Follistatin impacts Tumor Angiogenesis and Outcome in Thymic Epithelial Tumors

Tumor angiogenesis is a key factor in the progression of thymic epithelial tumors (TETs). Activin A, a member of the TGFβ family, and its antagonist Follistatin are involved in several human malignancies and angiogenesis. We investigated Activin A and Follistatin in serum and tumor tissue of patients with TETs in relation to microvessel density (MVD), WHO histology classification, tumor stage and outcome. Membranous Activin A expression was detected in all tumor tissues of TETs, while Follistatin staining was found in tumor nuclei and cytoplasm. Patients with TETs presented with significantly higher Activin A and Follistatin serum concentrations compared to healthy volunteers, respectively. Follistatin serum concentrations correlated significantly with tumor stage and decreased to physiologic values after complete tumor resection. Follistatin serum concentrations correlated further with MVD and were associated with significantly worse freedom from recurrence (FFR). Low numbers of immature tumor vessels represented even an independent worse prognostic factor for FFR at multivariable analysis. To conclude, the Activin A - Follistatin axis is involved in the pathogenesis of TETs. Further study of Follistatin and Activin A in TETs is warranted as the molecules may serve as targets to inhibit tumor angiogenesis and tumor progression.


Serum follistatin and Masaoka -Koga tumor stage. Preoperative Follistatin but not Activin A
serum concentration was significantly different according to Masaoka -Koga tumor stage (p = 0.037 and p = 0.988; respectively; Fig. 1C,D). In particular, Follistatin serum concentrations were 806.5 ± 791.2 pg/mL and 713.4 ± 589.1 pg/mL in stage III and IV TETs, which was significantly different compared to 334.3 ± 286.1 pg/mL and 285.9 ± 246.0 pg/mL in stage I and II TETs (p = 0.005) ( Table 1).

impact of myasthenia gravis on Activin A and follistatin serum concentrations. Activin A and
Follistatin serum concentrations did not significantly differ in patients with TAMG compared to TETs without MG (p = 0.751 and p = 0.211; respectively; Table 1). In addition, we analyzed serum samples of MG patients without thymic malignancy compared to sex-(p = 0.834) and age-matched (p = 0.079) healthy volunteers. Myasthenic patients showed also significantly elevated Activin A (p = 0.009; Fig. 1G) and Follistatin (p = 0.002; Fig. 1H) serum concentrations compared to controls (Table 1). Furthermore, we differentiated myasthenic patients based on their thymic pathology into patients with TETs, those with true thymic hyperplasia and those with follicular thymic hyperplasia. However, benign thymic pathology did not significantly impact Activin A and Follistatin serum concentration in patients with MG (p = 0.804 and p = 0.722; respectively).

follistatin und Activin A expression in thymic epithelial tumors. We performed Follistatin and
Activin A staining in 95 specimens of TETs (4 micronodular thymomas (MNT), 14 A, 14 AB, 10 B1, 21 B2, 17 B3 thymomas, 15 TCs; Fig. 2). Membranous Activin A expression was detected in all TETs without significant differences between thymomas and TCs or according to WHO subtype. Conversely, Follistatin staining was found in tumor nuclei and cytoplasm (p = 1.000). There was a trend towards absent nuclear Follistatin expression in TCs compared to thymomas. In particular, 75% of all TCs displayed absent Follistatin in nuclei, whereas absent nuclear Follistatin expression was found in 19% of all thymomas (p = 0.053). There were no semiquantitative differences in Follistatin staining intensity between thymomas and TCs.
Follistatin serum levels correspond with microvessel density. We performed Spearman correlation analyses to evaluate whether serum Follistatin impacts MVD and vascular architecture in TETs ( Fig. 3A-C). Follistatin serum levels showed a significant negative correlation with the number of immature (r: −0.443; p = 0.014) tumor vessels (Fig. 3D). In addition, there was also a significantly positive correlation between Follistatin and mature tumor vessels (r: 0.549; p = 0.002 Fig. 3E). As expected, immature tumor vessels correlated significantly with mature vessels (r: −0.440; p < 0.001) and total number of vessels (r: 0.932; p < 0.001), respectively.
