Article Text


Research paper
Cerebrospinal fluid amyloid-β and phenotypic heterogeneity in de novo Parkinson's disease
  1. Guido Alves1,2,
  2. Kenn Freddy Pedersen1,2,
  3. Bastiaan R Bloem3,
  4. Kaj Blennow4,
  5. Henrik Zetterberg4,
  6. George F Borm5,
  7. Turi O Dalaker1,6,
  8. Mona K Beyer1,6,
  9. Dag Aarsland7,
  10. Ulf Andreasson4,
  11. Johannes Lange1,
  12. Ole-Bjørn Tysnes8,
  13. Robert Zivadinov9,
  14. Jan Petter Larsen1
  1. 1The Norwegian Centre for Movement Disorders, Stavanger University Hospital, Stavanger, Norway
  2. 2Department of Neurology, Stavanger University Hospital, Stavanger, Norway
  3. 3Department of Neurology, Radboud University Nijmegen Medical Center, Donders Institute for Brain, Cognition and Behaviour, Nijmegen, The Netherlands
  4. 4Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, the Sahlgrenska Academy at University of Gothenburg, Mölndal, Sweden
  5. 5Department of Epidemiology, Biostatistics and HTA, Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands
  6. 6Department of Radiology, Stavanger University Hospital, Stavanger, Norway
  7. 7Department of Psychiatry, Stavanger University Hospital, Stavanger, Norway
  8. 8Department of Neurology, Haukeland University Hospital, Bergen, Norway
  9. 9Buffalo Neuroimaging Analysis Center, The Jakobs Neurological Institute, State University of New York, Buffalo, New York, USA
  1. Correspondence to Dr Guido Alves, The Norwegian Centre for Movement Disorders, Stavanger University Hospital, PO Box 8100, Stavanger N-4068, Norway; algu{at}


Background In Parkinson's disease (PD), the motor presentation characterised by postural instability/gait difficulties (PIGD) heralds accelerated motor, functional and cognitive decline, as compared with the more benign tremor-dominant (TD) variant. This makes the PIGD complex an attractive target for the discovery of prognostic biomarkers in PD.

Objective To explore in vivo whether variability in brain amyloid-β (Aβ) metabolism affects the initial motor presentation in PD.

Methods We quantified cerebrospinal fluid (CSF) concentrations and ratios of Aβ42, Aβ40 and Aβ38 using a triplex immunoassay in 99 patients with de novo PD with the PIGD phenotype (n=39) or the TD phenotype (n=60). All patients underwent standardised assessments of motor and neuropsychological function and cerebral MRI. 46 age-matched normal controls served as external reference.

Results Patients with PD with the PIGD phenotype had significantly reduced CSF Aβ42, Aβ38, Aβ42/40 and Aβ38/40 levels compared with patients with the TD phenotype and controls. CSF marker levels in patients with PD-TD did not differ from those in controls. Multivariate regression models demonstrated significant associations of CSF Aβ markers with severity of PIGD and lower limb bradykinesia in patients with PD, independently from age, MRI white matter hyperintensities and cognition. No associations were found between CSF markers and other motor features.

Conclusions Motor heterogeneity in de novo PD independently relates to CSF Aβ markers, with low levels found in patients with the PIGD presentation. This suggests that disturbed Aβ metabolism has an effect on PD beyond cognition and may contribute to the variable rate of motor and functional decline in PD.

  • Parkinson's Disease
  • GAIT
  • CSF
View Full Text

Statistics from


Parkinson's disease (PD) is the second most common neurodegenerative disorder after Alzheimer's disease. There is a remarkable interindividual variation in the pattern and progression of motor and cognitive deficits in PD. Ample evidence exists that the initial motor presentation is predictive of the evolution of PD. Most consistently, early axial motor involvement, often called PIGD (postural instability and gait difficulties), has been shown to herald accelerated motor and functional decline in PD.1 ,2 In addition, PIGD symptoms signal a faster rate of cognitive decline and increased risk for subsequent dementia in more advanced stages of the disease.3 ,4 A better understanding of the neurobiology of the PIGD complex, and motor heterogeneity, in general, might facilitate the discovery of prognostic biomarkers and new treatments and has therefore been earmarked as a prioritised research area in PD.5 ,6

