Interplay between Caveolin-1 and body and tumor size affects clinical outcomes in breast cancer

Highlights • Breast tumor CAV1 levels were assessed in relation to phenotype and prognosis.• CAV1’s prognostic impact depended on anthropometric and tumor factors.• Stromal CAV1 predicted high recurrence risk in ‘low-risk’ patients.• Stromal CAV1 nearly doubled locoregional recurrence risk in breast cancer.• Cytoplasmic CAV1 was a marker for metachronous contralateral breast cancer.


Background
Caveolin-1 (CAV1) is a protein located in cholesterol-rich plasma membrane raft domains, defined as caveolae, and functions as a master regulator of cell signaling and transport [1,2]. CAV1 modulates many cellular functions, including nutrient and drug internalization, tumor-stroma interactions, hypoxia response, inflammation, epithelial-mesenchymal transition (EMT), and cell cycle regulation [1][2][3]. CAV1 and caveolae have been implicated in cancer cell metabolic regulation, including mitochondrial bioenergetics and fatty acid metabolism [4]. Many of these cellular functions, which are important drivers of breast cancer aggressiveness, lack established biomarkers and targets [3,5]. This fact highlights the need to investigate new potentially relevant biomarkers, such as CAV1.
CAV1 is also expressed in the stromal compartment and can be considered a marker of the tumor microenvironment (TME) [3]. The importance of TME for tumor development, metastasis, and treatment resistance is increasingly recognized [6]. The TME may also link the host and the tumor, particularly for adiposity and metabolic-related effects. Different CAV1 genotypes were associated with metabolic and obesity-related factors, such as waist circumference [7]. Challenges remain in elucidating the role of CAV1 in the TME comprised of several distinct cell types, including immune cells, cancer-associated fibroblasts (CAF), endothelial cells, and adipocytes with different functions [8]. The TME's function depends on interactions with the tumor cells [9,10]. Therefore, there is a need for new markers to investigate how the TME combined with established prognostic markers might modulate prognosis and treatment prediction.
Despite compelling in vitro data, the clinical significance of CAV1 remains unclear, and smaller studies investigating its prognostic impact in breast cancer have provided conflicting results [11][12][13]. Since breast cancer is a heterogeneous disease with different tumor biology, receptor expression, and outcomes [14][15][16], CAV1's prognostic impact may be context-dependent. Loss of CAV1 in stroma indicated transformation of surrounding tissue into TME through tumor cell and stroma interactions mediated by various signaling pathways, including TGFβ in early breast cancer [1,3]. Conversely, other works have reported that upregulation of CAV1 in TME promotes invasion and metastasis at a later stage in breast cancer development [17].
The interplay between tumor size and stromal CAV1 is yet to be explored. CAV1 is rarely expressed in luminal cells in normal breast tissue but rather in myoepithelial cells [18,19]. In breast cancer, CAV1 interacts with both HER2 and ER [20], suggesting that CAV1 plays different roles depending on receptor expression and localization. Thus, we investigated the role of CAV1 in signaling pathways and different cellular compartments of breast cancer in relation to clinical outcomes overall and different patient subgroups.

Cohort description
The BCblood cohort is a population-based cohort consisting of primary breast cancer patients operated at Skåne University Hospital in Lund. Ethical approval has been granted by the Lund University Ethics Committee (Dnr 75-02, Dnr 37-08, Dnr 658-09, and amendments). All participants provided written informed consent. The cohort has been described in detail elsewhere [21,22]. In short, patients diagnosed with a first breast cancer and no other malignancies within 10 years before inclusion, before surgery, were included. At inclusion, the participants answered a questionnaire regarding lifestyle and anthropometric measurements were taken by research nurses. Medical records and registries were used to obtain clinical data. Exclusion criteria were carcinoma in situ, preoperative treatment, and distant metastasis within 0.3 years of inclusion. A final number of 1018 patients included October 2002 to June 2012, remained (Fig. 1). The patients were followed until June 30,