Furthermore, we explored the possibility that other systemic inflammatory markers, such as Fibrinogen, CRP, Heat Shock Proteins or molecules of the RAGE axis, may also correlate with specific vascular architecture of TETs. However, beside serum Follistatin none of the abovementioned proteins showed significant correlation with mature, immature or total tumor vessels ( Table 2). Nonetheless we found a significant correlation between Follistatin and CRP (r: 0.576; p = 0.006) and Follistatin with Fibrinogen (r: 0.512; p = 0.018). Masaoka-Koga tumor stage and microvessel density. MVD and vascular architecture showed significant differences according to Masaoka -Koga tumor stage. The highest number of mature tumor vessels (38.7 ± 9.7) was detected in stage IV TETs (p < 0.001; Fig. 3G), while the number of immature vessels decreased gradually from stage I TETs (109.6 ± 46.4) to stage IV TETs (28.9 ± 3.9; p < 0.001; Fig. 3F). Similar, highest mature to total vessel ratio of 0.53 ± 0.05 and lowest immature to total vessel ratio of 0.47 ± 0.05 was found in metastasized stage IV TETs (Table 3).
impact of microvessel density and serum parameters on outcome. Next www.nature.com/scientificreports www.nature.com/scientificreports/ used for dichotomizing patients into lower and higher subgroups, respectively. Low numbers of immature and total vessels were associated with significantly worse FFR (p = 0.001; p = 0.006) and CSS (p = 0.014; p = 0.017), respectively. High preoperative Follistatin serum concentrations were associated with significantly worse FFR (p < 0.001; Fig. 3I), but not CSS (p = 1.000), while Activin A serum levels did not impact outcome (Table 4).

Univariable and multivariable cox-Regression analyses. Finally we performed univariable and mul-
tivariable Cox-regression analyses to assess the prognostic power of following clinical characteristics on FFR and CSS: histology (TC vs. thymoma), Masaoka-Koga tumor stage (III-IV vs. I-II), MG (neg. vs. pos.), sex (female vs. male), number of immature (low vs. high) and mature (high vs. low) tumor vessels and total number of vessels (low. vs. high).
At univariable analysis, presence of TC (HR 6.10; p = 0.002), tumor stage III-IV (HR 6.10; p = 0.003), low number of immature (HR 14.8; p = 0.010) and low number of total tumor vessels (HR 6.48; p = 0.016) represented significantly worse prognostic factors for FFR. Among these variables, only low number of immature tumor vessels stayed an independent worse prognostic factor at multivariable analysis. In particular, low number of immature tumor vessels showed a 35-fold higher risk for tumor recurrence compared to TETs with high number of immature tumor vessels (HR 35.3; p = 0.021). With respect to CSS, only presence of TC (HR 15.2; p = 0.002) was a significant worse prognostic factor at univariable testing (Table 5).  Activin A serum levels in patients with malignant pleura mesothelioma 22 , lung adenocarcinoma 23 , breast cancer and prostate cancer 24 , hepatocellular carcinoma 25 , and endometrial and cervical carcinoma 26 . In our study, preoperative Activin A serum levels did not significantly correlate with Masaoka -Koga tumor stage, WHO classification or outcome, which was in contrast to literature, where increased Activin A serum levels were commonly associated with higher tumor stage and poor outcome.
Conversely to Activin A, elevated Follistatin serum concentrations were described only in few malignancies, including ovarian cancer 27 , lung adenocarcinoma 28 , hepatocellular carcinoma 29 , and prostate cancer 30 . In particular, high Follistatin serum levels were associated with shorter overall survival (OS) and poor prognosis in hepatocellular carcinoma 29 , emergence of bone metastases in prostate cancer 30 or tumor proliferation of lung adenocarcinoma cells 28 . This is in agreement with our data, showing that highest Follistatin serum concentrations were found in patients with advanced stage TETs and TCs, and that high preoperative serum levels were associated with significantly worse FFR.
