Tumor Co-Expression of Progranulin and Sortilin Identies a Highly Malignant Subgroup of Breast Cancer

The growth factor progranulin has been implicated in numerous biological processes such as wound healing, inammation and progressive tumorigenesis. Both progranulin and its receptor sortilin are known to be highly expressed in subgroups of breast cancer and have been associated with various clinical properties including tamoxifen resistance. Recent data further suggest that progranulin via its receptor sortilin can drive breast cancer stem cell propagation as well as increased metastasis formation in an in vivo breast cancer xenograft model. In this retrospective biomarker analysis, we aimed to determine whether tumor co-expression of progranulin and sortilin has prognostic and treatment predictive values for breast cancer patients. Patients and Methods We explored how established analyzing receiving either years tamoxifen or with a median follow-up of Breast cancer-specic were and Cox Proportional Hazards to and sortilin in relation to clinical markers. of patients including progranulin and sortilin co-expression as well as regular prognostic parameters available for the study. Results showed that high progranulin and sortilin co-expression together with


Abstract Background
The growth factor progranulin has been implicated in numerous biological processes such as wound healing, in ammation and progressive tumorigenesis. Both progranulin and its receptor sortilin are known to be highly expressed in subgroups of breast cancer and have been associated with various clinical properties including tamoxifen resistance. Recent data further suggest that progranulin via its receptor sortilin can drive breast cancer stem cell propagation as well as increased metastasis formation in an in vivo breast cancer xenograft model. In this retrospective biomarker analysis, we aimed to determine whether tumor co-expression of progranulin and sortilin has prognostic and treatment predictive values for breast cancer patients.

Patients and Methods
We explored how co-expression of progranulin and sortilin was associated with established clinical markers by analyzing a tissue microarray including 560 randomized premenopausal breast cancer patients receiving either two years of tamoxifen treatment or no adjuvant treatment, with a median follow-up time of 28 years. Breast cancer-speci c survival were analyzed with Kaplan-Meier and Cox Proportional Hazards regression models to assess the prognostic and predictive value of progranulin and sortilin in relation to known clinical markers.

Results
Co-expression of progranulin and sortilin was observed in 20% of the breast cancer samples. In untreated patients, prognostic considerations could be detailed separately from treatment prediction and the high progranulin and sortilin expressing subgroup was signi cantly associated with breast cancer-speci c death in multivariable analyses (HR=2.188, CI: 1.317-3.637, p=0.003) along with tumor size, high tumor grade and lymph node positivity. When comparing the untreated patients with tamoxifen treated patients in the ERα positive subgroup, co-expression of progranulin and sortilin was not linked to tamoxifen resistance.

Conclusion
Data suggests that co-expression of progranulin and its receptor sortilin is a novel prognostic biomarker combination identifying a highly malignant subgroup of breast cancer. Importantly, this subpopulation could potentially be targeted with anti-sortilin based therapies.

Background
Breast cancer is the most common cancer in women worldwide. Even though early detection rates and existing therapies contributes to a slightly enhanced survival, many patients experience metastasis and tumor relapse. As a result, breast cancer remains the leading cause of cancer death among the female population(1). Treatment failure and cancer relapses are multifactorial but most likely in uenced by drug resistance and self-renewal properties of the small population of tumor cells termed cancer stem cells (CSCs) (2). Further, breast cancer is in many aspects a heterogeneous disease, including subtypes with diverse phenotypes and clinical behaviors (3)(4)(5). Each of the subtypes is responsive to different treatment regimes. Patients that express the estrogen receptor alpha (ERα) are treated with endocrine adjuvant therapy, such as tamoxifen or an aromatase inhibitor. However, despite that endocrine therapy improves the survival of the ERα positive patient group, many patients experience tumor relapse or therapy resistance (6). Consequently, there is an obvious need to further identify key mediators involved in breast cancer progression in order to optimally distinguish subgroups of breast cancer patients that will bene t from speci c treatments or having tumors with inherent aggressive properties.
Moreover, sortilin is a known progranulin binding receptor (30), highly expressed in breast cancer cell lines compared to non-tumorigenic breast epithelial cells (31). Sortilin has also been associated with increased metastatic potential in breast cancer (31), suggesting that the progranulin receptor could also be involved in breast cancer progression.
Here, we investigated tumor speci c expression and potential clinical associations for progranulin and its associated receptor sortilin, using tissue microarrays (TMAs) from a randomized tamoxifen trial including 560 premenopausal breast cancer patients with the aim to identify optimal biomarkers for breast cancer progression and prognosis that could potentially be targeted by anti-sortilin based therapy. This study was performed according to REMARK guidelines presented in Additional le 1.