2019.
Most patients included before November 2005 had missing HER2 status. HER2 status for patients with missing status was obtained from dual gene protein staining of HER2 on the TMAs and showed 97.7% agreement with pathological assessment [23]. Following the Swedish clinical routine, the ER and PR positivity cut-offs were >10% stained nuclei. Anthropometric factors were dichotomized as follows, BMI ≥25 kg/m 2 , waist ≥80 cm, and breast volume ≥850 ml [24].
TMA construction, staining, and evaluation CAV1 staining on TMAs was performed as previously described [21,25]. In brief, duplicate 1 mm cores were stained with a primary rabbit polyclonal CAV1 antibody (diluted 1:1000; ab2910, Abcam, Cambridge, UK). Two evaluators (V.I.C. and M.Ba.) blinded to clinical data evaluated CAV1 as previously described [25]. Both stainings of the cytoplasm of invasive tumor cells and stromal cells were dichotomized: cytoplasm as positive (1+/2+/3+) or negative (0) and stroma as strong (3+) or not strong (0/1+/2+). The stromal and cytoplasmic categories were combined to create a joint cytoplasmic/stromal CAV1 status with four categories: negative/not strong, negative/strong, positive/not strong, and positive/strong (representative images in Fig. 1). There were 19 patients with bilateral invasive tumors. Scoring of both tumors was possible for ten patients. The highest category was used for the four cases where the categories differed. Clinicopathological information was collected from the corresponding side.

TCGA dataset
Gene-level RNA-sequence and reverse-phase protein array (RPPA) data for CAV1, other proteins involved in key signaling pathways in breast cancer [26], and corresponding clinical data were obtained and processed, as previously described from a subcohort of 809 patients [27] of TCGA (https://portal.gdc.cancer.gov).

Statistical analyzes
For statistical analyzes, STATA® version 17.0 (StataCorp, College Station, TX, US) was used. Mann-Whitney U and Kruskal-Wallis tests were used to determine whether categories of cytoplasmic, stromal, and combined CAV1 status differed according to time between surgery and staining (TBSAS). Both stromal and combined CAV1 status were negatively associated with TBSAS (both P < 0.001). Therefore, TBSAS was always included as a covariate in multivariable models, including stromal or combined CAV1. Cytoplasmic, stromal, and combined CAV1 in relation to patient and tumor characteristics were analyzed using logistic regression (simple or multinomial) adjusted for age at inclusion (continuous). The negative/not strong CAV1 status was used as reference.
Three main endpoints were used for survival; any first breast cancer recurrence, distant metastasis, and death. During exploratory analyzes, further survival analyzes were conducted for locoregional recurrence, contralateral breast cancer, and non-distant recurrence. Breast cancerfree interval (BCFI), locoregional recurrence-free interval (LRFI), contralateral breast cancer-free interval (CBCFI), non-distant recurrence-free interval (NDRFI), and distant metastasis-free interval (DMFI) were defined as the time from inclusion until the first event. Patients without recurrences were censored at the time of the last follow-up before emigration, death, or last follow-up by June 30, 2019. Overall survival (OS) was defined as the time until death or last follow-up by June 30, 2019.
Univariable survival analyzes were conducted with Kaplan-Meier curves and Log-rank tests. Cox proportional hazards models were used for multivariable survival analyzes. Two models were used. Model 1 was adjusted for age and tumor characteristics. Model 2 included model 1 and was further adjusted for postoperative treatments before any event.  Schoenfeld's residuals were used to test the proportional hazard assumption for stromal, cytoplasmic, and combined CAV1 status in model 2 for the three main endpoints. Survival analyzes with CBCFI as an endpoint were restricted to patients without bilateral tumors.
To examine potential effect modifications by anthropometric factors and tumor characteristics on the associations between cytoplasmic, stromal, and combined CAV1 and BCFI, DMFI, and OS, formal two-way interaction analyzes were performed in model 1. For adjuvant treatments, formal two-way interaction analyzes were performed in model 2. The interaction analyzes for tamoxifen and aromatase inhibitors were restricted to patients with ER + tumors.

