Supplementary Figure S1 shows the workflow of our study and Supplementary Figure S2 illustrates the mechanism of our study.

The FSTL3 expression in CRC was first investigated using the TIMER web tool (https://cistrome.shinyapps.io/timer/) (Li et al., 2017) and the Gene Expression Profiling Interactive Analysis (GEPIA) database (https://www.oncomine.org/) (Tang et al., 2017). The differential expression of FSTL3 was further confirmed using the TCGA-Colon Adenocarcinoma (COAD) cohort and GSE10950, GSE44861 datasets (Zhao et al., 2020; Chen and Ke, 2021).

The raw counts of RNA-sequencing data and corresponding clinical information regarding FSTL3 were obtained from the TCGA dataset (https://portal.gdc.cancer.gov/) in January 2020 (Wang et al., 2016). The univariate and multivariate cox regression analyses were performed to access the independent prognostic values of FSTL3. The forest plot represented the p-value, hazard ratio (HRs), and 95% confidence interval (CIs) of each variable through the “forestplot” R package.

The FSTL3 expression levels were analyzed based on the various classification parameters, such as the T (Tumor) stages, N (Node) stages, M (Metastasis) stages, pathological stages, and histological grades according to the TCGA-COAD data.

To reveal the value of FSTL3 on the prognosis of CRC patients, survival analyses such as the overall survival (OS), disease-free survival (DFS), and post-progression survival (PPS) were performed through GEPIA. The automatically selected best cutoff was selected for analysis. In addition, the progression-free survival (PFS), progression-free interval (PFI), disease-specific survival (DSS), and disease-free interval (DFI) were also estimated based on the colon cancer cases in TCGA-COAD.

The median cutoff of FSTL3 expression in the TCGA-COAD was used to define the groups with high and low FSTL3 expression. The “DESeq” R package was used to obtain the FSTL3 -correlated genes. The “Enrichr” database was used to perform the functional enrichment analysis to explore the potential functions of FSTL3(Kuleshov et al., 2016).

The gene set enrichment analysis (GSEA) was performed using the Broad Institute GSEA software 3.0 (Powers et al., 2018). The gene set “subset of GO” was downloaded from the Molecular Signatures Databases (http://www.gsea-msigdb.org/gsea/msigdb/index.jsp) and was used for the GO enrichment analysis (Powers et al., 2018). The FDR <0.1 was considered to be statistically significant. In addition, a single cell analysis was conducted based on the GSE146771 to seek more evidence on the potential function of FSTL3 (Cao et al., 2021).

The ssGSEA algorithm was initially performed to assess the correlation between the FSTL3 expression levels and the overall immune as well as stromal infiltration levels in CRC (Liu et al., 2020). Furthermore, a high-performance computational method for quantifying cellular components from bulk tissue gene expression profiles, CIBERSORT was used to estimate immune infiltrations reliably (Chen et al., 2018). Spearman’s rank correlation coefficient was calculated for pairwise correlation comparisons and p < 0.05 was considered statistically significant. All the results from the above analyses methods and R package were implemented by the “ggplot2” and “pheatmap” packages.

A complete list of reagents and antibodies is provided in Supplementary Tables. All the concentrations were chosen based on the previous studies or the manufacturer’s instructions. The experimental details are given in Supplementary material.

The human CRC cell lines, SW620, SW480, RKO, HT-29, LoVo, Caco2, human monocytic cells THP-1 and normal human colonic epithelial cell line, NCM460 were purchased from the cell bank of the Chinese Academy of Sciences (Shanghai, China). The CRC and THP-1 cells were cultured in the RPMI-1640 medium with 10% fetal bovine serum (FBS). NCM460 was cultured in DMEM with 10% FBS. All the cells were incubated in 5% CO₂ at 37°C.

The protocol for western blotting was based on the previous studies (Hnasko and Hnasko, 2015). The Target/β-actin bands were identified using a gel image processing system (ChemiDoc XRS+). Subsequently, the relative protein levels were calculated.

The study’s protocol was approved by the ethics committee of the Jiangsu Province Hospital of Chinese Medicine, and informed consent was obtained from clinicians and patients (2020NL-107-01). The CRC tissue and the adjacent healthy tissues (margin, 5 cm) were collected during surgery from 30 previously treatment-naïve patients with CRC at the Jiangsu Provincial Hospital of Traditional Chinese Medicine. The tumors were staged and graded using the 8th edition of the American Joint Committee on Cancer tumor-node-metastasis (TNM) staging system (Tong et al., 2018). After extraction, the tissue specimens were rinsed with cold phosphate-buffered saline and immediately placed in liquid nitrogen. The flash-frozen tissues were then transferred and stored at −80°C until further examination using immunohistochemistry (IHC) and western blot analysis. The preoperative serum samples were collected from all the patients, and serum was stored at −80°C for further studies.