Ethics Statement
The original study (SBII:2) was approved by the Ethics Committee at Lund and Linköping Universities, Sweden (Dnr LU 240-01 and for the continuation of the study: Dnr Linköping 01-134 and Dnr LU 2015-350). Randomization was performed by the Regional Oncological Centers and oral informed consent was registered for all patients. The data was analyzed anonymously.

Patients and tumor samples
This retrospective study includes an invasive breast cancer cohort consisting of 560 premenopausal patients enrolled in a randomized clinical trial from 1984-1991, where patients received either two years of tamoxifen treatment (n=275) or no systemic treatment (randomized untreated) (n=285). Each patient underwent surgery (either radical mastectomy or breast-conserving surgery) followed by radiotherapy and in a small number of cases adjuvant polychemotherapy (>2%). All patients were followed-up for breast cancer-speci c survival (BCSS) with up to 32 years of follow-up data. BCSS was calculated as the time from surgery of primary breast cancer to death from breast cancer. The median post-surgery follow-up time without a breast cancer-speci c death was 28.41 years. ERα status was determined by immunohistochemistry or enzyme immunoassay, progesterone receptor (PR) status by immunohistochemistry and human epidermal growth factor receptor 2 (HER2) status was determined by in situ hybridization and immunohistochemistry as previously described (32,33). A CONSORT diagram for the trial pro le is given in Additional le 2: Figure S1 and clinical and tumor characteristics for the two study groups are presented in Table 1. Additional details of the trial have been described previously (32,34,35).

Immunohistochemistry
Representative tumor areas of formalin-xed and para n-embedded tissue material was collected from 444 of the 560 patients and selected for TMA construction and sectioned, followed by depara nization and rehydration as previously described (34). Progranulin and sortilin expression were determined by immunohistochemistry using an Autostainer LINK 48 and the Envision FLEX+ detection system (DAKO).

Scoring
Scoring was performed independently by pathologists (Landberg) and a trained breast cancer surgeon (Rafnsdóttir) without knowledge of pathological or clinical data. The immunostaining scoring for progranulin and sortilin were implemented using an Allred scoring system, ranging from 1 (no/low staining), 2 (low/intermediate staining), 3 (intermediate/high staining) to 4 (high staining).

Statistical analysis
All statistical calculations and modelling were performed in SPSS software version 25 (SPSS, Chicago, IL), GraphPad Prism version 7.00 (GraphPad Software, San Diego, CA) or RStudio version 3.6.2 (packages stats, ggplot2, rms, survival and survminer). Spearman´s rank-order Correlation Coe cient was used to test the signi cance of the association between progranulin and sortilin scoring. The relationship between progranulin and sortilin scoring and various parameters were analyzed using Pearson´s Chisquare test for categorical variables and Kruskal-Wallis (or Man-Whitney U) test for continuous variables. Kaplan-Meier curves were used to estimate BCSS, and the log-rank test was used to compare BCSS among different staining and treatments, as well as to calculate Hazard Ratio (HR) and 95% Con dence Interval (CI) in these groups. Univariate and multivariable analysis was performed using Cox proportional hazard model for relative risk estimation of different variables, including tumor grade, size, age, lymph node status and ERα status to compare BCSS among different treatment groups. For univariate and multivariable analysis, HR and 95% CI were calculated. Performance of the multivariable models were measured using the concordance index (C-index) and the proportional hazards assumption was tested by Schoenfeld residuals. A 10-fold cross-validation, repeated 100 times, were performed to validate the cohort and estimate the prediction accuracy of the tted model. All p-values correspond to two-sided tests, and p-values of < 0.05 were considered statistically signi cant.

