The cytoplasmic expression of FSTL3 correlates with colorectal cancer progression, metastasis status and prognosis

Abstract Follistatin‐like (FSTL) family members are associated with cancer progression. However, differences between FSTL members with identical cancer types have not been systematically investigated. Among the most malignant tumours worldwide, colorectal cancer (CRC) has high metastatic potential and chemoresistance, which makes it challenging to treat. A systematic examination of the relationship between the expression of FSTL family members in CRC will provide valuable information for prognosis and therapeutic development. Based on large cohort survival analyses, we determined that FSTL3 was associated with a significantly worse prognosis in CRC at the RNA and protein levels. Immunohistochemistry staining of CRC specimens revealed that FSTL3 expression levels in the cytosol were significantly associated with a poor prognosis in terms of overall and disease‐free survival. Molecular simulation analysis showed that FSTL3 participated in multiple cell motility signalling pathways via the TGF‐β1/TWIST1 axis to control CRC metastasis. The findings provide evidence of the significance of FSTL3 in the oncogenesis and metastasis of CRC. FSTL3 may be useful as a diagnostic or prognostic biomarker, and as a potential therapeutic target.

stabilization of the tumour and suppression of further spread. 4 In the past, targeted therapies, such as those targeting vascular endothelial growth factor (VEGF), VEGF receptor, proto-oncogene B-Raf (BRAF), epidermal growth factor receptor and immune checkpoint inhibitors, have been developed to treat metastatic CRC. The effects of these approaches on survival have been limited. 5 According to 2019 cancer statistics, CRC patients have a 5-year survival rate of 64%, but when the disease has metastasized, the survival rate drops to 12%. 6 CRC will greatly benefit from unravelling the mechanism of tumour progression and discovering novel treatments and prognosticating targets.
Follistatin (FST)-like 3 (FSTL3) glycoprotein is a member of the follistatin-module-containing protein family. Despite its typical classification as a secreted protein, FSTL3 is transported into the nuclei of cells. 7 The secreted form of FSTL3 (isoform 1) exhibits a high degree of similarity to FST and is involved in the regulation of various biological effects by binding to and antagonizing members of the transforming growth factor-beta (TGFβ) superfamily, such as activin A, myostatin and bone morphogenetic protein 2 (BMP2). [8][9][10] Secreted FSTL3 plays a role in bone formation, haematopoiesis and leukemogenesis. [11][12][13] The function of the nuclear form of FSTL3 (isoform 2) is currently unknown but may be related to transcriptional regulation through interactions with MLLT10. 14,15 Aside from the biological regulation of FSTL3, there has been little information on the association between FSTL3 expression and malignancy, with only a few published studies. The expression of FSTL3 is higher in infiltrating ductal carcinomas of the breast than in normal tissue. 16,17 It has also been reported that the serum level of FSTL3 is higher in breast cancer patients. 18 However, in a subsequent study, FSTL3 levels in breast cancer were inversely related to tumour size and nuclear grade and did not correlate with disease survival. 19 The expression of FSTL3 in liver cancer is downregulated compared to normal liver tissue. 20 Increased FSTL3 expression in non-small-cell lung cancer tissues has been described, with the expression of FSTL3 associated with poor prognosis. 21 Chromosomal translocation of CCND1 to FSTL3 has been reported in a B-cell lymphoma case. 22 It remains unclear how alterations in FSTL3 may lead to tumour progression, and its role in the tumorigenesis of CRC is not well understood.
Here, we investigated FSTL3's role in the carcinogenesis of CRC.
Using both RNA and protein as biomarkers, we showed that FSTL3 has significant prognostic value in CRC. Immunohistochemistry (IHC) staining of clinical specimens revealed a significant correlation between patients with high cytosolic FSTL3 expression and poor prognosis. Using a simulated molecular interaction model and clinical correlation analysis, we identified the Twist Family BHLH Transcription Factor 1 (TWIST1)/TGFβ axis as a molecular mechanism regulated by FSTL3. The findings may guide the diagnosis and therapeutic development of patients with CRC.

| Evaluation of the prognostic value of FSTL3 by public databases
To determine the correlation between FSTL3 expression and patient outcomes, RNA expression microarray data for 177 patients with CRC were collected from the GSE database (GSE17536) (Table S5). These probes were used to measure the RNA expression levels of FST and FSTL proteins. The hazard ratio of FSTL family proteins and the survival curve of GSE 17536 were investigated using the PrognoScan online analysis platform. 23 In addition, the colon adenocarcinoma data set (COAD) of The Cancer Genome Atlas (TCGA) was used for survival and clinicopathological characteristic analyses (Table S3). Based on prognostic indices calculated from beta coefficients multiplied by gene expression values, CRC samples were divided into two groups of approximately equal size (Table S2). Using Table S1, heatmaps were generated to illustrate RNA-sequencing (RNA-seq) intensity RNA-seq by expectationmaximization (RSEM) values for different FSTL genes based on their highest, middle and lowest values in normal or tumour patients. To analyse the correlation between clinicopathological characteristics and FSTL3 expression, a chi-square test was employed. Kaplan-Meier analysis was used for statistical evaluation as described below.

