Follistatin-like protein 1 plays a tumor suppressor role in clear-cell renal cell carcinoma

Background We previously showed that the expression of follistatin-like protein 1 (FSTL1) was significantly down-regulated in metastatic clear-cell renal cell carcinoma (ccRCC). In this study, we aimed to characterize the role of FSTL1 in the development of ccRCC. Methods The effects of FSTL1 on cell activity and cell cycle were investigated in ccRCC cell lines with altered FSTL1 expression. Gene expression microarray assays were performed to identify the major signaling pathways affected by FSTL1 knockdown. The expression of FSTL1 in ccRCC and its effect on postoperative prognosis were estimated in a cohort with 89 patients. Results FSTL1 knockdown promoted anchorage-independent growth, migration, invasion, and cell cycle of ccRCC cell lines, whereas FSTL1 overexpression attenuated cell migration. FSTL1 knockdown up-regulated nuclear factor-κB (NF-κB) and hypoxia-inducible factor (HIF) signaling pathways, increased epithelial-to-mesenchymal transition, up-regulated interleukin-6 expression, and promoted tumor necrosis factor-α-induced degradation of NF-κB inhibitor (IκBα) in ccRCC cell lines. FSTL1 immunostaining was selectively positive in epithelial cytoplasm in the loop of Henle, and positive rate of FSTL1 was significantly lower in ccRCC tissues than in adjacent renal tissues (P < 0.001). The multivariate Cox regression analysis showed that the intratumoral FSTL1 expression conferred a favorable independent prognosis with a hazard ratio of 0.325 (95% confidence interval 0.118–0.894). HIF-2α expression was negatively correlated with FSTL1 expression in ccRCC specimens (r = − 0.229, P = 0.044). Intratumoral expression of HIF-2α, rather than HIF-1α, significantly predicted an unfavorable prognosis in ccRCC (log-rank, P = 0.038). Conclusions FSTL1 plays a tumor suppression role possibly via repressing the NF-κB and HIF-2α signaling pathways. To increase FSTL1 expression might be a candidate therapeutic strategy for metastatic ccRCC.

In our previous study investigating global profiling of gene expression in RCC cells with different metastatic potential, we found that follistatin-like protein 1 (FSTL1) was frequently down-regulated in metastatic ccRCC [19]; and the C allele of rs1259293 in the coding region of FSLT1 was associated with an increased risk and unfavorable postoperative prognosis of RCC, possibly by down-regulating FSTL1 expression in renal tissues [20]. FSTL1, a secreted glycoprotein encoded on chromosome 3 in humans, is widely expressed in cells of non-hematopoietic lineage, particularly in cells of the mesenchymal lineage [21]. FSTL1 is induced in response to inflammatory injuries and plays important roles in promoting the accumulation of myofibroblasts and subsequent fibrosis, promoting cardiac function, and reducing glomerular and tubulointerstitial inflammatory damage in the kidney via attenuating tumor necrosis factor alpha (TNFα)stimulated expression of proinflammatory cytokines [22][23][24]. The role of FSTL1 in cancers is controversial. During cancer metastasis from the primary site to the bone, FSTL1 mediates cancer cell invasion and expands a population of bone marrow-derived pluripotent mesenchymal stem-like cells [25]. In prostate cancer, the androgen-dependent up-regulation of FSTL1 promotes growth of cancer cells [26]. In colorectal cancer (CRC), FSTL1 is selectively expressed in cancer stroma and attenuates CRC cell proliferation [27]. Although FSTL1 is overexpressed in plasma and cancerous tissues of CRC patients, it has not been, thus far, implicated in prognosis [28]. In ovarian and endometrial cancers, FSTL1 functions as a tumor suppressor via inducing apoptosis [29]. However, the role of FSTL1 in RCC remains elusive.
We hypothesized that FISTL1 might play a role as tumor suppressor in ccRCC. In the present study, we aimed to clarify the effects of aberrant FSTL1 expression on the growth and aggressiveness of RCC cells, identify the signaling pathways that were affected by FSTL1, and validate the prognostic functions of FSTL1 with a cohort of RCC patients.