It is commonly assumed that the cancerogenic function of Follistatin depends primarily on its role as antagonist of Activin A 21 . Follistatin could prevent Activin A induced growth inhibition and apoptosis and therefore Follistatin fosters cell proliferation and tumor growth. Stove C. et al. 31 found that in-vitro melanoma cells express activin receptors and that treatment with Activin A led to growth inhibition and apoptosis, which could be counteracted by addition of Follistatin 31 . In accordance to that, Chen F. et al. 28 demonstrated that cultured lung adenocarcinoma cells secreted Follistatin, and that inhibition of Follistatin led to significantly augmented Activin A induced apoptosis 28 .
In addition, Follistatin also has effects on tumor-associated angiogenesis, which mostly depends on the inhibition of Activin A. Endothelial cells constitutively express activin receptors and VEGF, and binding of Activin A leads to inhibition of proliferation of cultured endothelial cells by downregulation of p21 and downregulation of VEGF expression 19,32 . In accordance to that, inhibition of p21 increased endothelial cell proliferation and resistance to Activin A mediated growth inhibition 33 . Moreover, Follistatin induces also angiogenesis by binding and activating Angiogenin, a potent activator of angiogenesis 34 .
Within our study, we could further link for the first time the pro-angiogenic function of serum Follistatin to MVD in TETs. Follistatin serum levels showed a significant negative correlation with absolute number of immature tumor vessels, and a significant positive correlation with mature tumor vessels. Herein, we defined maturity of tumor vessels as presence of vessel lumen that depends on the proliferation of endothelial cells. Therefore, the amount and characteristic pattern of tumor vessels in TETs most likely depend on proliferation of endothelial cells, which might be influenced by Follistatin.
Tumor angiogenesis represents one of the hallmarks of cancer and is essential for tumorigenesis, tumor growth and metastasis 35 38 could show that vascular architecture and expression profile of key angiogenic growth factors differ significantly in TETs. In particular, densest vascular networks of predominantly small vessels were found in A thymomas, while TCs showed the least dense vascular network with almost large capillaries and highest VEGF expression 38 .
Similar to that, patients with TCs showed the lowest MVD, but the highest amount of mature tumor vessels in our cohort. Furthermore, we could demonstrate that the number of immature vessels gradually decreased with level of invasiveness, while the amount of mature tumor vessels gradually increased with tumor stage. Indeed low numbers of immature tumor vessels represent a significantly worse prognostic factor for FFR with a 35-fold www.nature.com/scientificreports www.nature.com/scientificreports/ increased risk for recurrent disease. We managed to link the pro-angiogenic function of Follistatin with MVD and further to demonstrate that tumor vessel subtypes differ significantly according to WHO classification. Hence, manipulation of the Activin A and Follistatin system in TETs may represent a potential target to interfere with tumor angiogenesis, tumor growth and metastases in TETs.  www.nature.com/scientificreports www.nature.com/scientificreports/ According to data from the Unigene database, Follistatin is not expressed in normal (physiologic) thymus 21 . Herein, we could demonstrate that Follistatin and Activin A were generally expressed in TETs. We found constant membranous Activin A staining in all TETs, which was most likey caused by the usage of an anti-human Anti-Activin A receptor type IB antibody. Conversely, Follistatin as a soluble antagonist of Activin A, was detected in cytoplasm and nucleus of tumor cells. Soluble Follistatin molecules envelop Activin A and thus block the access of Activin A to type I and type II receptor binding sites 39 . Furthermore, Follistatin and Activin A binding complexes can interact with cell-surface proteoglycans followed by internalization and lysosomal degradation 40 , which could be an explanation of Follistatin expression in the cytoplasm of TETs. With respect to the nuclear function of Follistatin, Lin C. et al. 41 (2016) could demonstrate that nuclear Follistatin prevented cultured human lung epithelial cells and mouse lunge tissue from reactive oxygen induced apoptosis and that down-regulation of Follistatin promoted apoptosis 41 . Similar, in HeLa cells, nuclear expression of Follistatin delayed glucose deprivation-induced apoptosis by attenuating RNA synthesis, which represents a key process of cellular energy homeostasis and cell survival 42 . Altogether we assume that nuclear expression of Follistatin in malignant thymic epithelial cells might cause cell-survival and inhibition of apoptosis, while the cytoplasmic Follistatin expression might result from degradation of Follistatin -Activin A complexes.
