Article Text

Download PDFPDF
Characteristics, behaviour and role of biomarkers in metastatic triple-negative breast cancer
  1. Yutaro Goto1,2,
  2. Aye Aye Thike2,3,
  3. Clara Chong Hui Ong2,
  4. Johnathan Xiande Lim2,
  5. Nur Diyana Md Nasir2,
  6. Huihua Li4,5,
  7. Valerie Cui Yun Koh2,
  8. Xiao-Yang Chen2,6,
  9. Joe Poh Sheng Yeong2,7,
  10. Hironobu Sasano1,8,
  11. Puay Hoon Tan3,6,9
  1. 1 Anatomic Pathology, Tohoku University School of Medicine, Sendai, Miyagi, Japan
  2. 2 Anatomical Pathology, Singapore General Hospital, Singapore, Singapore
  3. 3 Duke-NUS Medical School, Singapore, Singapore
  4. 4 Health Services Research Unit, Singapore General Hospital, Singapore, Singapore
  5. 5 Centre for Quantitative Medicine, Duke-NUS Medical School, Singapore, Singapore
  6. 6 Anatomy, National University Singapore Yong Loo Lin School of Medicine, Singapore, Singapore
  7. 7 Integrative Biology for Theranostics, Institute of Molecular and Cell Biology, Singapore, Singapore
  8. 8 Pathology, Tohoku University Hospital, Sendai, Miyagi, Japan
  9. 9 Pathology, Singapore General Hospital, Singapore, Singapore
  1. Correspondence to Professor Puay Hoon Tan, Pathology, Singapore General Hospital, Singapore, Singapore; tan.puay.hoon{at}


Aims Characterising the factors responsible for metastatic triple-negative breast cancer (TNBC) is of significant importance, considering its high mortality rate and scant data. In this study, we evaluated the characteristics, clinical behaviour and role of biomarkers (androgen receptor (AR), oestrogen receptor beta (ERβ) and p53) in metastatic TNBC.

Methods Immunohistochemistry was performed for AR, ERβ and p53 on 125 primary TNBCs with known metastasis and correlated with clinicopathological parameters and outcome. AR and p53 mRNA profiling was also carried out on 34 tumours from the same series and correlated with outcomes.

Results In this cohort, grade 3 and pT2 tumours predominated. The most common site for metastasis was the lung and pleura (41, 32.8%), and 15 (12.0%) cases demonstrated metastasis in multiple sites. Among these, 92% of tumours metastasised without preceding local recurrences. Five- and ten-year overall survival (OS) rates were 27% and 7.2%, while 5- and 10- year survival rates after metastasis were 9.6% and 3.2% respectively. AR, ERβ and p53 protein expressions were observed in 16%, 96.8% and 58.1% of tumours, respectively. A combinational phenotype of AR-ERβ+p53+ tumours was associated with poorer OS (HR 1.543, 95%CI 1.030 to 2.310, p=0.035). Higher AR mRNA levels were significantly associated with favourable OS (p=0.015) and survival after metastasis (p=0.027).

Conclusions Metastatic TNBC harboured aggressive behaviour and displayed predominantly visceral metastasis with most metastatic events occurring without intervening local recurrences. A combinational phenotype of AR-ERβ+p53+ was significantly associated with poorer OS.

  • Metastatic triple negative breast cancer
  • protein and mRNA expression
  • outcomes
  • androgen receptor
  • estrogen receptor
  • p53

Statistics from


Tumour metastases are a significant clinical problem due to lack of effective treatment, accounting for more than 90% of cancer-related mortality in various cancers.1–3 In breast cancer, it was reported that 25%–50% of patients will eventually develop lethal distant metastasis after removal of the primary tumour,2 and the prognosis of these patients is generally unfavourable.2 3 Breast cancer has been traditionally classified into four molecular subtypes (luminal A, luminal B, human epidermal growth factor 2 (HER2) positive and triple negative/basal-like). Among them, triple-negative breast cancer (TNBC), which is defined by a lack of oestrogen receptor (ER), progesterone receptor (PR) and HER2, is known to be biologically aggressive with higher rates of local recurrence, early metastasis and shorter survival rate compared with other subtypes.4–8 It is reported that 14%–94% of TNBC eventually metastasise to different organs including the liver, brain, lungs and pleura, and bone with extremely low overall survival (OS) rates after metastasis.6 9