Growth, migration, and invasion assays
Anchorage-independent growth of RCC cells with aberrant FSTL1 expression was evaluated with a double-layered soft agarose culture system, as previously described [30]. Cell migration assay (without matrigel) and cell invasion assay (with matrigel) were performed using 8-μm pore size 24-well cell culture transwell plates (Corning, Corning, NY, USA). These experiments were performed in triplicate.

Cytometry
Cell cycle and cell surface markers of NRCC-shScramble and NRCC-shFSTL1 cells were examined using a flow cytometer (MACSQuant, Miltenyi Biotec, Bergisch Gladbach, Germany). The estimation of cell cycle was performed with propidium iodide (PI) staining as previously described [29]. To compare proportions of cells in different cell cycle phases, NRCC-shFSTL1 and NRCC-shScrambled cells were passaged synchronously. Cell markers were detected using anti-CD44-PE (1: Cell debris and fixation artifacts were excluded by appropriate gating. The acquisition process was stopped when 10,000 events for cell cycle analysis and 30,000 events for cell surface marker analysis were collected in the population gate. The FlowJo version 7.6 software (Flowjo, Ashland, NC, USA) was used for data acquisition and analysis. Assays for other details were as previously described [30]. Each flow cytometry assay was conducted in triplicate.

cDNA microarray analysis
Total RNAs from NRCC-shScramble, NRCC-shF-STL1-1, and NRCC-shFSTL1-2 cells were extracted using Trizol Reagent (Life technologies, Carlsbad, CA, USA) and then reversely transcribed, biotin-labeled, fragmented, and hybridized to Human Genome U133 plus 2.0 chips (Affymetrix, Santa Clara, CA, USA) according to the manufacturer's instructions. All hybridized microarrays were scanned by Affymetrix GeneChip ® Scanner 3000 (Affymetrix). Raw intensities were extracted by Command Console Software 3.1 (Affymetrix). Preprocessing was then performed by MAS5 integrated in Expression Console (Affymetrix). Probe sets with low intensities and more than two absent calls were filtered. Probe set-level data was transformed into log2 scale and summarized to gene-level data by averaging the intensities of probe sets annotating to the same gene. Genes containing probe sets whose fold changes were more than two and without absent calls were considered differentially expressed genes. Gene set enrichment analysis (GSEA) was applied to explore the enriched gene sets [32]. The gene expression differences between NRCC-shFSTL1 and NRCC-shScramble cells were computed as the statistic to rank the genes in the data. Normalized enrichment score (NES) were computed for every gene set in chemical and genetic perturbation part of Molecular Signatures Database v4.0 (MsigDB v4.0, http://software. broadinstitute.org/gsea). Gene sets were randomized 1000 times to obtain false discovery rate (FDR). Gene sets with FDR of < 0.25 were considered statistically significant. Microarray data were deposited in GEO with Accession No. GSE76948. All microarray data were analyzed using R programming (https://www.cran.r-project.org).

qRT-PCR and Western blotting
Total RNAs were isolated and reversely transcribed into cDNA, and subjected for qRT-PCR as previously described [9]. The primers for amplification of FSTL1, IL-6, E-cadherin, N-cadherin, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as well as their amplified conditions and locations are summarized in Table 1. Amplification efficiencies of all primers are present in Table 1, estimated based on standard curve assay as previously described [33]. Gene expression was relatively quantified using GAPDH as an internal control. qPT-PCR assays were controlled using no template controls (NTC) and "no-reverse transcription" (no-RT) controls. Quantification cycle (C q ) was no less than 35 and no peak of melt curve occurred in both controls. All qPCR reactions, including reactions of NTCs and "no-RT" controls, were run in triplicate and the average C q of each sample was calculated based on all three biological replicates. Each qRT-PCR assay was conducted in triplicate. Protein was extracted, quantified, and subjected to Western blotting according to standard protocols as previously described [34]. The primary antibodies used in this study were as follows: anti-FSTL1 (C-term) (1:1000 dilution; Abgent No. AP10534b, San Diego, CA, USA), anti-IκBα (1:1000 dilution; Cell Signaling Technology No. 9242), and anti-β-actin (1:1000 dilution; Cell Signaling Technology No. 3700). β-actin was used as the loading control for cellular proteins. The protein bands were detected using GBOX Chemi XL1.4 machine (Gene, Cambridge, UK) and GeneSys version 1.2.2.0 software. The densities of protein bands were quantified using Genetools software (version 4.02, Cambridge, UK).

