Article Text

Download PDFPDF
Stem cell biomarker ALDH1A1 in breast cancer shows an association with prognosis and clinicopathological variables that is highly cut-off dependent
  1. Martin Sjöström1,
  2. Linda Hartman1,
  3. Gabriella Honeth1,
  4. Dorthe Grabau1,2,
  5. Per Malmström1,3,
  6. Cecilia Hegardt1,
  7. Mårten Fernö1,
  8. Emma Niméus1,4
  1. 1Department of Clinical Sciences Lund, Oncology and Pathology, Lund University, Lund, Sweden
  2. 2Division of Pathology, Skåne University Hospital, Lund, Sweden
  3. 3Division of Oncology and Radiation Physics, Skåne University Hospital, Lund, Sweden
  4. 4Division of Surgery, Skåne University Hospital, Lund, Sweden
  1. Correspondence to Dr Emma Niméus, Department of Clinical Sciences Lund, Oncology and Pathology, Medicon Village, By 406, Lund SE-223 81, Sweden; Emma.Nimeus{at}


Aims Aldehyde dehydrogenase family 1 member A1 (ALDH1A1) is a putative marker of breast cancer stem cells (CSCs) with prognostic implications when expressed in cancer or stroma. However, previous results are contradictory and we therefore aimed to further evaluate the impact of ALDH1A1 on distant disease-free survival (DDFS) and correlation with clinicopathological variables in breast cancer, specifically by evaluating different cut-offs.

Methods Two breast cancer cohorts (N=216 and N=210) were evaluated with immunohistochemistry for ALDH1A1 on tissue microarrays with three different cut-offs in cancer cells and in stromal cells. The association of ALDH1A1 with DDFS and other clinicopathological variables was assessed. As further validation, gene expression levels of ALDH1A1 and association with survival were analysed in one of the cohorts and a separate cohort.

Results ALDH1A1 expression in cancer cells was associated with either a better or a worse prognosis, depending on cut-off. Considering weakly stained cancer cells as positive, ALDH1A1+ was associated with a better prognosis in both cohorts. Considering only strongly stained cells as positive, ALDH1A1+ was associated with oestrogen receptor and progesterone receptor negativity in both cohorts and worse prognosis in one of the cohorts. Stromal ALDH1A1 staining was associated with improved DDFS in one cohort. Gene expression analysis showed that a high ALDH1A1 expression was associated with a better prognosis.

Conclusions ALDH1A1 is associated with DDFS and clinicopathological variables, both in cancer cells and stroma, but is highly cut-off dependent. Only the strongly ALDH1A1-stained cells show a more aggressive phenotype typical for CSCs.


Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


According to the predominant theory, cancer stem cells (CSCs) are tumour-initiating cells with extensive self-renewal and self-replicative capabilities, constituting a small proportion of all tumour cells in an individual cancer. CSCs are crucial for cancer initiation and required for the cancer for metastatic capacity,1 and may contribute to treatment resistance.2 The proportion of CSCs differs greatly between different cancer types and also within a cancer type.3 Given the remarkable heterogeneity of breast cancer,4 breast tumours may exhibit different patterns of CSCs with implications for prognosis or other clinically important parameters.5 Several markers have been proposed to identify CSCs in breast cancer, including the aldehyde dehydrogenase family 1 member 1 (ALDH1A1) and the CD44 high and C24 low phenotype (CD44+/CD24).6

Aldehyde dehydrogenase (ALDH) is an enzyme required for the conversion of retinol (vitamin A) to retinoic acid. It is a key regulator in retinoic acid signalling important for vertebrate development and is important for the stem cell abilities of haematopoietic stem cells.7 Several ALDH isoforms exist and mainly the isoform ALDH1A1 is used as a marker of breast cancer stem cells,8 although questioned,9 and ALDH activity is predictive of successful engraftment from xenotransplants.10 Several studies have shown a correlation between ALDH1A1 expression and prognosis when analysed in primary tumours and in lymph node metastases.8 ,11 ,12 A recent meta-analysis showed that ALDH1A1 expression is correlated with clinicopathological variables typical of a more aggressive disease and poor prognosis. These findings were robust and not dependent on cut-off value, study centre or antibody used.13 However, various cut-offs have been used and several studies have shown discordant results.14–17 One study needed additional information on CD44 expression to get significant correlation with prognosis.18 Thus, the use of ALDH1A1 as a marker for CSCs has been questioned.19–21