Correlation between progranulin and sortilin expression and clinicopathological parameters
In order to validate the potential prognostic as well as treatment predictive value of progranulin and sortilin tumor expression, we analyzed 444 breast cancer samples arranged in TMAs that were successfully stained for progranulin and sortilin using immunohistochemistry (IHC). Clinicopathological and molecular parameters included in the study are summarized in Table 1. The median age of the patients was 45 years (range 25-57) and the median follow-up period was 28.41 years. At the last followup, 206 (46.4%) of the 444 patients analyzed had died of breast cancer. To predict the accuracy of the multivariable model, a 10-fold cross-validation, repeated 100 times, were performed and demonstrated equivalent results (C-index: 0.642 for all patients in the full model and mean C-index of the repeated testsets: 0.616). For the cohort studied, there was a signi cantly increased BCSS for tamoxifen treated patients having ERα positive cancer (p=0.031, n=384) (see Additional le 2: Figure S2).
Next, we investigated how progranulin and sortilin protein expression were associated with established clinicopathological parameters. In support for a biological association, progranulin and sortilin protein expression correlated signi cantly (r=0.112, p=0.026) (see Additional le 2: Table S1). Further, progranulin tumor expression was signi cantly linked to histological grade (p<0.001), where patients with high-grade tumors showed high progranulin expression as well as Ki67 (p=0.001) and the hypoxic marker hypoxiainducible factor 1-alpha (HIF1α) (p=0.002) (see Additional le 2: Table S1). There was also a signi cant association between progranulin and ERα status (p<0.001) as well as PR status (p=0.001) (see Additional le 2: Table S1), where ERα positive tumors tend to have lower progranulin expression, which is consistent with previous ndings (16). In addition, sortilin expression was signi cantly associated with ERα, where ERα positive tumors tend to have higher sortilin (p=0.004) and PR expression (p<0.001).
For the analysis of progranulin and sortilin expression in relation to BCSS, we initially concentrated on the randomized untreated patients in order to obtain prognostic information not affected by adjuvant tamoxifen treatment (all univariate data Table 2, left). Interestingly, the double high progranulin and sortilin group was signi cantly different from the remaining subgroups and also associated with worse outcome, as illustrated in Figure 2A (p=0.003, n=206). When indicating all four subgroups, the double high subgroup separated signi cantly from the two subgroups of low progranulin expression (p=0.021 and p=0.005) whereas there was a non-signi cant trend for a difference between the double high group and progranulin high group with low sortilin expression (p=0.170) ( Figure 2B). In order to clarify the signi cance of adding sortilin expression to progranulin, we performed multivariable Cox Proportional Hazard (CPH) regression analyses, only analyzing patients with high progranulin tumor expression. In support for an important additive function for sortilin in the progranulin high patient group, high tumor tissue expression of sortilin was signi cantly linked to BCSS (HR=3.013, 95% CI: 1.219-7.448, p=0.017) together with lymph node (LN) positivity (HR=3.854, 95% CI: 1.666-8.919, p=0.002) and tumor size (HR=1.089, 95% CI: 1.037-1.143, p=0.001) (C-index: 0.701) (see Additional le 2: Table S4).
The univariate analysis (Table 2), linking the double high progranulin and sortilin to reduced BCSS in untreated patients was further validated by multivariable CPH regression analysis in the two subsets of patients including progranulin and sortilin co-expression as well as regular prognostic parameters available for the study. Results showed that high progranulin and sortilin co-expression together with grade, tumor size and LN status were identi ed as signi cant risk factors for BCSS (double high: HR=2.188, 95% CI: 1.317-3.637, p=0.003, high grade: HR=1.737, 95% CI: 1.054-2.860, p=0.030, LN positivity; HR=2.250, 95% CI: 1.348-3.758, p=0.002, respectively) (C-index: 0.667) ( Table 2, right). Next, we included all patients available within the randomized study in order to increase the statistical power of the data. In this extended patient material, including tamoxifen treated patients, high co-expression of progranulin and sortilin was signi cantly linked to BCSS (p=0.003, n=395) as illustrated in Additional le 2: Figure S3. In addition, univariate and multivariable CPH analysis on all patients revealed comparable results as for the untreated patients (C-index: 0.642) (see Additional le 2: Table S5). Here, endocrine treatment with tamoxifen was also identi ed as an independent prognostic variable (HR=0.710, 95% CI: 0.517-0.974, p=0.034). Interestingly, the double high group was not associated with any of the established clinicopathological parameters, including grade (p=0.063) and Ki67 (p=0.066) (see Additional le 2: Table S6).