Because of the lack of ER, PR and HER2, patients with TNBC are not eligible for endocrine and HER2-targeted therapy, which can be applied for breast cancers of other molecular subtypes. Conventional chemotherapy remains the only treatment for TNBC even today.7 10 11 It is therefore important to identify and characterise the factors that drive progression, in particular after metastasis since more than 10% of patients with TNBC present with distant metastasis at the time of diagnosis.9

The prognostic and therapeutic roles of hormonal receptors in breast cancer are widely accepted. Oestrogen receptor alpha (ERα) is well characterised in breast carcinogenesis and serves as an effective endocrine therapeutic target of the luminal subtypes (luminal A and luminal B).12–15 In 1996, a second ER, named oestrogen receptor beta (ERβ) was discovered, harbouring high structural similarities with ERα at the DNA-binding domain (96%) and the ligand-binding domain (58%).16 Despite these structural similarities, ERα and ERβ possess distinct functions and patterns of expression.17 18 Although reports have described a dual role of ERβ in breast cancer, depending on the presence or absence of ERα, its precise role remains uncertain.19

Belonging to the same nuclear receptor family as ER and PR is the androgen receptor (AR).20 Previous studies on genomic profiling have disclosed different subtypes of TNBC, one of which is the luminal androgen receptor subtype,21 22 which is dependent on AR signalling,23 thereby making AR a potential therapeutic target for TNBC. However, studies investigating the prognostic value of AR in TNBC have mixed results, with AR positivity being associated with a better outcome,24 25 or poorer prognosis,23 or that it has no significant effect on survival.26

Another biomarker, p53, a known tumour suppressor and a regulator of the cell cycle, is overexpressed in invasive breast cancer, frequently mutated in TNBC and considered to be an indicator of biological aggressiveness.27–29 Studies exploring the prognostic value of p53 in TNBC have also yielded varied results, ranging from p53 positivity heralding a poorer outcome30 to it being a better prognostic factor,31 32 and to the status of p53 having no significant association with patient outcomes.33

To the best of our knowledge, studies evaluating biomarkers in metastatic TNBC are few. As TNBCs do not express ERα, PR and HER2, it would be worthwhile to explore the expression of other biomarkers in search of a potential therapeutic target. In this study, we evaluated clinicopathological characteristics, behaviour and outcomes of TNBC with metastases, and investigated the role of AR, ERβ and p53 in these tumours.

Materials and methods

Patients and tumours

Among 774 TNBCs diagnosed at the Department of Anatomical Pathology, Singapore General Hospital, from 1994 to 2012, 160 (20.7%) primary tumours were known to have metastasised according to case records, during a median follow-up period of 71 months. None were metastatic at diagnosis. Out of 160 cases, 125 had sufficient paraffinised material for tissue microarray (TMA) construction, and these cases became the study cohort. Cases without follow-up data and/or sufficient paraffinised tissues were excluded from the study. Clinicopathological parameters, including age, ethnicity, tumour size, histological grade, lymphovascular invasion, nodal stage and metastatic sites, were retrieved from pathology records.

TMA construction

Histological slides were retrieved and reviewed, and representative areas were marked out for TMA construction using the Manual Tissue Arrayer MTA-1 (Beecher Instruments, Sun Prairie, WI, USA) with 1 mm cores and three cores per case.34

Immunohistochemistry (IHC) and immunoscoring

Sections (4 µm) were cut from TMA blocks and then placed onto charged Bond Plus glass slides (Leica Biosystems, Richmond, IL, USA) in a similar orientation to facilitate faster evaluation. Antibodies to ER, PR and HER2 to reconfirm triple negativity, CK14, epidermal growth factor receptor (EGFR) and 34βE12 (to detect basal-like phenotype),34 AR, ERβ and p53 were applied to the sections, and IHC was performed using the Leica Bond-Max autostainer (Leica Biosystems, Melbourne, Australia). Details of antibodies are shown in table 1. Appropriate positive controls adopted from the antibody datasheets were applied.