ALDH1A1 expression has also been described in both spindle-shaped and round stromal cells, of which some, but not all, are lymphocytes.22 Interestingly, ALDH1A1 staining in stroma has been associated with increased survival,16 ,23 but contradictory results have also been reported.12 ,24

In this study, we therefore analysed ALDH1A1 expression in both cancer cells and stromal cells in two cohorts of breast cancers. We specifically assessed the importance of different cut-offs of ALDH1A1 staining and found an association with clinicopathological variables and distant disease-free survival (DDFS) that was highly cut-off dependent. As a validation of the results, we showed gene expression levels to be highly correlated with protein expression and found a strong association between high ALDH1A1 mRNA levels and improved prognosis.

Materials and methods

Patients and tissue microarray

Two well-defined breast cancer patient cohorts were selected, which have been previously described.25 In brief, cohort 1 consisted of 273 premenopausal and postmenopausal patients with stage II breast cancer. The patients were selected from two randomised clinical trials.26 ,27 All patients were operated with modified radical mastectomy or breast-conserving surgery with axillary lymph node dissection between 1985 and 1994. After breast-conserving surgery, radiotherapy (50 Gy) was given locally or locoregionally if the disease had spread to axillary lymph nodes. All patients were treated with tamoxifen for 2 years, irrespective of oestrogen receptor (ER) status, but no other systemic therapy was given. In cohort 1, we were able to stain and score 216 tumours for ALDH1A1 expression. Cohort 2 consisted of 237 premenopausal women with node-negative breast cancer.28 All patients were operated between 1991 and 1994, and 50% received postoperative radiotherapy, but mainly no systemic adjuvant treatment (4% received endocrine therapy and 10% received chemotherapy). From cohort 2, we were able to stain and score 210 tumours. In one tumour sample, no stroma was found and thus 209 tumours were scored for stroma in cohort 2. The dropout for both cohorts was due to technical problems with staining or no cancer tissue present on the core section. Evaluation of other clinicopathological variables in these cohorts has been previously described,26–30 and characteristics of the included patients are given in table 1. Tissue microarrays (TMA) were prepared from 0.6 mm cores from paraffin-embedded blocks as previously described.31

Table 1

Patient and tumour characteristics of cohorts 1 and 2

Immunohistochemical staining of ALDH1A1

De-paraffinisation and pretreatment was conducted in PTLink (Dako), and staining was performed with EnVision visualisation kit (DAKO, K801021-2) in an Autostainer Plus (Dako). A purified mouse anti-ALDH1A1 antibody (clone 44/ALDH, BD Biosciences, cat no 611194) was used as primary antibody at 1:100 dilution. Counter staining was performed with Mayer's haematoxylin for contrast. Liver tissue was used as positive control and negative tumour cores as negative controls. The antibody was previously shown to specifically identify the ALDH1A1 isoform.32

Scoring of immunohistochemical staining of ALDH1A1

ALDH1A1 levels were evaluated under the supervision of a pathologist in one or two core sections from each tumour, both in cancer cells and in stromal cells, and the mean value was used. In cancer cells, percentage of stained cells (0–100) and staining intensity (4 levels: 0, negative; 1, weak; 2, moderate; and 3, strong) were evaluated. A histoscore (HS) was calculated by multiplying percentage of stained cells and staining intensity resulting in a value between 0 and 300. For further analysis, we used three cut-offs: (1) HS>10 (tumours with more than 4–10% stained cells, depending on intensity, were regarded as positive), (2) HS>0 (tumours with any positive staining was regarded as positive) and (3) strong staining (tumours with strong staining in any number of cancer cells were regarded as positive) (table 2). For stroma, we evaluated the intensity in spindle-shaped cells found inside a tumour area in four levels (0, negative; 1, weak; 2, moderate; and 3, strong). We did not evaluate percentage of stromal cells as most of the tumours contained a modest number of stromal cells and usually a majority showed the same staining intensity. For statistical reasons regarding group size, and prior to any analysis, stromal negative and weak staining was combined, resulting in three levels (0, negative or weak; 1, moderate; and 2, strong). The association of ALDH1A1 expression in cancer cells with clinicopathological variables in cohort 1 has previously been published for the cut-off strong staining.33 In the present study, the expression was re-evaluated as we wanted the two cohorts to be scored by the same investigator and as the previously reported results did not evaluate other cut-offs, stromal staining or association with prognosis. The two evaluations showed high concordance and accordingly did not change our main findings (data not shown).

