Prediagnostic serum selenium levels in relation to breast cancer survival and tumor characteristics

Women with lower levels of serum selenium (Se) may have a worse survival in breast cancer than women with higher levels, despite no difference in incidence of the disease. Our study was conducted to test whether Se is associated with the aggressiveness of breast tumors. Both the risk of having a tumor characteristic associated with worse prognosis, as well as the overall and breast cancer‐specific mortality, were studied. We identified breast cancer cases and controls within the Malmö Diet and Cancer Study, a population‐based cohort with 17 035 women recruited between 1991 and 1996. Inclusion criteria were incident breast cancer. Exclusion criteria were carcinoma in situ and bilateral breast cancer. Controls were selected among breast cancer‐free women both from matching (n = 694) as well as randomization (n = 492). After exclusion, 1066 cases remained and were compared to controls regarding their prediagnostic serum Se levels and subsequent risk of having a certain tumor characteristic or intrinsic subtype. We also followed breast cancer patients regarding overall and breast cancer‐specific mortality, comparing different Se quartiles. No association between serum Se quartile and any tumor characteristic or intrinsic subtype was found. Lower overall mortality was found among women in the highest Se quartile compared to the lowest using an adjusted Cox proportional hazards model, hazard ratio 0.63 (95% confidence interval: 0.44‐0.89). Similar results were seen for breast cancer‐specific mortality, 0.60 (0.37‐0.98). The results of our study support that Se is associated with a lower mortality in breast cancer, not related to established prognostic factors.

increased breast cancer mortality among women in the lowest quartile of serum Se as well as Se intake, and thus a potential protective effect of Se might be seen in clinical outcome rather than in incidence. 4,5 Se is an essential mineral important in the anti-oxidation system and for thyroid hormone function, mediating its functions through selenoproteins, a group of proteins with incorporated Se. 6 Several possible mechanisms for how Se impacts breast cancer development and progression have been suggested. The selenoprotein glutathione peroxidase-1 (GPx-1), which has a role in anti-oxidation and DNA stability, could be the mediator of a protective effect. 7,8 Indeed, loss of heterozygosity in the GPx-1 gene is a common event in human breast cancer, and overexpression of GPx-1 has been shown to protect against chromosome damage in cell cultures. 9,10 However, the Se pathways are complex and another major selenoprotein, thioredoxin reductase (TRx), has on the contrary been suggested to support the growth of cancer cells. Inhibition of this protein has recently rendered promising results in animal studies on the use of TRx inhibition as a cancer therapy. 11 With these conflicting mechanisms in mind, the effect that Se might have on breast cancer prognosis is yet to be established.
Breast cancer should not be seen as a single entity. Tumor characteristics such as size and intrinsic subtype, together with clinical data such as lymph node positivity are essential factors to take into account when considering prognosis and treatment. [12][13][14] Our study is, to the best of our knowledge, the first one using prediagnostic serum Se levels to study the risk of developing breast cancer in relation to specific tumor characteristics or intrinsic subtype, as well as the first to study prediagnostic serum Se and breast cancer mortality. We hypothesized that higher Se levels would be associated with lower breast cancer mortality as well as a reduced risk of developing prognostically unfavorable tumors.

| Design
Our study applied a prospective design to follow-up survival among women diagnosed with breast cancer, as well as a nested case-control design to study the risk of specific breast tumor characteristics relevant to prognosis. This approach was followed within the populationbased cohort the Malmö Diet and Cancer Study (MDCS). The MDCS includes individuals born between 1923 and 1950 that were living in Malmö, Sweden, during the recruitment period, 1991 to 1996. A total of 17 035 women were included in the MDCS, representing 43% of eligible women. 15 The baseline examination included a questionnaire, blood sample collection and measurements of height and weight. 16

