Article Text
PDF
Abstract
Background Alzheimer's disease (AD) pathology is found in a considerable portion of patients with Parkinson's disease (PD), particularly those with early dementia (PDD). Altered cerebrospinal fluid (CSF) levels of amyloid-β (Aβ) and tau proteins have been found in PDD, with intermediate changes for Aβ42 in non-demented PD. The authors investigated whether AD-related CSF protein levels are altered and relate to neuropsychological performance in early, untreated PD.
Methods CSF concentrations of Aβ42, Aβ40 and Aβ38 were measured by electrochemiluminiscene and levels of total tau (T-tau) and phosphorylated tau (P-tau) by ELISA in 109 newly diagnosed, unmedicated, non-demented, community-based PD patients who had undergone comprehensive neuropsychological testing, and were compared with those of 36 age-matched normal controls and 20 subjects with mild AD.
Results PD patients displayed significant reductions in Aβ42 (19%; p=0.009), Aβ40 (15.5%; p=0.008) and Aβ38 (23%; p=0.004) but not T-tau (p=0.816) or P-tau (p=0.531) compared with controls. CSF Aβ42 reductions in PD were less marked than in AD (53%; p=0.002). Sequential regression analyses demonstrated significant associations between CSF levels of Aβ42 (β=0.205; p=0.019), Aβ40 (β=0.378; p<0.001) and Aβ38 (β=0.288; p=0.001) and memory impairment, but not executive-attentional or visuospatial dysfunction. Tau protein levels did not correlate with cognitive measures.
Conclusion CSF Aβ levels are altered in a subset of patients with early PD and relate to memory impairment. Our study suggests that alterations in Aβ protein metabolism may contribute to the heterogeneity in pattern and course of cognitive decline associated with PD. Longitudinal studies are needed to clarify the clinical significance of CSF Aβ peptides as prognostic biomarkers in PD.
- Parkinson's disease
- Alzheimer's disease
- cerebrospinal fluid (CSF)
- biomarkers
- cognition
- neurobiology
Statistics from Altmetric.com
Introduction
Dementia is a frequent non-motor complication in Parkinson's disease (PD) with important consequences for patients, care givers and community.1 In addition, a considerable portion of patients with PD exhibit less severe cognitive impairment.1 Neuropsychological deficits, including executive-attentional, visuospatial and memory dysfunction, may be found at the time of diagnosis,2–4 and constitute a risk factor for progression into more severe cognitive impairment, including dementia.5 6 However, not all patients with PD develop dementia, and in those who do, there is a remarkable interindividual heterogeneity in the onset, pattern and progression of cognitive deficits. Understanding this heterogeneity would provide valuable insights into pathogenesis, which is important for developing biologically relevant prognostic biomarkers in PD, and for patient management.
There is increasing awareness of pathological overlap in neurodegenerative diseases. Lewy body pathology as well as Alzheimer's disease (AD)-related pathological changes such as amyloid deposits and possibly neurofibrillar tangles (NFT) are thought to be implicated in the aetiology of dementia in PD (PDD).7–9 Amyloid deposits are considered downstream products of disturbed processing of the amyloid precursor protein (APP) with subsequent misfolding, fibrillisation and extracellular aggregation of synaptotoxic amyloid-β (Aβ).10 Both non-fibrillar and fibrillar Aβ deposits, mainly containing aggregates of Aβ species ending at residue 42 (Aβ42) or 40 (Aβ40), are found in a considerable portion of PD patients at autopsy, particularly those with early onset of dementia.8 9 11 12 In vitro, animal and clinicopathological studies in humans suggest that Aβ may accelerate cognitive decline in PD.8 13 These observations indicate that in vivo markers of disturbed amyloid metabolism may be useful early prognostic biomarkers of evolving dementia in PD.
Cerebrospinal fluid (CSF) Aβ and phosphorylated Tau (P-tau) are promising biomarkers for AD. Reduced CSF levels of Aβ42 are considered to reflect aggregation and deposition of Aβ. The shorter, C-terminally truncated Aβ peptides, Aβ38 and Aβ40, are less prone to aggregate as compared with Aβ42. Their CSF concentrations do not reflect amyloid deposition but rather reflect production of Aβ peptides from APP by β- and gamma-secretases.14 Increased P-tau is an accepted marker of neurofibrillar tangles (NFT), whereas increases in T-tau are thought to be associated with axonal degeneration.10 15
Reduced CSF Aβ42 and increased Tau protein levels have been demonstrated in PDD compared with healthy controls.16–18 Intermediate reductions of CSF Aβ42 have been found even in PD without dementia,16–18 although findings have been inconsistent. Mixed results in heterogeneous diseases such as PD frequently are due to limited sample size and selected patient cohorts. Therefore, large and representative samples are needed to adequately describe the profile of CSF markers in PD. In addition, few studies have explored the relations of these markers with neuropsychological functioning, and even fewer have investigated Aβ species others than Aβ42, such as Aβ40 and Aβ38, in PD without dementia. Since studies in newly diagnosed and untreated patients have not been reported, it remains unclear how early during the course of the disease alterations in AD-related CSF biomarkers might be detected in PD.
