Significance of unphosphorylated and phosphorylated heat shock protein 27 as a prognostic biomarker in pancreatic ductal adenocarcinoma

Purpose Few studies reported about the potential of unphosphorylated heat shock protein 27 (HSP27) and phosphorylated heat shock protein 27 (pHSP27) as a predictor for survival and gemcitabine resistance in pancreatic ductal adenocarcinoma (PDAC). In this study, we analysed the expression patterns of pHSP27 and HSP27 in a patient population after surgery and correlated the immunohistochemical results with clinicopathological data and long-term outcome of the patients. Methods HSP27 and pHSP27 (Ser-15, Ser-78 and Ser-82) protein expression were analysed by immunohistochemistry using the immunoreactive score (IRS) from paraffin-embedded tissue of 106 patients with PDAC who underwent surgery. Immunohistochemical results were correlated with clinicopathological data, disease-free (DFS) and overall survival (OS). Results HSP27 expression was significantly lower in patients with a shorter OS (p = 0.006) and DFS (p < 0.0001). A higher HSP27 expression was associated with a better response to gemcitabine in the resected, non-metastasised patients group (p = 0.001). Furthermore, HSP27 was downregulated in patients suffering from metastases at time of surgery (p < 0.001) and in undifferentiated tumours (p = 0.007). In contrast, pHSP27-Ser15, -Ser78 and -Ser82 were not associated with any survival data of the study population. Conclusion HSP27 seems to be a strong indicator for the prediction of OS and DFS. Moreover, HSP27 could play a role in the formation and migration of liver metastases of PDAC.


Introduction
PDAC remains a challenging disease with a poor overall 5-year survival rate between 4 and 8% (Ilic and Ilic 2016;Ferlay et al. 2013). The best chance for cure with long-term survival is surgical resection. Unfortunately, the majority of the patients display hepatic metastasis at time of diagnosis and only 15-20% are eligible for surgery (Siegel et al. 2018;Vincent et al. 2011).
Heat shock proteins (HSPs) were first discovered by Ritossa et al. as a family of proteins induced by heat shock and other stimuli (Lindquist and Craig 1988;Michel and Starka 1986). These proteins have been characterized as chaperones which are capable of modifying structures of several proteins and can prevent accumulation of misfolded proteins (Ritossa 1996;Gaestel et al. 1989). HSP27 belongs to the small heat shock proteins and can occur in various types depending on post-translational modifications such as phosphorylation at three Serin-sites (Ser-15, Ser-78 and Ser-82) on C-terminal region. The phosphorylation is performed by several kinases including MAPK activated protein kinase 2 (MAPKAPK-2) (Schäfer et al. 1998;Pietersma et al. 1997). Furthermore, several studies report that HSP27 is a major target of protein kinases A, B, C and D (Kostenko and Moens 2009;Gaestel et al. 1991;Butt et al. 2001;Döppler et al. 2005). The effect of phosphorylation is a conformational change of HSP27 which leads to dissociation of large oligomers and the formation of dimers presenting the active leading conformation (Jovcevski et al. 2015).
The phosphorylated and unphosphorylated form of HSP27 has been linked to tumour progression and cell migration (Cayado-Gutiérrez et al. 2013;Zoubeidi and Gleave 2012). Therefore, the expression of HSP27 has been shown to be related to tumour characteristics. Consecutively, the potential of HSP27 in diagnostic, prognostic or treatment implications have been examined in multiple tumour entities. Overexpression of HSP27 correlates with a shorter overall survival (OS) or disease-free survival (DFS) in ovarian, gastric, prostate cancer, and osteosarcoma (Piura et al. 2002;Elpek et al. 2003;Takeno et al. 2001;Bostwick 2000;Uozaki et al. 2000). In contrast, a higher expression is associated with a longer OS in endometrium, oesophageal cancer and malignant fibrous histiocytoma (Geisler et al. 1999;Nakajima et al. 2002;Têtu et al. 1992). These results indicate that HSP27 might serve as a reliable biomarker in human cancer but must be considered separately for every tumour entity (Ciocca and Calderwood 2005).
Only a few clinical studies are available which investigated the potential of HSP27 and pHSP27 to predict survival in PDAC. However, the results are inconsistent or even contradictory (Schäfer et al. 2012;Kawano et al. 2018;Tsiaousidou et al. 2013;Okuno et al. 2016). Therefore, it is still questionable whether HSP27 could be used as a biomarker in PDAC.
Furthermore, weak points of these published studies are reduced numbers of patients and lack of clinical data. Since almost all patients with PDAC receive adjuvant chemotherapy, survival could also be influenced by chemotherapeutic resistance. HSP27 and pHSP27 are shown to be involved in tumour suppression and resistance to chemotherapy with gemcitabine in pancreatic cancer (Guo et al. 2015;Mori-Iwamoto et al. 2007;Kuramitsu et al. 2012;Liu et al. 2012;Taba et al. 2011). pHSP27, which is induced by gemcitabine, is reported to play an important role in the suppression of cancer cell growth in pancreatic cancer cell lines (Taba et al. 2010;Nakashima et al. 2011). However, it is well known that adjuvant therapy with modified FOLFIRINOX results in a longer OS when compared with gemcitabine (Conroy et al. 2018;Yang et al. 2019). Recent evidence indicates that HSP27 could take a significant role in treatment response to mFOLFIRINOX in PDAC. As the mFOLFIRINOX regiment consists of oxaliplatin, irinotecan, 5-fluorouracil and leucovorin, previous studies showed a higher sensitivity for irinotecan and 5-fluorouracil in colon cancer when HSP27 is suppressed (Choi et al. 2007;Hayashi et al. 2012;Shamada et al. 2018). In addition, the inhibition of HSP90 improves the efficacy of oxaliplatin in p53-deficient colon cancer cells (Moser et al. 2007).
In the present study, both HSP27 and pHSP27 were analysed in a patient population diagnosed with PDAC and its expression was correlated with clinicopathological data including OS and DFS.