Table 1

Details of antibodies

Immunohistochemical staining was scored by two observers (YG and AAT). Nuclear staining for ER, PR, AR, ERβ and p53, cytoplasmic membrane decoration for HER2 and EGFR, and cytoplasmic immunoreactivity for CK14 and 34βE12 were recorded. Staining intensity was scored as 0, 1+, 2+ and 3+ (nil, weak, moderate and strong staining, respectively). The percentage of positively stained tumour cells was assessed as a proportion of the total number of tumour cells in the section. Any unequivocal staining of at least 1% of tumour cells was regarded as a positive biomarker expression for ER, PR,35 CK14, EGFR and 34βE12. HER2 positivity was defined when >10% of tumour cells expressed 3+ cytoplasmic membrane staining.36 For this study, the cut-off value of 10% defined AR and p53 positivity. Differential cut-off values were tested to define ERβ positivity, and 25% nuclear expression of ERβ reached a significant level.

RNA extraction and NanoString nCounter analysis

Of the 125 cases that formed the study cohort, 34 were subjected to AR and p53 gene expression analysis. RNA was extracted from formalin-fixed paraffin-embedded (FFPE) standard sections of 10 µm thickness using the RNeasy FFPE kit (Qiagen, Hilden, Germany) on a QIAcube automated sample preparation system (Qiagen) and was quantified with the Agilent 2100 Bioanalyzer system (Agilent, Santa Clara, CA, USA). A total of 100 ng of functional RNA (>300 nucleotides) was assayed on the nCounter Custom Code Set (NanoString Technologies, Seattle, WA, USA). Normalisation of NanoString counts was done using positive control probes and housekeeping genes (online supplementary table 1).

Supplemental material


Follow-up data were obtained from patients’ clinical case notes. OS was defined as the time from the date of diagnosis to the date of death/date of the last follow-up, while survival after metastasis (postmetastatic survival) was defined as the time from the date of first evidence of metastasis to the date of death/date of the last follow-up.

Statistical analysis

Findings were analysed using the statistical software SPSS for Windows, V.25. The relationship between clinicopathological parameters and protein expression was analysed using χ2 and Fisher’s exact tests. Survival outcomes were evaluated using Kaplan-Meier analysis and were compared by log-rank test. Cox regression was carried out to evaluate the effect of biomarkers on outcomes. Linear regression was also performed to evaluate the relationship between protein expression and mRNA levels. A p value of <0.05 was considered statistically significant.


Clinicopathological characteristics

Clinicopathological characteristics of 125 patients are summarised in table 2. The age ranged from 30 to 87 years old (mean 54.4 years old, median 54 years old). Among this cohort, Chinese predominated with 101 (80.8%) individuals, Malay with 18 (14.4%) and Indian with 3 (2.4%), with the remaining 3 (2.4%) from other ethnic groups. Most tumours were histological grade 3 (96, 76.8%) and pT2 (100, 80.0%). The most common histological subtype was invasive ductal carcinoma not otherwise specified (NOS) (122, 97.6%). Lymphovascular invasion was present in 59 (47.2%) tumours, and axillary lymph node positivity was observed in 75 (60%) cases. A tripanel of CK14/EGFR/34βE1234 detected 104 (83.2%) tumours to be of basal-like phenotype.

Table 2

Clinicopathological characteristics of metastatic triple-negative breast cancer (n=125)

The most common site for metastasis in this series was the lung and pleura (41, 32.8%), followed by lymph node (22, 17.6%), central nervous system (CNS) (19, 15.2%), bone (11, 8.8%), liver (10, 8.0%), soft tissue (5, 4.0%), pericardium (1, 0.8%) and peritoneum (1, 0.8%), and 15 (12.0%) cases showed multiple metastases. Sites of metastases are summarised in table 3. In this study, we noticed that 115 out of 125 (92%) tumours metastasised without preceding local recurrences.

Table 3

Sites of metastasis

Expression of biomarkers/combinational phenotypes and association with clinicopathological parameters

Individually, AR positivity was observed in 20 (16%), ERβ in 121 (96.8%) and p53 in 72 (58.1%) tumours, respectively. One case of p53 was not available for analysis. Positive and negative expressions of these biomarkers are shown in figure 1. ERβ+p53+ combinational phenotype was observed in 71 (57.3%) cases, AR-ERβ+ phenotype in 101 (80.8%), AR-p53+ phenotype in 59 (47.6%) and AR-ERβ+p53+ phenotype in 58 (46.8%) cases, respectively (table 4). Individual biomarkers and combinational phenotypes showed no correlation with clinicopathological parameters in this series, with the exception of the AR-ERβ+ phenotype, which was significantly associated with tumour size (p=0.028) (online supplementary tables 2 and 3).