The aetiology of PIGD in PD is not fully elucidated. Both PD-related and comorbid pathologies have been proposed as potential underlying causes of PIGD in PD. PIGD symptoms do not respond well to dopaminergic treatment and are therefore considered to reflect mainly non-dopaminergic pathologies. In particular, impaired attention and executive function arising from cholinergic deficits has emerged as potential source of gait impairment and related falls in PD.7 In support of this, treatment with cholinesterase inhibitors significantly reduced fall frequency in a recent randomised controlled trial, yet many patients with PD receiving active treatment showed no, or only mild, improvement.8 This indicates that other, possibly yet unrecognised, neurobiological factors may contribute to gait and balance problems in PD.

Although PD is an α-synucleinopathy, amyloid-β (Aβ) pathology is increasingly acknowledged as an important determinant of clinical variability in PD.9 Clinicopathological studies in PD suggest a close temporal relationship between PIGD and subsequent dementia, and early dementia development with increased Aβ deposition at autopsy.9–12 Because substantial evidence suggests that Aβ changes precede the onset of related cognitive decline,13 this raises the possibility that PIGD symptoms—at least in part—might be an early clinical manifestation of Aβ brain mis-metabolism in PD. However, post mortem data do not allow any firm conclusions.

Cerebrospinal fluid (CSF) is a promising source for biomarker discovery and—in contrast to brain tissue obtained at autopsy—allows monitoring biochemical processes in vivo. CSF Aβ42 is one of the most validated biomarkers in neurodegenerative research, with lowered levels reflecting cerebral Aβ aggregation and deposition.13 Other Aβ species, such as Aβ38 and Aβ40, have been less studied but may serve as complementary biomarkers of Aβ pathology. These markers are thought to reflect disturbed Aβ production rather than deposition.14 Several recent independent large-scale studies find CSF Aβ42 alterations in patients with PD with dementia and also in a proportion of non-demented, even cognitively unimpaired, subjects.15–17 In the first CSF study facilitating a population-based, incident PD cohort, we previously reported that CSF levels of Aβ42, Aβ40 and Aβ38 are related to neuropsychological measures of memory impairment and are lowered in a subset of patients with newly diagnosed PD, yet not as much as in Alzheimer's disease.18

We here explored in vivo whether variability in Aβ brain mis-metabolism has impact beyond cognition and is an independent determinant of motor heterogeneity from the earliest clinical stages of PD.

Subjects and methods

Subjects with PD

Subjects were newly diagnosed and untreated patients with PD who are participating in the ongoing Norwegian ParkWest project, a population-based study of the incidence, neurobiology and prognosis of PD.19 The recruitment and diagnostic procedures to establish a population-representative incident PD cohort have been described elsewhere in detail.18 ,19 A diagnosis of PD was made by neurologists experienced in movement disorders, based upon medical history, detailed clinical, neuropsychiatric and neuropsychological examination at study entry and repetitive clinical assessments to ensure that subjects did not develop atypical features, including dementia during the first year of motor onset.19 In addition, neuroimaging with cerebral MRI and, where appropriate, dopamine transporter imaging, was performed (see below).19 Exclusion criteria for this CSF study were (1) symptomatic treatment within 2 weeks before study entry; (2) incomplete motor examination at baseline; (3) non-progressive parkinsonism or negative dopamine transporter imaging or (4) clinical or MRI signs indicating atypical or secondary parkinsonism.19 Of the 109 patients included in our previous study,18 106 fulfilled accepted diagnostic criteria of PD20 and had no exclusion criteria. Written informed consent was obtained from all participants. The study protocol was approved by the regional committee for medical research ethics, Western Norway.

Examination programme

Data reported here were obtained during the baseline assessments of the ParkWest study, which were conducted an average of 1.2 months from diagnosis. The clinical and neuropsychological examination programme and lumbar puncture were performed within two consecutive days before the start of treatment or after at least 2 weeks of drug withdrawal following a short (< 1 week) treatment period (n=3). In addition, cerebral MRI was conducted near baseline, as follows.

