PEDF regulates plasticity of a novel lipid–MTOC axis in prostate cancer-associated fibroblasts

Cancer-associated fibroblasts (CAFs), a specialized group of activated fibroblasts, are important contributors in tumor–stroma crosstalk by facilitating the proliferation of cancer cells and by remodeling the extracellular matrix to support tumor invasion (Franco and Hayward, 2012; Gandellini et al., 2015; Banerjee et al., 2017). Despite their pro-tumorigenic role, most studies have documented epigenetic changes in CAFs rather than a gain of an oncogene or loss of a tumor suppressor, which is more common in their neighboring tumor epithelial cells (Hu et al., 2005; Qiu et al., 2008; Du and Che, 2017). Unlike normal prostate fibroblasts (NFs) and wound healing myofibroblasts, CAFs remain chronically active owing to the influx of soluble tumor-derived factors, and they strongly resist reversion to a normal quiescent phenotype (Li et al., 2007; De Waver et al., 2008; Franco and Hayward, 2012; Kalluri, 2016). In addition to the crucial role CAFs have in remodeling the microenvironment, a limited number of studies have shown that CAFs can serve as a local source of energy to help tumors in sustaining growth. In breast cancer, CAFs assist in meeting the energy demands of tumor cells by secreting lactate and pyruvate as energy metabolites through the glycolytic pathway, a well-described metabolic adaptation within the tumor microenvironment (TME) known as the Warburg effect (Warburg, 1956; Pavlides et al., 2009; Gonzalez et al., 2014). Cancer cells can, in turn, utilize these energy metabolites in the mitochondrial tricarboxylic acid (TCA) cycle, thereby, promoting energy production to increase their proliferative capacity. This cooperation between CAFs and cancer cells requires some intrinsic metabolic sensors to detect energy deficits to mobilize nutrients and accelerate the cell cycle for rapid growth. Although glucose is often the first source for metabolic needs, more recent studies revealed that tumor cells actively mobilize intracellular lipid stores to support their growth (Gazi et al., 2007; Nieman et al., 2011). To date, however, no studies have focused on metabolic re-programming in CAFs or the impact of a lipid-rich microenvironment on the phenotype of CAFs.

Intracellular lipid metabolism and homeostasis are regulated by selective proteins and highly dynamic organelles called lipid droplets (LDs). LDs are sites of lipid storage, membrane synthesis and trafficking of cargo proteins throughout the cytoplasmic compartment (Murphy, 2001). They are formed by a core of neutral lipids containing triacylglycerol (TAG) and cholesterol esters (CEs) that are surrounded by a phospholipid monolayer, with proteins either embedded in this monolayer or attached to its surface (Zehmer et al., 2009; Khor et al., 2013). LDs participate in lipid flux by undergoing an active cycle of lipolysis; this metabolic process involves several proteins, most of which are localized on the surface of the LD. This group of surface proteins includes adipose triglyceride lipase (ATGL; officially known as PNPLA2) (Zechner et al., 2009), comparative gene identification 58 (CGI-58; officially known as ABHD5) (Young and Zechner, 2013; Boeszoermenyi et al., 2015), members of the perilipin family (Khor et al., 2013) and pigment epithelium-derived factor (PEDF; officially known as SERPINF1) (Chung et al., 2008; Borg et al., 2011; Zhang et al., 2015), and they stimulate lipolysis and the release of free fatty acids (FFAs). Factors that regulate lipogenesis tend to reside in the cytoplasm but crosstalk between other TAG pathway members is essential to maintain the net lipid balance in normal cells. For example, while diacylglycerol O-acyltransferase 1 (DGAT1) promotes TAG synthesis (Sachdev et al., 2016), the G0/G1 switch protein 2 (G0S2) acts as a potent inhibitor of ATGL, and increased activity of one or both of these proteins favors ectopic lipid accumulation in many cell types (Harris et al., 2011; Schweiger et al., 2012; Khor et al., 2013; Cerk et al., 2014). Many tumors, including prostate cancer, have a significantly lower level of PEDF (Halin et al., 2010), while some head and neck tumors have a mutation in G0S2 (Tokumaru et al., 2004), suggesting that the pathologic imbalance in the TAG pathway is a common mechanism for dysregulated lipid metabolism. Owing to the multifunctional properties of the TAG-related proteins, altered expression of these molecules can negatively impact other fundamental processes, including mitosis, angiogenesis and apoptosis (Yamagishi and Matsui, 2014; Wang et al., 2015; Zagani et al., 2015; Grace et al., 2017).

LDs utilize microtubules (MTs) as tracks for directional movement (Bostrom et al., 2005; Orlicky et al., 2013; Welte, 2015) and one study found that cytoplasmic LDs tend to localize near the microtubule-organizing center (MTOC) in HEK293 cells (Orlicky et al., 2013); however, no biological explanation was proposed regarding any specific function LDs might exert close to an MTOC area. The best-studied MTOC is the centrosomal MTOC (cMTOC), which is located in the perinuclear area, generates a radial MT array and the mitotic spindles during mitosis (Bornens, 2002, 2012). More recent studies have expanded the view on MTOCs when differentiated cells were found to often develop alternative MTOCs defined as non-centrosomal MTOCs (ncMTOCs) (Sanchez and Feldman, 2017). These structures are located throughout the cytoplasm and, unlike cMTOCs, generate a dynamic and disorganized MT array (Sanchez and Feldman, 2017). These findings suggest that ncMTOCs are involved in non-mitotic processes, such as cell polarity, migration, invasion and intracellular trafficking (Bartolini and Gundersen, 2006). All of these processes are critically important in the TME. We postulate that direct crosstalk exists between lipid storage organelles and MTOCs, and that changes in neutral lipid content within CAFs and TAG-regulating proteins, such as PEDF, result in aberrant MTOC amplification and create a tumor permissive microenvironment. In the present study, we show that the deficiency of PEDF in prostate CAFs results in an increase in intracellular LDs, as well as in an MTOC amplification phenotype that consists of multiple MTOCs, consistent with ncMTOCs, dispersed throughout the cytoplasm. A plasticity of MTOC amplification in CAFs was discovered when restoration of PEDF normalized the number of MTOCs. PEDF had a similar activity in reducing MTOCs in prostate cancer cells. A novel lipid–MTOC signaling axis was observed when DGAT1 was inhibited to suppress intracellular LD density, concurrently also reducing the number of MTOCs. An unexpected interaction between MTOCs and LDs was noticed when LDs were found to carry centrosomal proteins, i.e. pericentrin and γ-tubulin, suggesting that LDs in CAFs can acquire a MTOC-like phenotype. Taken together, these data suggest that lipid-laden CAFs can modulate MTOC distribution and number through a new PEDF-dependent lipid–MTOC axis.