Ethics approval
All patients' data were fully anonymised, and the study was performed, according to the standards set in the Declaration of Helsinki 1975. The tumour tissue used was leftover material that had initially been collected for diagnostic purposes. All diagnostic procedures have already been fully completed when samples were retrieved for the study. The study was approved by the Ethics Committee of the Medical Chamber ("Ärztekammer"), Hamburg, Germany (approval number PV5510).

Patients
A total of 106 patients (female, n = 50; male, n = 56; median age, 67.2 years [IQR 56.9-76.4 years]) who had been diagnosed with PDAC and went through surgery between 2010 and 2018 were included. Patients' demographic and clinicopathological characteristics are shown in Table 2. The diagnosis was histologically confirmed and TNM classification was assessed according to AJCC 7th edition. The R-status was obtained pathologically via circumferential resection margin. All patients had a follow-up either until death (n = 76) or until their most recent contact (n = 28) on August 31, 2019.

Methods of evaluation
An immunoreactive score (IRS) was implemented for the evaluation of the HSP27 expression based on the intensity and quantity of immune staining in pancreatic cancer cells. The IRS score was applied as first described by Kaemmerer et al. (2012) and Remmele and Stegner (1987). The intensity of staining was graded as negative (0), mild (1), moderate (2) and intense (3). The percentage of positive cells was evaluated as 0 (no positive cells), 1 (< 10% positive cells), 2 (10-50% positive cells), 3 (51-80% positive cells) and 4 (> 80% positive cells). The IRS score was obtained by multiplying these two individual scores. As a result, every tissue sample was classified into negative (IRS points 0-1), weak (2-3), mild (4-8) or strong (9-12). Breast carcinoma was used as positive control and brain tissue as negative control. The evaluation of the protein expression was performed by two independent reviewers without knowledge of the patient characteristics.

Statistical analysis
Differences in continuous variables were analysed with Mann-Whitney U test and differences in proportions with chi-square-test or Fisher exact test. DFS and OS was analysed using the Kaplan-Meier method. Univariate and multivariate Cox proportional hazards models were used to assess the effects of variables on OS and DFS. A two-sided p-value less than 0.05 was considered as statistically significant. All analyses were performed using SPSS Inc. (Chicago, IL, USA).