Table 2

ALDH1A1 staining (numbers and percent) in cancer cells for three different cut-offs: (1) HS>10, (2) HS>0 and (3) strong staining in any number of cells

Analysis of gene expression data

For 67 of the tumours in cohort 1, we had gene expression levels (mRNA) available from cDNA microarray experiments as previously described.34 For evaluation of gene expression levels of ALDH1A1 in a larger cohort, we used the freely available online tool ‘Gene expression-based outcome for breast cancer online’ (GOBO) ( V. Briefly, it uses 11 publicly available gene expression data sets and plots Kaplan–Meier survival curves for selected genes. We used default settings with all patients (N=1881), three quantiles for stratification of patient groups, censoring at 10 years and distant metastasis-free survival (DMFS) as end point for gene symbol 212224_at (ALDH1A1).

Statistical analysis

DDFS, meaning time from randomisation (cohort 1) or diagnosis (cohort 2) to the first of the events distant recurrence or breast cancer death, was used as end point and visualised using the Kaplan–Meier method. The influence of ALDH1A1 on DDFS was tested using a log-rank test or log-rank test for trend (for categorisations with more than two levels). The Cox proportional hazards model was used for obtaining HRs and for multivariable modelling. Proportional hazards assumptions were checked both graphically and using Schoenfeld's test.36 Associations between ALDH1A1 expression and other variables were analysed using Pearson's χ2 test or Fisher's exact test when variables were binary; otherwise, linear regression with test for zero slope was used, equalling a χ2 test for trend. All p values correspond to two-sided tests. p Values <0.05 were considered statistically significant. All calculations were performed using Stata V.12.1 (StataCorp, College Station, Texas, USA).


Staining results and association of ALDH1A1 with clinicopathological variables

Staining was predominantly observed in the cytoplasm in both cancer cells and stromal cells (figure 1). Strong staining was usually observed in a minority of cancer cells, but weak and moderate staining could be observed in a majority of cancer cells (tables 2 and 3).

Table 3

ALDH1A1 staining in stromal cells

Figure 1

Representative images of aldehyde dehydrogenase family 1 member 1 (ALDH1A1) staining intensity of breast cancer cells and stromal cells. Bars are 70 μm. (A–D) Cancer cell staining (A, negative; B, weak; C, moderate; D, strong). (E–H) Stromal cell staining (A, negative; B, weak; C, moderate; D, strong).

In cohort 1, strong ALDH1A1 staining was associated with ER and progesterone receptor (PR) negativity, human epidermal growth factor receptor 2 (HER2) positivity and increased age. HS>0 was associated with small tumour size. No other significant association with clinicopathological variables was observed when using cut-offs HS>10 or HS>0 in cancer cells, or with stromal staining (see online supplementary table S1).

A similar association between strong ALDH1A1 staining and ER and PR negativity was observed in cohort 2, but no association with HER2 overexpression or age. For the other cut-offs in cancer cells, only HS>10 and H>0 (borderline) was significantly associated with a lower histological grade. In cohort 2, stromal staining was significantly associated with ER and PR positivity, increased age and lower histological grade (see online supplementary table S2).

ALDH1A1 levels and DDFS

When analysing ALDH1A1 protein expression in cancer cells, HS>10 was significantly associated with an improved DDFS in both cohorts (p=0.02 and 0.04) (figure 2A, B). This was diminished in both cohorts when the cut-off was set to HS>0 (figure 2C, D). With the cut-off set to include only tumours with strongly stained cancer cells as ALDH1A1+ , ALDH1A1+ was associated with a lower DDFS in cohort 1 (p=0.03), but not in cohort 2 (p=0.54) (figure 2E, F).

Figure 2

Distant disease-free survival in relation to aldehyde dehydrogenase family 1 member 1 (ALDH1A1) levels in breast cancer cells. ALDH1A1 levels are assessed with immunohistochemistry on two cohorts with three different cut-offs. (A and B) Histoscore (HS)>10 as cut-off. (C and D) HS>0 (any positive intensity) as cut-off. (E and F) ALDH1A1 strong staining (strong staining in any number of cells) as cut-off. Plots are created using the Kaplan–Meier method, p values are obtained with the log-rank test and HRs(with 95% CI) with univariable Cox regression modelling.