Clinical examinations

Clinical examinations included a full medical history and a general medical and standardised neurological examination by a study neurologist experienced in movement disorders. Disease stage was determined by the (modified) Hoehn and Yahr staging.21 Severity of parkinsonism was assessed by the Unified Parkinson's Disease Rating Scale (UPDRS).22 Measures of tremor, rigidity, bradykinesia and PIGD were derived from the UPDRS activities of daily living and motor sections. Based on the relative severity of tremor (mean of UPDRS items 16, 20–21) vs PIGD symptoms (mean of UPDRS items 13–15, 29–30), motor phenotype was determined as either tremor-dominant (TD, n=60) or postural instability/gait difficulty (PIGD, n=39) phenotype, following the classification algorithm proposed by Jankovic et al.1 Motor subtype remained indeterminate in seven subjects, who were excluded, leaving 99 patients eligible for this study.

Neuropsychological examinations

Patients with PD underwent comprehensive neuropsychological assessments that included the Mini-Mental State Examination (MMSE)23 as a measure of global cognitive performance and neuropsychological examination of verbal memory (California Verbal Learning Test II), executive function and attention (Stroop, semantic verbal fluency, serial-7) and visuospatial function (Visual Object and Space Perception battery). Composite z-scores, corrected for age, sex and education, were calculated for each cognitive domain (memory, attentional-executive, visuospatial) using locally collected normative data.24

Magnetic resonance imaging

MRI was conducted in 89 of the 99 patients with PD using a standardised protocol on average 30 days after study entry.25 MRI acquisition was performed on a 1.5T (n=71) or 1.0T (n=18) system. MRI scans were missing in 10 subjects owing to contraindications (n=3), claustrophobia (n=2), incomplete examination (n=1) and unknown reasons (n=4). None of these 10 subjects had a history or clinical signs of cerebrovascular disease. All available scans were visually inspected for artefacts or focal pathology before analyses, with no evidence of cortical infarction on MRI in any patient. MRI analyses were conducted at the Buffalo Neuroimaging Analysis Centre, Department of Neurology, University at Buffalo, Buffalo, New York, USA, by operators blinded to all clinical patient characteristics. White matter hyperintensities (WMH) were outlined on each axial FLAIR (fluid attenuation inversion recovery) image slice using a reproducible, semiautomated local threshold technique with Java Image software (V.3.0, Xinapse Systems, Northants, UK, and calculated by a single rater (TOD).

Control subjects

We obtained CSF from 46 normal control subjects (NC) as external reference for CSF marker analyses in patients with PD. These subjects had undergone diagnostic investigation at the neurological department or elective knee or hip surgery at the department of orthopaedic surgery, Stavanger University Hospital, Norway. None of these subjects had a history or clinical evidence of a central nervous system disease, cognitive impairment (MMSE≥28), or psychiatric disorder.

CSF procedures

Lumbar puncture was performed in all subjects after an overnight fast and between 7:00 and 10:00 to minimise diurnal variation of CSF marker levels. Samples were sent immediately to the laboratory, centrifuged and supernatants stored in polypropylene vials at −80° C. CSF levels of Aβ42, Aβ40 and Aβ38 were quantified by an Aβ triplex assay (human Aβ peptide ultra-sensitive kit, Meso Scale Discovery, Gaithersburg, Maryland, USA) which uses C-terminus specific antibodies to capture the different Aβ peptides and a SULFO-TAG–labelled anti-Aβ antibody (4G8) for detection with electrochemiluminescence. All CSF samples were randomised and run in duplicate at one laboratory (Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden) with operators blinded to all clinical data, according to the manufacturer's instructions. Further details on the analytical procedures and assay performances have been described previously.18


Statistical analyses were run in SPSS V.18.0 (PASW statistics). In the first step, we segregated the PD cohort according to clinical phenotype, TD or PIGD and compared characteristics of these groups and normal controls (NCs) using t tests, Mann–Whitney tests and χ2 tests as appropriate. CSF marker levels of Aβ42, Aβ40 and Aβ38 were expressed as absolute values (pg/ml). In addition, we calculated the ratios of Aβ42/40 and Aβ38/40, given that the relative content of Aβ peptides to Aβ40 (as a proxy of the total CSF Aβ content) has been proposed as a more precise measure of Aβ metabolism than absolute CSF Aβ marker concentrations.26

In the second step, we used linear regression models to explore in more detail the relation of CSF marker levels (independent variable) with severity in specific motor domains (dependent variable) in patients with PD. Standardised regression coefficients were calculated. CSF, MRI and motor measures were log-transformed as appropriate to correct for skewness and to reduce the potential impact of incidental outliers. Separate analyses were conducted for each CSF marker. In all regression models, standardised residuals were approximately normally distributed and between ±3.0. There was no evidence of cases exerting undue influence in any model, as evidenced by Cook's distances (D<1). Owing to the explorative nature of this study, we did not adjust for multiple comparisons, and two-tailed p values <0.05 were considered statistically significant.