| Case and control selection
Information about breast cancer diagnosis was gained through linking MDCS participants to the Swedish Cancer Registry using their Swedish personal identity number. Excluding those with prevalent breast cancer at baseline, all women in the MDCS diagnosed with breast cancer for the first time from baseline up until December 31, 2013, were identified and eligible as cases in our study (n = 1186). An equal number of controls (n = 1186) were selected. The controls were selected using two different methods. One group of controls (n = 694) was selected based on a previous breast cancer study in the MDCS by Almquist et al that used incidence density matching, matching for age, menopausal status and time of inclusion. 17 All unique controls from that study that still remained free from breast cancer up until December 31, 2013, were included as controls in the present study. The remaining controls (n = 492) were randomized from the female population of the cardiovascular (CV) subcohort in the MDCS. The CV-subcohort was created to be an extension of the MDCS with extra baseline examinations, including extra blood sampling and ultrasound of the carotid arteries, and 50% of those invited to the MDCS between 1991 to 1994 were also invited to the CV-subcohort. 18,19 In total, 3531 women accepted the invitation, however, among those, not everyone completed the MDCS baseline examination and was thus never included in the MDCS cohort. Our randomization was performed among those 3363 women who were also among the 17 035 complete participants in the MDCS. 18,19 The reason for selecting our control group using the described methods was that future studies will investigate thyroid hormones (analyzed in the study by Almquist et al) and genetic data (available in the CV subcohort). 17 After including eligible cases and controls, a total of 120 cases were excluded due to bilateral breast cancer (n = 20) or carcinoma in situ (n = 100). Furthermore, 63 patients were excluded from risk analysis of tumor characteristics due to missing tumor material for pathology re-evaluation at the time of data collection. The inclusions and exclusions are visualized in Figure 1.

| Selenium analysis
Blood samples were taken at baseline and stored at −80 C. Serum Se was analyzed from these samples for all participants in the present study during October 2015 by ALS Scandinavia AB, Luleå, Sweden.

What's new?
While selenium may have little effect on the overall risk of breast cancer, its effect on breast cancer prognosis is yet to be established. Here, the authors studied the association between pre-diagnostic serum selenium level and breast cancer aggressiveness and mortality in 1,066 breast cancer patients compared to controls. They found better breast cancer-specific and overall survival among women with higher pre-diagnostic selenium levels. However, no correlation between selenium status and specific breast cancer characteristics or intrinsic subtype was found. Altogether, the findings suggest that selenium is an independent prognostic factor for mortality in breast cancer patients.
Single element standards traceable to the National Institute of Standards and Technology were used on an ICP-SFMS (Thermo Element 2). Every sample was diluted to 10 mL from 0.15 mL of serum by adding an alkali solution containing 0.1% NH 3 and 0.005% EDTA/ Triton-X. Reference samples were included in all batches (Seronorm; Sero AS, Norway, Lot 0608414). The inter-batch coefficient of variation was 0.03, and the limit of detection was 4 ng/mL.

| Baseline and patient characteristics
Descriptive data such as age as well as life-style and reproductive factors were collected from the baseline questionnaire. BMI was calculated from measured height and weight. Menopausal status was defined based on information from the questionnaire as described previously. 17 Data regarding lymph node status, distant metastasis, as well as oncological and surgical treatment were collected from medical records.

| Tumor characteristics
Information about tumor characteristics was collected in three different time periods. During all periods, tumor size, nodal status and information on metastasis were collected from medical records. Tumors from cases diagnosed from 1991 to December 31, 2004, were reevaluated regarding histological grade, and a tissue micro-array (TMA) was constructed to assess proliferation (Ki67), HER2− and hormone receptor status as described by Borgquist et al 20  In an additional analysis, the same methods as above were used to compare mortality between Se quartiles among the controls.
Adjustment of the Cox proportional hazards model was made for factors with a theoretical effect on survival or Se status in our cohort: age, BMI, smoking, year and the season the sample was taken.
Finally, to validate the intrinsic subtype variable, mortality was compared between different quartiles. Additional factors included in these analyses were age at baseline, age at diagnosis, BMI, lymph node status and tumor size.

