Article Text

Download PDFPDF

Myeloid-derived suppressor cells at diagnosis may discriminate between benign and malignant ovarian tumors
  1. An Coosemans1,2,
  2. Thaïs Baert1,3,
  3. Jolien Ceusters1,
  4. Pieter Busschaert4,
  5. Chiara Landolfo1,5,
  6. Tina Verschuere6,
  7. Anne-Sophie Van Rompuy7,
  8. Adriaan Vanderstichele2,4,
  9. Wouter Froyman2,8,
  10. Patrick Neven2,4,
  11. Ben Van Calster8,9,
  12. Ignace Vergote1,2,4 and
  13. Dirk Timmerman2,8
  1. 1 Department of Oncology, Laboratory of Tumor Immunology and Immunotherapy, ImmunOvar Research Group, Katholieke Universiteit Leuven, Leuven, Belgium
  2. 2 Department of Gynecology and Obstetrics, Leuven Cancer Institute, University Hospital Leuven, Leuven, Belgium
  3. 3 Department of Gynecology and Gynecologic Oncology, Kliniken Essen-Mitte gGmbH, Essen, Germany
  4. 4 Department of Oncology, Laboratory of Gynecologic Oncology, Katholieke Universiteit Leuven, Leuven, Belgium
  5. 5 Department of Gynecology and Obstetrics, Queen Charlotte's and Chelsea Hospital, London, UK
  6. 6 European Organisation for Research and Treatment of Cancer, Brussels, Belgium
  7. 7 Department of Pathology, Katholieke Universiteit Leuven, Leuven, Belgium
  8. 8 Department of Development and Regeneration, Women and Child, Katholieke Universiteit Leuven, Leuven, Belgium
  9. 9 Department of Biomedical Data Sciences, Leiden University Medical Center, Leiden, The Netherlands
  1. Correspondence to Dr An Coosemans, Department of Oncology, Laboratory of Tumor Immunology and Immunotherapy, ImmunOvar Research Group, Katholieke Universiteit Leuven, Leuven 3000, Belgium; an.coosemans{at}kuleuven.be

Abstract

Background The behavior of the immune system as a driver in the progression of ovarian cancer has barely been studied. Our knowledge is mainly limited to the intra-tumoral adaptive immune system. Because of the widespread metastases of ovarian cancer, an assessment of the circulating immune system seems more accurate.

To demonstrate the presence of immune cells in blood samples of patients with ovarian neoplasms.

Methods In this exploratory prospective cohort study, peripheral blood mononuclear cells were collected at diagnosis from 143 women, including 62 patients with benign cysts, 13 with borderline tumor, 41 with invasive ovarian cancer, and 27 age-matched healthy controls. Immune profile analyses, based on the presence of CD4 (cluster of differentiation), CD8, natural killer cells, myeloid-derived suppressor cells, and regulatory T cells, were performed by fluorescence activated cell sorting.

Results In a multivariable analysis, six immune cells (activated regulatory T cells, natural killer cells, myeloid-derived suppressor cells, monocytic myeloid-derived suppressor cells, exhausted monocytic myeloid-derived suppressor cells, and total myeloid cells) were selected as independent predictors of malignancy, with an optimism-corrected area under the receiver operating characteristic curve (AUC) of 0.858. In contrast, a profile based on CD8 and regulatory T cells, the current standard in ovarian cancer immunology, resulted in an AUC of 0.639.

Conclusions Our immune profile in blood suggests an involvement of innate immunosuppression driven by myeloid-derived suppressor cells in the development of ovarian cancer. This finding could contribute to clinical management of patients and in selection of immunotherapy.

  • ovarian cancer

Statistics from Altmetric.com

HIGHLIGHTS

  • Assessment of immune cells in blood may be an alternative for immune monitoring in ovarian cancer.

  • Myeloid-derived suppressor cells play an important role in discriminating benign from malignant ovarian tumors.

  • Regulatory T cells and CD8 together, still the main focus in ovarian cancer, do not have a similar potential.