Clinical and demographic characteristics

Characteristics of patients with PD with PIGD and the TD phenotype are presented in table 1. As expected, the PIGD group displayed higher PIGD (p<0.001) and lower tremor (p<0.001) scores and a higher UPDRS activities of daily living score (p=0.015) than the TD group. No statistical differences between the two groups were seen for age, gender distribution, disease duration, UPDRS motor score, Hoehn and Yahr stage, WMH or cognitive parameters.

Table 1

Characteristics of patients with Parkinson's disease with PIGD (postural instability/gait difficulty) or tremor-dominant (TD) phenotype at diagnosis

CSF Aβ and motor phenotype

CSF marker levels in the PD groups are shown in table 2. Patients with the PIGD phenotype displayed significantly decreased concentrations of Aβ42 (307 pg/ml vs 408 pg/ml, 24.8% reduction, p=0.011) and Aβ38 (390 pg/ml vs 578 pg/ml, 32.5% reduction, p=0.004) but not Aβ40 (5664 pg/ml vs 6453 pg/ml, 12.2% reduction, p=0.126) compared with patients with the TD phenotype. Similar differences were seen between the two groups using the ratios of Aβ42/40 (0.054 vs 0.063, 14.3% reduction, p=0.039) and Aβ38/40 (0.067 vs 0.085, 21.2% reduction, p=0.001).

Table 2

CSF characteristics in patients with Parkinson's disease with PIGD or tremor-dominant (TD) phenotype at diagnosis

Figure 1 shows the CSF marker levels in the two PD groups in relation to 46 NCs (mean (SD) age 64.7 (10.5) years, MMSE score 29.2 (0.8)). Whereas the PIGD group displayed significantly lower levels in all CSF Aβ markers, no differences were found for the PD-TD group, as compared with NCs, although there was a considerable overlap in CSF marker levels between the PD groups.

Figure 1

Box plots illustrating cerebrospinal fluid values (pg/ml) of (A) Aβ42, (B) Aβ38 and ratios of (C) Aβ42/40 and (D) Aβ38/40 in patients with Parkinson's disease with postural instability/gait difficulties (PD-PIGD) or tremor-dominant (TD) phenotype at diagnosis and normal controls (NC). Boxes indicate IQR, with bars representing medians. Circles represent outliers.

CSF Aβ and motor pattern

Because the assignment of motor phenotypes was based on a ratio of tremor to PIGD scores, we first included these scores individually as dependent variables in linear regression models. Tremor severity was not associated with any CSF marker (p>0.05). In contrast, PIGD severity was inversely associated with CSF levels of Aβ42 (β=–0.278, p=0.005), Aβ42/40 (β=–0.352, p<0.001), Aβ38 (β=–0.212, p=0.035) and Aβ38/40 (β=–0.300, p=0.003). Scatter plots illustrating the relation of CSF Aβ42 levels with tremor and PIGD severity are shown in figure 2 (top row).

Figure 2

Scatter plots illustrating the relation of cerebrospinal fluid (CSF) Aβ42 levels (pg/ml) with severity of (A) postural instability/gait difficulties (PIGD), (B) tremor and bradykinesia in (C) lower and (D) upper limbs in patients with Parkinson's disease. Raw CSF levels and motor scores with fit line are displayed. For statistical models, these variables were entered after logarithmic transformation.

Multivariate linear regression models with adjustment for potential confounders (table 3) confirmed these findings and additionally showed significant associations of attentional-executive dysfunction on PIGD severity, with similar effect sizes as seen for CSF Aβ markers. When we added the centred cross-product of Aβ markers and attentional-executive dysfunction to the models, no significant interaction effect on PIGD severity was observed (data not shown).