It is well known that PEDF exerts anti-tumor and anti-angiogenic functions in cancer (Crawford et al., 2001; Filleur et al., 2009; Becerra and Notario, 2013). PEDF can also regulate the lipid content through ATGL in other cell types (Chung et al., 2008; Borg et al., 2011) and functions as a Wnt inhibitor (Protiva et al., 2015). At baseline (i.e. in untreated control cells), PEDF (50 kD) was always highly expressed in NFs, whereas in CAFs little to no detectable PEDF protein was consistently found (Fig. 1A,B). Next, we determined whether a pro-lipogenic (oleic acid, OA) or anti-lipogenic (DGAT1 inhibitor) environment can alter the LD content by modifying the levels of PEDF. When NFs were treated with OA (lipid stimulus), PEDF protein levels were markedly reduced (Fig. 1A,B), suggesting a mechanism that allows stromal cells to increase lipid storage. When CAFs were tested under any condition, PEDF was not detectable by western blot. Only low levels of the ATGL inhibitor G0S2 were detectable in CAFs (Fig. 1A,C). Western blotting was performed to also assess the levels of ATGL (56 kD) and CGI-58 (40 kD) in NFs versus CAFs in response to different treatments (Fig. 1D). ATGL and CGI-58 are important proteins that are located on the surface of LDs and regulate intracellular lipid metabolism by promoting lipolysis (Eichmann et al., 2015; Lord et al., 2016). Similar to the pattern observed with PEDF, NFs expressed high levels of ATGL at baseline and the protein was reduced by treatment with OA [122.7±1.2 vs 62.7±2.4 (control vs OA); P<0.01; Fig. 1E]. At baseline, CAFs expressed less ATGL than NFs (64.2±7.0 vs 122.7±1.2; P<0.05) and the amount was relatively unchanged after inhibition of DGAT1 (Fig. 1D,E). Interestingly, when NFs or CAFs were lipid stimulated with OA, a second, more intense, band of higher molecular mass appeared above the ATGL band, which persisted after DGAT1 inhibitor was added to OA. The less-intense band of less molecular mass disappeared in CAFs treated with OA+DGAT1 inhibitor, while it persisted in NFs in response to the same treatment. However, the slight increase in intensity of the band at the higher molecular mass in CAFs could be a shift of the lower molecular weight protein (Fig. 1D). Whether the higher molecular weight protein observed in OA-stimulated cells represents another isoform of ATGL is not clear. Unlike PEDF and ATGL, levels of the ATGL activator CGI-58 remained relatively constant in untreated NFs and CAFs (controls), with only a modest reduction within the treatment groups (Fig. 1D,F). These results suggest that PEDF, ATGL and G0S2 in stromal fibroblasts are sensitive to microenvironmental stimuli, and are likely to contribute to the net lipid content.

To determine whether changes in TAG-modulating proteins result in dysregulated lipid storage in prostate stromal cells, we quantified the number of baseline LDs in NFs and CAF, and tested whether the LD density does change in response to various stimuli. The cells were stained with Oil-Red-O, which specifically stains neutral lipids, constituting the core of LDs (red staining in Fig. 2A,B). Under control conditions, both cell types revealed intracellular LDs, although they were different in number and distribution. CAFs showed a significantly higher mean LD density compared to that of NFs (188.5±17.8 vs 66.8±6.8; P<0.0001) (Fig. 2C). To simulate a lipogenic microenvironment of ectopic lipid accumulation, such as observed in obesity, NFs and CAFs were subjected to a 24 h treatment with 200 µM OA. This treatment regimen increased the mean number of LDs >2-fold in NFs (141.4±8.0 vs 66.8±6.8; P<0.0001), whereas an almost 1.5-fold increase was evident in CAFs (280.8±30.1 vs 188.5±17.8; P<0.05) (Fig. 2C). LD density in NFs never reached the baseline LD level observed in CAFs. To assess whether the lipid-rich phenotype in CAFs or NFs can be suppressed, DGAT1 inhibitor was used to block lipogenesis. This treatment drastically decreased the mean LD density in both cell types, whether added alone (NFs: 2.6±0.4 vs 66.8±6.8, P<0.0001; CAFs: 3.3±0.5 vs 188.5±17.8, P<0.0001) or together with OA (NFs: 60.1±11.3 vs 141.4±8.0, P<0.0001; CAFs: 88.4±12.7 vs 280.8±30.1, P<0.0001) (Fig. 2C). To determine whether or not suppression of lipid resulted in a change in the activation of CAFs, we confirmed that CAFs treated with OA still express alpha smooth muscle actin (α-SMA) and vimentin (data not shown).

LD localization and size have been shown to influence LD intracellular trafficking and signaling, and the patterns are cell type specific (Nielsen et al., 2017). The control group of NFs revealed a consistent phenotype, with LDs concentrated in the perinuclear region of the cell showing a ring-like configuration. This perinuclear distribution did not change, even when LD density increased after the treatment with OA. However, when NFs were treated with DGAT1 inhibitor, the size of the perinuclear LDs was reduced (Fig. 2A). At baseline, LDs in CAFs had a significantly different distribution pattern from NFs. LDs were diffusely distributed throughout the cytoplasm from the perinuclear region to the leading edge of the cells, and this pattern persisted after the treatment with OA. When CAFs were treated with DGAT1 inhibitor, the few remaining LDs were mainly located in the perinuclear region although some were also found in the peripheral zone (Fig. 2B).

