP and WPMY-1 cells, but not in only one of these cell types (Fig. 6A,B). Similar to the migration result, addition of hrFST rescued the decreased proliferation of PC3-GFP cells in co-culture when FST was knocked down in both cell types (Fig. S13A-D).
In order to ascertain the broader applicability of these findings, we interrogated the role of FST in other prostate epithelial (BPH-1) or prostate cancer (DU145, LNCaP and 22RV1) cell lines. Using siRNA, we obtained strong FST knockdown efficiencies in BPH-1 and 22RV1 cells (Fig. S14A,B). The migration of these two cell lines was then assayed in mono-and WPMY-1 co-cultures, using the same single and double knockdown approach as described previously (Fig. 4). The results were consistent with the effects for PC3-GFP + WPMY-1 co-culture, in that co-culture with WPMY-1 cells significantly increased the motility of BPH-1-GFP and 22RV1-mCherry cells compared to their mono-cultures. Likewise, the migratory abilities of BPH-1-GFP and 22RV1-mCherry cells in WPMY-1 co-culture decreased only when FST was knocked down in both cell types (Fig. 7, Fig. S15A,B). In summary, these results indicate that a marked increase in FST production occurs when prostate cancer cells are co-cultured with prostate fibroblasts, both cell types generate functional FST and this factor promotes both proliferation and migration of co-cultured prostate cancer cells.

Determination of the role of FST in regulating tumour growth
To the best of our knowledge, the role of FST in prostate tumour growth has not been characterised using a mouse xenograft model. To expand on the in vitro findings, a mouse sub-renal graft model was employed to test whether FST KD could have an impact on tumour growth following implantation of PC3/ WPMY-1 cell recombinants (Fig. 8A).
Importantly, co-injection with fibroblasts significantly enhanced both tumour size and weight (Fig. 8B-D). In addition, FST knockdown in the WPMY-1 cells did not affect these parameters. However, unlike the in vitro assays for cell proliferation and migration, knockdown of FST in either the PC3 cells, or both cell types, reduced tumour growth to a similar degree (Fig. 8B-D). This likely reflects differences between 2D co-culture in vitro and 3D growth in vivo, and the presence of the host environment. It also indicates that the FST signal in vivo is predominantly from the PC3 cancer cells, so that depletion of FST in the PC3 cells leads to decreased tumour growth, and that in this model, the WPMY-1 cells promote tumour growth by a FST-independent mechanism.