Table 3

Associations of PIGD severity with CSF levels of Aβ42, Aβ42/40 ratio, Aβ38 and Aβ38/40 ratio in univariate and multivariate linear regression models with adjustment for potential confounders.

To further investigate the association of CSF markers with motor features beyond tremor and PIGD, we extracted subscores for rigidity (item 22) and bradykinesia (items 23–26, 31) from the UPDRS motor section. No significant associations (p>0.05) were found between these scores and any CSF marker. However, when we calculated scores separately for lower and upper extremities, lower limb bradykinesia was significantly associated with all four CSF markers (Aβ42 (β–0.257, p=0.010), Aβ42/40 (β–0.330, p=0.001), Aβ38 (β–0.200, p=0.047), Aβ38/40 (β –0.289, p=0.004)). These associations remained significant after adjustment for potential confounders (data not shown). In contrast, no associations (p>0.05) were found between CSF Aβ markers and upper limb bradykinesia (figure 2, bottom row), nor upper or lower limb rigidity.


Although aggregation of α-synuclein and formation of Lewy bodies constitute the pathological hallmarks of PD, Aβ pathology is often seen in patients with PD.9 ,11 ,12 In this respect, most previous studies have focused on patients with more advanced disease and dementia. However, emerging evidence indicates that Aβ pathology may play a role in PD at earlier disease stages than previously thought.18 ,27 Recent post mortem studies report increased Aβ deposition in patients with non-tremor dominant disease onset in which PIGD and bradykinesia are the prominent motor symptoms.9 ,12As Aβ changes are likely to develop over years or possibly decades,13 we explored whether similar differences could be detected in vivo using a large, population-based and untreated cohort of patients with incident PD who underwent lumbar puncture at diagnosis.

Consistent with post mortem data, this study demonstrates that the initial motor presentation in PD relates to Aβ mis-metabolism in vivo, as measured in the CSF. Patients with the PIGD phenotype at diagnosis displayed decreased CSF levels of Aβ42 and Aβ38, whereas Aβ peptide levels in patients with TD onset did not differ from those in NCs. Similar differences were observed using the ratios of Aβ42/40 and Aβ38/40. Furthermore, we found significant associations between these CSF markers and severity of PIGD and lower limb bradykinesia. These associations were independent of cognitive impairment and other potential confounders, such as age and WMH. This suggests that altered brain Aβ metabolism has impact beyond cognition and contributes to motor heterogeneity in PD from the earliest stages of the disease.

While we found that the severity of PIGD symptoms and lower limb bradykinesia were associated with a CSF marker profile indicating Aβ mis-metabolism, no such associations were found for bradykinesia in upper limbs, or for tremor or rigidity. This is striking, given that recent evidence demonstrates that axial symptoms and lower limb bradykinesia progress significantly more rapidly than upper limb bradykinesia, rigidity and tremor, and are the major determinants of motor progression in PD.28 Furthermore, gait and balance problems in PD herald accelerated cognitive decline and increased risk for dementia,3 ,4 suggesting common underlying pathological mechanisms. Aβ promotes α-synuclein aggregation and accelerates motor and cognitive decline in PD animal models,29 ,30 and clinicopathological studies provide evidence for a strong correlation between brain Aβ load and time to dementia in PD.10 ,11 Combined, these observations indicate that Aβ pathology may be a common denominator of PIGD and its associated motor and cognitive decline.

Whereas CSF Aβ42 has been widely studied in various neurodegenerative diseases, this is less true for other Aβ species. We found significant reductions of CSF Aβ38 levels in the PIGD group, which were more pronounced than those seen for Aβ42. CSF Aβ38 is thought to reflect the production of Aβ peptides from amyloid precursor protein by β- and γ-secretases rather than Aβ deposition.14 Therefore, and because PIGD symptoms in PD probably derive from subcortical or brain stem pathologies, significant cortical retention of the fibrillar amyloid PET marker Pittsburgh Compound B (PiB) might not be expected in these subjects. This is in line with the negative findings of PiB-PET studies in patients with PD without dementia.31 Nevertheless, future studies should explore the functional and structural brain correlates of PIGD and the temporal relation between changes in CSF Aβ, PiB-PET and cognition in PD.