Considering the importance of LD size for cargo proteins and signaling, LD size differences were evaluated. Specifically, LDs were divided in three groups based on their surface area: small LDs (S-LDs, 0.1–1.0 µm²), medium LDs (M-LDs, 1.1–2.5 µm²) and large LDs (S-LDs, >2.5 µm²). Under baseline conditions, both NFs and CAFs contained mainly small and some medium-sized LDs. Compared to NFs, the mean number of S-LDs and M-LDs in CAFs was higher (S-LDs: 138.6±15.3 vs 62.0±7.1, P<0.0001; M-LDs: 48.6±11.9 vs 4.2±0.8, P<0.0001) (Fig. 2D). In control CAFs and NFs no L-LDs were observed; but, unsurprisingly, treatment with OA increased the number of M-LDs and L-LDs in both NFs and CAFs compared to control cells (NF M-LDs: 71.5±5.6 vs 4.2±0.8, P<0.000; NF L-LDs: 13.8±1.3 vs 0.5±0.2, P<0.0001; CAF M-LDs: 133.9±20.5 vs 48.6±11.9, P<0.01; and CAF L-LDs: 24.1±6.1 vs 1.3±0.3; P<0.01) (Fig. 2D). However, the mean number of S- and M-LDs in CAFs was still higher than in NFs (S-LDs: 106.9±18.0 vs 57.7±4.8, P<0.01; M-LDs: 133.9±20.5 vs 71.5±5.6, P<0.001). Finally, to determine whether the DGAT1 inhibitor impacts on LD size, both NFs and CAFs were treated with either DGAT1 inhibitor alone, or DGAT1 inhibitor with or without OA. Treatment with DGAT1 inhibitor decreased the size of LDs significantly, especially those of S-LDs in both NFs and CAFs (Fig. 2D). A concurrent increase in lipolysis due, in part, to the marked expression of pro-lipolytic proteins (Fig. 1A,D) could be one mechanism responsible for the size reduction in LDs and the decrease in LD density observed.

CAFs are known to proliferate at a higher rate than NFs (Kalluri, 2016). Given that our data showed elevated levels of stored lipids in CAFs versus NFs, it was unclear whether modulation of intracellular lipid would alter cellular proliferation. To assess whether reduction of LD density impacts on proliferation of NFs and CAFs, cells were treated for 24 h with DGAT1 inhibitor (that inhibits lipogenesis). Cell proliferation was then analyzed based on the percentage of cells that stained positivity of the proliferating cell nuclear antigen (PCNA); cells treated with DGAT1 inhibitor were compared to untreated control cells (Fig. 3A,B). Our results showed that treatment with DGAT1 inhibitor reduced cell growth in both NFs (1.2±0.1 vs 2.1±0.3; P<0.05) and CAFs (1.5±0.2 vs 4.3±0.5; P<0.001) (Fig. 3B). Moreover, in support of other studies, untreated CAFs demonstrated increased proliferation rate compared to NFs (4.3±0.5 vs 2.1±0.3; P<0.001). Growth of NFs and CAFs was also analyzed by using the MTT Cell Proliferation Assay (Fig. 3C) and, again, our data showed that treatment with DGAT1 inhibitor, significantly decreased proliferation of both NFs (OD: 0.20±0.00 vs 0.23±0.00; P<0.05) and CAFs (OD: 0.39±0.00 vs 0.50±0.00; P<0.0001) when compared to untreated controls (Fig. 3C). These results suggest that the higher proliferative baseline capacity of CAFs is due to ready access to stored lipids and/or loss of growth inhibition by PEDF. Moreover, elevation of PEDF in NFs could be one mechanism to suppress growth when lipogenesis is blocked by DGAT1 inhibitor (Fig. 1A).

Amplification of the centrosome – or, in general, abnormalities of the MTOC – has been observed in many tumors and can disrupt astral microtubule organization within cells, promote aneuploidy and/or directly promote tumorigenesis, possibly independent of genomic instability (Zyss and Gergely, 2009). Since CAFs, unlike cancer cells, are cells that are genetically relatively stable within the TME, we sought to determine whether CAFs had acquired any MTOC aberrations similar to those of tumor cells and to test whether a microenvironmental stimulus, such as a lipid-inducing challenge is capable of modulating MTOC number. To assess MTOC density in CAFs compared to NFs, we performed immunofluorescence staining of the pericentriolar matrix protein pericentrin, commonly used to denote the presence of both types of MTOC, i.e. cMTOC and ncMTOC (Dictenberg et al., 1998). As expected, all NFs revealed one or two pericentrin-stained foci located very close to the nucleus (Fig. 4A). The MTOC number in NFs remained the same despite any exogenous treatment regimen. In contrast, the MTOC density and distribution in CAFs were strikingly different. At baseline, CAFs demonstrated a significant increase in pericentrin foci (Fig. 4B,C) compared to NFs. These MTOCs localized not only to the perinuclear region but were also spread to the cytoplasmic edge of the cell (Fig. 4B,C,E). We discovered that the MTOC number in CAFs varied significantly and appeared to act as a metabolic sensor because alteration of lipid content in response to a microenviromental stimulus changed MTOC density. Administration of OA markedly increased MTOC density (156.7±16.2 vs 69.1±10.6; P<0.0001) and treatment with DGAT1 inhibitor simultaneously reduced the density of LDs and ncMTOCs (40.5±7.0 vs 69.1±10.6; P<0.05) (Fig. 4C,D). This MTOC sensitivity to intracellular lipid content suggests a new signaling pathway between MTOCs and LDs. In addition to the number of MTOCs, the composition of the MT network in CAFs was different from the pattern observed in NFs. The MT network in CAFs appeared to be less organized, especially in perinuclear and peripheral regions, compared with the more organized MTs in NFs (Fig. 4A,B).