High tumor co-expression of progranulin and sortilin was not associated with tamoxifen resistance
Since the analyzed cohort include randomized untreated and tamoxifen treated patients we could de ne a potential tamoxifen response or resistance in the subgroup of patients with high co-expression of progranulin and sortilin. These analyses were restricted to patients with ERα positive breast cancer where high tumor co-expression of progranulin and sortilin demonstrated a signi cantly worse BCSS compared to mixed groups (p=0.005, n=279) ( Figure 3A) similar to all samples described above. Multivariable analysis revealed that the double high group (HR=1.980, 95% CI: 1.308-2.996, p=0.001) as well as grade (HR=1.612, 95% CI: 1.086-2.394, p=0.018) and HER2 positivity (HR=1.716, 95% CI: 1.027-2.867, p=0.039) were signi cantly associated with reduced BCSS in ERα positive breast cancer and that tamoxifen treatment signi cantly improved ERα positive BCSS (HR=0.628, 95% CI: 0.431-0.915, p=0.015) (C-index: 0.637) (see Additional le 2: Table S7). When analyzing the response to tamoxifen treatment, the ERα positive patients with double high expression revealed no signi cant improvement in BCSS comparing untreated patients with the tamoxifen treated group (p=0.231, n=55) ( Figure 3B) in contrast to the remaining group with mixed progranulin and sortilin expression (p=0.046, n=224) ( Figure 3C). This potential difference in tamoxifen response was nevertheless not signi cant in an interaction analysis (p=0.971), suggesting that despite the lack of signi cant response to tamoxifen treatment in univariate analysis, the double high group was not resistant to tamoxifen treatment. Altogether, these results suggest that high co-expression of progranulin and sortilin recognizes an ERα positive patient group that could bene t from complementary therapy, targeting sortilin.