| Patients and tissue samples
During the study period from 1998 to 2008, 236 patients with CRC who underwent tumour resection at the Wan Fang Hospital of Taipei Medical University were enrolled. As previously described, CRC specimens were used to construct tissue microarrays (TMAs). 24 The protocol for the collection of tissue samples and clinical records The period of time until the last follow-up or until death related to CRC was recorded from the date of treatment. The time was used to determine disease-specific survival (DSS), which was defined as the percentage of time between the date of treatment and the date of local recurrence or newly diagnosed metastasis. In this study, all patients were followed up for at least 5 years or until death.

| Immunohistochemistry
Immunohistochemistry staining of samples was performed as previously described. 24,25 Following deparaffinization and rehydration of the tissue sections, the sections were blocked with 3% hydrogen peroxide. Tris-EDTA buffer (pH 9.0) was used for heat-induced antigen retrieval. Incubation with primary antibody was performed Sigma-Aldrich) was used.

| IHC score assessment
Two pathologists who were blinded to the clinical parameters reviewed and scored the TMA sections stained with FSTL3. FSTL3 staining was assessed semi-quantitatively, based on the intensity and percentage of positively stained cells. There were four levels of staining intensity: 0, none; 1, weak; 2, moderate and 3, strong.
In each case, a final staining score of 0-1 or 2-3 indicated a low or high level of expression, respectively. The interpretations of the two pathologists did not differ significantly.

| Molecular regulation simulation signalling network analysis
An analysis of the clinically relevant molecules of FSTL3 in COAD was conducted using three TCGA databases: PanCancer Atlas (https://gdc. qiagen.com/plugi ns/ingen uity-pathw ay-analy sis/) to simulate the possible molecular regulation network. IPA is a molecular regulation simulation signalling network formed by combining a variety of omics data and known literature or databases to study the potential regulatory relationships between molecules of interest. Molecular regulatory relationships can be identified using the prediction legend.

| Statistical analysis
A chi-square test was used for categorical data and the Student's t test was used for continuous variables to analyse the relationships between clinicopathological characteristics and FSTL3 ex-

| FSTL family mRNA distribution in CRC
The related expression levels of the FSTL components from TCGA-COAD datasets were used to compare the expression levels of the FSTL family in CRC ( Figure 1A, Table S1). Based on the comparison of the solid tissue normal and primary tumour groups, it appeared that the FSTL family and FSTL3 exhibited significant differences (p < 0.0001) ( Figure 1B). Using different data sources (TIMER2.0), we also observed a highly significant level of FSTL3 expression in CRC ( Figure S1). FSTL3 was expressed more frequently in the tumour group than in the normal group. Based on a comparison of the same cases, the expression of FSTL3 was significantly elevated in tumour tissues (p = 0.0038; Figure 1C, Table S2). Furthermore, FSTL3 was compared according to the differences between pathological manifestations (Table S3) lymph nodes (p < 0.0005), and metastasis (p = 0.0429) ( Figure 1D).
Overall, these comparative analyses demonstrate that FSTL3 is a unique marker of the FSTL family in CRC and that there is a positive relationship between FSTL3 and the disease process.

| Expression level of FSTL3 mRNA is a prognostic factor for CRC
Using the mRNA expression data of 177 samples from CRC patients from GSE17536 (Table S5) Figure S2). This finding was further confirmed by analysing TCGA-COAD data for the expression of FSTL3.  Table S4).
Analysis of FSTL3 expression and the clinicopathological characteristics of the patients revealed a significant association between high FSTL3 expression and tumour size (T status, p < 0.001), lymph node invasion (N status, p < 0.001), and tumour stage (p < 0.001) ( Figure 2G). Taken together, these findings suggest an increased risk of poor health outcomes associated with high expression of FSTL3 in patients with CRC who experience tumour recurrence.