Discussion
The current study explored the intercellular communication between prostate stromal fibroblasts and cancer cells, which is a well-documented driver of tumour progression in prostate cancer [7,8]. By systematically profiling the conditioned media from both individual cultures or co-culture of prostate fibroblasts and prostate epithelial or cancer cells, marked changes in coculture secretomes were revealed, highlighting specific TME-derived secreted factors as candidate therapeutic targets. Additionally, the study identified the critical role of the secreted glycoprotein FST, characterising it for the first time to our knowledge in the context of prostate cancer/PE crosstalk with fibroblasts, and demonstrating that it enhances prostate cancer cell proliferation and migration in vitro and also tumour growth in vivo.
Characterisation of secreted factors in co-cultures, rather than the corresponding mono-cultures, is a relatively uncharted area, and the co-culture may only be compared to one cell type, as in a recent study on IL-6 in gastric cancer [36]. A key feature of our study is the discovery of candidate secreted factors through an antibody array approach in which we not only characterise the secretomes from prostate epithelial/cancer cells, or CAF and NPF mono-cultures, but also compare them with those from corresponding cocultures. This determined that while the different mono-cultures could exhibit distinct secretion patterns, the secretomes of the co-cultures could be completely different, and this was not a simple additive effect. Consequently, the levels of particular secreted factors in the co-culture cannot be predicted based on the mono-cultures since intercellular communication between the cell types affects the co-culture secretome. This communication may be mediated via soluble or ECM factors, or via direct cell-cell contact. In our model, cell-cell contact, or at least very close cell proximity, was necessary to enhance the production of FST. Similar findings have been reported in other cell systems. For example, chronic inflammatory diseases feature massive up-regulation of IL-1 and TNF induced by direct cellular contact of T cells and monocytes/macrophages [37] and direct co-culture of splenocytes and adipocytes leads to a greater increase in IL-6 and MCP-1 levels compared to indirect co-culture and individual culture [38]. Furthermore, direct cell-cell contact of pancreatic cancer cells and fibroblasts induces calcium oscillations, NF-jB activation, and activin A secretion, leading to increased EMT of cancer cells [39]. Follistatin is associated with the development, progression and metastasis of various cancers. The primary function of FST is as an antagonist of specific TGF-b superfamily proteins, including activins (activin A, B and C), bone morphogenetic proteins (e.g. BMP-4, BMP-7, BMP-11) [40,41], Myostatin [40] and TGF-b3 [42]. FST has a high affinity for activin A [18], and increasing evidence reveals that dysregulation of the FST/activin A system could lead to alterations of the normal homeostasis of prostate tissue and cause development and progression of prostate cancer [43][44][45][46].
For example, activin A's inhibition of prostate cancer cell proliferation is completely reversed by FST [44,47,48]. Furthermore, serum levels of FST are increased in prostate cancer patients [19] and overexpression of FST occurs in tumours compared to normal prostate tissues [49], supporting a model where FST suppresses activin signalling-mediated cell growth inhibition and promotes prostate cancer development and progression [50,51]. However, recent data indicate that activin A signalling, and hence potentially FST, may play stage-specific roles in prostate cancer, with non-canonical activin A signalling maintaining epithelial quiescence in the normal prostate [45] but activin A positively regulating metastasis in pre-clinical models and associating with poor prognosis in prostate cancer patients [46].
Despite high serum levels of FST associating with poor prognosis in hepatocellular carcinoma [52,53] and high tumour FST also representing a negative prognostic factor in lung, ovarian and gastric cancer [49], survival data from the TCGA database indicates that high tumoural expression of FST does not correlate with altered disease-free survival of prostate cancer patients (Fig. S16). However, it should be noted that the TCGA data correspond to FST mRNA not protein, and it's possible that FST expression is regulated at the post-transcriptional level and/or at the level of secretion. In addition, it is unclear from the TCGA data whether gene expression relates to the cancer cells, CAFs, or both. Moreover, FST could have opposing effects at different stages of disease progression, as indicated for activin A.
In addition to FST, other factors also exhibited increased expression in co-culture conditioned media, and although their knockdown did not reduce the migratory ability of PC3 cells in the functional screen, these factors might exhibit other biological roles. Three of them (CXCL10, CXCL6 and CXCL16) belong to the chemokine (C-X-C motif) ligand (CXCL) family, which are known to play key roles in inflammatory diseases, neoplastic transformation and tumour growth regulation [54]. CXCL10 increases the proliferation of mouse breast cancer stem cells and breast cancer cells, and is significantly associated with triple negative breast cancer versus HER2 + , Luminal A and Luminal B breast cancer [55]. CXCL6 production is up-regulated in aged prostate stroma and promotes proliferation of both prostate stromal fibroblasts and epithelium [56], and increased CXCL6 expression occurs in prostate cancers with high Notch1 levels [57]. In addition, CXCL16 can promote the proliferation of PC3 cells and high expression of CXCL16 correlates with high-stage and high-grade prostate cancer [58,59]. The CXCR6-CXCL16 axis also promotes docetaxel resistance through phosphorylation of GSK-3b, NF-jB and ERK1/2 [60], as well as angiogenesis in prostate cancer via AKT/mTOR-mediated regulation of VEGF and IL-8 [61]. Consequently, these factors exhibit crucial biological roles in cancer development and progression, and are likely to emerge as 'hits' in screens that incorporate other biological endpoints.

Conclusions
In summary, our work highlights potent intercellular communication between prostate stromal fibroblasts and epithelial/cancer cells that results in a reprogrammed combined secretome. It also reveals FST as a key factor involved in prostate cancer-stroma interaction in vitro and in vivo, suggesting that the role of FST should be revisited in the context of different stages and mutational profiles of prostate cancer, which may provide insights from a precision oncology perspective and identify potential strategies for disease management. Fig. S2. Secretome of co-cultured prostate epithelial/cancer cells with NPFs/CAFs. Fig. S3. Schematic of data visualisation by the X-plot. Fig. S4. X-plot approach to visualise how secretomes of co-cultures of BPH-1 with NPFs or CAFs differ from corresponding mono-cultures. Fig. S5. X-Plot highlighting proteins with differential expression between NPF and CAF co-cultures. Fig. S6. Secretome comparison between co-cultures of PC3 with NPFs or CAFs and corresponding monocultures. Fig. S7. X-Plot highlighting proteins with differential expression changes in PC3 co-cultures with NPFs or CAFs. Fig. S8. Expression levels of secreted factors in primary and secondary cytokine/chemokine array screens. Fig. S9. A random cell migration assay identifies FST as a critical regulator in the co-culture system. Fig. S10. FST knockdown in WPMY-1 and PC3-GFP cells by CRISPRi. Fig. S11. Stable knockdown of FST in both cell types by CRISPRi impairs migration of prostate cancer cells in co-culture.
Fig. S12. Human recombinant FST rescues the decreased migration of prostate cancer cells in co-culture caused by stable FST knockdown. Fig. S13. Human recombinant FST rescues the impaired proliferation of prostate cancer cells in coculture mediated by stable FST knockdown. Fig. S14. FST expression and knockdown in different prostate epithelial/cancer cell lines. Fig. S15. FST regulates migration of additional prostate epithelial/cancer cell lines in co-culture with fibroblasts. Fig. S16. Relationship of tumoural FST expression to disease-free survival of patients with prostate cancer. Table S2. siRNA sequences used. Table S3. Summary of sgRNA sequences and primers. Table S4. Real-time PCR primers. Table S1. Clinicopathological features of tumours of origin for primary fibroblast cell lines. Table S5. List of proteins for the 200 human antibody array analysis. Table S6. Expression of targets in the first and secondary customised antibody array analysis.