Significant cardiac morbidity and mortality exists in patients with COPD. Shared risk factors include age, smoking history and exposure to air pollution and passive smoke. Although the inappropriate under-prescribing of β-blockers contributes, it is now appreciated that the observed cardiac risk is not only due to smoking and conventional cardiovascular risk factors, but also other independent factors. A number of hypotheses exist for the increased cardiovascular morbidity and mortality seen in COPD including inflammation, pulmonary hypertension, lung hyperinflation and shared genetics models. Mounting evidence from large randomised controlled trials suggests that COPD treatment may be cardio-protective. We review the current evidence supporting the aforementioned hypotheses and how their modulation may prevent cardiovascular morbidity and mortality in COPD. The persisting underdiagnosis of COPD may have significant consequences. Further mechanistic studies identifying the onset and impact of individual interventions will develop our understanding of this emerging and highly relevant clinical field.
- Imaging and diagnostics
- chronic lung disease
- myocardial disease
- cardiomyopathy apical
- cardiomyopathy hypertrophic
- myocardial fibrosis
- ventricular hypertrophy
Statistics from Altmetric.com
- Imaging and diagnostics
- chronic lung disease
- myocardial disease
- cardiomyopathy apical
- cardiomyopathy hypertrophic
- myocardial fibrosis
- ventricular hypertrophy
Increased cardiovascular morbidity and mortality in COPD
Chronic obstructive pulmonary disease (COPD) is predicted to become the sixth leading cause of disability and the third most common cause of death by 2020.1–3 COPD is commonly underdiagnosed in the UK4 and abroad.5–7 Its 1% prevalence across all ages rises steeply to 9%–10% for those over 40.8 A large proportion of morbidity and mortality in COPD is associated with cardiovascular complications.9
Reduced pulmonary function, no matter what cause, is associated with increases in all-cause and cardiac mortality, myocardial infarction and arrhythmia.10–13 Forced expiratory volume in one second (FEV1) is ranked second to smoking and above blood pressure and cholesterol as a predictor of all-cause and cardiovascular mortality.14 It has been suggested that a reduction in FEV1 combined with a smoking history better predicts cardiovascular mortality than cholesterol.15
In COPD, a large proportion of patients succumb to cardiovascular causes rather than respiratory failure (table 1). The likelihood of cardiac mortality, ventricular arrhythmias, coronary artery disease (CAD) or congestive cardiac failure increases with worsening FEV1, the rate of FEV1 loss independently predicting CAD mortality.23 ,28 Newer assays, such as the highly sensitive Troponin, have further highlighted the subclinical myocardial damage occurring during COPD exacerbations and are predictive of mortality.29
The impact of shared exposures
It may be that the relationship between COPD and cardiovascular disease (CVD) is solely due to shared common risk factors.
It is well accepted that smoking causes both COPD and CVD. Passive smoke inhalation, in a dose–response manner, also increases the likelihood of fatal and non-fatal myocardial infarction with an estimated pooled OR from a meta-analysis of 1.22 (1.04–1.41) and 1.32 (1.04–1.67), respectively. Similarly, increased risks have been observed for COPD mortality with ORs related to spousal smoking of 1.67 in never-smoking male subjects.30
Recently, there has been a significant focus on international public health strategies to combat the adverse effects of smoking, and secondhand smoke in particular. Although the impact of smoke-free legislation (SFL) on respiratory symptoms is well studied in normal individuals31–33 data on lung function are less robust since the changes noted were arguably of little clinical relevance,34 ,35 suffered from poor participant compliance36 and were confounded by seasonal variation in temperature.37 The impact of SFL on COPD is unclear since the few studies that have assessed the rate of admissions for COPD exacerbations show conflicting results.38 ,39 This is in contrast to the data for CVD. A recent meta-analysis assessing the impact of SFL revealed a reduction of acute coronary syndrome risk in 30 of 35 estimates with a 10% (95% CI 6 to 14, p<0.001) pooled RR reduction,40 supporting the notion that smoke exposure has a significant independent role in cardiovascular morbidity, although whether those patients with concomitant COPD achieved equivalent improvements in coronary events is unclear.