Table 4

The expression of biomarkers and combinational phenotypes

Figure 1

Immunohistochemical protein expression of AR, ERβ and p53. AR, androgen receptor; ERβ, oestrogen receptor beta.

mRNA Assays

In a subset of 34 tumours that underwent gene expression analysis based on mRNA levels, the mean level of AR was 211.5, with a median of 41.7, ranging between 10.2 and 1700.5. The mRNA levels of p53 ranged from 69.7 to 555.3 (mean 217.4 and median 148.4). Among this cohort, 7 (20.6%) tumours showed high AR mRNA expression, while 12 (35.3%) tumours harboured high p53 expression using mean cut-off values. Linear regression showed a significant association between AR protein expression and AR mRNA levels (r=0.813, p<0.001) (figure 2). It indicated that AR mRNA levels increased by about 13.9 units (95% CI 9.62 to 18.22) with every unit increase of AR IHC (p<0.001). However, there was no significant correlation between p53 protein expression and p53 mRNA levels (r=0.348, p=0.088)

Figure 2

Linear regression showing significant association between AR protein expression and AR mRNA levels (r=0.813, p<0.001). AR, androgen receptor; IHC, immunohistochemistry.

Survival analysis

Follow-up of this cohort ranged from 11.9 to 202.9 months, with a mean of 44.8 months and a median of 37.1 months. Breast cancer-specific death occurred in 98 (78.4%) patients. The OS (mean 61.6 months, median 41.0 months) and survival after metastasis (mean 28.2 months, median 11.7 months) are summarised in table 5. The survival rates of 5 and 10 years of the whole series and those of different metastatic sites are summarised in table 6. Metastasis to bone showed slightly better survival compared with metastasis to the CNS and the liver.

Table 5

Mean and median for OS and survival time after metastasis

Table 6

Survival rates (5 and 10 years)

In this series, individual AR and p53 protein expressions had no impact on outcomes. Additionally, patients whose tumours expressed ERβ showed a trend for poorer OS, although the observed difference was not statistically significant (p=0.230) (figure 3A). Furthermore, patients with a combinatorial phenotype of AR-ERβ+p53+ disclosed poorer OS (p=0.034) (figure 3B), although other combinational phenotypes did not show any significant results (online supplementary figure 1). On univariate Cox regression analysis, this particular combinational phenotype had a significant impact on OS (HR 1.543, 95% CI 1.030 to 2.310, p=0.035) (table 7). However, multivariate analysis failed to indicate significant results.

Table 7

Univariate Cox regression analysis on overall survival

Figure 3

Kaplan-Meier plots of (A) ERβ and (B) combinatorial phenotype. AR, androgen receptor; Cum, cumulative; ERβ, oestrogen receptor beta; OS, overall survival.

On Kaplan-Meier analysis, patients whose tumours had higher AR mRNA levels disclosed favourable OS (p=0.015) and survival after metastasis (p=0.027) (figure 4). On univariate Cox regression analysis, tumours with higher mRNA levels had an impact on OS (HR 0.187, 95% CI 0.042 to 0.835, p=0.027) but not on survival after metastasis. Multivariate analysis showed no significant results. In this series, p53 mRNA assays did not have any impact on survivals.

Figure 4

Kaplan-Meier curves disclosing both (A) poorer OS (p=0.015) and (B) poorer survival after metastasis (p=0.027) for patients with low A R mRNA levels compared with patients with high A R mRNA levels. AR, androgen receptor; ERβ, oestrogen receptor beta; OS, overall survival.


Metastatic breast cancer remains incurable, and evolving understanding of cancer biology will present novel opportunities to translate into clinically relevant therapy. To date, the mainstay of treatment for metastatic breast cancer is cytotoxic chemotherapy, which has been shown to improve survival. Between 20% and 30% of women with early-stage invasive breast cancer progress to distant metastasis.37 At our institution, the metastatic rate of TNBC was 20.6%, while the rate varies from 14% to 94% in previous reports,6 9 depending on the size and type of study cohort, length of follow-up period, ethnicity, age group and geographical distribution.