| Cytosolic FSTL3 in CRC correlates with tumour stage, distant metastasis and tumour recurrence
We examined the mRNA expression profiles using data from public data sets. A negative association was evident between FSTL3 TA B L E 2 Analysis of disease-specific survival (DSS) and disease-free survival (DFS) of colorectal cancer patients by univariate and multivariate models.  expression and poor prognosis among CRC patients (Figures 1 and 2).
Therefore, we investigated the protein expression levels of FSTL3 in clinical CRC samples and examined whether they were correlated with patient outcomes. A total of 236 CRC patients were analysed for the expression of FSTL3. As shown in Figure 2B, FSTL3 was expressed in normal colon tissues to a greater or lesser extent (0 to +1). FSTL3 expression varied in cancer tissues ( Figure 3A), ranging from no staining to strong cytoplasmic staining. In addition, nuclear staining for FSTL3 was detected in these samples ( Figure 3A). The cytosolic presence of

| TGFβ signalling is involved in FSTL3-mediated CRC progression
Molecular regulation simulation signalling network analysis was performed as previously described 26 to determine the molecular mechanism through which FSTL3 mediates CRC progression. TCGA CRC data sets were downloaded to identify common molecules related to FSTL3 using a Venn diagram ( Figure 4A, Table S6). The Venn diagram revealed a Spearman's correlation of more than ±0.3 related to FSTL3 ( Figure 4B). IPA assessment was performed on these consistent molecules to generate simulated molecular networks as-  Figure S3 displays the molecular map of TGF-β1 signalling that may be activated by FSTL3 ( Figure 3A) and the molecules regu- ( Figure S4, Table S8). The collective findings demonstrate that FSTL3 can contribute to CRC metastasis through EMT-related signalling and possibly through the TGF-β1 pathway.

| TWIST1 as a transcription regulator connects the TGFβ/FSTL3 axis in CRC
To further investigate the important transcriptional regulators that influence EMT-related signalling through FSTL3, a Venn diagram analysis was performed on the molecules affected by these EMTrelated signalling mechanisms ( Figure 5A, Table S9). IPA identified many transcription regulators related to FSTL3 regulation, including TWIST1, ERG, SNAI1, RUNX3, HIF1A, TP63, KLF11, EOMES, and SMAD2 ( Figure 5B). A strong relationship was evident between these nine transcription factors and the eight molecules that are regulated by FSTL3 (R = 0.88) ( Figure 5C). The finding supported the unbiased analysis results ( Figure 4D,E). Statistically, these EMTrelated molecules, including SNAI1, VIM and TGF-β1, were mostly affected by FSTL3 ( Figure 5D, Table S10). These factors were correlated with the TWIST1 and FSTL3 transcription regulators, which were the most correlated molecules (R = 0.64) ( Figure 5B).
TWIST1 regulated the molecular network of FSTL3-related molecules ( Figure 5E). A combined analysis of the IPA results ( Figures 4D and 5E, Figure S3) was performed to determine the molecules regulated by TGFβ, FSTL3 and TWIST1 ( Figure 6A). The correlation of these molecules with FSTL3, TGF-B1 and TWIST1 was evident ( Figure 6B). This is noteworthy given that each of these molecules   Table 2). The present results will inform the interpretation of FSTL3 expression in future clinical prognostication efforts.

| DISCUSS ION
A key factor in metastasis is the ability of a cell to undergo EMT. has previously been correlated with TGF-β1 and is reflected in the prognostic data. 32 Therefore, our analysis is sufficient to reflect past observations and to provide evidence of a possible regulatory relationship between FSTL3 and TWIST1/TGF-β1.
TGFβ has previously been shown to regulate FSTL1 expression.
As we previously discovered, FSTL1 can regulate the progression of lung cancer by interacting with secreted phosphoprotein 1 (SPP1), and the distribution of molecular molecules associated with BMP and TGFβ signalling is the opposite. In gut tissues, fibrosis and carcinogenesis are highly correlated with the stability of TGFβ signalling. 37 Based on this study, FSTL3 appeared to be positively correlated with TGFβ in both tissue and cellular data. Consequently, different FSTL family members have disparate roles and are regulated differently in different tissues. It remains to be determined whether FSTL3 also participates in BMP signalling in CRC and whether it cross-talks with TGFβ signalling.
In conclusion, the findings of the present study confirm the clinical significance of cytosolic FSTL3 in CRC progression and explain the molecular mechanism of FSTL3. The findings also highlight the potential of FSTL3 as a specific target through big data screening and validation of clinical organization. project administration (equal); resources (equal); software (equal).

ACK N O WLE D G E M ENTS
For the Aperio digital pathology analyses, we would like to express our sincere appreciation to GRC Instrument Core Facilities.

This study was supported by the Ministry of Science and Technology
(MOST-111-2314-B-001-011-MY3) to MH.

CO N FLI C T O F I NTE R E S T S TATE M E NT
There are no competing interests declared by the authors.

DATA AVA I L A B I L I T Y S TAT E M E N T
RNA-seq data can be obtained from Gene Expression Omnibus submission GSE17536.