Discussion
Despite valuable traditional cancer therapies, many breast cancer patients experience relapse and therapy resistance. Thus, it is vital to continue to search for mediators driving tumor progression as well as identify biomarkers that better predict high-risk patients with a more aggressive clinical behavior as well as potential resistance to various therapies. The interest in progranulin has emerged over the last years, with publications demonstrating an overexpression of progranulin in different cancer types and associations with poor prognosis and survival (9,12,(21)(22)(23)(24)(25)(26)(27)(28). Further, the progranulin receptor sortilin has been linked to breast cancer aggressiveness as well as being expressed in other types of cancer, such as prostate and ovarian cancer (31,36,37). Recent studies from our group have emphasized the stem cell propagating effect by progranulin through its receptor sortilin (7), indicating that this pathway could be central in mediating CSC properties during tumor progression. The existence of a targetable receptor further suggest that future cancer therapies could be developed, selectively targeting CSC propagation via sortilin.
Here, we analyzed the expression levels of both progranulin and its receptor sortilin in a large and unique randomized clinical trial with long-term follow-up in order to clarify if tumor co-expression of the activator and the receptor de ne any speci c breast cancer type in relation to clinical aggressiveness. The results indeed showed that high co-expression of progranulin and sortilin could be detected in 20% of the patients and was associated with decreased BCSS. In support for an important function of progranulin and sortilin activation in breast cancer progression, multivariable regression analysis identi ed high coexpression of progranulin and sortilin, as well as histological grade and lymph node status as independent risk factors.
Previous reports have associated high progranulin expression with ERα negative patients(16) as well as a predictive marker for recurrence in ERα positive breast cancer (29). We recently showed that progranulin secretion increased in ERα positive breast cancer when cells were subjected to hypoxia, whereas ERα negative breast cancer cells had constitutive high secretion of progranulin (7). Here we observed a signi cant link between progranulin and HIF1α, where tumors with high expression of HIF1α tend to express high progranulin. The positive link of progranulin expression with HIF1α suggests a hypoxic in uence on progranulin expression, which is in line with previous published data (7). Further, we observed that high progranulin expression tend to associate with ERα negative status. In contrast, high sortilin expression was associated with ERα positive tumors, which suggests that even though progranulin associated with sortilin, their respective link to ERα are different. Although, the clinical relevance of hypoxic driven progranulin induced CSC propagation in different breast cancer subtypes needs to be studied further.
Sortilin has previously been associated with breast cancer aggressiveness and contributes to tumor cell adhesion and invasion (38). We recently published that a small molecule inhibitor of sortilin (AF38469) (39) block progranulin induced breast cancer progression in vivo (7) . In this study, orally administration of AF38469 signi cantly reduced the development of metastasis, which suggest that sortilin may function as a therapeutic target in breast cancer.
Here, in this cohort, high sortilin tumor expression on its own demonstrated no signi cant reduction of BCSS in either all patients or in the randomized control group (a Kaplan-Meier plot of the sortilin expression is shown in Additional le 3). However, the combination of high tumor co-expression of progranulin and sortilin demonstrated a signi cantly worse BCSS. Importantly, multivariable analysis revealed that when analyzing the progranulin high subgroup separately, high sortilin expression was identi ed as a signi cant prognostic variate linked to worse BCSS together with lymph node positivity and tumor size. This support the fact that sortilin add prognostic information when combined with progranulin.
Current treatment for patients with ERα positive tumors includes endocrine therapy such as tamoxifen and a previous report described progranulin to be associated with resistance towards tamoxifen therapy (8). Further, another report suggested that progranulin levels predicted recurrence in patients with hormone receptor positive breast cancer during tamoxifen treatment (28). Here, multivariable interaction analysis identi ed that the double progranulin/sortilin high ERα positive group was not resistance to tamoxifen treatment, even though univariate analysis demonstrated no signi cant improvement in BCSS in the tamoxifen treated group. Notably, the tamoxifen treatment in this cohort is only two years of adjuvant treatment and BCSS may also have been affected by later therapies at disease recurrence.

Conclusion
In conclusion, we have shown that a combination of high progranulin and high sortilin tumor tissue expression de nes a novel and highly malignant subgroup of breast cancer patients that can be targeted by anti-sortilin based therapies.

Role of the Funder/Sponsor
The funders had no role in the design and conduct of the study; data collection and analysis; decision to publish, or preparation of the manuscript.

Additional Contributions
We thank Björn Andersson and Jari Martikainen from the Bioinformatics Core Facilities at the Sahlgrenska Academy for statistical/bioinformatics support.

Availability of data and materials
The dataset analyzed during the current study are available from the corresponding author on reasonable request.

Ethics approval and consent to participate
The original study (SBII:2) was approved by the Ethics Committee at Lund and Linköping Universities, Sweden (Dnr LU 240-01 and for the continuation of the study: Dnr Linköping 01-134 and Dnr LU 2015-350). Randomization was performed by the Regional Oncological Centers and oral informed consent was registered for all patients.

Consent for publication
All authors read and approved the nal manuscript.