Study patients and follow-up
Formalin-fixed paraffin-embedded specimens of surgically removed tissues were collected from Changhai Hospital Affiliated to Second Military Medical University (Shanghai, China) between December 1998 and November 2011. All specimens were pathologically confirmed as ccRCC at the enrollment. We excluded ccRCC patients who refused to be enrolled in the study and those who did not adhere to the follow-up examination. All patients enrolled in this study were re-examined at our hospital within 1 month after surgery. The follow-up examination was performed every 6 months at our outpatient clinics or through phone. The last date of follow-up was May 5, 2015. Death from ccRCC relapse was defined as an event. Patients alive at the last follow-up and died of other causes were censored. Disease-specific survival (DSS) was defined as the duration from the date of receiving surgery to the date when patient died of ccRCC or when patient received the last follow-up. This study was approved by the institutional review board of Second Military Medical University. All experiments were performed in accordance with relevant guidelines. Informed consent was obtained from all subjects.

Statistical analysis
The comparative threshold cycle (Ct) method was employed to quantify the relative change in expression of target genes. Student t test was performed for two-group comparisons, and one-way ANOVA was performed for three-group comparisons. Paired Wilcoxon test was applied to evaluate the degree of positive immunostaining for FSTL1. Non-parametric analysis of

The growth and aggressiveness of ccRCC cells with altered expression of FSTL1
qRT-PCR was applied to determine the gene expression level of FSTL1 in ccRCC cell lines (ACHN, NRCC, MRCC, and 786-O) and HEK 293T cells (Fig. 1). It was shown that FSTL1 mRNA level was significantly lower in ACHN, NRCC, and MRCC cells than in HEK 293T cells (P values were 0.022, 0.004, and 0.007, respectively); however, FSTL1 mRNA level was significantly higher in 786-O cells than in HEK 293T cells (P = 0.002).
Retrovirus-mediated shRNA constructs targeting different regions of FSTL1 mRNA was applied to generate two stable FSTL1-knockdown cell lines termed NRCC-shFSTL1-1 and NRCC-shFSTL1-2 that were derived from NRCC cells. FSTL1 mRNA level was down-regulated by approximately 80% in NRCC-shFSTL1-1 and NRCC-shFSTL1-2 cells compared with that in shScramble control NRCC cells (Fig. 2a). Altered expression of FSTL1 was confirmed on protein levels by Western blotting (Fig. 2a). We selected NRCC-shFSTL1-2 cells in subsequent cell activity and cell cycle percentage assays. FSTL1 knockdown significantly promoted anchorageindependent growth (Fig. 2b), migration (Fig. 2c), and invasion (Fig. 2d). Transient transfection of shFSTL1 significantly reduced FSTL1 transcription in 786-O and ACHN cells, resulting in significantly increased anchorage-independent growth, migration, and invasion (Fig. 2). Transient transfection of FSTL1 cDNA significantly increased FSTL1 expression in ACHN cells and only attenuated cell migration ability (Fig. 2). MRCC was resistant to the stable or transient transfection of shFSTL1.
In cell cycle assay, NRCC-shFSTL1 and NRCC-shScramble cells were cultured and passaged synchronously. Flow cytometry was applied to determine cell cycle and cellular markers. It was found that the percentage of cells gated in the S and G 2 /M phases were higher in NRCC-shFSTL1 cells than in NRCC-shScramble cells (P = 0.048, P = 0.011; Fig. 3a); the percentage of CD99positive cells was increased (P = 0.026; Fig. 3b), whereas the percentage of CD24-positive cells was decreased in NRCC-shFSTL1 cells (P = 0.013; Fig. 3c). NRCC cells were nearly 100% positive for CD44 but almost all negative for CD133, CD105, EpCAM, and vimentin (data not shown), as measured by flow cytometry.