Follistatin serum concentrations decreased significantly after radical tumor resection, indicating that Follistatin serum levels are at least partially related to the presence of malignant thymic epithelial cells. Whether Follistatin is secreted directly by tumor cells, as demonstrated for human lung adenocarcinoma cells 28 , or indirectly by immune cells in response to tumor cells or as acute phase reaction 43 , remains elusive. Therefore, although our data are interesting, further studies are necessary to clarify the molecular and cellular effects of Activin A and Follistatin in TETs.
Our study has some limitations due to the mostly retrospective study design, the single center experience and the low number of patients, which limit our ability to draw conclusions. Moreover, although our data are interesting, we could not provide experimental data of Follistatin and Activin A in TET cell lines to further confirm our results. However, the strength of the study is that we could link the pro-angiogenic function of Follistatin to MVD in TETs, which corresponds to Masaoka-Koga tumor stage and outcome.
Taken together, we could demonstrate that preoperative Activin A and Follistatin serum concentrations were significantly elevated in patients with TETs and MG compared to controls. Follistatin serum levels were highest in patients with TCs and advanced tumor stage. Most importantly, Follistatin correlated significantly to MVD, which in turn was significantly different according to WHO subtype and Masaoka -Koga tumor stage. Number of immature tumor vessels represented a significant prognostic factor for FFR. Hence, manipulation of the Activin A and Follistatin system in TETs may represent a potential target to interfere with tumor angiogenesis and tumor progression in TETs.   months), respectively. Within oncologic follow-up, periodic chest CT-scans were performed postoperatively every 3 to 6 months for the first three years followed by annual CT-scans. Thirteen (13.5%) patients experienced recurrent disease (local: n = 2; regional: n = 5; distant: n = 6) and six patients died from TETs (6.3%). According to the recommendations of the International Thymic Malignancy Interest Group (ITMIG), we used CSS and FFR as main oncologic outcome parameters within this study 12 . CSS was calculated in all patients and was defined as time from date of surgery to date of death from a TET, while unrelated deaths or unknown causes of death were censored. In contrast, FFR was only calculated in patients after radical tumor resection (R0) and was defined as time from date of surgery to date of recurrence.
therapy. Among those 95 TETs who underwent surgical tumor resection, completeness of surgical resection could be achieved in 83 cases (87.4%), while R1 and R2 status were obtained in 9 (9.4%) and 3 (3.2%) of TETs, respectively. Forty-six patients (48.4%) underwent only surgical tumor resection, while the remaining 49 patients (51.6%) received multimodal therapy. In particular, 11 patients (11.6%) underwent neoadjuvant chemotherapy (ChT; n = 5) or concomitant radiochemotherapy (RChT; n = 6), while 42 patients (44.2%) received adjuvant therapy, including adjuvant ChT (n = 3), RT (n = 31) and RChT (n = 8), respectively. preparation of specimens and immunohistochemistry. In our clinical institute of pathology, tumor sections are processed from the capsule, every 1-2 cm of tumor tissue and from tumor regions, which appear different on gross inspection. However, our immunohistochemical analyses were done on selected slides from one block per patient that represented the main tumor component with a very homogenous histology.