Analysing stroma, we found that higher expression of ALDH1A1 in stromal cells was significantly associated with an improved DDFS in cohort 2 (p=0.05), but not in cohort 1 (p=0.83) (figure 3). For cohort 2, Schoenfeld's test indicated (p=0.01) departure from the assumption of proportional hazards, why the HR should be interpreted as a mean HR over the time period studied rather than being considered constant at all times.

Figure 3

Distant disease-free survival in relation to aldehyde dehydrogenase family 1 member 1 (ALDH1A1) levels in stromal cells. Plots are created using the Kaplan–Meier method, p values are obtained with the log-rank test for trend and HRs(with 95% CI) with univariable Cox regression modelling.

Multivariable analysis

To verify whether the effect of ALDH1A1 on survival was independent of the other known markers available (nodal status, histological grade, ER, age, size and HER2 (PR was left out as it is highly co-linear with ER)), we performed a multivariable Cox regression for the cut-offs where ALDH1A1 was significantly associated with DDFS in univariable analysis. The use of cut-off HS>10 in cohort 1 was an independent prognostic marker in a multivariable model (table 4) while the other cut-offs in cancer cells, as well as stroma, failed to be significant independent prognostic factors (data not shown).

Table 4

Multivariable analysis of DDFS in cohort 1

Comparison with gene expression data

For the 67 patients in cohort 1 with available gene expression data, ALDH1A1 mRNA was highly associated with protein expression in cancer cells, both as percentage of stained cells and intensity (p=0.002 and <0.001). In stroma, it showed a trend towards association (p=0.06). We split the data in median gene expression value and assessed the association with DDFS, which showed a trend towards increased DDFS with high levels of ALDH1A1 mRNA, but was not significant (HR=0.75, 95% CI 0.32 to 1.8, p=0.52).

To test this in a larger cohort, we used the freely available online tool ‘GOBO’ (, which analyses the impact of gene expression levels on DMFS for selected genes in 11 publicly available gene expression data sets. Analysis with GOBO showed that a high gene expression level was highly associated with increased DMFS (p<0.00001) (figure 4).

Figure 4

Aldehyde dehydrogenase family 1 member 1 (ALDH1A1) mRNA expression levels and distant metastasis-free survival (DMFS) analysed in publicly available datasets by the freely available online tool GOBO. ( Gene expression levels are split by three quantiles and outcome plotted with the Kaplan–Meier method.


In this study, we showed that ALDH1A1 protein expression in breast cancer cells and stromal cells is associated with clinicopathological variables and prognosis, but the association is highly cut-off dependent. Only strong staining in cancer cells, regardless of number of cells, was associated with the expected characteristics of a more aggressive disease. On the contrary, a lower expression in a higher number of cancer cells, thus a higher total expression, was associated with an improved DDFS. Gene expression levels of ALDH1A1 were highly correlated with protein expression, and high gene expression levels in a separate cohort were associated with an increased DMFS.

Most previous studies have reported a minority of cells in a tumour to be ALDH1A1+ and a minority of breast tumours to contain ALDH1A1+ cells. This was the case in our study when we used strong staining as cut-off and is in line with the idea that stem cells only make up a small proportion of the tumour cell population.8 On the other hand, Gong et al19 reported a majority of cancer cells to be ALDH1A1+ and failed to detect any significant correlation to prognosis in inflammatory breast cancer. Morimoto et al used >10% stained cells as cut-off and failed to detect a correlation with outcome.14 In ovarian cancer, Chang et al37 showed ALDH1A1 to be a factor of good prognosis, in a material where almost half of the tumours were considered ALDH1A1+. Notably this study used a rather high cut-off of >20% stained cells. Further, two studies using the Aldefluor assay failed to correlate ALDH activity to increased stem cell behaviour in adipose-derived stem cells and endothelial progenitor cells, respectively.38 ,39 Lehmann et al20 failed to detect an association between tumourigenicity and stem cell markers, and reported that the weakly tumourigenic cells have a higher fraction of ALDH1A1+ cells compared with the highly tumourigenic cells. The authors suggested that they failed to detect the ‘true’ ALDH1A1+ population. Interestingly, we detected weak or moderate staining in a majority of tumours, and when we considered also these tumours as ALDH1A1+ (HS>0 or HS>10), we found an association with improved DDFS, or no association, depending on cohort and cut-off. Thus, our data suggest that the previous discordant immunohistochemistry (IHC) results may partially be explained by that different evaluators have different cut-offs for considering a stained cell ‘positive’.