To assess whether restoration of PEDF normalized MTOC number or the MT network in CAFs, cells were treated for 48 h with 10 nM PEDEF, PEDF+OA or PEDEF+DGAT1 inhibitor. The MTOC density and MT morphology were analyzed (Fig. 5A,B) and revealed that treatment of CAFs with exogenous PEDF, compared to untreated CAFs, significantly reduced the number of pericentrin-stained MTOCs by 98.1% (1.3±0.1 vs 69.1±10.6; P<0.0001) (Fig. 5C). When lipid-stimulated CAFs were treated with PEDF, the MTOC number normalized to close to one MTOC (1.4±0.2 vs 156.7±16.2; P<0.0001) (Fig. 5A,C). To determine if the addition of DGAT1 inhibitor synergized with PEDF in modulating the MTOC number, both agents were used in one experimental group. No significant advantage was observed beyond the ‘normalization’ of MTOC number by treatment with PEDF alone (Fig. 5A,C). Moreover, the few [1–2 MTOCs remaining post treatment tended to be slightly larger and were located in the perinuclear region of the cell, the typical site of cMTOCs (Fig. 5B)]. Assessment of the pericentrin ‘dot’ area in control CAFs revealed a variation in size with a mean area of 13.9±1.0 µm² (Fig. S1). When CAFs were treated with PEDF the 1–2 pericentrin dots left appeared more uniform in size and overall larger with a mean area of 23.1±3.6 µm². These larger pericentrin dots were often irregular and multi-lobular in shape, suggesting an aggregation of several MTOCs. These data suggest that PEDF treatment not only normalizes the number of MTOCs but also promotes clustering or aggregation of several MTOCs. Treatment with PEDF also triggered an unexpected but consistent structural change in the MT network. The PEDF-deficient CAF phenotype of dense and disorganized MTs was remodeled to a less dense and more radial MT configuration in the perinuclear region when PEDF was restored (area occupied by MTs: 43.0%±0.6 vs 35.8%±0.8; P<0.001) (Fig. 5A,D). The peripheral MT component in CAFs that remained post-treatment was less dense but retained higher complexity.

To assess if restoration of PEDF in CAFs reduced neutral lipid content, CAFs were treated with 10 nM PEDF for 48 h, fixed and LDs stained with Oil-Red-O. Adding exogenous PEDF promoted a 73.3% decrease in the number of LDs per cell compared to CAF baseline (50.4±5.3 vs 188.5±17.8; P<0.0001) (Fig. 5G). Interestingly, when PEDF treatment was combined with addition of DGAT1 inhibitor, CAFs showed increased lipid catabolism with a significant decrease in the number of LDs per cell compared to treatment with PEDF alone (2.1±0.5 vs 50.4±5.3; P<0.0001) or DGAT1 inhibitor alone (2.1±0.5 vs 4.2±0.7; P<0.05) (data not shown). These data suggest that PEDF can synergize with the DGAT1 inhibitor to decrease LD density in CAFs. PEDF treatment of CAFs markedly reduced the number of MTOCs to 1–2 per cell, whereas this same treatment was only able to suppress LD density to a mean of 50 LDs/cell.

Centrosome number and MTOC abnormalities have long been implicated in cancer; they can cause chromosomal instability and loss of tissue architecture, two of the most common phenotypes observed during tumorigenesis (Nigg, 2006). To assess MTOC density in prostate cancer cells, immunofluorescence staining was performed in LNCaP and PC3 cells to detect the presence of the pericentriolar matrix protein pericentrin. Both LNCaP and PC3 cells demonstrated an increase in pericentrin-positive foci (within the range of 3–10 per cell), with many of the ncMTOCs diffusely located throughout the cytoplasm of the cell (Fig. 5E,F). Since addition of PEDF to CAFs resulted in a significant change in MTOC amplification by normalizing their numbers back to ∼1 perinuclear centrosome, we investigated whether PEDF had the same effect in cancer cells. The analysis revealed that, compared to untreated control cells, treatment with exogenous PEDF significantly reduced the number of pericentrin-positive MTOCs in both LNCaP cells (1.4±0.1 vs 5.2±0.5; P<0.0001) and PC3 cells (1.3±0.1 vs 4.7±0.6; P<0.0001) (Fig. 5E,F).

To determine whether the amplified MTOC population in CAFs exhibited typical features of centrosomes, immunofluorescence staining for pericentrin and centrin was performed (Fig. 6A). NF control cells consistently had a single pericentrin-positive/centrin-positive focus. In contrast, CAF, LNCaP and PC3 control cells demonstrated three or more structures with dual staining for centrin and pericentrin supporting that these are consistent with cMTOCs (CAFs versus NFs: 4.0±0.2 vs 1.2±0.1, P<0.0001; LNCaP versus NFs: 3.4±0.2 vs 1.2±0.1, P<0.0001; PC3 vs NFs: 4.1±0.2 vs 1.2±0.1, P<0.0001) (Fig. 6A,B). Some other structures in these cells stained only positive for pericentrin and most were most likely to represent ncMTOCs and/or pericentrin-positive LDs. When CAFs were treated with exogenous PEDF, the number of foci positive for centrin and pericentrin reduced to 1–2 (CAFs+PEDF versus CAFs CTR: 1.3±0.1 vs 4.0±0.2; P<0.0001), demonstrating that PEDF is also able to normalize the cMTOCs in CAFs (Fig. 6A,C). To assess MT nucleation, re-growth assays were performed and analyzed in control and PEDF-treated cells (Fig. 6D). At baseline, after 30 s of regrowth, NFs showed a single MT array that appeared organized, whereas CAFs exhibited several f MT nucleation foci and their MT arrays appeared less symmetrical (Fig. 6D magnification). Treatment with PEDF reduced the number of MTOCs in CAFs to 1 and, in this group, the MTs appeared shorter (Fig. 6D), which concurs with our other finding demonstrating reduced MT density in PEDF-treated cells (Fig. 5A,D). Since the analysis of MTOC amplification in CAFs revealed more pericentrin-stained ncMTOCs than expected and the distribution patterns involved regions away from the nucleus, we investigated whether or not LDs interact more directly with MTOCs and, possibly, carry MTOC matrix proteins, such as pericentrin or γ-tubulin, on their surface. Immunofluorescence staining was performed to assess a possible colocalization between γ-tubulin and LDs in untreated CAFs (Fig. 6E), showing that a subset of LDs appears to be positive for centrosomal proteins. It is possible that pericentrin and/or γ-tubulin positivity in a subset of LDs (30–40%) represents cargo proteins on the LD surface. Although this would be the first observation of a centrosomal protein associated with LDs, these organelles are known to actively shuttle a variety of proteins within the cytoplasm of several cell types (Murphy, 2001).