The development of COPD as a result of damage caused by air pollution is well documented.41 ,42 Day-to-day changes in pollutant levels and longer term exposure also affect CVD risk, and all commonly measured pollutants are positively associated with increased mortality and hospital admissions for CVD. Hypotheses relate to direct pollutant effects of oxidative stress, pulmonary and systemic inflammation or pro-coagulant reactions which, as will be discussed later, have a significant overlap with the postulated effects of COPD. Although observational studies have found associations between cardiac autonomic dysfunction and air pollution, this has not been borne out in controlled conditions and further mechanistic explanations are needed for this discrepancy.43–45
The impact of differential prescribing practices
COPD patients have been excluded from clinical trials involving β-blockers due to historical concerns over bronchoconstriction. Despite database studies persistently demonstrating survival benefit for COPD patients receiving β-blockers in CAD and congestive cardiac failure, and meta-analyses confirming the safety of cardioselective β-blockers in COPD of all severities, underuse still occurs due to lingering concerns over bronchoconstriction in a population which, based on the above evidence, arguably have the most to gain.46–48
Although the underprescribing of key medication contributes to cardiac risk, it is now appreciated that COPD is a multi-system disorder that exerts ‘systemic effects’. Despite the aforementioned shared risk, the intriguing finding that increased cardiovascular morbidity and mortality in COPD are independent of smoking as well as other conventional cardiovascular risk factors, age and gender is driving research into understanding the underlying mechanisms of this increased risk and to what extent it is modifiable.26 ,28 ,48–53
Models explaining the interaction between COPD and CVD
The hypotheses relating to why COPD patients are at increased risk of developing or succumbing to CVD are best summarised through the following models:
Pulmonary hypertension (PH) model
Lung hyperinflation model
Shared genetics model.
Evidence is growing that COPD and CVD may be linked through inflammatory processes contributing to atherosclerosis.
Inflammation and CAD
The involvement of inflammation in the development of atheromatous plaques is well established.54 ,55 The majority of cells are monocytes which gain entry, along with T cells and mast cells,56 ,57 via vessel adhesion molecules after exposure to irritative stimuli including cytokines, dyslipidaemia and hypertension. The monocytes develop into macrophages which themselves have an array of pro-inflammatory actions when studied in mice models.58 Thrombotic complications are not always at the most severely narrowed site but at recently disrupted fibrous plaques, which typically contain macrophages,59 whose caps have been thinned by inflammatory and immune processes.60 Thus, the fate of an atherosclerotic lesion depends on the balance between pro-inflammatory and anti-inflammatory cytokines which in turn determines the magnitude of the inflammatory response.61
C reactive protein (CRP) is an acute phase reactant and a marker of systemic inflammation. CRP and interleukin (IL)-6, the precursor to CRP, are increased during unstable angina62 ,63 and are markers of poor prognosis.64 In situations of coronary vasospasm rather than plaque rupture CRP is normal, suggesting vessel inflammation and not myocardial ischaemia causes the inflammatory response.65 CRP measured on a highly sensitive immunoassay can predict myocardial infarction and mortality in apparently healthy individuals.66 ,67 Although association does not prove causality, many are convinced that further investigation is warranted.
Hallmarks of atherosclerosis include endothelial dysfunction with reduced arterial compliance, increased central arterial pressure, left ventricular (LV) afterload and reduced diastolic coronary artery filling. A measure of arterial stiffness, known as aortic pulse wave velocity (PWV), can be measured non-invasively using applanation tonometry of the radial artery.68 Repeatedly, aortic PWV has been shown to be an independent predictor of CAD in hypertensives, diabetics and the general population including the older populations.69–74 PWV is also associated with disease duration and the presence of inflammatory mediators CRP and IL-6, independent of atherosclerosis, in rheumatoid arthritis and other chronic inflammatory conditions.75 ,76
Statins have potent lipid lowering but also anti-inflammatory effects. The JUPITER trial, a randomised double-blind placebo-controlled trial, demonstrated a 1.2% absolute risk reduction and a 44% RR reduction in combined cardiovascular end points in healthy individuals when treated with rosuvastatin versus placebo. Significantly, participants had elevated HsCRP but lipid levels below treatment threshold, suggesting benefit due to anti-inflammatory rather than lipid lowering effects. Much controversy remains as to whether the population received care consistent with current standards77 and confounding lipid-lowering actions are hard to ignore. Future studies involving methotrexate and canakinumab, potential cardiovascular therapeutic agents without confounding effects, are planned to help establish the contribution of inflammation. Respiratory therapeutics may also have a role to play.60 ,78
Inflammation and COPD
COPD causes an airway inflammatory process, distinct from asthma, involving neutrophils, macrophages, T lymphocytes and augmented concentrations of leukotriene B4, IL-1, IL-6 and IL-8b and tumour necrosis factor α.79–84 There is evidence suggesting inflammation may spill-over from the lungs to the systemic circulation, most likely caused by alveolar macrophages which clear inhaled particles and initiate local and systemic inflammation.85 ,86 Animal studies have shown that movement of lung proteins to the systemic circulation is dependent on a number of factors including lung inflammation,87–89 whereas human studies provide circumstantial evidence in the form of increased circulating levels of tumour necrosis factor α, CRP and fibrinogen.90 Furthermore, positron emission tomography studies have demonstrated excessive aortic inflammation in COPD,91 whose associated comorbidities include skeletal muscle dysfunction,92 osteoporosis93 and diabetes,84 which all have inflammatory aetiologies.