Immunohistochemical analysis in PDAC and normal tissue
A positive IRS score of HSP27 with various interindividual intensity was found in the cytoplasm of pancreatic cancer cells in 65 patients (61.3%). Nuclear staining was not detected. Expression of HSP27 was classified into four grades according to the IRS score (negative, mild, moderate and strongly positive; Table 1). 23 patients (21.7%) had a strongly positive staining. We found a significant difference of HSP27 expression between cancer cells and normal pancreatic tissue (p < 0.001). In 79 (74.5%) of the corresponding 106 normal pancreatic tissues, HSP27 expression was positive whereof 43 (40.6%) showed a strongly positive staining. In contrast, the majority of patients had a negative IRS score for pHSP27-Ser15 (82 patients, 77.4%) and pHSP27-Ser78 (69 patients, 65.1%). Of the corresponding normal pancreatic tissue, 67 patients (63.2%) were pHSP27-Ser15-negative (p = 0.012) and 52 patients (49.1%) pHSP27-Ser78-negative (p > 0.001). However, only 48 patients (45.3%) had a negative IRS score for pHSP27-Ser82 in pancreatic cancer cells with 35 patients (33.0%) being negative in corresponding normal pancreatic tissue (p > 0.001).
Furthermore, a lower IRS score for HSP27 was significantly correlated with a bigger tumour size (p = 0.044), a less differentiated tumour (p = 0.007) and lymphatic invasion (p = 0.013). The group of patients with a positive IRS score Table 2 Association between IRS score of unphosphorylated and phosphorylated HSP27 and clinicopathological features of the study population IRS score is classified as 0 (negative), 1 (positive, weak expression), 2 (positive, mild expression) and 3 (positive, strong expression). All statistically significant variables are highlighted in bold. Statistical analysis: Chi-square test a Data not available for 22 patients (20.7%) Characteristic n = 106 HSP27 pHSP27-Ser15 pHSP27-Ser78 pHSP27-Ser82 was more likely to achieve a negative resection margin in surgery (p = 0.003). Immunohistochemical staining of pHSP (Ser-15, Ser-78 and Ser-82) showed a significantly higher IRS score for pHSP27-Ser82 compared to the other sites Ser-78 and Ser-15 in all patients (p < 0.001). When correlating the expression of each Serin-site of pHSP27 with clinicopathological characteristics, we observed a significantly lower pHSP-Ser82 expression in patients with a higher pathological T-stage (p = 0.025). In addition, the pHSP27-Ser82 expression was significantly associated with a negative resection margin and the pHSP27-Ser78 expression correlated with the tumour localization (p = 0.016, Table 2). We could not find a significant correlation between the pHSP27-Ser15 expression and any clinicopathological feature.

Expression of HSP27 in liver metastasis
26 PDAC patients with synchronous liver metastasis were analysed. Liver metastases as the corresponding primaries showed a negative or weakly positive IRS score for HSP27 expression. Most importantly, we observed similar HSP27 expression in the metastases compared to the primary tumours (p = 0.821). None of the examined metastases showed a mildly or strongly positive expression. Similar results were found for all forms of pHSP27.

HSP27 and long-term outcome
On August 31, 2019, 28 patients were still alive. In 77 patients, data regarding time and type of recurrence were available.
Using Kaplan-Meier analysis, we found a significant correlation between the HSP27 expression and outcome of the patients. Patients with a mildly or strongly positive HSP27 tumour had a significant better outcome regarding the OS (Fig. 1a, p < 0.0001) and DFS (Fig. 2a, p = 0.006) than those with a negative or weakly positive expression.
The median OS of patients with a mildly and strongly positive expression was 28.4 months and 28.0 months, respectively. In contrast, patients with HSP27-negative tumours had a median OS of 10.0 months. Nearly similar results were observed for the DFS with a median of 9.0 months in HSP27-negative tumours and 7.0 months in patients with weakly positive HSP27 expression. The median DFS was significantly longer in patients with mildly positive (25.0 months) and strongly positive (19.0 months) HSP27 expression. We could not observe significant differences for any phosphorylated form of HSP27 (Fig. 1b-d,  Fig. 2b-d).
We performed multivariate analysis using Cox proportional hazards model to investigate whether the expression of HSP27 or pHSP27 was an independent factor for the outcome after surgery. The results showed that only HSP27 functions as an independent marker for OS (p = 0.029, Table 3) and DFS (p = 0.015, Table 4). Furthermore, the analysis revealed a significant impact of the tumour pathological stage, resection margin, and vascular invasion for the DFS (Table 4). Interestingly, the patients with HSP27negative tumours had mainly a different type of recurrence compared to those patients with a positive expression. The group with no expression of HSP27 and a significant shorter DFS formed liver metastasis as type of recurrence in 80.0% of the cases compared to 53.3% in the patients group with a positive HSP27 expression (p = 0.029).