PEDF is known to be an inhibitor of the Wnt/β-catenin pathway, which regulates several processes including angiogenesis, inflammation and fibrosis (Park et al., 2011; Protiva et al., 2015). In prostate cancer, β-catenin is known to be overexpressed within the nuclear compartment, where it can chronically activate the transduction of several genes implicated in tumor growth (Kypta and Waxman, 2012). More recently, β-catenin has been recognized to have additional activities related to centrosomes. β-catenin was found to colocalize with pericentrin during interphase and mitosis, and to be involved in centrosome segregation and mitotic spindle orientation (Bahmanyar et al., 2010; Mbom et al., 2013). To assess whether PEDF can change MTOCs was through inhibition of the Wnt pathway, β-catenin levels were analyzed by western blotting and immunofluorescence. Fig. 7A shows the levels of β-catenin (94 kD) in NF and CAF control cells. Compared to those in CAFs, NFs had lower levels of total-β-catenin. To investigate whether addition of PEDF inhibits active β-catenin, immunofluorescence staining was performed, showing intracellular localization of active β-catenin in CAFs (Fig. 7B). In control cells, β-catenin was located in both the cytoplasm and nucleus. The presence of nuclear β-catenin was confirmed using z-stack analysis and by performing vertical and horizontal cuts through the nucleus. After treatment with exogenous PEDF, levels of β-catenin – the functional modulator of centrosome morphology and distribution – were markedly decreased in both the cytoplasmic and nuclear compartments of CAFs, which was confirmed by quantification of total β-catenin and β-catenin phosphorylated at Y142 (phospho-β-catenin) in western blots (CAF control cells vs CAFs treated with PEDF) (Fig. 7C). These data suggest a new mechanism for MTOC amplification in CAFs that is related to the loss of the PEDF Wnt inhibitory action, thus increasing β-catenin (Mbom et al., 2013; Protiva et al., 2015).

Restoration of PEDF in CAFs reduces the number of LDs and normalizes MTOC density, thus, more resembling the phenotype of NFs. To assess the impact of PEDF reduction in NFs on MTOC biology, we performed RNA interference (RNAi), in which PEDF was decreased by >50%; pericentrin- and centrin-positive MTOC density was analyzed by immunofluorescence staining (Fig. 7E). Although there was no amplification of MTOC as observed in CAFs, PEDF-deficient NFs showed a significantly higher percentage of cells with double centrosomes (showing centrin- and pericentrin-positive staining) when compared to the control group (62.8±3.5 vs 13.5±2.9, P<0.0001) (Fig. 7F). These data suggest that PEDF deficiency results in the loss of a crucial inhibitory signal that is related to cell cycle progression or the early stages of centrosomal amplification. It is not surprising that deficiency of PEDF only was insufficient to induce amplification because other studies have shown that single knockdown of even p53 was unable to trigger amplification (Nigg and Holland, 2018).

Metabolic adaptations are known to occur with high frequency in tumor epithelial cells to help meet the demands of their high proliferative capacity (Levine and Puzio-Kuter, 2010; Satoh et al., 2017). Cancer cells can revert to lipid stores to obtain the needed energy and to alter cell cycle regulators, such as centrosomes, to fuel progression. Measurement of metabolites and lipids associated with the metabolic re-programming switch in prostate cancer has the potential diagnostic use of distinguishing cancer from normal tissue (Banerjee et al., 2014). Much less is known about the spectrum of metabolic changes made within stromal fibroblasts, specifically CAFs, in the TME. In this study, we have demonstrated that human prostate-derived CAFs store more neutral lipid in LDs than normal prostate fibroblasts (NFs). Each cell type had a unique LD distribution pattern and size range, which could impact signaling (Welte and Gould, 2017; Bombrun et al., 2017). Normal fibroblasts had a distinct concentric ring of neutral lipids residing close to the nucleus, whereas LDs in CAFs were dispersed throughout the cytoplasm. Intracellular lipid content and LD size were sensitive to exogenous stimuli, such as lipid challenge with OA, since both cell types augmented their lipid stores and increased their mean LD size; however, the mean LD area was always greater in CAFs. To assess whether lipid storage in CAFs is linked to growth, we blocked the lipogenesis enzyme DGAT1. Both growth and lipid stores of NFs and CAFs were suppressed, although the activation of CAFs was not affected. These results suggest that prostate CAFs are simultaneously keeping pace with their tumor cell partners by making pro-lipogenic metabolic adaptations and that the metabolic switch has a broader effect on the TME (Banerjee et al., 2014). In the TME of obese patients, where lipid flux is abnormal, it is possible that CAFs acquire an even higher net lipid content to expand their cell population and promote tumor progression (Engin, 2017).