FSTL1 knockdown up-regulated the NF-κB and HIF signaling pathways
To explore the molecular mechanism by which FSTL1 affects the progression of ccRCC, we examined the gene expression profiles of the two NRCC-shFSTL1 cell lines with NRCC-shScramble cells using cDNA microarray assays (GEO Accession No. GSE76948). After filtered with absent/present calls, 105 differentially expressed genes with a fold change of > 2 were identified in FSTL1 knockdown NRCC cells, including 57 up-regulated genes and 48 down-regulated genes. The spectrum and absolute gene expression levels of differentially expressed genes identified were consistent between NRCC-shFSTL1-1 and NRCC-shFSTL1-2 cells (Fig. 4a, b). We applied GSEA software package to enrich the gene sets from the global gene expression in response to FSTL1 knockdown in NRCC cells. Taking FDR of < 25% as threshold, 12 functional gene sets were enriched by the down-regulated genes and 23 were enriched by the up-regulated genes. According to NES, the top 11 gene sets enriched by the up-regulated genes were applied to construct a network (Fig. 4c). In this network, NF-κB-and HIF-related functional gene sets formed two obvious subnetworks. The representative gene sets with the highest NES score in the NF-κB-and the HIF-related signaling subnetworks were named "Hinata NF-κB targets keratinocyte up" [35] and "Elvidge HIF1α and HIF2α targets up" [36], respectively (Fig. 4d, e).

FSTL1 knockdown promoted IL-6 expression, epithelial-to-mesenchymal transition (EMT), and TNFα-induced degradation of NF-κB inhibitor (IκBα) in ccRCC cells
Our qRT-PCR assays indicated that IL-6 transcription was significantly up-regulated following FSTL1 knockdown in ACHN, 786-O, and NRCC cells (P < 0.05) (Fig. 5a). The transcription of E-cadherin was significantly down-regulated whereas N-cadherin was significantly up-regulated following FSTL1 knockdown in NRCC cells; transient transfection of shFSTL1 also down-regulated E-cadherin and up-regulated N-cadherin in ACHN cells. Furthermore, transient transfection of FSTL1 cDNA significantly up-regulated E-cadherin and down-regulated N-cadherin in ACHN cells (Fig. 5b, c). We used Western blotting to analyze the expression of IκBα protein in response to TNFα (10 ng/mL) treatment for 0, 15, 30, 60, and 90 min. The relative IκBα levels were calculated based on its baseline expression at the 0 min and adjusted by expression levels of β-actin. It was found that FSTL1 knockdown repressed the recovery of IκBα after TNFα treatment (Fig. 5d).