| RESULTS
In Table 1, we present descriptive data among included and excluded women. The mean serum Se level was 91.3 ng/mL among the controls, 91.2 ng/mL among the breast cancer cases with invasive tumors with missing material, and 92.2 ng/mL among the rest of the cases.
The women with breast cancer were younger than the controls at baseline, more likely to have used oral contraceptives or HRT, and F I G U R E 2 Kaplan-Meier curves of survival comparing selenium quartiles more likely to be obese but less likely to be manual workers. In Table S1, we present descriptive data for the different Se quartiles.
The women with the highest Se levels (Q4) were older compared to those having the lowest Se levels (Q1). They were also more likely to have had their blood sample collected during the winter and to be postmenopausal, and were also less likely to be current smokers or manual workers. In total, 262 women had missing Se values and they were older, less educated, less likely to have used oral contraceptives and more likely to be obese.
In Table 2 In Figure 2A, the breast cancer-specific mortality is visualized in a Kaplan-Meier curve, showing a higher mortality in the lower quartiles compared to the higher quartiles. Results from unadjusted and adjusted Cox's regression analyses are presented in Tables 4 and S4.
There was a lower breast cancer-specific and overall mortality in Se  Figure 2B and   Table S6).
In sensitivity analyses when using the quartiles based only on controls, similar differences in risk and mortality were seen in all analyses as when using quartiles based on both cases and controls. The adjusted OR for invasive breast cancer in Q4 vs Q1 among cases and controls was 1.03 (0.79-1.35) and the HR for breast cancer-specific mortality among cases was 0.59 (0.36-0.98).