Synthesis or mobilization of stored lipid, such as TAG, requires precise coordination of various enzymes to maintain lipid homeostasis (Zechner et al., 2017). The rate-limiting enzyme ATGL and its activator CGI-58 control lipolysis via the TAG pathway, and these activities are counterbalanced by the lipogenesis enzyme DGAT1 (Harris et al., 2011). Other TAG-regulating proteins that influence net lipid content include PEDF, which binds to ATGL (Chung et al., 2008; Borg et al., 2011) and G0S2, which acts as an inhibitor to ATGL (Schweiger et al., 2012; Cerk et al., 2014). Mutations in ATGL, CGI-58 or PEDF can cause a range of human diseases, including systemic metabolic disorders or defects in bone development (Marini et al., 2014). Recently, loss of CGI-58 in prostate cancer cells was found to promote an aggressive phenotype (Chen et al., 2017a,b). In many cancers, including prostate cancer, there is a stepwise decrease in PEDF expression when tumors become more aggressive (Halin et al., 2010; Becerra and Notario, 2013). Much less is known about the influence PEDF or other factors have in the regulation of the TAG pathway in the stromal compartment of the prostate. PEDF is a 50 kD glycoprotein that is expressed in almost all healthy (normal) cells. It has a broad spectrum of functions including anti-inflammatory, anti-angiogenic and anti-tumorigenic activities, and acts as a Wnt inhibitor (Dawson et al., 1999; Crawford et al., 2001; Chung et al., 2008; Filleur et al., 2009; Becerra and Notario, 2013; Protiva et al., 2015). PEDF influences systemic fatty acid metabolism by enhancing lipolysis, and by promoting lipid accumulation in skeletal muscle and liver. The lipolytic activity of PEDF was discovered when PEDF null mice demonstrated hepatic steatosis, and when human hepatocytes required interactions between PEDF and ATGL to regulate their lipid content (Chung et al., 2008; Borg et al., 2011). In our study here, ATGL and PEDF were found to be strongly expressed in NFs, an observation that was in contrast to the much reduced levels observed for the lipolysis inhibitor protein G0S2. Prostate-derived CAFs had minimal to undetectable levels of PEDF and ATGL, which were lower than those recovered in NFs. The limited availability of the key lipolysis-regulating proteins PEDF and ATGL are likely to contribute to the net gain of stored lipid observed in CAFs, because restoration of PEDF effectively decreased intracytoplasmic lipid content. In one study investigating melanoma, CAFs expressed low levels of PEDF and loss of this protein contributed to tumor progression; however, lipid metabolism was not explored (Nwani et al., 2016). Our data, demonstrating modulation of PEDF, ATGL and G0S2 in CAFs, add new lipid mediators for consideration when evaluating aberrant lipid signaling in tumor–stroma crosstalk.

LDs not only store lipid but they can also carry cargo and facilitate the movement of proteins to various locations within the cell. Studies involving Drosophila were some of the first to recognize that LD movement is dependent on intact MTs. In Drosophila embryos and some normal human cells, kinesin 1 and cytoplasmic dynein, i.e. MT-associated motor proteins, were shown to physically interact with LDs and direct LD motility (Welte, 2015). This cooperation between LDs and MTs could have broader implications in the control of cellular functions in the TME. MTOCs operate as sites in order to locate MT minus ends, and function as the point of MT nucleation and stabilization by using pericentriolar matrix proteins, such as pericentrin (Dictenberg et al., 1998; Bornens, 2002). The plus-ends of MTs are more dynamic compared to the slower remodeling minus ends (Voter and Erickson, 1984). Correct MTOC orientation is crucial for several processes, including mitosis, cellular differentiation and secretion (Vertii et al., 2016; Muroyama and Lechler, 2017; Sanchez and Feldman, 2017). The centrosome is one of the best-studied MTOCs (Bornens, 2012). Normal cells have one – occasionally two, when dividing – centrosome, the cMTOC, close to the nucleus. It is a non-membrane bound organelle surrounded by pericentriolar matrix proteins, such as pericentrin and γ-tubulin, and responsible for generating a radial array of MTs (Dictenberg et al., 1998). Pericentrin controls the nucleation of MTs by anchoring the γ-tubulin ring complex to the MTOC (Bornens, 2002). In the cMTOC, precision regarding nucleation is essential for bipolar spindle formation and chromosome assembly during mitosis, in order to ensure correct cell cycle progression. In support of the importance of pericentrin, its downregulation in peripheral blood leukocytes has been shown to disrupt mitotic checkpoints and to cause arrest at the G2/M checkpoint, leading to cell death (Unal et al., 2014).

Supernumerary centrosomes, represented by several pericentrin-positive foci that were diffusely located in the cytoplasm, have also been associated with increased tumor aggression, raising the notion that additional centrosomes are advantageous for tumor growth (Pihan et al., 2001; Chan, 2011; Rivera-Rivera and Saavedra, 2016). In support of this concept, centrosome overexpression was found to increase MT nucleation and resulted in the formation of protrusions and cytoplasmic extensions that invaded the surrounding matrix, thus, enabling tumors to invade and metastasize (Godinho et al., 2014). Cancers of the breast, ovary, liver and prostate were shown to have structural abnormalities of the centrosome regarding size, shape and number, although centrosomes have not been studied in the stromal cell population within the TME (Lingle et al., 1998; Pihan et al., 2001; Kim et al., 2008). In our study, untreated control CAFs showed amplification of cMTOCs and ncMTOCs compared to the 1–2 perinuclear MTOCs consistently observed in NFs. Additionally, we observed that restoration of PEDF in CAFs not only normalized the number of MTOCs but also reduced MT density and levels of β-catenin. A similar reduction in the number of cMTOCs and ncMTOCs was found when cells of the prostate cancer cell lines LNCaP and PC3 were treated with PEDF. One potential mechanism for these activities included the ability of PEDF to act as a potent Wnt inhibitor and reduce levels of activated β-catenin. Overexpression of β-catenin in prostate cancer is well-documented (Kypta and Waxman, 2012). β-catenin has also been shown to localize to the centrosome, and it appears to be important in mitotic spindle assembly and microtubule dynamics (Mbom et al., 2013). Its role in ncMTOCs has not been explored. The ability of PEDF to modulate MTs has been observed by another group, where PEDF-mediated disassembly of the MT network was a mechanism responsible for increased permeability in vascular endothelial cells (He et al., 2015). Our data, demonstrating a new function of PEDF to normalize MTOC number in CAFs, highlight a previously unrecognized plasticity in cMTOCS and/or MTOC biology and a possible function of MTOCs as metabolic sensors, especially ncMTOCs. Furthermore, blockade of lipogenesis in response to DGAT1 inhibitor reduced both lipid content and MTOC number. This discovery of a new lipid–MTOC signaling axis (see model proposed in Fig. 8) that is highly responsive to PEDF and DGAT1 inhibitors could prove to be an attractive therapeutic target to stabilize tumor growth by simultaneously reducing stored lipid and the numbers of cMTOCS and ncMTOCs.