Expression patterns of FSTL1, HIF-1α, and HIF-2α in human ccRCC tissues and adjacent renal tissues as well as their roles in predicting postoperative prognosis
A total of 89 ccRCC patients (men, 59; women, 30) were involved in the final analysis. The median age of the 89 ccRCC patients was 59 years (range 25-85 years). Sixtythree (70.8%) patients had stage I-II ccRCC. The median follow-up time was 82.3 months (interquartile range 58.8-102.2 months). All patients involved in this study had tumor tissue samples and 67 had the paired adjacent renal tissues. FSTL1 expression was examined by using IHC in all 89 tumor tissues, and 67 adjacent renal tissues. However, HIF-1α and HIF-2α expression was examined in only 85 and 78 tumor tissues, respectively, because some tissue specimens were insufficient for IHC assays of all target proteins. FSTL1 was positive in 65.2% (58/89) of ccRCC tissues and 94.0% (63/67) of the adjacent non-cancer tissues. In the adjacent tissues, FSTL1 was selectively positive in epithelial cytoplasm of the loop of Henle; HIF-1α was negative whereas HIF-2α was positive in the loop of Henle (Fig. 6a-c). In ccRCC tissues, FSTL1 was negative or detected in the cytoplasm of cancer cells; HIF-1α and HIF-2α were positive in 52.9% (45/85) and 66.7% (52/78) of ccRCC cases, respectively ( Fig. 6d-f ). The positive rate of FSTL1 protein was significantly lower in ccRCC tissues than in the adjacent normal tissues among 67 patients with paired ccRCC specimens (P < 0.001, Table 2). In ccRCC tissues, FSTL1 expression was positively correlated with HIF-1α (Spearmen r = 0.216, P = 0.047) but negatively correlated with HIF-2α expression (Spearmen r = − 0.229, P = 0.044) ( Table 3).
A total of 89 ccRCC patients who had IHC scores of FSTL1 expression in tumor tissues were successfully followed up after surgery. Up to the date of last visit, 17 patients died of ccRCC relapse. Kaplan-Meier analysis indicated that intratumoral FSTL1 expression predicted a favorable postoperative DSS (Fig. 6g). The Kaplan-Meier survival curves for FSTL1-positive and FSTL1-negative patients were not cross, suggesting that the proportional hazards assumption was reasonable in this case. Compared with the patients with AJCC stages III-IV ccRCC, those with stages I-II diseases had a good prognosis (log-rank test, P = 0.004). No significant correlation was observed between intratumoral FSTL1 expression and AJCC stage of ccRCC patients (Table 3). Multivariate Cox regression analysis including age, gender, and AJCC stage showed that intratumoral FSTL1 expression conferred a favorable postoperative prognosis independently (Table 4). Moreover, Kaplan-Meier analysis showed that  Intersection genes (green dots) with fold change > 2 were plotted to connect the gene sets. The gene sets fell into nuclear factor-кB (NF-κB)-and hypoxia-inducible factor (HIF)-related signaling subnetworks, respectively, which is connected by a histone deacetylase 1 (HDAC1)-related gene sets. d Enrichment plot of the representative gene set with the highest NES score in the NF-κB-related signaling subnetwork. e Enrichment plot of the representative gene set with the highest NES score in the HIF-related signaling subnetwork. The vertical dashed line in each plot denoted the point at which NES reached its maximum high expression of HIF-2α, rather than HIF-1α, in tumor tissues predicted an unfavorable postoperative prognosis (Fig. 6h, i).

