High expression of HSP90 is associated with poor prognosis in patients with colorectal cancer

Patients

Formalin-fixed paraffin-embedded tumor specimens from surgical resections of the primary tumor were obtained from 99 patients with CRC who underwent curative surgery at our hospital from September 2008 to August 2010. The ethical approval of the study was obtained from the ethics committee of National Cancer Center/Cancer Hospital of Chinese Academy of Medical Sciences and Peking Union Medical College (NCC2018-071). These patients had confirmed pathological diagnosis and complete clinical data, and did not receive radiotherapy or chemotherapy before the surgery. Survival data of all these patients were followed up by telephone interview until October 31, 2018.

Immunohistochemistry and evaluation of staining

Immunohistochemistry assay was performed with four µm sections of paraffin-embedded tissues on polylysine coated slides. Slides were deparaffinized with xylene and rehydrated through descending gradient ethanol (100–95–85–75%). Then these slides were washed with phosphate-buffered saline solution (PBS, 0.01M, pH 7.0) for three times. Antigen retrieval was performed by boiling in 10 mM sodium citrate buffer (pH 6.0) for 15 mins and then cooled down to room temperature (RT) naturally. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide at RT for 10 mins. After being washed for three times with PBS, the sections were incubated in 5% bovine serum albumin at RT for 30 min to block nonspecific binding. Then they were incubated with anti-HSP90 mAb (1:500) (cat: ab59459; Abcam, Cambridge, MA, USA) overnight at 4 °C. After washing with PBS for three times, the specimens were incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG secondary antibodies (1:500) (cat: no. WLA023a; Wanleibio Co., Ltd., Shenyang, China) for 1 h at RT and visualized by DAB development with Dako EnVison kit (Dako, Glostrup, Denmark). Finally, all slides were counterstained with hematoxylin, dehydrated and mounted with glycerol gelatin. Sections were observed with a microscope.

The HSP90 expression-intensity scores were independently determined by two experienced pathologists who were blinded to the information of patients. According to the percentage of the HSP90 positive tumor cells and intensity of staining, the expression area was graded into four levels: 0 (0), 1 (<1/10), 2 (1/10–1/3), 3 (1/3–2/3), and 4 (≥2/3). The staining intensity score was defined as follows: no staining, 0; weak staining, 1; moderate staining, 2; strong staining, 3. The total HSP90 expression index was generated by multiplying the score of staining intensity and the percentage of HSP90 expression, which ranged from 0 to 12. We defined scores of 0–3 as the low expression group, and scores of 4–12 as the high expression group.

Analysis of HSP90 genes expression level in different cancers

Online analysis of HSP90 mRNA levels of different cancers was done by GEPIA, http://gepia.cancer-pku.cn/index.html), a web server for cancer and normal gene expression profiling analysis. It worked based on database of the The Cancer Genome Atlas (TCGA) and the GTEx project including 9,736 tumors and 8,587 normal samples RNA-Seq data and could be used to analyze gene expression through a standard processing pipeline.

Statistical analysis

Nonparametric Mann–Whitney U test was used to detect the HSP90 expression difference between cancer tissues and adjacent normal tissues. Chi-square test or Fisher’s exact test was performed to analyze the relationship between HSP90 status and CRC patients’ clinicopathological features. Survival curves were determined with the Kaplan–Meier method, and different survival rates between groups were compared with the log-rank test. The significance of variables for survival was conducted with the Cox proportional hazards model in univariate and multivariate analysis. All statistical analysis was performed with the SPSS 13.0 statistical software (IBM Corp., Armonk, NY, USA). P < 0.05 (two-tailed) was considered to indicate a statistically significant difference.

Bioinformatics analyses

We firstly downloaded RNA Seq V2 RSEM data which including 17,989 gene expression of 382 CRC tissues from https://www.cbioportal.org/ by using R “cgdsr” package. By calculating Pearson’s correlation coefficients, we screened genes which positively correlated with the expression of HSP90 (Pearson’s correlation ≥ 0. 4, P < 0.0001). Then GO analysis and KEGG analysis were performed using edgeR on OmicShare, an online platform for data analysis (www.omicshare.com/tools). Q value < 0.05 was used as the thresholds in selecting significant GO and KEGG pathways.