A spill-over of pulmonary inflammation may lead to downstream effects such as arterial stiffness which could provide a mechanistic link among COPD, CAD and diastolic dysfunction (DD).94–96 Studies have indeed shown that arterial stiffness is raised in COPD and related to the extent of airflow obstruction and emphysema.97 ,98
Although PH is an established prognostic factor, its true prevalence in COPD is unknown due to selection bias in studies mainly assessing hospitalised patients.99 The pathogenesis of pulmonary vascular disease is multifactorial, relating to alterations in vascular and pulmonary biology, gas exchange, structural changes in the pulmonary vasculature as well as mechanical factors.100 Pulmonary artery pressure (PAP) depends on pulmonary artery wedge pressure, cardiac output and pulmonary vascular resistance (PVR). Many factors increase PVR in COPD (table 2) and alveolar hypoxia which results in vascular remodelling and hypoxic vasoconstriction localised to the small precapillary arteries is classically considered to be the most important.102
The development of PH in COPD is classically attributed to severe hypoxia in end-stage disease causing raised pulmonary pressures, right ventricular hypertrophy, dilatation and systolic failure, or ‘chronic cor pulmonale’ resulting in reduced exercise tolerance and survival. This simple hypothesis remains controversial;103 vascular remodelling is seen in patients with milder forms of the disease100 and pulmonary endothelial dysfunction develops in COPD before the onset of hypoxaemia which is thought to be due to the combination of the direct effects of cigarette smoke and that of local inflammation, and is considered one of the initiating events in the progression to PH.104
It is argued that the severity of PH in most COPD patients cannot account for all the manifestations; although transient rises in PAP occur during nocturnal hypoxaemia105 and exacerbations,106 only 1%–4% have a PAP >40 mm Hg at rest, significantly lower than other forms of PH,107 and in sufficiently severe COPD to be considered for lung volume reduction surgery, the PAP may only be moderately elevated at baseline (26.3±5.2 mm Hg).108 ,109
These observations have led to alternative explanations and a better appreciation of the contribution of DD to morbidity.110 The chronically pressure-overloaded right ventricle (RV) causes leftward septal shift, ventricular interdependence and LV DD despite normal LV systolic function.111 The impact of DD should not be discounted; in a single-centre echocardiographic out-patient study, including patients with COPD, moderate to severe DD independently predicted mortality in patients with normal systolic function.112 Although PH associated DD contributes to the cardiac morbidity of COPD, other mechanisms must also contribute since subclinical DD exists in milder airflow obstruction without PH.113 ,114
Lung hyperinflation model
Lung hyperinflation results from the loss of elastic recoil combined with expiratory flow limitation, promoting increased expiratory lung volume (or ‘air trapping’) and intrinsic positive end-expiratory pressure with significant clinical consequences in those with a more severe phenotype, including a twofold increase in all-cause mortality.115 ,116 It exerts its effects via two mechanisms.