Missing data
Most (931, >99%) patients included in the survival analyzes had no missing values. However, some had missing data on BMI (n = 27) and HER2 (n = 34; 93% complete cases). Based on the pattern of missing data, the missing values were assumed to be 'missing at random'. Therefore, missing values for all variables in the survival analyzes, including BMI and HER2, were imputed using chained equations. Onehundred datasets, including 935 patients, were created from ten iterations each. Pooled results of the multiple imputation were used for the survival analyzes with further adjustment for BMI ≥25 kg/m 2 and HER2.
In the TCGA database, correlations were assessed using Spearman's rank (R s ) for all patients. No survival analyzes were conducted due to scarce follow-up (median follow-up 1.3 years for patients at risk). Also, CAV1 mRNA expression was investigated in a panel of 51 human breast cancer cell lines and the tumors of 1881 patients in the Gene expressionbased Outcome for Breast cancer (GOBO) platform [28,29]. The cell lines were classified according to Neve et al. [30]. Moreover, functional protein associations for CAV1 were explored in the STRING database [31].
Power calculations were performed using PS Power and Sample Size Calculation program version 3.1.16 (Vanderbilt University, TN, USA) [32]. For the power calculation, we assumed that with 855 patients (380 with positive cytoplasmic staining and 330 with strong stromal staining), a 10-year accrual time with an additional 7-year follow-up, true HRs of ≤0.785 or ≥1.299 and ≤0.781 or ≥1.306, respectively, would be detectable with 80% power and α of 0.05.
The REporting recommendations for Tumor MARKer prognostic studies (REMARK) were followed [33]. P-values were considered as the level of evidence against the null hypothesis, and all P-values are two-tailed. Nominal P-values are presented without adjustment for multiple testing due to the exploratory nature of this study [34]. Positive cytoplasmic CAV1 was associated with anthropometric factors related to a poor metabolic profile, such as large breast volumes (P adj = 0.016) and somewhat larger waists (P adj = 0.092). Strong stromal CAV1 was associated with younger age (P adj < 0.001) and more hormonal intrauterine device use (P adj = 0.039). Likewise, both negative/strong and positive/ strong CAV1 status were associated with younger age (both P adj ≤ 0.009). CAV1 positive/not strong showed stronger associations with large breast volumes (P adj = 0.003) and waists (P adj = 0.058) than positive cytoplasmic CAV1, irrespective of stromal CAV1.

Tumor characteristics in relation to CAV1 levels
Positive cytoplasmic CAV1 was associated with several unfavorable tumor characteristics: ER -, PR -, TNBC, histological grade III, and lower frequency of lobular-type tumors (all P adj ≤0.002) but was inversely associated with ALNI (P adj = 0.036). Conversely, strong stromal CAV1 was associated with several favorable tumor characteristics: ER + (P adj = 0.002), non-TNBC (P adj = 0.048), and lower frequency of histological grade III (P adj = 0.002). For combined CAV1 status, the positive/not strong group was associated with unfavorable tumor characteristics (ER -, PR -, TNBC, histological grade III, and lower frequency of lobulartype tumors; all P adj ≤ 0.001).

CAV1 signaling pathways in TCGA, STRING, and GOBO
CAV1 mRNA and protein levels were correlated (R s = 0.47) in TCGA. Both CAV1 mRNA and protein levels were positively correlated with genes and proteins associated with non-luminal subtypes (particularly basal), cell cycle control, inflammation, EMT, and the IGF/Insulin system, with correlations (R s ≥ 0.3; Fig. 2A). Analyzes in GOBO yielded similar results; CAV1 was associated with non-luminal subtypes (Fig. 2B, C). In GOBO, CAV1 mRNA was positively associated with gene modules related to lipid metabolism, stroma interactions, early response to growth factors, and basal pathways while being negatively associated with mitotic regulation (Fig. 2D). Functional networks in STRING showed CAV1's strong associations with tyrosine kinases, inflammatory markers, TGFβ pathway/EMT, and IGF/Insulin system (Fig. 2E).