One of the most consistent results when comparing ALDH1A1 levels and clinicopathological variables is the association with ER negativity.8 ,13 ,14 ,17 ,33 ER-negative tumours are usually of a more aggressive phenotype, and thus in line with the idea that CSCs contribute to a more aggressive and treatment resistant tumour. The association of ALDH1A1 expression and ER negativity was confirmed in our study in both cohorts, but only when using strong staining as cut-off. Further, strong ALDH1A1 staining was in one cohort associated with worse DDFS. We therefore suggest that the strongly stained cells more accurately identify cells with a more aggressive phenotype, typical of CSCs.

The identification of breast CSCs is controversial and several other biomarkers, particularly the CD44+/CD24 phenotype, have also been studied for their ability to identify CSCs. It is appealing to compare these different phenotypes. Our group has previously shown that the CD44+/CD24 phenotype is enriched in basal-like breast cancer, although this phenotype did not correlate to ALDH1A1 expression when analysing overall CD44+/CD24 expression. However, the CD44 molecule can be alternatively spliced and interestingly ALDH1A1 expression was correlated to the standard isoform of CD44 (CD44S).33 ,34 Zhong et al40 also found no correlation of ALDH1A1 and the CD44+/CD24 phenotype, and suggested that ALDH1A1 is a better clinical biomarker than CD44+/CD24. Neumeister et al18 found that the CD44+/CD24 phenotype and ALDH1A1 expression have a certain overlap but do not identify the exact same subpopulation of cells or tumours, and when using both markers they found the tumours that conferred the worst prognosis. Ginestier et al8 saw similar results with only a small proportion of breast tumours showing an overlap in expression of CD44+/CD24 and ALDH1A1, and Ricardo et al17 showed some overlap of ALDH1A1 cells and the CD44+/CD24 phenotype and argue that the markers identify cells at a distinct level of differentiation. Taken together, these data suggest that the markers identify different subpopulations of breast cancer cells with aggressive/CSC-like phenotype, but that none of the markers unambiguously identify all CSCs. Our data suggest that future studies aimed at identifying CSCs with various markers should take the cut-off dependency of ALDH1A1 into account.

We found several similarities between our two cohorts, for example, the distribution of ALDH1A1 levels and that HS>10 consistently was associated with an improved outcome, while the outcome was diminished gradually when using cut-off HS>0 and strong staining. Importantly, strong staining was associated with ER and PR negativity in both cohorts. However, some results were discrepant between the two cohorts. Notably, strong ALDH1A1 was associated with lower DDFS in cohort 1, but not in cohort 2. This may be explained by different patient characteristics between the two cohorts, for example, patients in cohort 1 have stage 2 disease, while patients in cohort 2 have node negative disease and thus a better general survival. The better general survival in cohort 2 could make it harder to detect a patient group with a low DDFS, as other prognostic factors possibly are stronger than ALDH1A1. Furthermore, the treatment is different between the cohorts, with cohort 1 being treated with tamoxifen, while cohort 2 is largely systemically untreated. There is also an age difference between the two cohorts with cohort 1 being postmenopausal and cohort 2 being either premenopausal or postmenopausal, and the expression and prognostic effect of ALDH1A1 have been suggested to vary with age.14 ,41

The evaluation of stroma on TMA is not optimal and is ideally performed on whole tissue sections. However, we could identify stromal cells in almost all of our tumours, and the stromal staining was found to be significantly associated with increased survival in one of the cohorts. Interestingly, stromal expression has been reported to be present in all cases of benign breast tissue and also to be more common in benign than malignant breast lesions. This may suggest that strong stromal ALDH1A1 expression more resembles normal breast tissue, while the loss of ALDH1A1 represents a more tumour-promoting milieu.22 ,42 ALDH1A1 expression has been reported in both spindle-shaped cells and round stromal cells, the latter presumed to be mainly lymphocytes.22 We therefore chose to evaluate the spindle-shaped cells, presumed to be fibroblasts, as the immune response and number of infiltrating lymphocytes are known prognostic factors. The exact role of ALDH1A1 expression in stromal cells is not clear, but it underlines the importance of the tumour microenvironment.