INTRODUCTION

Ovarian cancer has the fifth highest mortality rate among women diagnosed with cancer in Europe. It is a silent killer, metastasizing throughout the abdomen before causing symptoms. Consequently, 63% of patients are diagnosed with FIGO (International Federation of Gynecology and Obstetrics) stage III or IV, leading to poor prognosis, with a median survival of 36–53 months (stage III) and 20 months (stage IV). The vast majority of women are diagnosed with a high-grade serous ovarian cancer sub-type.1

Although advances have been made in the diagnosis of ovarian cancer using International Ovarian Tumor Analysis methods,2 there are still shortcomings. Despite good overall performance, the current models face problems with ovarian cysts, which may be hard to categorize. It is well established that the immune system is an important factor in the onset and development of cancer.3 Based on both the adaptive and innate immune response, tumor cells may be eliminated. It can occur that rare cancer cell variants are not destroyed. These cells enter the equilibrium phase, in which the adaptive immune system keeps the cancer cells in a dormant state by continuous editing of tumor immunogenicity. Based on a series of events resulting from this immunological pressure, including loss of tumor antigen expression, tumor cells will escape immune surveillance.4

Immunosuppressive cells are attracted towards the tumor (eg, myeloid-derived suppressor cells, regulatory T cells, M2 macrophages), suppress the existing adaptive immune system (e.g secretion of interleukin 10, transforming growth factor β, increasing number of immune checkpoint molecules…), and therefore promote tumor growth.5 The exact mechanism is probably different and specific for each tumor, which can also explain the success rate of immune checkpoint inhibitors, focusing on the adaptive immune system, in some tumors and not in others.6

Knowledge of the immune system in ovarian cancer is mainly limited to its intra-tumoral behavior at diagnosis. It is known that the presence of intra-tumoral T cells improves survival, that the presence of regulatory T cells decreases overall survival independent of stage, and that an increasing ratio of CD8+ (cluster of differentiation)/regulatory T cells improves overall survival.7–9 However, it was recently demonstrated that the immune system behaves differently in different metastatic implants.10 Therefore, it might be advisable to not only limit research to the intra-tumoral immune system, but also to look into the systemic counterpart. However, studies on the immune system in blood (as a liquid biopsy) are rare and contradictory and, in general, focus mainly on the adaptive immune system, as shown in Table 1. Borderline tumors are under-represented in these studies, even though they constitute one of the pitfalls in the diagnostic investigation of ovarian pathology.

Table 1

Immune profiling in blood of patients with ovarian cancer: literature overview

In this study, we explore the variation of a broad range of innate and adaptive immune cells in blood samples of patients with benign cysts, borderline tumors, or invasive cancer, and compare these with values in age-matched healthy controls. In addition, we evaluate the prognostic potential of circulating immune cells. The goal of this study was to demonstrate the presence of immune cells in blood samples of patients with an ovarian tumor that are most promising for further investigation and to include them in future diagnostic and prognostic models

METHODS

For this exploratory prospective cohort study, patients, diagnosed with transvaginal ultrasound and planned for surgical removal of an ovarian mass, were consecutively enrolled in two prospective studies (OV-IMM-2014 (Ovarian-Immunology-2014) and TRANS-IOTA (Translational- International Ovarian Tumor Analysis) at the University Hospitals Leuven (Belgium) between June 2014 and February 2017. Healthy controls (not operated) were included separately. Studies were approved by the local ethics committee (s50887 (for healthy controls), s51375 and S59207 (for TRANS-IOTA), and s56311 (for OV-IMM-2014)). OV-IMM-2014 recruited only patients with primary invasive ovarian cancer, following them up over 10 years, starting from diagnosis. TRANS-IOTA consecutively recruited patients with ovarian masses at diagnosis. All patients underwent surgery. Based on subsequent histopathological examination, patients could be categorized as having a benign cyst, a borderline tumor, an invasive ovarian cancer, or a metastatic tumor to the ovary. Healthy controls, aged ≥40 years, age-matched with the patients and having a benign cyst, were included consecutively, after transvaginal ultrasound, demonstrating two normal ovaries. They had all consulted for non-ovarian-related gynecological complaints. Exclusion criteria included women with active therapy for non-ovarian cancer at the point of inclusion, presence of immune disease, treatment with immunomodulators, pregnancy, age <18 years, surgery of the suspected mass elsewhere before inclusion, and infectious serology (human immunodeficiency viruses, hepatitis B and C).