The protocol used for IHC was based on earlier studies (Nizioł et al., 2021). The images were captured using a NIKON Eclipse Ni-E microscope (NIKON, Japan) (original magnification, ×400). The H-SCORE (range 0–300, higher scores indicating stronger positive staining) was calculated as described previously (Yang et al., 2019).

The lentiviral vectors were used for overexpressing and knocking down FSTL3. The viruses were designed, synthesized, and produced by the GeneChem Corporation. Transfection was performed according to the supplier’s protocol. The HT-29 and RKO cells were transduced with the recombinant lentivirus using 2 μg/ml polybrene for 24 h. Subsequently, the stably transfected GFP-expressing cells were identified using 1.5 μg/ml puromycin. The FSTL3 overexpression and knockdown and transduction efficiency were assessed using western blots and GFP-expression.

The CCK8 assay was performed using a CCK8 kit following the manufacturer’s protocol. Briefly, CRC cells were plated into 96-well plates (5 × 10³ cells per well) in 100 μl of culture medium or serum-free condition for 12, 24, 48 h at 37°C. CCK-8 solution (100 μl/well) was added for another 2 h and then incubated for 12, 24, and 48 h. Then, the optical density (OD) was measured at 450 nm with a microplate reader (BioTek Synergy HT).

The cell supernatants or patients’ serum was examined for FSTL3/FN1/α5β1 expression using the ELISA Kit based on the given instruction manual. A microplate reader (BioTek Synergy HT) was used to examine the optical density at 450 nm.

The clonogenic ability of the cells was assessed using a clone formation assay, performed as described previously (Grover et al., 2016). The number of colonies was counted using a compound light microscope (Olympus BX53, Japan).

All the animal experiments were approved by the ethics committee of the Jiangsu Province Hospital of Chinese Medicine (2021-5-062). Twenty-four 4-week-old male BALB/c nude mice were obtained from the Beijing Institute of Biomedicine (Beijing, China) (Certificate No. SYXK 2019-0010). The RKO cells transfected with sh-FSTL3, oe-FSTL3, and NC and control cells (4 × 10⁶ cell/mouse) were injected subcutaneously into the right armpit region (n = 6 per group). Seven days later, tumor formation was observed beneath the skin. The maximum 1) and minimum tumor diameter 2) were measured twice weekly. On day 28, the mice were euthanized and all the tumors were collected. The tumor volume was calculated (V = 1/2ab²), and the growth curves of the subcutaneous xenografts were drawn.

The protocol used for the wound healing assay was based on earlier studies (Han et al., 2020). Cell migration towards the scratch zone was photographed using an inverted fluorescence microscope (Olympus CKX-41, Japan) (×200 magnification).

Cells were dissociated to produce single-cell suspensions and were seeded in six-well ultralow-attachment plates at a density of 5 × 10 ³ cells/well. They were cultured in serum-free medium DMEM with FGF (20 ng/ml), EGF (20 ng/ml), and 2%B27. After culturing for 7 days, the size and number of tumor spheres were evaluated using light microscopy (Olympus BX53, Japan) (×40 magnification).

The cell migration and invasion were assessed using a transwell assay based on a previously published protocol (Wang et al., 2021). The membrane in the chamber was cut and imaged using light microscopy (Olympus BX53, Japan) (×200 magnification), and the cell counts were obtained using ImageJ software.

The protocol used for immunofluorescence staining was based on earlier studies (Donaldson, 2015). The immunofluorescence staining was observed using epifluorescence microscopy (Olympus, BX60-32FB2-A03) and different filters were used for capturing images using an Olympus, DP50 camera (×400 magnification).

The THP-1 cells (1 × 10⁵ cells/ml) were treated with phorbol 12-myristate 13-acetate (PMA) (10 ng/ml) for 48 h to allow the induction of macrophage differentiation (Genin et al., 2015). The PMA-containing medium was replaced with the serum-free medium, and the cells were cultured for 24 h. Two days before the co-culture experiment, the cells (1 × 10⁵ cells/ml) from the control, knock-down (sh-FSTL3), overexpression (oe-FSTL3), and negative control (NC) groups were seeded onto the 0.4 μM transwell inserts. For co-culture, the culture medium in the inserts with the CRC cells was removed and transferred to the top of the pates with differentiated THP-1 cells. After 48 h of further co-culturing, the cells were obtained, and immunofluorescence staining was performed.