To orthogonally validate our results from the protein level, we also analysed the gene expression levels of ALDH1A1. We showed that the levels of ALDH1A1 mRNA were correlated with the levels of protein expression and that high mRNA levels in a separate cohort conferred a better prognosis. The high levels of mRNA may represent a low level of expression in many cells (as the gene expression levels are analysed in a homogenate of the tumour). This could be analogous to the weak staining we observed with IHC in a majority of tumours, reflecting a higher total expression than a few cells with strong staining, and thus supports the result that a low ALDH1A1 expression in a majority of cells is associated with an improved outcome. It could also represent expression in stromal cells, which is associated with improved outcome in one of our cohorts, or lymphocyte infiltration, which is a known prognostic factor in cancer.43

As ALDH1A1 is an enzyme-regulating retinoic acid signalling and thus important for tissue development as well as embryogenesis, it is plausible that high levels of ALDH1A1, and associated increased retinoic acid signalling, are required for a stem cell phenotype. On the other hand, the weak ALDH1A1 staining in breast cancer cells, which we herein found to be associated with increased DDFS, is more intriguing. Various normal tissues express different levels of ALDH1A1, and this is correlated with expression in cancers from the same organ.44 For some organs, such as liver or colon, a high expression in normal tissue could mean that a cancer expressing ALDH1A1 more resembles the differentiated phenotype and thus a less aggressive disease with better survival, which has been suggested in lung cancer.45 ,46 Although not previously supported in breast cancer, our results show that patients with tumours containing weakly stained breast cancer cells have a better DDFS, and we thus speculate that these cells are of a more differentiated phenotype.

In conclusion, our finding that the association of ALDH1A1 with prognosis and clinicopathological variables is highly cut-off dependent may contribute to the understanding why previous reports have been contradictory. We found that only strong ALDH1A1 staining was correlated with the expected tumour characteristics of a more aggressive disease, while a lower ALDH1A1 expression in cancer cells, but possibly in more cells and thus a higher total expression, was associated with a less aggressive disease and a better prognosis. This was supported by our results on the gene expression level, which analyses the entire tumour, including the stroma. Our data suggest that only strong staining may represent a subpopulation of cells with a more aggressive phenotype, typical of CSCs and a cut-off only separating cells with or without ALDH1A1 expression may be suboptimal; a better approach may be to use strong staining as cut-off.

Take home messages

  • Expression of the putative cancer stem cell marker aldehyde dehydrogenase family 1 member A1 (ALDH1A1) in cancer cells and stroma is associated with outcome and clinicopathological variables in breast cancer.

  • The association of ALDH1A1 with outcome is highly cut-off dependent as considering weakly stained cells as positive is associated with an improved outcome, while only strongly stained cells are associated with an aggressive disease or poor outcome.

  • Gene expression data show that high gene expression levels are associated with improved outcome, potentially reflecting weak staining in many cancer cells (and thus a higher total expression), or stromal components.

  • We propose that only cancer cells with strong ALDH1A1 staining have a more aggressive phenotype typical of breast cancer stem cells.


We gratefully thank Kristina Lövgren for expert technical assistance with immunohistochemistry staining and Dr Johan Staaf for valuable help with the GOBO tool. We also thank the BioCare strategic research school for providing an excellent research environment.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Handling editor Runjan Chetty

  • Contributors MS and DG did the IHC scoring. MS and LH did the statistical analysis. MS, GH, CH, MF and EN conceived and designed the study. PM, DG and MF provided the samples and patient follow-up. All authors contributed to the interpretation of the results, helped drafting and revising the manuscript, read and approved the final manuscript, and contributed to the writing of the article.

  • Funding This project was supported by grants from the Governmental funding of clinical research within the NHS (National Health Services, ALF), the Swedish Breast Cancer Association (BRO), the Mrs Berta Kamprad Foundation, the Skåne University Health Care, the Swedish Cancer Society, the Gunnar Nilsson Cancer Foundation and the Anna and Edwin Berger Foundation.

  • Competing interests None declared.

  • Ethics approval This project was approved by the ethical committee at Lund University (Lund, Sweden; LU 240-01), waiving further requirement for informed consent for the study.

  • Provenance and peer review Not commissioned; externally peer reviewed.