A blood sample (Vacuette NH sodium heparin tube, BD, ref 455051) was taken at diagnosis, both in OV-IMM-2014 and TRANS-IOTA patients. Peripheral blood mononuclear cells were isolated using Ficoll density gradient centrifugation. Cells were counted, frozen to −80°C using a slow-freeze protocol (max −1 °C/min), and then transferred to liquid nitrogen for storage. In addition, available tumor tissue from the primary tumor and/or metastases (resected for diagnostic purposes or at the time of debulking surgery) was used for immunohistochemistry.

Peripheral blood mononuclear cells were defrosted in batches. A total of 11 patient samples were surface stained simultaneously, using a 96-well plate. In addition, a technical control sample was added to each plate as a 12th sample. After staining, cells were acquired using LSRFortessa Cell Analyser (BD Biosciences, Franklin Lakes, New Jersey, USA). Cells were surface stained using a nine-parameter color panel, according to the manufacturer’s protocol using three separate panels as shown in online supplementary table S1. Zombie yellow (Biolegend; ref 423104) was used to exclude dead cells from the analysis. Analysis of samples was performed using FlowJo Software (FlowJo, LLC, Ashland, Oregon, USA). Samples were excluded from statistical analysis if the technical control sample differed significantly from the standard (z-score >2). We considered 19 immune cell parameters as predictors Natural killer cells, (activated/exhausted) CD4+ and CD8+ (activated implies CD69 positive, exhausted implies programmed cell death 1 positive), (activated/exhausted) regulatory T cells (activated implies CD69 positive, exhausted implies programmed cell death 1 positive or CD152 positive), and regulatory T cells positive for human leucocyte antigen – DR isotype) were evaluated relative to the number of live CD45+ cells. Total myeloid-derived suppressor cells were evaluated relative to the number of live cells. (Exhausted) monocytic and granulocytic myeloid-derived suppressor cells (exhausted implies programmed death ligand 1 positive) were used as relative numbers to total myeloid-derived suppressor cells. Total myeloid cells were based on forward and side scatter, as also described.11 Furthermore, in analogy to immunohistochemical analysis, the CD8+/regulatory T cell ratio was also evaluated. Tumor tissue from primary and metastatic disease was stained for the presence of CD8 and Forkhead box P3 (FoxP3). For description of the method, see online supplementary file M1.

Supplemental material

Immune cells were log-transformed (with base 2) for all statistical analyses, but descriptive statistics and box plots were based on original values. To correlate immune cells with age, Spearman correlations were used in the healthy control group. To compare the different patient groups, the area under the receiver operating characteristic curve (AUC), with 95% CI, was calculated, based on the logit transform method.12 An AUC of 0.5 indicates no discrimination between groups, and an AUC of 1 indicates perfect discrimination. In addition, to investigate which cells might be most interesting, we applied a ridge logistic regression with backward elimination using the apparent AUC as selection criterion.13 To obtain the optimism-corrected AUC, we started by generating 100 bootstrap samples that had the same sample size as the original dataset. These bootstrap samples were constructed by randomly selecting patients from the original data. These patients were selected ‘with replacement’, meaning that a patient could be selected more than once or not at all for a given bootstrap sample. On each bootstrap sample, ridge regression with backward elimination (selection criterion: AUC) was applied. The difference between the AUC of the model based on the bootstrap sample and the apparent AUC of the model based on the original dataset with the same number of variables (=optimism) was computed. Over all the bootstrap samples, median optimism for each step in the selection procedure was calculated. This median optimism was subtracted from the corresponding apparent AUC, resulting in an optimism-corrected AUC.

Univariable Cox proportional hazards analysis with Firth bias correction was applied to relate immune cells to progression free survival. We encountered missing values for the immune cells due to technical error of the Fluorescence Activated Cell Sorting machine or too high inter-analysis variability of the technical control sample (z-score ≥2). We assume that this can be classified as ‘missing completely at random’.14 For univariable analysis, we decided to continue with available data. For the multivariable analysis to predict whether tumors where benign or malignant, we imputed missing values to avoid that too many patients had to be excluded.14 We used single stochastic imputation using multivariate imputation chained equations.15 The multivariable analysis was performed on this completed dataset. All statistical analyses are performed in R version 3.4.1, using packages AUROC, logistf, penalized, mice, and coxphf.