More recently, there has been more interest in characterizing the pericentrin- or γ-tubulin-stained structures distributed in the cytoplasm and peripheral regions of the cell. In addition to the well-described perinuclear centrosome, another type of structure – the ncMTOC, which can vary depending on the cell type –has been described (Sanchez and Feldman, 2017). Studies are emerging to better characterize the structure and function of ncMTOCs. It appears that ncMTOCs contain proteins that interact with MT minus-ends and help to anchor proteins; however, details of their function in cellular processes within the TME remain to be elucidated. Cellular differentiation is one bioactivity that has been linked to activation of ncMTOCs (Sanchez and Feldman, 2017) and that is relevant to tumor aggressiveness. Some groups have reported that other organelles, such as mitochondria or the Golgi complex, can simulate or convert to ncMTOCs through modulation of various proteins (Zhu and Kaverina, 2013; Chen et al., 2017a,b). Recently, a mouse embryonic study revealed a new function for ncMTOCs in directing intracellular transport of molecules, such as E-cadherin (Zenker et al., 2017). In our current study of lipid metabolism in CAFs, we noticed an excessive number of pericentrin-expressing cytoplasmic structures outside of the typical perinuclear location of a centrosome. Centrin positivity and evidence of MT nucleation were observed in four or more structures, supporting the concept that both cMTOCs and ncMTOCs are amplified in CAFs. The group of ncMTOCs that did not show MT nucleation might participate in some of their other proposed – non-mitotic – functions, such as cell differentiation and polarity. NFs contained endogenous lipid and, when stained for pericentrin, only one centrosome was found to be localized close to the nucleus. Occasionally, a dividing cell would have two centrosomes. When NFs were challenged with a lipid-promoting stimulus, the centrosome number remained constant at 1. In contrast, untreated control CAFs harbored a significantly higher number of cytoplasmic structures that stained positive for pericentrin and these increased in response to the lipid challenge. The mean number of pericentrin-positive structures was more than three times higher than the reported number of centrosomes in amplified tumors (Lingle et al., 1998; Pihan et al., 2001). Our results prompted us to test the hypothesis that crosstalk exists between a microenvironmental lipid challenge and the modulation of centrosomal structures or ncMTOCs. To test this, we blocked lipogenesis by using the DGAT1 inhibitor and found that this treatment not only reduced the lipid content but, concurrently, also reduced pericentrin-positive ncMTOCs in the cytoplasm. To further explore this signaling network and test the hypothesis that a microenvironmental lipid challenge elevates ncMTOC density, we asked whether pericentrin is one of the cargo proteins on LDs. Since LDs move along MTs, ready access to a MT matrix protein might facilitate modulation of MTs in a more efficient manner. We found both the MTOC matrix proteins pericentrin and γ-tubulin on the surface of a subset of LDs, distributed throughout the cytoplasm in CAFs, thus, emphasizing the importance of testing the functional capacity of pericentrin-positive structures in the TME. It is unclear whether carrying these matrix proteins indicates that LDs can acquire an ncMTOC-like phenotype within a lipid-rich environment, such as in cells of obese individuals, or whether easy access of these proteins on the surface of LDs enables them to rapidly remodel MTs to facilitate movement. Additional studies are required to assess the function of MTOC proteins on the LD surface and to investigate whether these proteins still perform their usual MT-related functions. In summary, our data support that prostate CAFs have acquired a tumor-like phenotype by re-programming lipid metabolism and amplifying MTOCs. Normalization of MTOCs by restoring PEDF or by blocking lipogenesis in CAFs highlights a previously unrecognized plasticity in cMTOC and ncMTOC biology.

In this study the experiments were performed by using two primary human normal fibroblast (NF) cultures and three primary cancer-associated fibroblast (CAFs) cultures isolated, respectively, from normal prostate and radical prostatectomy specimens of different patients. CAFs were evaluated and tested in an in-vivo tissue recombination model to confirm their pro-tumorigenic potential (Franco et al., 2011). Human prostate cancer cell lines LNCaP and PC3 were purchased from ATCC (Manassas, VA). NFs, CAFs and LNCaP cells were cultured at 37°C under 5% CO₂ in RPMI medium (Thermo Fisher Scientific, Waltham, MA) containing 10% fetal bovine serum (FBS; Sigma, St Louis, MO) and 1% penicillin/streptomycin, whereas PC3 cells were cultured in DMEM (Thermo Fisher Scientific, Waltham, MA) containing again 10% FBS and 1% penicillin/streptomycin. After reaching 80–90% confluence, cells were harvested with 0.25% trypsin-EDTA (Thermo Fisher Scientific, Waltham, MA) and passaged at a ratio of 1:2 (confluent cells in solution to fresh medium). After seeding overnight (at ∼60% confluence), cells were exposed to several treatments: 200 µM oleic acid (OA; Sigma) with or without 1 µM DGAT1 inhibitor (A-922500, Cayman Chemical, Ann Arbor, MI) for 24 h; 10 nM PEDF (BioProducts MD, Middletown, MD) with or without 200 µM OA for 48 h; and 10 nM PEDF with or without 1 µM DGAT1 inhibitor for 48 h.