CAV1 levels and prognosis
In the BCblood cohort, the patients were followed for up to 15 years. The median follow-up for the 668 patients still at risk was 9.0 years (interquartile range 7.0-11.1 years). There were 184 patients with any recurrence during the follow-up (116 with distant metastasis). Onehundred-seventy-six patients died during follow-up, of which 96 had a prior recurrence.
The hazards were proportional for cytoplasmic, stromal, and combined CAV1 status for the three main endpoints (all P s ≥ 0.1). Positive cytoplasmic CAV1 was weakly associated with higher recurrence risk in the univariable analysis but not after adjustment (Supplementary Fig. 1; Supplementary Table 2). Further, cytoplasmic CAV1 was not associated with DMFI. However, positive cytoplasmic CAV1 conferred a borderline lower risk of death in the multivariate analyzes (Supplementary Table 2). Neither stromal nor combined CAV1 status was associated with BCFI, DMFI, or OS, neither in the univariable nor multivariable analyzes.
The results indicate that strong stromal CAV1 is associated with increased recurrence risk among low-risk patients (Table 2).
Moreover, the impact of stromal and cytoplasmic CAV1 on DMFI was modified by TNBC (P interaction = 0.013) and HER2 status (P interaction = 0.034). There were also interactions between combined CAV1 status and tamoxifen-treatment in patients with ER + tumors with regards to both DMFI (P interaction = 0.022) and OS (P interaction = 0.005).
In the analyzes with multiple imputation and further adjustments for HER2 and BMI, the results remained essentially the same, except for the interaction between tamoxifen and combined CAV1 status for which the interaction became weaker (P interaction =0.081).