The mechanical effects of hyperinflation
Lung hyperinflation increases mediastinal pressures through stiffening of the parietal pleura and cardiac fossa, with potentially significant cardiovascular physiological consequences including left and right DD,117 an assertion supported by a 2816 participant lung substudy of ‘Multi-ethnic study of atherosclerosis’, a prospective cohort study designed to identify subclinical measures of CVD. Significant associations were found between CT-quantified levels of emphysema and MRI measures of LV end-diastolic and end-systolic volumes.118 Although recent echocardiographic studies have shown hyperinflation to independently predict cardiac chamber size, the unaffected isovolumetric relaxation time suggests reduced preload conditions rather than a stiff myocardium as the cause for DD, a concept supported by Jorgensen et al who found associations between decreased intrathoracic blood volume and LV performance in severe emphysema.119 ,120
Neuro-humoral effects of hyperinflation
Activation of the sympathetic nervous system (SNS) in normal individuals causes cardiovascular remodelling, with altered PWV, arterial compliance and DD, which may occur through direct effects on tone, modulation of baro-receptor sensitivity or activation of the renin-angiotensin system.121–124 SNS activation in COPD is evidenced through reduced heart rate variability, increased norepinephrine turnover and increased plasma–renin activity.125 Hyperinflation, as well as its mechanical restrictions, may also impact on the SNS via a lung inflation reflex, increased hypoxia and neural-respiratory drive, which drives skeletal muscle contractility with the subsequent ischaemic release of free radicals.126
Shared genetics model
Genome wide association studies
There may be more to the relationship between COPD and CVD than simply sharing smoking as a risk factor. Parallel pathological processes, as a result of shared genetics, may mediate the relationship between COPD and CVD via gene–environment interactions causing connective tissue remodelling through modification of elastin content and proteolytic enzyme activity. There is extensive literature on the role of genomics and genome wide association studies in the understanding of the complex polygenic interactions of COPD and CVD and the interested reader is referred to two well-written reviews on the subject.127 ,128 Two examples are provided here to illustrate this large area of research. Matrix metalloproteases (MMPs) are proteolytic enzymes that degrade collagen and elastin within the lung matrix and have other pro-inflammatory roles in smoking-induced emphysema models. MMPs also contribute to vascular remodelling in PH and it is widely accepted that MMP polymorphisms are pivotal to non-pulmonary vascular remodelling in atherosclerosis.129–131 Polymorphisms of epoxide hydrolase and glutathione-s-transferase, involved in oxidative stress, are strongly associated with upper lobe emphysema and are risk factors for MI and atherosclerosis, respectively.132 ,133
Telomere function and the ageing process
‘The senescence hypothesis’ is the shortening of telomere length, a marker of cellular ageing, caused by the failure to protect DNA against oxidative injury and reactive oxygen species.134 Telomere shortening accelerates end-stage heart failure135 and several studies in diverse populations have shown an association, which exists before the development of clinical disease, between shorter telomeres in circulating leucocytes and both CAD and coronary artery calcification.136 ,137 The oxidative damage and chronic inflammation in COPD, originating from environmental irritants such as cigarette smoking and air pollution, may reduce telomere length in circulating cells138 ,139 with important downstream effects on auto-immunity,140 potentially explaining the persistence of lung inflammation in COPD despite many years of smoking cessation, driving telomere attrition and increasing cardiac risk over and above the risk associated with smoking. Inflammation and oxidative stress are involved in virtually all cardiovascular and associated diseases which may explain the wide number of conditions associated with telomere length141 and the emerging evidence points towards the potential use of telomere length as a marker of both cardiovascular and obstructive airways disease or as a therapeutic target.142
Evidence that COPD treatment reduces cardiovascular morbidity and mortality