NFs and CAFs were grown on glass coverslips coated with poly-L-lysine (Sigma) and cultured and treated as described above. Cells were then washed three times with PBS, fixed in 10% formalin (30 min at room temperature) and stained with Oil-Red-O (Oil-Red-O Stain, propylene glycol; Newcomer Supply, Middleton, WI) to visualize neutral lipids. Coverslips were mounted on glass slides and sealed with permaslip. Pictures were taken of representative fields for each treatment using a 100× objective to count single intracellular LDs.

NFs and CAFs cultured on dishes were treated as described above and after 24–48 h washed with PBS, scraped and lysed in M-PER+protease inhibitor buffer (Sigma). Cell lysates were centrifuged at 8050 g for 20 min at 4°C; the protein concentration was then determined using Pierce 660 nm Protein Assay Reagent (Thermo Fisher Scientific, Waltham, MA) and compared with that of a standard. Proteins separated by pre-cast gels (Mini-PROTEAN TGX Stain-Free Gel; Bio-Rad, Des Plaines, IL) were transferred onto 0.2 μm polyvinylidene difluoride (PVDF) membranes (Bio-Rad), blocked in 7% milk and incubated with primary and secondary antibodies. Antibodies against PEDF (1:1000, Cat. No. AB-PEDF4, BioProducts MD, Middletown, MD), ATGL (1:200, Cat. No. 10006409, Cayman Chemical), CGI-58 (ABHD5, 1:250, Cat. No. sc-376931, Santa Cruz Biotechnology, Dallas, TX), G0S2 (1:500, Cat. No. H00050486-B01P, Novus Biologicals, Littleton, CO), total β-catenin (1 µg/ml, Cat. No. ab16051, Abcam, Cambridge, UK) and β-catenin phosphorylated at Y142 (0.25 µg/ml, Cat. No. ab83295, Abcam) were used. Antibody staining against GAPDH (1:1000, Cat. No. 2118, Cell Signaling Technology, Danvers, MA) was for normalization. Membranes were incubated at room temperature with enhanced chemiluminescence (ECL) detection reagents (Bio-Rad) and the protein bands were visualized by chemiluminescence using the ChemiDoc Touch Imaging System (Bio-Rad).

Proliferation analysis was performed using two methods: the proliferating cell nuclear antigen (PCNA) staining and the Vibrant MTT Cell Proliferation Assay Kit (Molecular Probes, Eugene, OR). In the first case, NFs and CAFs were grown on glass coverslips and treated with DGAT1 inhibitor for 24 h. After treatment, cells were washed three times with PBS, fixed in 4% paraformaldehyde for 20 min and stained with PCNA antibody (1:40, Cat. No. M0879, Dako, Denmark). After staining, coverslips were mounted on glass slides and the percentage of positive cells in the DNA synthesis phase was determined. Regarding the MTT Cell Proliferation Assay, a total of 10⁴ cells (CAFs) per well was seeded into 96-well culture plates and treated for 24 h with DGAT1 inhibitor. After 24 h, the medium was replaced with 100 µl of fresh medium and 10 µl of 12 mM MTT was added to each well. After incubation of 4 h at 37°C, 100 µl of 0.01 M SDS-HCl solution containing 0.1 g/ml SDS was added to solubilize the formazan dye product for 4 h incubation at 37°C. The optical density was then determined using a spectrophotometer at a wavelength of 570 nm.

Cells were grown and treated on glass coverslips. After the treatments, cells were washed three times with PBS, fixed in 4% paraformaldehyde for 20 min, and permeabilized with 0.1% Triton X-100 for 15 min. Cells were first incubated for 20 min in 5% normal horse serum, and then with primary and secondary antibodies. Mouse anti-γ-tubulin (1 µg/ml, Cat. No. ab27074, Abcam), rabbit anti-α-tubulin (1:100, Cat. No. ab15246, Abcam), mouse anti-pericentrin (1 µg/ml, Cat. No. ab28144, Abcam) and rabbit anti-centrin 1 (1:500, Cat. No. 12794-1-AP, Proteintech, Chicago, IL) antibodies were used to observe the MT point of nucleation, MTs, MTOCs and centrosomes, respectively, whereas rabbit anti-β-catenin antibody (1:100, Cat. No. 95826, Cell Signaling Technology) was used to stain protein inside the cells. Intracellular LDs were stained using BODIPY for 20 min (Thermo Fisher Scientific, Waltham, MA). After staining, the coverslips were mounted on glass slides using ProLong Gold antifade reagent with DAPI (Invitrogen, Carlsbad, CA) and sealed with permaslip. Cells were then imaged using confocal microscopy (Nikon Eclipse TE 2000-U microscope equipped with NIS-Element version 4 software).

Cells were grown on glass coverslips and, after reaching 60–70% of confluence, 5 µM nocodazole was added to the medium for 3 h at 37°C to depolymerize the microtubules. After the incubation the cells were washed with fresh medium and incubated at 37°C to allow regrowth. Cells were then fixed at intervals of 30 s, 1 min and 5 min in 4% paraformaldehyde for immunofluorescence. MTs and MTOCs were stained using rabbit anti-α-tubulin and mouse anti-pericentrin antibody, respectively.

To reduce PEDF levels in human normal fibroblasts, we transfected normal fibroblasts according to the manufacturer's instructions (Thermo Fisher Scientific, Waltham, MA) with commercial small interfering RNA (siRNA) constructs targeting PEDF (Cat. No. 4392420). After 72 h of transfection, cells were harvested and cultured for 48 h in fresh medium lacking siRNA in a dish or on poly-L-lysine-coated coverslips to perform western blotting or immunofluorescence staining, respectively.

To determine differences between groups we used Student's t-test, where the differences were considered statistically significant when P<0.05. This analysis was performed using GraphPad Prism, version 7.03. ImageJ software was used to analyze LD number and size and to quantify the protein levels by densitometry. Microtubule density analysis was performed using ImageJ by calculating the area (in %) occupied by MT.

We acknowledge the technical expertise provided by Dr Lijun Huang NorthShore University Research Institute, Evanston, IL.