Data were reported as mean ± standard deviation. The t-tests and one-way ANOVA were used to perform comparisons between the two groups and among the multiple groups, respectively. All data were analyzed using the SPSS 26.0 (SPSS Inc., USA) and illustrated using the GraphPad Prism 8.0 (GraphPad Software, Inc., USA). All the experiments were carried out at least thrice. **p < 0.01 and *p < 0.05 were defined to be statistically significant.

The TCGA-COAD data showed that the FSTL3 expression is higher in the CRC tissues than in the normal tissues (Figure 1A; p < 0.05). Moreover, the TIMER and TCGA data analyses also showed higher FSTL3 expression in the CRC than in the normal tissues (Figure 1B).

The expression of FSTL3 was further explored in the CRC tissues using data from the Gene Expression Omnibus (GEO) database. Data from the GSE10950 and GSE44861 datasets indicated a significant difference in the FSTL3 expression between the CRC tissues and adjacent tissues (Figure 1C). Western blot, ELISA, and IHC staining revealed that FSTL3 was overexpressed in the CRC cells and tissues. The mean H-SCOREs for FSTL3 expression in the CRC and paracancerous tissues were 95.53 ± 15.96 and 11.70 ± 3.55, respectively (Figures 1D–G) (p < 0.01, ANOVA). The FSTL3 protein expression in CRC was further verified using the IHC data from The Human Protein Atlas, which revealed that FSTL3 to be primarily expressed in the cell membrane and cytoplasm (Figure 1H).

The multivariate hazard ratios for the different variables were then calculated using a Cox regression model based on the TCGA-COAD. Univariate analysis results (OS) demonstrated that the FSTL3 overexpression (p = 0.002), T classification (p = 0.004), N classification (p < 0.001), and M classification (p < 0.001) were all closely correlated with a poor prognosis (Figure 2A). The multivariate analyses revealed that the T classification (p = 0.021) and M classification (p < 0.001) were all independent predictors of an unfavorable prognosis (Figure 2D).

Furthermore, the results of the univariate analysis (DSS) demonstrated that FSTL3 overexpression (p < 0.001), T classification (p = 0.004), N classification (p < 0.001), and M classification (p = 0.004) were in close correlation with a poor prognosis (Figure 2B). The multivariate analyses revealed soverexpression (p = 0.039), and M classification (p < 0.01) to independently predict unfavorable prognosis (Figure 2E).

Finally, the results of the univariate analysis (PFI) demonstrated FSTL3 overexpression (p < 0.001), T classification (p < 0.001), N classification (p < 0.001), and M classification (p < 0.001) to be closely correlated with a poor prognosis (Figure 2C). The multivariate analyses revealed FSTL3 overexpression (p = 0.012), and M classification (p < 0.001) to independently predict the unfavorable prognosis (Figure 2F).

The FSTL3 levels are related to the clinicopathological characteristics of the CRC patients, including the T/N/M stage, pathological stage, BMI, weight, age, sex, and race, as illustrated in Figure 2G.

To evaluate the association of FSTL3 expression levels with the survival in CRC patients, the GEPIA database was used. Interestingly, the expression of FSTL3 showed a significant negative correlation with the patient survival (GEPIA: Overall survival (OS), p = 0.006; Disease-Free Survival (DFS), p = 0.018) (Figure 2H).

The results based on the TCGA-COAD revealed the increasing FSTL3 expression levels to be associated with a worse prognosis (OS, p = 0.01; PFI, p = 0.0073; DSS, p = 0.0045) in the KRAS wild type (Figure 2I) and mutant type (Figure 2J).

