Expression of FSCN1 and FOXM1 are associated with poor prognosis of adrenocortical carcinoma patients

ACC clinical data and RNA-seq data from the TCGA project were analysed by GEPIA [22]. GEPIA used one-way ANOVA and the limma method in the differential expression analysis. The Kaplan-Meier curve method and Log-rank test were used in the survival analyses. This study meets the publication guidelines provided by TCGA (http:// cancergenome.​nih.​gov/​publications/​publicationguide​lines). Microarray data (GSE12368) from the Gene Expression Omnibus database (GEO, http://​www.​ncbi.​nlm.​nih.​gov/​geo) were used to compare expression of candidate genes in ACC and benign adrenocortical adenoma (ACA) tissues [23]. Differentially expressed genes (DEGs) were analysed using GEO2R query and limma R packages from the Bioconductor project (http://​www.​bioconductor.​org). Genes with an adjusted P value < 0.05 and a log2 fold change (logFC) > 1 were considered DEGs.

TIMER (https://​cistrome.​shinyapps.​io/​timer/​) was used to estimate the potential association among candidate genes, immune cell infiltration and clinical parameters. The correlation between candidate genes with tumour microenvironment status was calculated based on six immune infiltrates (B cells, CD4+ T cells, CD8+ T cells, neutrophils, macrophages and dendritic cells) by Spearman’s correlation method, which was also validated using pathological estimations in the TIMER project [24]. Additionally, the candidate gene and immune signature correlation analysis was also validated in the GEPIA 2 platform (http://​gepia2.​cancer-pku.​cn/​#index) using the Pearson correlation coefficient method. The signatures were evaluated mainly based on specific markers in different types of immune cells. For example, the effector T cell signature depends on the expression level of CX3CR1, FGFBP2 and FCGR3A, and the effector memory T cell signature depends on PDCD1, DUSP4, GZMK, GZMA and IFNG.

Patients underwent resection for tumours at the West China Hospital and those that were pathologically confirmed as ACC from 2009 to 2016 were analysed. A total of 51 patients were enrolled in this study. The method and criteria of clinical record extraction and long-term follow-up were the same as our previous report [25]. Recurrent disease was diagnosed on the basis of clinical, radiographic, and laboratory evidence, including local recurrence, peritoneal carcinomatosis and distant metastases. Gender, age, grade, stage, treatment, R status, Ki67 index, and clinical follow-up data were updated. The corresponding formalin-fixed, paraffin-embedded (FFPE) tissues in our institutional biobank were retrospectively collected. This research was approved by the West China Hospital of Sichuan University Biomedical Research Ethics Committee following the ethical guidelines as required by the Declaration of Helsinki.

Serial FFPE tissue sections of 4-μm thickness were subjected to immunohistochemistry (IHC) analysis following standard protocols. Briefly, sections were deparaffinized in xylene and rehydrated through a graded ethanol series, then placed in 3% H2O2 for 15 min at room temperature. After the heat-mediated retrieval using sodium citrate or EDTA, slides were incubated with different primary antibodies: mouse anti-human Fascin (55 K-2) monoclonal antibody (#99978, Cell Signaling Technology, Danvers, MA, USA); rabbit anti-human FOXM1 monoclonal antibody (ab207298, Abcam, Cambridge, MA, USA), overnight at 4 °C. SignalStain® Boost IHC Detection Reagent (HRP, rabbit, CST) was applied for 30 min at room temperature according to the manufacturer’s instructions.

Immunostaining results were independently evaluated by two investigators blinded to the clinical data (F.Z. and C.Z.) and the inter-observer agreement was evaluated through the Cohen k coefficient value (0.86). A semi-quantitative H-score was calculated according to previous research [11] by multiplying the intensity score by the proportion score in which membrane and cytoplasmic staining intensity was evaluated with a score of 0 (negative), 1 (weak), 2 (moderate) or 3 (strong) and the proportion score was calculated as 0, 0.1, 0.5 or 1, respectively corresponding to 0%, 1–9%, 10–49% or > 50% of the positive tumour cells in each specimen. For the nuclear positive FOXM1, the proportion score was calculated as 1 or 2 according to the percentage of nuclear positive cells, as follows: 1, < 30% of cells positive; 2, ≥ 30% of cells positive. The cut-off value distinguishing high or low expression of candidate markers was H score ≥ 1 or < 1 (FSCN1) and H score > 1 or = 1 (FoxM1) in this research.

First, the mRNA expression levels of FSCN1 and FOXM1 in the TCGA ACC cohort were analysed. Both FSCN1 (logFC = 1.573, adjusted P < 0.001) and FOXM1 (logFC = 1.733, adjusted P < 0.001) were found to be over-expressed in ACC tissues (n = 77) compared to normal tissues (n = 128, Fig. 1a). Furthermore, patients at higher pathological stages were more likely to have a higher expression level of FSCN1 and FOXM1 (P < 0.001, Fig. 1b). The general characteristics of TCGA ACC patients were summarized in a previous report [21]. In addition, based on the microarray data (GSE12368) in ACCs (n = 12) and ACAs (n = 16), we also checked the different expression levels of FSCN1 and FOXM1. FSCN1 and FOXM1 were both overexpressed in ACCs (logFC = 1.28, adjusted P value = 0.00261 and logFC = 3.958, adjusted P value < 0.001, respectively).