The majority of TNBC that metastasise are of high nuclear and histological grade, with histological subtype of infiltrative ductal carcinoma (NOS) and a larger tumour size. Our series shows similar findings. However, low-grade tumours also have the potential to progress to distant metastasis. In our study, there were three grade 1 tumours. For the first case, the primary tumour was an invasive ductal carcinoma (NOS) with metastasis to one axillary lymph node at the time of initial diagnosis for which segmental mastectomy and axillary clearance were performed. After 7 years, lung metastasis developed without prior local recurrence. The second case was diagnosed as an invasive lobular carcinoma, which was confirmed by E-cadherin IHC with metastases to 7 of 18 axillary lymph nodes, 3 with extranodal extension. After mastectomy and axillary clearance, bone marrow metastasis occurred 3.5 years later. For case 3, the primary tumour was grade 1 invasive ductal carcinoma (NOS) on excision, and there was subsequent recurrence of grade 2 invasive ductal carcinoma, which was also triple negative after 8 years. The patient developed metastasis to the brain 9 years after local recurrence. We observed that distant metastasis could occur even in initial low-grade primary tumours, which were accompanied by axillary lymph node metastases at the time of diagnosis; distant metastases could be detected without prior local recurrences.

TNBCs have a predilection for visceral metastasis to the brain and lungs, unlike non-TNBCs, which have a higher rate of metastasis to the bone and axillary lymph nodes.8 38 Our study follows this trend of TNBC progression, with the most common metastatic site being the lung and pleura. Non-axillary lymph node metastasis in our study included cervical, mediastinal and supraclavicular lymph nodes. In our series, 12% of patients developed synchronous metastasis of multiple sites at the time of initial metastasis, which is in agreement with Tseng et al 6 (12% vs 18%), who also reported that brain and liver metastases were significantly correlated with poorer postmetastatic OS when compared with bone metastasis.

It was suggested that TNBC has a tendency to disseminate via the haematogenous route rather than through the lymphatic system. The development of distant metastasis in TNBC is a complex process that is not well understood. It involves numerous processes such as genetic and epigenetic aberrations, angiogenesis, interactions between the tumour and stroma, intravasation through the basement membrane, survival in the circulatory system, and extravasation into distant organs.1 39 During this event, tumour cells of TNBC have a predilection for visceral organs. Dent et al 8 explained that specific protein factors based on molecular studies may underlie the propensity towards visceral metastases, or perhaps may make it less likely for the tumour cells to adhere to bone.

From recent data, relative survival rates for women diagnosed with all types of breast cancer are 89% at 5 years postdiagnosis and 83% after 10 years.37 As our series is based on metastatic TNBC, OS rates of 5 and 10 years are much reduced, and survival rates of 5 and 10 years after metastasis (postmetastatic survival) are less than 10% after 5 years and about 3% after 10 years. To our best knowledge, scant data are available on metastatic TNBC. Survival rates for visceral metastasis are shorter compared with bone metastasis. There are no patients who survived after 10 years in cases of CNS, liver and multiple metastases. One Canadian study reported that the median survival time from metastasis to death was 9 months, which is not much different from our Asian series (9 months vs 11.7 months).8 After metastasis, rapid progression to death is the usual course in TNBC.40

We evaluated the impact of combinational phenotypes on survival and their relationship with clinicopathological parameters. Only a combinational phenotype of AR-ERβ+was significantly associated with larger tumours, which were likely to be of higher grade. Although AR is associated with favourable pathological parameters and p53 is correlated with biologically aggressive features,27 28 34 we were unable to duplicate these results in our cohort, likely due to a selection bias for larger, high-grade tumours and the smaller sample size.

In this study, individual biomarkers had no impact on survival outcomes, although AR was a significant independent predictor of better survival in our previous large TNBC cohort,25 with p53 as a poorer prognostic indicator.29 However, in this series, higher AR mRNA levels indicated favourable outcomes in patients with metastatic TNBC supporting AR as a better prognostic indicator. Previous reports have documented that another ER, ERβ, is expressed in TNBC.14 15 Although ERα and ERβ are encoded by two different genes, they share 96% and 58%–60% homology in the DNA-binding domain and the ligand-binding domain of ERα, respectively.16 41 Despite sharing similar structures and mechanisms of action, the two ER subtypes evoke distinct transcriptional responses and therefore influence cancer cellular processes differently, suggesting independent roles in therapy resistance.42