The genes positively or negatively correlated with FSTL3 in TCGA-COAD were obtained based on the “DESeq” R package (Figure 3A). A total of 21 differentially expressed genes (DEGs) were imported into the DEGs PPI, and then the Cytotype cytoHubba was applied for further analysis demonstratingFN1 to be the highest-scoring hub gene (Figure 3B). The correlation analyses between the serum FN1, ITGA5 (gene symbol for α5β1), and serum FSTL3 in the patients by comparing between the two groups (area within the purple box) and TCGA-COAD-based correlation analyses between FN1, ITGA5, and FSTL3 (green box). The results showed a highly significant positive correlation between the expression of FSTL3 and FN1 and ITGA5 (Figure 3C). FN1 was overexpressed in COAD (Figure 3D) but was not correlated with the prognosis of COAD patients (Figures 3E–G). The analysis of the functional enrichment showed FSTL3 to be involved in “extracellular matrix organization,” “extracellular structure organization,” “collagen-containing extracellular matrix,” “platelet-derived growth factor binding,” “Protein digestion and absorption,” ECM-receptor interaction,” and “Focal adhesion.” Most results have been correlated to the EMT or extracellular matrix (ECM) leading to the invading phenotype of the carcinoma cells. (Figures 3H–K).

In the single-cell level study, the FSTL3 and FN1 were more inclined to express on the fibroblasts (Figures 3L–Q), which are important players in the EMT.

The expression level of FSTL3 was the highest in the RKO and HT-29 cell lines, therefore these were selected for further experiments. The transfection efficiency was verified by GFP expression and western blot ((Figure 4A) (p < 0.01). FSTL3 (either overexpression or knockdown) does not have an obvious effect on cell proliferation as well as cell viability under either with serum or without serum condition for 48 h (Supplementary Figure S3). FSTL3 silencing decreased the capacity of forming the tumor cell clone as well as the sphere-forming abilities (Figures 4B,C). Moreover, the stable FSTL3 overexpression in the RKO cells promoted the formation of the subcutaneous xenograft tumors in vivo (Figures 4D–F) (p < 0.01). Based on the TIMER, the data, and patient’s serum (n = 30) in Figure 3 C was the FSTL3 expression was positively correlated with that of FN1, and ITGA5. Subsequent the in vitro experiments using the western blot revealed a reduction in the levels of FN1, ITGA5 after FSTL3 silencing (p < 0.05) (Figure 4G), and the opposite trend was observed when FSTL3 was overexpressed. The GSEA for FSTL3 revealed the potential role of FSTL3 in “epithelial-mesenchymal transition in CRC” (Figure 4H). Based on this result, the EMT markers and EMT-related phenotypes were studied and the FSTL3 overexpression was found to increase the EMT-related phenotypes. However, this effect was attenuated after treatment with the integrin ITGA5 -specific inhibitor, ATN-161 (Figures 4I–N). The cell viability assays showed the cell proliferation to be unaffected by the selected concentration of ATN-161 (Supplementary material).

GSEA also revealed the functional enrichment for FSTL3 under the “regulation of actin cytoskeleton” and “regulation of microtubule cytoskeleton” domains (Figure 5A). Subsequent in vitro experiments revealed that FSTL3 overexpression makes the cell pseudopodia longer and more obvious (Figures 5B,C) and the upregulation of F-actin (Figures 5D,E). Consistent with the prior experiments, this effect was attenuated after treatment with ATN-161 (p < 0.01).

The ssGSEA algorithm was used to calculate the correlation between the FSTL3 expression and infiltration degree of several immune cells (Figure 6A). Considering FSTL3 to be associated with the abundance of macrophages, the relationship between the FSTL3 and macrophage abundance was further assessed using the GSE10950 and GSE44861 datasets. After excluding the normal samples from these two datasets, 22 immune cell profiles were obtained for the CRC samples (Figure 6B). Subsequently, the GSE10950 and GSE44861 datasets were used to assess the relationship between the FSTL3 expression and the macrophage infiltration The FSTL3 expression levels were positively correlated with the M2 macrophage abundance (Figure 6C). Therefore, the correlation between the FSTL3 and the M2 surface markers were calculated using the TIMER database and a positive correlation was observed between the FSTL3 expression and MRC1 (CD206) (R = 0.36, p < 0.001) and CD163 (R = 0.47, p < 0.001) expression (Figure 6D). This series of results suggested a positive association between the FSTL3 expression and M2 macrophage infiltration. To further investigate the influence of FSTL3 overexpression on the M2 macrophage abundance in CRC, a tumor–macrophage cell co-culture model was established using a transwell non-contact co-culture unit (Figure 6E). The FSTL3 overexpression was found to significantly upregulate the surface markers of the M2 tumor-associated macrophages (TAMs) (CD206 and CD163) (Figures 6F,G).

Hence, the findings of this study confirmed the FSTL3 levels to be positively correlated with the abundance of M2 macrophage infiltration.