Discussion
In the present study, we presented a series of data to identify that FSTL1 functioned as a novel tumor suppressor in ccRCC. However, related evidence was mostly obtained from FSTL1 knockdown assays. FSTL1 knockdown facilitated ccRCC cell growth, migration, invasion, and promoted cell cycle, possibly because a low level of FSTL1 was unable to repress some cancer-promoting force(s); overexpression of FSTL1 did not inhibit the growth and invasion, possibly because FSTL1 expression at the normal level was enough to repress the cancerpromoting force(s), thus it should be important to restore the FSTL1 expression level to the normal range for attenuating the growth advantage of ccRCC. We found that MRCC and ACHN, the two cell lines derived from metastatic ccRCC tissues, were resistant to the infection with recombinant retrovirus; furthermore, MRCC was also resistant to the transient transfection, which is possibly due to the "stem-like" property of metastatic ccRCC cells [30]. Thus, we gave up our intention to up-regulate the expression of FSTL1 in MRCC. Our cytometry results indicated that FSTL1 knockdown up-regulated CD99, a cellular marker related to ccRCC aggressiveness [37], and down-regulated CD24 expression in NRCC cells. We found that CD44 was nearly 100% positive in NRCC. High CD44 expression in tumors is a poor prognostic marker of RCC [12]. CD44 high CD24 low signature usually determines the cancer stem cell (CSC) and EMT phenotype in oral cancer [38]. Additionally, CD44 high CD24 low signature is mostly believed to be a CSC-like phenotype in cancers such as prostate cancer [39] and breast cancer [40]. Thus, it is possible that FSTL1 knockdown increases CSC-like properties of ccRCC cells. We also found that FSTL1 knockdown promoted EMT process via up-regulating N-cadherin and down-regulating E-cadherin in ccRCC cells. Thus, FSTL1 might be a novel tumor suppressor that attenuates EMT process in ccRCC cells.
Using global gene expression profiling, we found NF-κB-and HIF-related functional gene sets were the predominant signaling pathways affected by FSTL1 knockdown. Interestingly, a histone deacetylase 1 (HDAC1)-related pathway bridges the NF-κB-related signaling subnetwork and HIF-related signaling subnetwork. This is supported by previous findings [41]. HDAC1 may link the NF-κB and HIF signaling pathways. The treatment with HDAC1 inhibitors might be an option of advanced RCC [42]. The two pathways work synergistically in tumor cells that survive under both inflammatory and hypoxia microenvironments. Thus, we chose to validate the activity of the NF-κB signaling pathway in ccRCC cells and the expression of HIF1/2α in clinical samples. FSTL1 knockdown up-regulated the expression of IL-6 in ccRCC cells. Elevated level of IL-6 expression is associated with cancer cell proliferation, angiogenesis, and metastasis via fueling signal transducer and activator of transcription 3 (STAT3), mitogen-activated protein kinase (MAPK), and Akt signaling, thus promoting EMT and subsequent cancer metastasis [43]. FSTL1 knockdown promotes the NF-κB signaling pathway in ccRCC cells (Fig. 5d), whereas activation of NF-κB promotes the development of RCC [44]. FSTL1 knockdown also affect the HIF signaling pathway. HIF-1α and HIF-2α have opposing effects in ccRCC biology, with HIF-1α acting as a tumor suppressor and HIF-2α acting as an oncogene [45]. The somatic mutation of von Hippel-Lindau (VHL) gene, a tumor suppressor gene, can lead to increased expression of HIF-2α in both sporadic and familial ccRCCs [46]. HIF-2α activation is involved in the generation of RCC-derived, CXCR4-positive CSCs [47]. Therefore, we believe that FSTL1 knockdown may derepress NF-κB and HIF-2α signaling in ccRCC cells, thus promoting cancer invasion and metastasis. Interestingly, both FSTL1 and HIF-2α were locally expressed at the same locations but FSTL1 expression was negatively associated with HIF-2α expression in ccRCC tissues, implying that FSTL1 in situ inhibit HIF-2α expression in renal tissues. The mechanisms by which FSTL1 in situ inhibit HIF-2α expression remains to be elucidated.
Our IHC analyses indicated that FSTL1 was locally expressed in epithelial cytoplasm in the loop of Henle, which is consistent with the results of a previous study [32]. However, it is mainly expressed in epithelial cells, rather than in mesenchymal cells as previously believed [22], indicating that the role of FSTL1 in renal tissues is  quite special. FSTL1 expression is significantly lower in ccRCC than in adjacent renal tissues, but the intratumoral FSTL1 expression is not associated with tumor stage. Importantly, intratumoral FSTL1 expression confers a favorable independent postoperative prognosis in ccRCC; whereas high expression of HIF-2α, rather than HIF-1α, in ccRCC tissues predicts a poor postoperative prognosis. Thus, intratumoral FSTL1 and HIF-2α expression are both prognostic factors for ccRCC; however, they have inverse effects. Our study has limitations. First, the cancer inhibition effect of FSTL1 was investigated only in vitro. The evidence generated from animal experiments are needed to further clarify the function of FSTL1 on the growth and metastatic potential of ccRCC cells. Second, only 89 patients were enrolled in this study. The value of FSTL1 for prognosis prediction remains to be re-confirmed in large cohorts.

Conclusions
This present study clarifies the important nature of FSTL1 in ccRCC. FSTL1 functions as a tumor suppressor possibly via repressing NF-κB and HIF-2α signaling pathways. Intratumoral FSTL1 expression in ccRCC tissues conferred a favorable independent postoperative prognosis to ccRCC patients. Treatment approaches that increasing FSTL1 expression in tumors might lead to effective therapy for advanced ccRCC. Authors' contributions YL, XT, XC, and WL contributed to acquisition of whole data independently and presented the same results. YY, YL, XT, and XH were responsible for pathological and immunohistochemical assays. JH, DS, and LW were involved in the diagnosis and the recruitment of the patients in our affiliated hospitals and follow-up study. HZ, JY, YD, and WL contributed to statistical analysis and data interpretation. TT contributed to revising it critically for important intellectual content. GC and XT designed and organized the study. GC wrote the manuscript. All authors read and approved the final manuscript.