Formaldehyde-fixed and paraffin embedded human specimens of all patients (n = 95) with TETs were avail- Microvessel density. As already published and described by Weidner et al. 1992, we assessed microvessel density (MVD) using the "hotspot" method 47 . Briefly, slides were screened at low magnification to identify the areas with the greatest number of CD34 stained vessels ("hotspot"). MVD was determined by counting all CD34 positive vessels at 200x magnification (corresponding to 0.95 mm²) in 95 TETs. One specimen was removed due to lack of evaluable remaining thymic tissue. Three representative hotspots were calculated for each specimen. Tumor vessels were further differentiated into immature and mature vessels according to the classification of Gee MG et al. based on the presence of perfused lumen and perivascular cells 48 . Accordingly, immature vessels are differentiated from intermediate and mature vessels by the presence or absence of vessel lumen, whereas intermediate and mature vessels are characterized by the presence or absence of perivascular cells. Herein, we used a simplified method and differentiated only between immature and mature vessels. Within the Results section, total number of immature, mature and all tumor vessels are indicated as well as following ratios: Ratio 1 (immature/ mature), Ratio 2 (immature/total) and Ratio 3 (mature/total).

Enzyme-linked immunosorbent assays (ELISA). Preoperative serum samples were available of 46
patients with TETs (17 TAMG, 29 non-TAMG) and 30 patients with MG for analysis of Activin A and Follistatin blood levels. Forty-nine sex and age-matched healthy volunteers were used as controls. Pre -and postoperative (2019) 9:17359 | https://doi.org/10.1038/s41598-019-53671-8 www.nature.com/scientificreports www.nature.com/scientificreports/ serum samples were available in a subset of 30 patients with TETs. Postoperative serum samples were collected in patients who underwent primary surgery 6 to 12 months ago, who did not receive adjuvant therapy within the last month, and who did not have signs of recurrence or a 2 nd malignancy.
To assess Activin A and Follistatin serum concentrations, we used the commercial available human Activin A ELISA kit (R&D Systems, Minneapolis, MN, USA, DuoSet ® human Activin A, DY338) and human Follistatin ELISA kit (R&D Systems, Minneapolis, MN, USA, DuoSet ® human Follistatin, DY669). All tests were performed according to the Manufacture's protocols. Samples were measured in duplicates and researches performing the assays were blinded to the groups associated with each sample. Additionally to Activin A and Follistatin serum concentrations, Fibrinogen, CRP, heat shock protein 27 and 70, and molecules of the RAGE axis (sRAGE, esRAGE, HMGB1) as previously shown, were correlated with MVD 3,14,46,49 . Statistical methods. Statistical analysis of data was performed using SPSS software (version 21; IBM SPSS Inc., IL, USA). The type of test used is indicated in the table and/or the result section. All data are reported as mean ± standard deviation (SD) within result section. Chi-square test was used to investigate the association between nominal variables. Unpaired Student's t-test and one-way ANOVA were used to compare means of two or more than two independent groups with normal (Gaussian) distributions, respectively. Post-hoc Tukey's -B and Bonferroni correction were used in case of multiple testing. Mann-Whitney-U test and Krusky-Wallis test were performed to analyze non-normal distributed variables with two or more than two groups, respectively. Paired Student's t-test was performed for analyzing means of two dependent groups. Pearson correlation (r) was done to analyze linear relationships between two numerical measurements, while Spearman correlation analysis was applied for analysis of ranked variables. Kaplan-Meier analyses and Log-rank test were assessed for survival analysis. Uni -and multivariable Cox-regression analysis was used to evaluate the prognostic impact of different clinical variables on CSS and FFR. The receiver operating characteristic (ROC) curve and Youden-Index were calculated to find optimal cut-off values of 483.3 pg/mL for Activin A and 776.6 pg/mL for Follistatin for predicting tumor recurrence. ethics statement. Ethical approval was obtained from the Ethics Committee of the Medical University of Vienna (302/2011). All participating patients and control subjects gave their written informed consent, and all experiments were performed in accordance with the approved ethical guidelines.