Effects of smoking cessation
Smoking cessation results in an almost immediate improvement in blood pressure and heart rate, with the risk of stroke and CAD normalised to that of never smokers between 5 and 15 years after cessation. During the 5 years follow-up of The Lung Health Study, involving 5887 smokers with mild to moderate COPD, smoking cessation was associated with a 32% reduction in all-cause mortality, the trend persisting for 14.5 years of follow-up, and a 45% reduction in cardiovascular mortality likely through a direct reduction of cardiovascular events, but probably also through reducing the rate of COPD decline.143
Effects of oxygen therapy
In the Nocturnal Oxygen Therapy Trial, 203 COPD patients with evidence of chronic hypoxaemia were randomised to receive oxygen supplementation. After a minimum follow-up of 12 months, the RR of death with nocturnal versus continuous oxygen was 1.94 (95% CI 1.17 to 3.24).144 This protective effect was also seen in the Medical Research Council study where 5-year mortality for those receiving oxygen was half that of the control group.145 The benefit of oxygen in COPD patients with transient hypoxaemia related to sleep or exercise remains to be determined and is the subject of the ‘Long-term Oxygen Therapy Trial’.146
Effects of pulmonary rehabilitation
It is unclear to what extent the reductions in hospital admissions, exacerbations and mortality seen following pulmonary rehabilitation147 in COPD are a result of reduced cardiac morbidity. Endurance training results in adaptive responses of the heart including resting bradycardia, increasing end-diastolic dimensions, non-pathological cardiac hypertrophy, improved ventricular function and resistance to ischaemic insult.148 Significant improvements (10.3±0.7 to 9.2±0.8 m/s) in carotid-radial PWV occurred in an isolated exercise intervention. A case-control study of a 7-week multidisciplinary pulmonary rehabilitation course demonstrated significant reductions in PWV from 9.8 (±3) to 9.3 (±2.7) due to a reduction of 10 mm Hg in systolic blood pressure, the magnitude of which, if sustained, being sufficient to reduce cardiovascular morbidity and mortality by at least 13%.149 ,150 Improvements in dynamic hyperinflation (progressive air trapping during exercise), neuromuscular coupling or tolerance to dyspnogenic stimuli may have contributed to PWV improvement.151
Effects of inhaled COPD therapies
Two recent multi-centre, randomised, double-blind, parallel-group, placebo-controlled trials have raised the possibility that pharmacotherapy could affect survival. The 3-year TOwards a Revolution in COPD Heath study was conducted at 444 centres in 42 countries and randomised 6184 patients. Comparing salmeterol, fluticasone propionate and salmeterol–fluticasone propionate (SFC) combination inhaler with placebo, a reduction in all-cause mortality, the primary efficacy end point, despite a large differential dropout from the placebo arm would have reached statistical significance (p=0.04) but for an interim analysis for safety resulting in a corrected p value of 0.052.18
In the 4-year Understanding Potential Long-Term Impacts on Function with Tiotropium trial, 5993 patients were randomised to receive tiotropium or placebo. Hospitalisations and mortality 1-month following the completion of the trial (and cessation of study drug) was the prespecified secondary outcome measure. Although a reduction in all-cause mortality versus placebo on an intention to treat basis was demonstrated, this did not reach statistical significance (HR 0.89, CI 0.79 to 1.02). However post hoc, there was a significant reduction in all-cause mortality while on treatment after the protocol-defined treatment period of 4 years (HR 0.84, CI 0.73 to 0.97 p=0.016).152
Data on adverse events in these trials further raise the possibility of cardioprotective effects. A post hoc analysis in TOwards a Revolution in COPD Heath showed fewer cardiovascular events within the SFC group after adjusting for smoking status and gender (SFC 11.3%, fluticasone propionate 13.8%, salmeterol 13.4%, placebo 14.6%). In Understanding Potential Long-Term Impacts on Function with Tiotropium trial, an RR reduction in cardiac mortality (RR 0.86 95% CI 0.75 to 0.99), congestive heart failure (RR 0.59 95% CI 0.37 to 0.96) and MI (RR 0.71, 95% CI 0.52 to 0.99) was observed.153 ,154
Could the COPD models explain the beneficial effects of COPD treatment?