RESULTS

In total, 62 patients had a benign cyst, 13 a borderline tumor, and 41 invasive ovarian cancer (Figure 1). Healthy controls ≥40 years (n=27), age-matched with the benign cysts, were used as a comparison (online supplementary figure S1, A). Descriptive statistics of the immune cell parameters are displayed in online supplementary figure S1 B&C, Figure 2, online supplementary table S2, and online supplementary table S3. In concordance with the literature, granulocytic myeloid-derived suppressor cells were the cell type most susceptible to decay.16

Figure 1

Distribution of samples. Pie chart showing the histological and FIGO (International Federation of Gynecology and Obstetrics) stage distribution among the different patient groups: (A) Benign tumors (n=62); (B) Borderline tumors (n=13) (C) Invasive tumors (n=41). Combinations: endometrioma + mucinous cystadenofibroma (n=1); serous cystadenoma + fibroma (n=1); fibroma + endometrioma (n=1); others: Leydig cell (n=1), hydrosalpinx (n=1), adeno-acanthofibroma (n=1), struma ovarii (n=1), fibrothecoma (n=1). MMMT, malignant mixed Müllerian tumor.

Univariable AUCs varied between 0.53 and 0.76 (Table 2, online supplementary figure S2). AUCs ≥0.70 were observed only in the innate immune system (natural killer cells (AUC 0.70), monocytic myeloid-derived suppressor cells (AUC 0.71), and exhausted granulocytic myeloid derived suppressor cells (AUC 0.76)). Based on ridge logistic regression (Figure 3 and online supplementary figure 3S), the combination of all 19 markers resulted in a corrected AUC of 0.842. Based on logistic regression with backward elimination, the best AUC value was obtained using 10 immune markers, resulting in an optimism-corrected AUC of 0.864. The curve started to bend at the level of activated regulatory T cells, implying that the most important discriminating roles are reserved for activated regulatory T cells, natural killer cells, myeloid-derived suppressor cells, monocytic myeloid-derived suppressor cells, exhausted monocytic myeloid-derived suppressor cells, and total myeloid cells. This panel of six is dominated by immunosuppression driven by myeloid-derived suppressor cells, resulting in an optimism-corrected AUC of 0.858. Of note, an immune profile in blood based on the current standard in ovarian cancer immunology, CD8 and regulatory T cells, would result in an AUC of 0.639.

Four types of adaptive immune cells displayed AUCs >0.7 in discriminating healthy controls from benign ovarian cysts: activated and exhausted CD8+ T cells, activated CD4+ T cells, and exhausted regulatory T cells (Figure 2). Total CD4+ T cells, activated and exhausted (programmed cell death 1) regulatory T cells, regulatory T cells positive for human leucocyte antigen – DR isotype, and exhausted granulocytic myeloid-derived suppressor cells reached AUCs ≥0.90 in discriminating borderline tumors from stage I–II invasive ovarian cancer (online supplementary figure S4). An increase of immunosuppression (symbolized here by activated regulatory T cells (AUC 0.72), exhausted monocytic (AUC 0.74), and granulocytic myeloid-derived suppressor cells (AUC 0.88)) and an increase in natural killer cells (AUC 0.71) was noted in advanced-stage ovarian cancer (online supplementary figure S5) compared with early-stage ovarian cancer. Exhausted CD8+ cells had the highest potential to discriminate between non-high-grade serous ovarian cancer and high-grade serous ovarian cancer (online supplementary figure S6) (AUC 0.90) (Table 2).