HSP27 as a predictor for gemcitabine sensitivity
We created a subpopulation of 66 resected, non-metastasised and gemcitabine-treated patients. In this patients group, we investigated the influence of the HSP27 expression on the OS (Fig. 3a-d). Kaplan-Meier analysis revealed a significant better outcome of patients with a mildly or strongly positive HSP27 expression compared to those with a negative or weakly positive IRS score (p = 0.001). In addition, we applied Kaplan-Meier analysis for all forms of pHSP27 (Ser-15, Ser-78 and Ser-82) but could not find any significant differences (Fig. 3b, c).

Discussion
Previous studies investigated the potential of HSP27 and pHSP27 as a prognostic biomarker in several tumour entities and found different results (Love and King 1994;Têtu et al. 1995;Eto et al. 2016;Elpek et al. 2003;Takeno et al. 2001;Uozaki et al. 2000). Therefore, HSP27 and pHSP27 expression and their role as a prognostic biomarker should be investigated separately for each cancer type. Particularly

3
The aforementioned study by Schäfer et al. (2012) found a significant longer OS in 86 patients with a higher expression of unphosphorylated form of HSP27. In addition, Okuno et al. (2016) correlated a higher pHSP27-Ser82 expression with a longer OS in 49 patients. Our study, does confirm the findings of the study by Schäfer et al. as we found a significant longer OS in patients with mildly or strongly positive HSP27 expression. However, we could not find any significant correlation between the expression of pHSP27 and outcome of the patients. Furthermore, our study revealed a significant shorter DFS of patients with negative or weakly positive expression of the unphosphorylated HSP27 and multivariate analyses determined the HSP27 expression as an independent factor for DFS. Taking a closer look at the patients group suffering from a recurrence shows that patients with HSP27-negative tumours had a significantly higher incidence of liver metastases as a type of recurrence. To our knowledge, no study ever before had investigated the link between HSP27 and time as well as type of recurrence.
Therefore, the HSP27 expression seems to have the potential to predict not only the long-term outcome regarding OS and DFS but also the type of recurrence. Combining these results with those of Schäfer et al. concerning the Fig. 1 Kaplan-Meier curve of overall survival for HSP27 and pHSP27 expression. a HSP27: Strongly and mildly positive expression of HSP27 predicts a favourable OS compared to a negative or weakly positive protein expression (log-rank p value < 0.0001). The median OS was 28.2 months in the group with mildly and strongly positive expression. In contrast, the patients group with negative and weakly positive expression had a median OS of 11.0 months. b pHSP27-Ser15: The median OS was 16.0 months in patients with a negative expression of pHSP27-Ser15, as compared with 27.4 months for a weakly positive expression and 31.0 months for a mildly positive expression (log-rank p value: 0.318). c pHSP27-Ser78: No significant difference was found for the OS of pHSP27-Ser78 expression with a comparable median survival (negative: 20.4 months, weakly positive: 25.0 months, mildly positive: 24.6 months, strongly positive: 23.4 months, log-rank p value: 0.879). d pHSP27-Ser82: The median OS was 24.8 months in patients with a negative expression, as compared with 24.2 months for a strongly positive expression with a nonsignificant log rank p value of 0.403 findings for the OS it is obvious that tumours with a lower HSP27 expression are accompanied with a poor outcome of patients with PDAC.
One reason for a longer survival of patients with HSP27positive tumours could be a higher sensitivity against adjuvant chemotherapy with gemcitabine. Mori-Iwamato et al. (2007) investigated HSP27 as one of the target proteins of gemcitabine using proteomic analysis. Furthermore, there is evidence for pancreatic cancer cell lines that overexpression of HSP27 enhanced the sensitivity towards gemcitabine (Guo et al. 2015). While Guo et al. found the unphosphorylated form of HSP27 as an essential component for gemcitabine sensitivity, another study by Kang et al. (2015) described the ratio of phosphorylated to non-phosphorylated HSP27 as more important than the cellular level of HSP27 itself. In addition, Nakashima et al. (2011) suggest the phosphorylation status of HSP27 