Dataset analysis indicated a significant different expression of HSP90 between cancer tissues and normal tissues

Gene Expression Profiling Interactive Analysis is a commodious and intuitive online tool for gene analysis, a web-based tool based on TCGA and GTEx data. It provides key interactive and customizable functions including differential genes expression analysis, profiling plotting, similar gene detection, correlation analysis, and dimensionality reduction analysis (Tang et al., 2017). We analyzed the HSP90 mRNA expression profile across all tumor samples and normal tissues in GEPIA. From these results showed in bar plot (Fig. 1A), we found HSP90 had a different expression between some kinds of cancers and normal tissues. Statistical analysis showed in dot plot indicated significant higher expression in 10 kinds of cancer tissues: breast invasive carcinoma, cervical squamous cell carcinoma and endocervical adenocarcinoma, colon adenocarcinoma (COAD), lymphoid neoplasm diffuse large B-cell lymphoma, esophageal carcinoma, pancreatic adenocarcinoma, rectum adenocarcinoma (READ), skin cutaneous melanoma (SKCM), stomach adenocarcinoma, and thymoma (THYM) than corresponding normal tissues (Fig. 1B). Detailed analyses of the expression of HSP90 of COAD and READ with TCGA data were shown with number of samples in Fig. 1C.

HSP90 had a higher expression in colon cancer tissues

Immunohistochemistry was done to evaluate and compare the expression of HSP90 between 99 CRC tissues and 81 adjacent tissues. The expression of HSP90 mainly located in cytoplasm, which had no significant difference between cancer tissues and adjacent tissues (Fig. 2). The expressions of HSP90, evaluated by the HSP90 expression score index, were less than four points in most normal tissues and were lower than cancer tissues.

In order to make a deeper analysis, we defined score index 0 to 3 as low-level group (including negative expression), 4 to 12 as high-level group. In 99 CRC tissues, 55 samples (55.6%) were in the high-level group and the rest (44.4%) were in the low-level group. In contrary, 78 out of 81 (96.3%) adjacent tissues were in the low-level group. Furthermore, 65 out of 78 (80.2%) adjacent tissues were negative expression. Statistical analysis showed a significantly higher expression of HSP90 in cancer tissues (Nonparametric Mann–Whitney U test, P < 0.0001) (Fig. 2B).

Association between HSP90 expression and clinicopathological characteristics of CRC patients

Then we collected clinicopathological parameters of CRC patients and investigated correlations with HSP90 expression (Table 1). As for general clinical characteristics, no significant expression difference was found in sex groups (P = 0.223) or age groups (P = 0.438). Chi-square test and Fisher’s exact test were also used to analyze the relationship between HSP90 and some pathological variables which were important for evaluating prognosis. However, no obvious association was found between HSP90 expression levels in CRC tissues and tumor location (P = 0.604), pathological differentiation (P = 0.621), T phase (P = 0.505), lymph node metastasis (P = 0.608) or American Joint Committee on Cancer (AJCC) stage (P = 0. 749).

High expression of HSP90 predicts poor prognosis of CRC patients

Survival analysis was conducted to explore whether HSP90 expression influences clinical progression in CRC patients. Here, we did not analyze the relationship between M category (describes the presence or otherwise of distant metastatic spread) and prognosis as only three cases had distal metastasis. Besides lymph node metastasis (P < 0.001), AJCC stage (P < 0.001) and pathological differentiation (P = 0.033), high HSP90 expression was associated with poor prognosis (P = 0.039) in CRC patients as well. HSP90 high-expression group have a significantly worse overall mean of survival (OS) than low-expression group (51.6 vs. 66.0 months) (Fig. 3).

Next, Cox’s proportional hazards model was also done for further analysis of prognostic value of HSP90 expression and other clinical parameters (Table 2). These results revealed that variables including lymph node metastasis (HR = 2.192, 95% CI [1.544–3.112], P < 0.001), pathological differentiation (HR = 1.737, 95% CI [1.035–2.915], P = 0.037) and high HSP90 expression (HR = 1.738, 95% CI [1.018–2.966], P = 0.043) linked to poor prognosis (Fig. 4A). Then, multivariate analysis of Cox’s proportional hazards regression model was further done to exclude the false positive results caused by the interaction of different factors. Multivariate analysis results revealed that, in addition to the acknowledgeable factors including lymph node metastasis (HR = 2.459, 95% CI [1.652–3.661], P < 0.001) and pathological differentiation (HR = 1.749, 95% CI [1.017–3.007], P = 0.043), HSP90 expression level (HR = 1.930, 95% CI [1.113–3.349], P = 0.019) could be an independent prognostic factor for CRC patient prognosis prediction (Fig. 4B).

HSP90-related signaling pathways based on TCGA database

We performed functional enrichment analysis to explore the biological pathway and process correlated with HSP90. Firstly, the Pearson’s correlation coefficients between HSP90 expression values and other mRNA expression values were calculated in the TCGA dataset. A total of 85 genes were positively correlated with the expression of HSP90 (defined as Pearson’s correlation coefficient ≥ 0. 4, as shown in Data S1) and included in the GO analysis (Fig. 5) and the KEGG analysis, respectively (as shown in Supplemental Data). The GO analysis results showed that it mainly existed as an intracellular molecular (Fig. 5D), which were involved in metabolic process (Fig. 5B) and exerted binding and catalytic activity (Fig. 5C). Besides, KEGG analysis enriched these genes in pathways of Proteasome and Base excision repair (Fig. 6).