Figure 2

Immune cells in different patient groups and in healthy controls. Box plot visualization of the distribution of the studied 19 immune parameters throughout the different histologies in comparison with healthy controls. Each dot represents one patient. BOT, borderline tumor; CD, cluster of differentiation; gMDSC, granulocytic MDSC; MDSC, myeloid-derived suppressor cells; mMDSC, monocytic MDSC; NK, natural killer cell; Treg, regulatory T cells; PD(L)1 (programmed death (ligand)1) implies an exhausted cell type; CD69+ implies an activated cell type; CD152+ implies an exhausted cell type; Treg_HLA-DR+, regulatory T cells positive for human leucocyte antigen – DR isotype.

Figure 3

Results of the multivariable ridge logistic regression with backward elimination. The plot shows the optimism-corrected area under the receiver operating characteristic curve (AUC) through the backward elimination process (starting with a full model including 19 immune cells on the left, and eliminating cells one by one until one is left). The table shows in which order cells are eliminated, and the optimism-corrected AUC at each step. CD, cluster of differentiation; gMDSC, granulocytic MDSC; MDSC: myeloid-derived suppressor cells; mMDSC, monocytic MDSC; NK, natural killer cell; Treg, regulatory T cells; PD(L)1 (programmed death (ligand)1) implies an exhausted cell type; CD69+ implies an activated cell type; CD152+ implies an exhausted cell type; Treg_HLA-DR+, regulatory T cells positive for human lLeucocyte antigen – DR isotype.

Table 2

Statistical changes in immune cells between groups and relation between immune cells in invasive tumors andsurvival

Since immunohistochemistry is still the most common way of studying the immune system in ovarian cancer, diagnostic samples of primary and metastatic tumors were stained for CD8 and FoxP3.17 Tumor tissue was available for CD8 and FoxP3 staining from 15 primary tumors at diagnosis and 22 metastases at diagnosis. In six patients, matching primary and metastatic samples were available. Nearly all (n=13/15 (87%)) metastases studied displayed an immune-excluded phenotype, whereas all early-stage tumors from which tumor tissue could be obtained (n=4) had an inflamed phenotype. In this small sample size, there was no indication of a strong correlation between the amount of CD8 and FoxP3 and progression-free survival (results not shown). Comparison of the results in blood with the results in tissue for CD8 and FoxP3 showed an inverse correlation (Spearman rank order correlation coefficients used) for CD8 (ρ −0.33) when considering the primary tumor. For regulatory T cells, we observed a correlation close to 0 (ρ −0.08) (online supplementary figure S7).

We observed 30 progression-free survival events among 41 patients with invasive ovarian cancer after a median follow-up of 23 months (range 1.5–53). Univariable Cox regression analysis gave an observed HR >1 for 16/19 cell parameters, suggesting that increased values may be related to inferior prognosis (Table 2). However, confidence intervals are wide.

DISCUSSION

The goal of this study was to explore differences at diagnosis in both the innate and the adaptive immune system in the blood of patients with an ovarian tumor, in comparison with healthy controls. In addition, the presence of immune cells in patients with invasive ovarian cancer was also linked to prognosis. The rationale for this study design was to find a method for immune profiling, which matches the clinical scenario of widespread metastatic ovarian cancer, compared with the existing intra-tumoral assessments of the immune system.

The key finding is an immune profile relying on myeloid-derived suppressor cells to discriminate between benign and malignant ovarian cysts. From a clinical point of view, this comparison is most relevant. In contrast to the focus on the adaptive immune system in ovarian cancer since 2003, this study clearly indicates a role for the innate immune system, as was also demonstrated by our group in an ovarian cancer mouse model.18 In subgroup analyses, a substantial role for the adaptive immune system seems to be maintained.

A comparison between ovarian cancer and healthy controls (Table 1) has little clinical impact. Winkler et al19 and Chatterjee et al20 were the only groups who had a patient population comparable to ours, including healthy controls versus the three groups of ovarian masses (benign, borderline, and invasive). Our results are in line with Cannioto et al but not with Lutgendorf et al. 21 22 Wu et al and Okla et al had already evaluated the presence of monocytic myeloid-derived suppressor cells in ovarian cancer, but only compared healthy controls with malignant cases.23 24 They concluded that cells were increased in patients with malignant tumors and that their presence correlated negatively with prognosis. In our series, we observed a similar negative prognostic role, and we could also attribute an important role to this cell subtype in discriminating between benign and malignant cysts. In line with our results, emphasizing the relevance of programmed death ligand 1 expression on myeloid-derived suppressor cells, Chatterjee et al also underlined the importance of programmed death ligand 1 on monocytes in discriminating invasive ovarian cancer from benign cysts.20