as an essential determinant for gemcitabine-induced suppression Fig. 2 Correlation of HSP27 expression with disease-free survival. a HSP27: The median DFS was significantly longer in patients with a strongly (19.0 months) and mildly (25.0 months) positive expression when compared with a negative (9.0 months) and weakly positive (7.0 months) HSP27 expression (log-rank p value: 0.006). b pHSP27-Ser15: The median DFS was 9.4 months in patients with a negative as well as mildly positive expression of pHSP27-Ser15. Furthermore, the patients group with a weakly positive expression had a median DFS of 21.0 months (log-rank p value: 0.489). c pHSP27-Ser78: The expression of pHSP27-Ser78 is no predictor for the DFS with a comparable median DFS for all expression groups (negative: 10.0 months, weakly positive: 5.8 months, mildly positive: 11.0 months, strongly positive, 10.0 months, log-rank p value: 0.794) d pHSP27-Ser82: The median DFS was 11.0 months in patients with a negative expression, as compared with 10.2 months for a strongly positive expression with a non-significant log rank p value of 0.180 of pancreatic cancer cells. In our study, we investigated a subpopulation of 66 non-metastasised and resected patients which were treated with gemcitabine as adjuvant chemotherapy. The Kaplan-Meier analysis clearly shows a worse survival in patients with negative or weakly positive HSP27 expression compared to those with a mildly or strongly positive protein expression. Similar results could already be shown by Schäfer et al. (2012) where a negative HSP27 expression was associated with a shorter OS in gemcitabine-treated patients. Therefore, the influence of HSP27 towards the sensitivity of gemcitabine has a huge impact on the survival of resected patients. Even though Nakashima et al. and Kang et al. reported about the significance of phosphorylated forms of HSP27 regarding the effect of gemcitabine, our study does not show a significant correlation between pHSP27 and survival in the gemcitabine-treated subpopulation.
Another finding of our study was the downregulation of HSP27 in patients suffering from simultaneous metastases. The downregulation was also found in their relating liver metastases. These results indicate a possible role of HSP27 in the formation of liver metastases. Taken together with the correlation of a low HSP27 expression and liver metastases as a type of an early disease recurrence, we hypothesize that the downregulation of HSP27 could be one of the underlying mechanisms for the migration of pancreatic cancer cells in the liver. Previous studies already tried to investigate the specific impact of HSP27 in formation of metastasis of different cancer types. It is reported that HSP27 can increase transforming growth factor b (TGF-b)-stimulated MMP2 activity and therefore promotes cell invasion in prostate cancer (Xu et al. 2006). In addition, HSP27 modulates the expression of phosphatase and tensin homolog (PTEN) by which the PI3-kinase pathway is influenced in breast cancer (Cayado-Gutiérrez et al. 2013). Furthermore, Vahid et al. (2016) revealed an interaction between HSP27 and the prometastatic Hippo pathway in prostate, breast and lung cancer but the impact of this regulation on migration of metastasis needs to be elucidated. At this time, we do not understand the specific mechanisms how the activity and expression of HSP27 is regulated during the formation of metastasis and, consequently which targets are influenced (Wu et al. 2017). Nevertheless, our results show that HSP27 could act as an important factor in metastatic invasion which comes along with a poor outcome of patients with PDAC.

Conclusion
To conclude, our study demonstrates that a downregulation of HSP27 is associated with a poor outcome for patients with PDAC and therefore HSP27 could serve as a predictor for the OS and DFS. Furthermore, patients with stronger HSP27 expression seem to have a higher sensitivity for gemcitabine resulting in a longer OS. Beside this, patients with a negative HSP27 expression were either presenting simultaneous liver metastasis at time of surgery or developed mainly liver metastasis after surgical resection of the primary tumour within a short time. Therefore, the impact of HSP27 in formation of liver metastasis is obvious but the underlying mechanisms and involvement of specific pathways need to be elucidated in further studies.