The role of ERβ in breast cancer progression is poorly understood. Nonetheless, some studies have demonstrated ERβ positivity to be correlated with a more aggressive clinical outcome15 and with the proliferation marker, Ki-67.43–45 ERβ has also been identified as a novel activator of wild-type p53-dependent transcription, which leads to poorer survival of luminal breast cancer cells.42 46–50

On the other hand, some studies discovered ERβ to be a better prognostic indicator in patients with breast cancer.51–53 Ström et al 51 showed that the growth of luminal T47D cells expressing mutant p53 was inhibited by ERβ, both in vitro and in vivo. It has also been suggested that ERβ exerts both antiproliferative and proliferative effects in breast cancer cells, depending on the presence or the absence of ERα, respectively.19

ERβ positivity (96.8%) in our series is much higher than that in previous studies, which reported rates of 25% in triple-negative cancers54 and 37.6% in basal-like breast cancers.55 To the best of our knowledge, there have been no reports of ERβ on purely metastatic TNBC. In our study, which focused on triple-negative cancers that had metastasised, 78.4% of patients died of a breast cancer-specific cause, with the higher rate of ERβ positivity suggesting a possible tumour-promoting role of ERβ. When we used a cut-off value of 25% to define ERβ positivity, we were able to demonstrate significant survival differences for ERβ in combination with AR and p53, but not for ERβ alone. The clone of ERβ used in our current study is different from that described in other studies and could account partly for the differences in positive staining rates.54 55 Our findings may help in determining specific endocrine therapeutic choices in patients with aggressive, ERα-negative metastatic TNBC.

Our study indicates ERβ as a poorer prognostic indicator in combination with negative AR expression and positive p53 expression. ERβ may have a dual role in breast cancer exhibiting different functions in the presence and absence of ERα.19 56 A deeper understanding of the contribution of ERβ in TNBC, especially in metastatic disease progression, prognosis and treatment response, is crucial for further exploration of this receptor with regard to diagnostic and therapeutic approaches. However, we acknowledge that the number of patients in this cohort is small and is a limitation of the study. Additionally, although RNA was extracted for more than 34 cases, quality check of the RNA extracted was not successful due to the age of the FFPE materials. Studies with a larger sample size, as well as those with more recent FFPE materials, will be beneficial in affirming the results of this study.

In conclusion, we observed that ERβ expression, combined with negative AR and positive p53 expression, was a poorer prognostic indicator in patients with metastatic TNBC. However, protein expression of AR, ERβ and p53 alone did not predict demise in these patients. Future work to elucidate the extent of involvement of these biomarkers in metastatic triple-negative tumours, as well as to explore the mechanisms that govern cell cycling and dormancy, will allow a deeper understanding of tumour biology and support the development of novel therapeutic targets.

Take home messages

  • A combinational phenotype of AR-ERβ+p53+ was an indicator of poorer overall survival in patients with metastatic triple-negative breast cancer (TNBC).

  • Individual protein expression of AR, ERβ and p53 had no impact on survival outcomes.

  • Higher levels of AR mRNA disclosed favourable survival outcomes in patients with metastatic TNBC.


This study was supported by the National Medical Research Council, Singapore (SMPO201302). YG is a medical student intern on an elective posting from Tohoku University, Sendai, Japan.



  • Handling editor Cheok Soon Lee.

  • Contributors PHT conceived and designed the study. YG and AAT wrote the manuscript. CCHO, JXL and NDMN performed the TMA construction, cutting and staining of tissue sections. YG and AAT quantified the immunoscoring. AAT, HL and VCYK performed the data analyses. AAT, VCYK, X-YC, JPSY, HS and PHT revised the manuscript critically for content.

  • Funding This study was funded by the National Medical Research Council, Singapore (SMPO201302).

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval All procedures performed in this study were in accordance with the ethical standards of the SingHealth Centralised Institutional Review Board (CIRB) (Ref: 2013/664/F). The SingHealth CIRB operates in accordance with the ICH Guideline for Good Clinical Practice and with the applicable regulatory requirements. This article does not contain any studies with animals performed by any of the authors. A waiver for informed consent was approved by the SingHealth CIRB.

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

  • Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information.

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.