Discussion
We found that the prognostic impact of CAV1 was highly dependent on anthropometric factors associated with a poor metabolic profile and tumor characteristics. Strong stromal CAV1 was associated with a substantially worse prognosis only in patients with low BMI, small waist, and small tumors, factors that indicate low recurrence risk [35][36][37]. It has been shown that upregulation of CAV1 expression in stroma contributes to metastasis and invasion [17]; and that CAV1 is involved in cellular regulation and inflammation, lipid metabolism, and EMT through various pathways [1][2][3], which we could also confirm in TCGA, STRING, and GOBO. These pathways are considered markers of an active TME [6].
Moreover, CAV1 protein levels in TCGA were associated with two reactive breast cancer subgroups with an activated TME [14]. We hypothesize stromal CAV1 to be a marker of activated TME with a larger role in metastasis of less aggressive tumors. Larger body sizes have been associated with more aggressive tumor characteristics [24,38]. If so, this could explain the interactions between stromal CAV1 and anthropometric factors or tumor size, where a higher recurrence risk was observed in low-risk patients. Interestingly, in vitro and in vivo studies have shown that statins inhibit and lower CAV1 levels, hindering its oncogenic role [39,40]. A recent study reported post-diagnostic statin use to protect against distant but not locoregional recurrences [41], but CAV1's role in this setting is unknown.
CAV1's role in hypoxia might contribute to the observed increased locoregional recurrence risk associated with strong stromal CAV1 [3,40,42]. Hypoxia decreases the efficacy of radiotherapy used to achieve local control [43]. In vitro, hypoxia-induced factor 1 (HIF1) elevated CAV1 levels increased invasiveness through the NF-κB pathway [44]. CAV1 has been associated with radioresistance in both lung and prostate cancer [45,46]. Boxplots of CAV1 expression across subtypes as defined by Neve et al. [30]. and according to receptor status. (C) CAV1 expression (log2) across 51 individual breast cancer cell lines grouped according to Basal A (red), Basal B (gray), and Luminal (Blue) as defined by Neve et al. [30]. (D) CAV1 correlation within 1 881 breast tumors with eight gene modules (Stroma, Lipid, Immune response, mitotic checkpoint, mitotic progression, basal, early response, and steroid response) and corresponding correlation P-values from GOBO [28,29]. (E) STRING network analysis of the closest functional biological associations for CAV1 [30].
Cytoplasmic CAV1 was associated with a more malignant tumor phenotype, including a strong association with ER negativity, which is consistent with previous studies [11,12,18,19]. We found CAV1 in TCGA and GOBO to be associated with basal subtype, EMT, EGFR expression, and BRCAness, in line with others [18,19,47,48], all common in ERbreast cancer. Despite this, cytoplasmic CAV1 was borderline associated with longer OS but not with overall recurrence risk in multivariable models.
However, cytoplasmic CAV1 was associated with increased contralateral breast cancer risk. Provided the association between higher CAV1 and BRCA1 deficiency or basal phenotype [18,19,47], that are both risk factors for contralateral breast cancer [49], cytoplasmic CAV1 may be a proxy marker for these factors, representing a tumor phenotype that tends to occur in the contralateral breast. The role of CAV1 in contralateral breast cancer merits further study as there are few established risk factors [49] used to tailor preventative measures for patients at increased risk for contralateral disease.
Interestingly, CAV1 has been shown to modulate both trastuzumab uptake and HER2 expression in vitro [25,40]. We found effect modifications between HER2 and cytoplasmic CAV1 but also TNBC and stromal CAV1 on distant metastasis-risk. Unfortunately, the HER2 + and TNBC subgroups were too small to adequately assess the CAV1's impact on prognosis. Further investigation in larger HER2 + and TNBC cohorts is warranted to assess the prognostic impact of CAV1 in these subgroups. The impact of combined CAV1 status on overall survival was modified by tamoxifen-treatment. Others reported CAV1 co-expression with ER to potentiate downstream signaling, decreasing the efficacy of tamoxifen [50], and potentially explaining our findings. The crosstalk between CAV1 and non-genomic rapid ER signaling may also contribute to tamoxifen-resistance [20].
This study has several strengths, incorporating reliable clinicopathological and anthropometric data combined with tumor tissue from patients in a large population-based cohort considered representative for its catchment area [22]. Furthermore, quality-controlled data from TCGA, GOBO, and STRING databases were used [14,28,29,31]. The antibody used for IHC staining has been previously validated [42]. To date, there is no standardized way of assessing CAV1 IHC staining, reducing comparability between studies. The CAV1 staining was deemed to be homogenous, as previously reported [12,13].
In conclusion, the prognostic impact of CAV1 was highly dependent on its localization, anthropometric, and tumor factors. Stromal CAV1 predicted high recurrence risk in a group of supposedly 'low-risk' patients. In all patients, stromal CAV1 also doubled locoregional recurrence risk. Cytoplasmic CAV1 harbors potential as a new predictive marker for metachronous contralateral disease. If confirmed, CAV1 could be used as treatment target and for further risk-stratification.

Funding
The Swedish Cancer Society (CAN 20 0763), the Faculty of Medicine at Lund University, the Mrs Berta Kamprad Foundation, the South Swedish Health Care Region (Region Skåne ALF 40,620), and the Skåne University Hospital fund. AB holds a young researcher award from ALF (Region Skåne). HT was funded by Region Skåne ST-ALF. The funders had no role in study design and conduct of the study, data collection and analysis, data interpretation, or manuscript preparation and decision to submit the manuscript for publication.

Data availability
TCGA data can be downloaded from https://portal.gdc.cancer.gov. Data from GOBO is available at http://co.bmc.lu.se/gobo/ and data from STRING is available at https://string-db.org. Clinical data is not publicly available due to privacy laws. Questions regarding data can be directed to the corresponding author.

Declaration of Competing Interest
Ana Bosch is co-founder and board chair for SACRA therapeutics. She has received travel support from Roche, lecture fees from Eli-Lilly and has participated in Advisory Boards for Pfizer and Novartis. The other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Missing data for four patients for at least one variable. TBSAS -Time between surgery and staining.