In addition to CSF Aβ markers, attentional-executive impairment was a significant and independent predictor of PIGD severity in multivariate analyses. This is in keeping with increasing evidence from clinical, neuropsychological and neuroimaging studies that have linked alterations in executive and attentional function to gait impairment in PD.32–35 Attentional-executive impairment in PD has been related to dopaminergic deficiency in frontostriatal pathways, and degeneration of cholinergic projections from subcortical and brain stem nuclei to frontal and temporal lobes,7 ,36 even in early PD.37 We found no evidence of an effect of interaction between Aβ markers and attentional-executive dysfunction on PIGD severity, suggesting distinct pathways. Because the effect sizes of Aβ markers and attentional-executive dysfunction on PIGD severity were similar, this suggests that targeting both pathologies might be a promising strategy in the symptomatic treatment or prevention of PIGD in PD.

Our study, in line with previous clinicopathological9 ,12 and CSF38 research, clearly demonstrates that phenotypic heterogeneity in PD reflects variations in certain biological changes. Still, the observed overlap of CSF markers between phenotypes, and the linear associations of PIGD severity with CSF markers and attentional-executive function, do not support the notion of biologically distinct entities. Our findings rather suggest that these phenotypes are part of a complex continuous spectrum characterised by variable amounts of different pathologies, including Aβ mis-metabolism and attentional-executive deficits. This continuum is unlikely to be limited to PD but probably comprises the entire spectrum of Lewy body disorders. Indeed, the findings here may indirectly explain the over-representation of the PIGD phenotype in PD with dementia3 ,4 and dementia with Lewy bodies,4 where both Aβ pathology and attentional-executive deficits are more prominent than in PD without dementia.10 ,39 Erroneous inclusion of subjects with dementia with Lewy bodies and other atypical parkinsonian disorders, such as multiple system atrophy or progressive supranuclear palsy, would have led to potential confounding in this study. In this respect, the study design was robust, with detailed clinical, neuroimaging and neuropsychological assessments at study entry and repetitive clinical examinations to confirm the initial diagnosis of PD. Still, some evidence suggests that emerging biomarkers such as total40 and oligomeric41 α-synuclein and specific τ isoforms42 ,43 might aid in the differential diagnosis of synucleinopathies and tauopathies, and these markers could therefore be considered for inclusion in future studies of early PD and related disorders.

Limitations of this study include the cross-sectional results and the lack of more sophisticated assessments of gait and postural impairment. On the other hand, our study has a number of strengths which make it unique and support its validity, including (1) the careful recruitment of a large and population-based incident PD cohort in order to cover the substantial clinical and biological heterogeneity with sufficient statistical power; (2) assessment of all patients unmedicated, thereby eliminating potential bias by dopaminergic treatment on motor function, cognition and CSF marker levels; (3) the use of objective and reproducible motor measures which have previously been shown to be predictive for the motor and cognitive course in PD; (4) the consideration of various potential confounders and (5) the standardised CSF procedures, which included CSF marker analyses at a single, distinguished centre using a highly sensitive electrochemiluminescence detection method. Notably, while this method provides lower absolute CSF Aβ levels as compared with conventional ELISA, the observed Aβ42 concentrations in both patients with PD and NCs are consistent with those found by others, as detected by luminescence using Luminex xMAP technology.16 ,44

The results here provide new insights into the neurobiology of motor heterogeneity in early PD, with significant implications for future research and clinical management. Our study shows a clinically meaningful and therefore important link between a set of quantitative in vivo CSF biomarkers and the PIGD complex, one of the most consistent clinical predictors of an unfavourable disease course in PD. While previous evidence suggests that low CSF Aβ predicts cognitive decline in PD,17 our findings indicate that Aβ pathology may have a broader impact on disease progression in PD. Future studies should explore whether CSF Aβ markers might help to identify patients at high risk for accelerated motor and functional decline in PD.


We are grateful to all patients and control subjects for their participation in this study and thank the members of the Norwegian ParkWest study group and all other personnel involved in this study for their contributions.


View Abstract


  • Funding The Norwegian ParkWest study is partly funded by the Research Council of Norway (grant# 177 966), the Western Norway Regional Health Authority (grant# 911 218) and the Norwegian Parkinson's Disease Association.

  • Competing interests None.

  • Ethics approval Regional committee for medical research ethics, Western Norway.

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

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.

Linked Articles