Effects of inhaled COPD therapies
Inhaled fluticasone with or without long acting β2-agonists reduce exacerbations and improve health status in COPD.18 Observational studies155 as well as double-blind placebo-controlled biopsy studies have confirmed modulation of inflammation within the airway following administration of inhaled corticosteroids.156 ,157 Research data are conflicting with regard to whether inhaled therapies can reduce levels of systemic inflammation by reducing the spill-over of lung inflammation. Pinto-Plata et al found a significantly raised CRP level in COPD patients, which was 20% lower in those taking inhaled corticosteroids.158 An open-label prospective study demonstrated that SAL and fluticasone significantly reduced circulating CRP levels compared with a combination of an anticholinergic and albuterol.159 Furthermore, a proof of concept double-blind placebo-controlled clinical trial by Man and Sin found that withdrawing inhaled steroids increased CRP levels and re-introduction of inhaled fluticasone for 2 weeks resulted in 50% and 26% reductions in CRP and IL-6, respectively.160 However, when repeated as an 11-centre randomised, placebo-controlled, double-blind trial, no significant changes in CRP or IL-6 were identified, which may be due to a higher proportion of smokers in the latter study or the use of long-term inhaled corticosteroids modifying inflammatory processes within the lung.161 In any case, it is argued that the lack of correlation between inflammatory biomarkers in the airway and the systemic circulation does not necessarily exclude the possibility of ‘spill-over’ since many factors will influence concentrations in different body compartments; for example, receptor binding influences the free concentrations of many cytokines, with potential downstream spill-over effects on the cardiovascular system without noticeable changes to the intermediary biomarkers.90
One randomised control trial has looked at the effect of inhaled therapies on PWV in COPD. In all, 249 patients were randomised to receive fluticasone–SAL combination inhaler (Seretide) or placebo with a primary outcome measure of the change in PWV at 12 weeks, which did not reach statistical significance on an intention to treat basis. However, in those patients who remained on the study drug throughout the trial, a significant, although modest, reduction (0.4 m/s) in PWV was achieved, which was found in a post hoc analysis to be more marked in those with a PWV of >10.9 m/s at baseline. Although an increase in aortic PWV of 1 m/s corresponds to an adjusted increased risk of 15% in CV mortality,162 the reverse is less clear. Clinically significant reductions are yet to be established and the effect of these reductions on cardiac structure and function are, until such studies are carried out, unknown.
The effect of PDE4 inhibitors
Phosphodiesterase-4 (PDE4), a member of the PDE enzyme superfamily that inactivates cyclic AMP and cyclic guanosine monophosphate, is the main isoenzyme involved in inflammatory airway disease.163 The PDE4 inhibitors are a new anti-inflammatory drug with efficacy and acceptable tolerability in preclinical studies in COPD with no effect on cardiac repolarisation.164 ,165 A Cochrane review involving 23 randomised control trials concluded that PDE4 inhibitors, including roflumilast and cilomilast, were associated with significant improvements in FEV over placebo regardless of severity or concomitant treatment, and reduced the likelihood of exacerbations.166 Given that it is administered orally, it seems likely that it will exert a systemic effect on inflammation thereby potentially having a dual benefit on to the cardiovascular system. In vitro studies have pointed towards a third cardioprotective effect through prevention of nitric oxide induced myocyte apoptosis, potentially protecting against myocardial ischaemia and ischaemic reperfusion injury.167
Lung hyperinflation model: the effect of lung deflation
Bronchodilators, whether β-agonist or anticholinergic, improve air trapping in mild and more severe forms of COPD. Lung deflation results in improved exertional dyspnoea and exercise tolerance due to a combination of reduced inspiratory muscle loading and dynamic hyperinflation.168 ,169 Some of these improvements may also be due to an improvement in cardiovascular function. Jorgensen demonstrated increased LV end-diastolic dimensions and filling with significant improvements in cardiac index following lung volume reduction surgery. In the National Emphysema Treatment Trial cardiovascular substudy, there were improvements in capillary wedge pressure postlung volume reduction surgery. At 6 months, there was an improvement in ‘O2 pulse’, the total oxygen consumed during maximal effort per heart beat and a surrogate for stroke volume, in a subsequent post hoc analysis, which, interestingly, was also seen in a proportion of patients treated with medical therapy alone.170–172 Although this offers an insight into how lung deflation may improve cardiac function, the impact and exact onset is unclear since prospective randomised control trial data are lacking.
It is likely that a combination of the postulated mechanisms in the PH, inflammation and hyperinflation models contribute to the cardiovascular morbidity seen in COPD. A growing body of evidence is emerging that this risk is modifiable and the persisting underdiagnosis of COPD may have significant consequences with regard to cardiovascular morbidity and mortality. Further mechanistic studies identifying the onset and impact of individual interventions will help further our understanding of this emerging and highly relevant clinical field.
This work forms part of the research themes contributing to the research portfolio of Barts and the London Cardiovascular Biomedical Research Unit, supported and funded by the National Institute for Health Research.
Funding ISS is an employee of Barts and the London NHS trust who has received a research grant from GlaxoSmithKline. NCB is directly funded by Barts and the London NHS trust. SEP was directly funded by the Barts and The London National Institute for Health Research Cardiovascular Biomedical Research Unit.
Competing interests None.
Provenance and peer review Commissioned; externally peer reviewed.
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.