| DISCUSSION
In the present study, we found evidence of lower mortality in women who had the highest prediagnostic serum Se levels. This was seen for both breast cancer-specific and overall mortality. However, no correlation between Se status and specific breast cancer characteristics or intrinsic subtype was found, suggesting that Se might be an independent prognostic factor for mortality in breast cancer patients.
Our study is the largest to date examining serum Se and breast cancer prognosis, with 1066 included cases and 179 breast cancerspecific deaths. The only previous study on serum Se and breast cancer mortality is a case-control study by Lubinski et al, 5 including 546 cases with 58 deaths from breast cancer and serum Se levels after diagnosis. They found, similarly to us, that women with lower Se had worse survival. In three studies regarding dietary Se and breast cancer mortality, one found a strong correlation between a higher Se intake and a lower breast cancer-specific mortality while the other two studies reported weak evidence of such an association. 4,5,25,26 There are several potential biological mechanisms that may link anti-oxidative proteins. 7 GPx-1 has also been shown to increase DNA stability, although the exact pathway is not known. 10 Specific polymorphisms in the GPx-1 gene have been shown both to be more common in breast cancer tissue compared to normal breast tissue as well as to reduce GPx activity and the effect of Se supplementation. 9,28,29 Also, Se is important for normal thyroid function as well as for thyroid hormone conversion through the thyroid deiodinases. 30 Thyroid hormones as well as thyroid hormone receptors have been linked to the development and prognosis of breast cancer. [31][32][33][34] Thus, Se may play an indirect part in that pathway to affect breast cancer.
Additionally, selenium binding protein-1 is involved in cell-growth regulation and has been found to be correlated with poor survival in breast cancer. 35 All the above-mentioned mechanisms offer potential explanations for why Se could have a protective effect against breast cancer mortality among women with higher Se values. However, several of these mechanisms are not specific to breast cancer, which may explain why we also found the highest overall mortality among the controls with the lowest Se values in our study. One biological example is the loss of heterozygosity at a specific locus in the GPx-1 gene that is found in both lung cancer and breast cancer. 9,36 Indeed, previous epidemiologic data suggest a link between low Se and an increased mortality in other cancer types than breast cancer, for example, esophageal and gastric cancer. 37 Table S6).
In this present study, the Se levels were only measured once from a single blood sample, taken prediagnostically. The reported values could therefore have been affected by short-term Se intake prior to blood sampling, although some evidence supports that this is an adequate way of measuring Se status and that long-term ranking between individuals is maintained. 42,43 The fact that we used controls based on incidence density matching in another study, and added new by randomization, adds two considerations. First, the controls selected by incidence density matching are probably more similar to the cases regarding risk factors than anyone selected at random. Also, the randomization process was performed only on the CV subcohort, only including individuals recruited during 1991-1994 while the cases were selected from the whole recruitment period of the MDCS, 1991-1996. This could present a skewness in the data, for example, due to different sample storage times between cases and controls. However, all our analyses were adjusted for year of inclusion, correcting any potential difference. The finding that the controls were older than the cases at baseline should be commented on. One possible reason is that inclusion during later years was extended to younger women, and most controls were recruited during the first part of the study as they were selected from the CV cohort. 44 The controls were defined following the end of follow-up, and the higher age at baseline seen among the controls may be an effect of competing mortality since women who were older at baseline had a higher risk of dying from causes other than breast cancer.
However, age was adjusted for in the analyses and we do not believe that our selection of controls led to any strong bias.
Descriptive data on cases and controls are presented in Table 1.
The validity and completeness of the Swedish cancer registry, which was used in the present study to identify women with breast cancer, has been reported to be high and the Swedish cause of death registry has been reported as nearly complete. 45,46 Regarding the study population, only 43% of eligible women were included in the MDCS; however, the participants had the same demographic structure as those in a study carried out in the same population with a participation rate of 74.6%. 15 By comparing many prognostic factors we created a risk of producing a type 1 error. However, in the present study, this was not the case since we could not find any evidence at all that the prognostic factors differed between the Se quartiles. Regarding some tumor subgroups in our study, such as HER2+, distant metastasis at diagnosis and luminal B-like tumors, evidence of a difference in risk between quartiles could be masked due to poor statistical power for breast cancers with such characteristics. However, if there would be a true such difference, we found no evidence in our data to suggest that a difference in risk of such characteristics is the driver of the difference in mortality between the different selenium quartiles. However, a limitation to any observational study, including the present one, is the effect of unmeasured confounding. We hypothesis from a biological and epidemiological viewpoint that Se is protective regarding breast cancer, but if Se is truly a protective factor in the association, or just a result of unmeasured confounding, cannot be concluded from our study alone.
In Sweden, as in the rest of Europe, Se levels are relatively low compared to other countries such as the US. 47,48 Supplementation studies in China suggest that a plasma Se level of around 90 ng/mL is sufficient for full activity of selenoproteins, while lower Se leads to less activity of selenoproteins, and higher levels appear redundant. 49,50 The same study also concludes that those serum levels can be achieved in a population by an intake of 49-75 μg/day depending on mean weight of the population. 49 However, an optimal serum or intake level, or if full activity of selenoproteins is needed for good health, cannot be concluded from these studies alone and could vary depending on what effects and what side-effects are studied. Daily intake levels as low as 20 μg/day is sufficient to prevent the Se deficiency disease Keshan disease and the World Health Organization based their calculations regarding their recommendation of minimum daily requirement of Se on those levels. 51 In our study, quartiles 1 and 2 (≤90.5 ng/mL) thus theoretically represent women with unsaturated plasma Se proteins and quartiles 3 and 4 (>90.5 ng/mL) should represent women with full activity of plasma Se proteins. When looking at the results regarding mortality for the different quartiles with this in mind, there seems to be a close relation in the results between Q1 and Q2 as well as between Q3 and Q4. The best way of measuring Se status may be to use specific Se proteins, such as glutathione peroxidase-3 and selenoprotein P, instead of total Se since that also represents ingested selenomethionine. 52 However, total Se correlates well with those proteins, and as these specific proteins are more affected by storage time, we believe that assessing total Se is a valid and useful measurement. 53

| CONCLUSION
Lower overall and breast cancer-specific mortality was found among women in the highest Se quartile (≥100.01 ng/mL) as compared to the lowest quartile (≤81.00 ng/mL) in prediagnostically measured serum Se levels. There were no associations between Se levels and tumor characteristics, including intrinsic subtypes. This suggests that Se is associated with breast cancer survival, not related to classic prognostic factors.