Next, we validated the abnormal expression of Fascin (FSCN1) and FoxM1 (FOXM1) using ACC cases from the WCH cohort (n = 51) from 2009 to 2016 (Fig. 1b). The clinicopathological characteristics were analysed in Table 1. In the patient cohort, 4 (7.84%) were ENSAT stage I, 17 (33.3%) ENSAT stage II, 20 (39.22%) ENSAT stage III, and 10 (19.61%) ENSAT stage IV. The treatment and follow-up information were exhibited in Table 2. In this cohort, 34 (66.7%) patients underwent open adrenalectomy and 17 (33.3%) patients underwent laparoscopic adrenalectomy. The median age of WCH ACC patients was 44 (range from 3 to 79). Thirty-three (64.71%) patients died during follow-up, and 24 (47.06%) patients suffered from disease recurrence. Median overall follow-up was 723 (117–3210) days. The positive expression rate of Fascin and FoxM1 were 92.2% (47/51) and 78.4% (40/51), respectively. Representative sections and the different staining intensity grades of Fascin and FoxM1 were depicted in Fig. 1b and Fig. 1c. The correlation between the two markers and clinicopathological parameters were compared. Notably, patients at stage III/IV were more likely to have a high fascin expression status than those who were diagnosed at stage I/II (76.67% vs. 47.62%, P = 0.04). A high expression level of FoxM1 was more common in functional ACCs (82.61% vs. 32.14%, P < 0.001).

To explore their potential correlation with the immune environment in ACC, the relationships between FOXM1/FSCN1 and tumour immunity were evaluated based on the TCGA-ACC cohort in TIMER. FSCN1 expression was found to be positively correlated with ACC tumour purity (r = 0.349, p = 2.33e− 03), and negatively correlated with the infiltration signature of CD8+ T cells (r = − 0.372, p = 1.19e – 03, Fig. 2a), while FOXM1 showed a weak correlation with B cell and dendritic cell signatures (Fig. 2b). Given that an association between FOXM1/FSCN1 and immune signature was observed in our previous study [26], we further explored the potential links between FSCN1/FOXM1 and several immune markers, such as CD276 and the innate immune marker, CD161 (KLRB1) [26‐28]. Notably, Both FSCN1 and FOXM1 were found to be negatively correlated with innate immune signature, KLRB1(r = − 0.539, P value = 2.9e-07, and r = − 0.343, P value = 2.0e-03, respectively), and positively correlated with CD276 (r = 0.29, P value = 9.7e-03 and r = 0.326, P value = 3.52e-03, respectively, Fig. 2c and d). To validate the immune correlation, we further performed GEPIA 2 analysis. A negative correlation was observed between FSCN1 signature and effector T cell signature and effector memory T cell signature (Fig. 2e). We also merged the FSCN1 and FOXM1 signature and verified its correlation with CD276 and KLRB1 (Fig. 2f). As expected, the CD276 signature was significantly correlated with the FSCN1/FOXM1 signature (r = 0.52, P value = 1.2e-06). The KLRB1 signature was negatively associated with the FSCN1/FOXM1 signature (r = − 0.23, P value = 0.043). These results indicated that the FSCN1- and FOXM1-signatures might be involved in immune activities in the ACC microenvironment.

In the TCGA ACC cohort, patients with a high expression level of FOXM1 and FSCN1 had worse overall survival (OS, n = 76, HR = 4.9, P = 4e-04 and HR = 9.4, P = 4.1e-05, respectively, Fig. 3a) and disease-free survival (DFS, n = 76, HR = 3.2, P = 0.0016 and HR = 8.1, P = 1.2e-06, respectively, Fig. 3b). Based on the multivariable Cox proportional hazard model in TIMER, we explored the clinical relevance of FSCN1 and FOXM1 with the flexibility to correct for multiple covariates including gender, age, race, pathologic stage, CD8+ T cell signature, CD276 and KLRB1 (Table 3). The results suggested that FSCN1 and FOXM1 had independent prognostic effects on ACC patients (P < 0.001 and P = 0.001, respectively).

Moreover, the prognostic significance of Fascin (FSCN1) and FoxM1 (FOXM1) were also observed in the WCH cohort. High- and low- expression of two proteins were classified according to the expression score. As a result, high Fascin expression and high FoxM1 expression in the tumour tissues were both remarkably correlated with worse OS (HR = 4.69, P = 0.002 and HR = 3.95, P = 0.001, respectively) and DFS (HR = 3.92, P = 0.007 and HR = 3.34, P = 0.009, respectively). In the multivariate Cox model, Fascin score (HR = 2.86, 95%CI: 1.06–7.7, P = 0.038), FoxM1 score (HR = 3.17, 95%CI: 1.37–7.33, P = 0.007) and Ki67-index (HR = 2.39, 95%CI: 1.15–4.99, P = 0.02) were verified as independent risk factors for OS. Fascin score (HR = 2.98, 95%CI: 1.02–8.7, P = 0.046), FoxM1 score (HR = 2.98, 95%CI: 1.14–7.8, P = 0.026) and symptoms (HR = 4.10, 95%CI: 1.11–15.19, P = 0.035) were verified as independent risk factors for DFS (Table 4).

In addition, the prognostic difference in the Fascin and FoxM1 high−/low- groups stratified by stage, treatment, margin status (R), additional adjuvant therapy and Ki67 index were also evaluated using the univariate Cox proportional hazards regression method. Patients were divided into these subgroups to check if the Fascin/FoxM1-related prognostic effects could be observed. Accordingly, both Fascin and FoxM1 were found to correlate with OS in early (I/II) or late (III/IV) stage ACCs (Fig. 3c). This OS-correlation relationship was also shown in the open surgery subgroup, R0 subgroup, no-additional adjuvant therapy treatment subgroup and low-Ki67 subgroup (Fig. 3c). On the other hand, the DFS-correlation of Fascin and FoxM1 were respectively observed in early stage and late stage ACCs. The same effects were also found in the open surgery subgroup, no-additional adjuvant therapy treatment subgroup and low-Ki67 subgroup (Fig. 3d).