Increasing evidence shows the importance of the intratumor and intertumor heterogeneity and the more than likely sampling error when only examining tumor biopsies for clinical decision-making.25 Ovarian cancer is peculiar in this regard, because most often it is a widespread metastatic disease and the immune microenvironment is different in each metastasis.10 Our study focuses on blood samples. Our results point towards a prominent role for innate immunosuppression, with a leading role for myeloid-derived suppressor cells, as has also been shown in other cancers.23 26 27 In our study, the myeloid lineage seems to have a role in discriminating benign cysts from malignant counterparts. However, this needs to be confirmed in larger series. Granulocytic myeloid-derived suppressor cells, on the other hand, seem to behave in the opposite way, which is not a surprise since both cell types are technically calculated from the same background (ie, CD11b+ cells and one cell cannot be both monocytic and granulocytic).28 Moreover, these cells are the most susceptible to freezing and thawing,16 are difficult to discriminate by fluorescence activated cell sorting from neutrophils, tend to be less immunosuppressive, and their ontogenesis is often questioned.29 Based on the existing literature underlining a role for monocytic myeloid-derived suppressor cells in cancer,23 30 we propose further investigation is needed on their role in immune research in ovarian cancer. A practical problem resulting from this probably important effect of the innate immune system on progression of ovarian cancer is that the current immunotherapies—which mainly focus on immune checkpoint inhibition—are probably insufficient to reverse immunosuppression. Therapy focusing on manipulating this innate immune suppression could be of interest. Myeloid-derived suppressor cells may be targeted in four ways: (1) depletion of myeloid-derived suppressor cells; (2) blocking recruitment; (3) myeloid-derived suppressor cell differentiation; and (4) inhibition of immunosuppression mediated by these cells. Interesting targets can be all-trans retinoic acid that induces an up-regulation of gluthathione in myeloid-derived suppressor cells. This reduces the level of reactive oxygen species and therefore induces maturation of myeloid-derived suppressor cells into macrophages.31 One may also target migration of myeloid-derived suppressor cells by blocking the inflammatory mediator prostaglandin E2, which is responsible for inducing among others, C-X-C motif chemokine 12, involved in the migration of myeloid-derived suppressor cells towards ovarian cancer ascites.32 Phosphodiesterase-5 inhibitors reduce the immune suppressive activity of myeloid-derived suppressor cells by down-regulation of the interleukin-4 receptor α expression on tumor-infiltrating myeloid-derived suppressor cells, thereby reducing the expression of arginase 1.33

In conclusion, this is the first study describing both the innate and the adaptive immune system within the total range of ovarian pathology, at the systemic level. Immune profiling based on myeloid-derived suppressor cells reaches an AUC of 0.858, discriminating between benign and malignant cysts. In contrast to the prevailing thought, our data suggest an important role for the innate immune system in diagnosis and prognosis of ovarian tumors.

Acknowledgments

The authors would like to thank Anaïs Van Hoylandt, Gitte Thirion, and Godelieve Verbist for their outstanding technical contribution (isolation of blood samples, fluorescence activated cell sorting, immunohistochemistry).

References

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.

Footnotes

  • Contributors All authors have provided substantial contribution to the study and/or the manuscript and are in agreement will all aspects of the final manuscript.

  • Funding This work was supported by Kom Op Tegen Kanker (Stand Up to Cancer), the Flemish Cancer Society under grant 2016/10728/2603 to AC; the Olivia Fund under grant 2017/LUF/00135 to AC; Amgen Chair for Therapeutic Advances in Ovarian Cancer under grant 2017/LUF/00069 to IV; Internal Funds KU Leuven under grant C24/15/037 to BVC and DT; Research Foundation – Flanders (FWO) under grants G0B4716N to BVC and DT; 12F3114N to AC; 1803415N to DT; 12N4415N to TV; and Linbury Trust Grant under grant LIN2600 to CL.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • 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.