Against this background, we performed an analysis of baseline data from the Norwegian ParkWest study to investigate whether CSF markers of AD-related pathologies are altered and relate to neuropsychological functioning in a large, well characterised, non-demented, untreated and population-based sample of patients with newly diagnosed PD. We hypothesised that alterations in CSF Aβ levels might be detectable in a subset of patients with PD at time of diagnosis, and that these alterations would be associated with impaired memory, the key cognitive deficit of AD,10 rather than executive-attentional or visuospatial impairment, which are more closely related to Lewy body disease.19
Methods
Subjects
Case ascertainment and diagnostic procedures in PD
Patients with newly diagnosed PD (mean (SD) 1.2 (1.6) months since diagnosis) were recruited from the Norwegian ParkWest study, a population-based longitudinal cohort study of the incidence, neurobiology and prognosis of PD in Western and Southern Norway.20 All subjects signed written informed consent in study participation. The study is approved by the Regional Committee for Medical Research Ethics, University of Bergen, Norway.
Case ascertainment and diagnostic procedures have been described recently in detail.20 Briefly, using multiple sources of case identification and a multistep diagnostic procedure, we established a population-based cohort of 207 unmedicated, newly diagnosed patients who fulfilled accepted diagnostic research criteria of PD.21 To compensate for the lack of pathological confirmation at this early stage, we restricted inclusion to PD patients whose diagnosis remained unchanged until their latest clinical visit, on average 28'months after inclusion.20 Subjects who met criteria for possible or probable Dementia with Lewy Bodies (DLB),22 AD,23 vascular parkinsonism, or atypical parkinsonian disorders including multiple system atrophy, progressive supranuclear palsy, and corticobasal degeneration, had been excluded.20 In addition, for the purposes of this study, we excluded seven patients with typical PD, with no history of cognitive impairment during the first year after motor onset, but who met diagnostic criteria of PDD,24 based on clinical history and examination, and comprehensive neuropsychological examination, as recently described.2 Thus, 200 unmedicated, non-demented patients with newly diagnosed PD were eligible for this study. Of these, CSF was obtained at time of diagnosis from 109 subjects who had provided separate informed written consent in lumbar puncture (LP). One hundred and six of the 109 PD patients were ever drug-naïve, whereas another three subjects had been assessed at least 2 weeks after withdrawal of antiparkinsonian drugs. None of these patients had clinical signs or a history of fluctuating cognition, persistent visual hallucinations or antipsychotic drug use.
Clinical and neuropsychological assessments in PD
All PD patients were examined unmedicated at time of diagnosis by experienced study neurologists and research nurses. Clinical and neuropsychological assessments were conducted the day before or the same day the patients underwent lumbar puncture and provided blood samples.
Severity of parkinsonism was measured using the Unified Parkinson's Disease Rating Scale (UPDRS) motor subscale.25 The disease stage was assessed by the Hoehn and Yahr scale.26 Severity of depressive symptoms, which may affect cognition in PD, was assessed by the Montgomery and Aasberg Depression Rating Scale (MADRS).27 Neuropsychological assessments included tests on memory, executive-attentional impairment and visuospatial function as follows:
The California Verbal Learning Test II (CVLT-2) (list A, trial 1 to 5, List A retest, and List A delayed recall) was applied to assess both encoding and retention memory processes.
Executive and attentional dysfunction was assessed using three tests: Semantic verbal fluency as a measure of initiation was administered by asking the patients to generate as many names of animals as possible within 1 min. Working memory was assessed using the serial-7 test derived from the MMSE. In the Stroop test, the patient was timed while reading as many words as possible in 45 s in the first part, telling the colour of the words in the second part, and in the third part, the interference task (Stroop Colour–Word), telling the print-colour of the word, which challenges the ability to inhibit prelearnt automated responses.
To measure visuospatial functioning, the Silhouettes test from the Visual Object and Space Perception Battery (VOSP), which offers a form perception task while minimising the influence of motor, attentional, mnestic and executive functions, was administered.
Expected neuropsychological test scores in PD subjects, based on age-, sex- and education, were calculated by regression estimations using data from a large cohort of age-, sex- and education-matched control subjects who had undergone the same neuropsychological test battery. Compositae z-scores were constructed for the domains of memory, executive-attentional and visuospatial function, as described previously.2
CSF controls
CSF peptide marker levels in PD patients were compared with those from 36 age-matched normal controls without brain disease or cognitive impairment (MMSE ≥28) who underwent elective neurological examination or orthopaedic surgery or at Stavanger University Hospital. Twenty patients with mild AD (MMSE score ≥20) with similar age and disease duration compared with the PD cohort served as positive controls. These subjects were diagnosed as having AD according to accepted diagnostic criteria23 by experienced old-age psychiatrists or geriatricians.
Preanalytical treatment of CSF
In all subjects, CSF samples were obtained between 7 and 10 am after overnight fasting by lumbar puncture in the L3/L4 or L4/L5 interspace. Samples were immediately sent to the laboratory where they were routinely assayed for cell counts, levels of glucose and protein to exclude relevant CSF abnormalities. Thus, the first 3–4 ml was used for routine analyses. Study samples were collected in polypropylene (PP) tubes and centrifuged, and the supernatants were stored in PP tubes at −80°C. For aliquoting purposes, all samples were thawed and refrozen once before analysis.
CSF analysis
CSF levels of Aβ42, Aβ40, and Aβ38 were determined in 165 subjects (109 PD, 36 NC, 20 AD) by two of the authors (EM, UA) using the Aβ triplex assay (Human Aβ peptide Ultra-Sensitive Kits) developed by Meso Scale Discovery (Gaithersburg, Maryland). This assay uses C-terminus specific antibodies to capture the different Aβ peptides and a SULFO-TAG-labelled anti-Aβ antibody (4G8) for detection by electrochemiluminiscence. The standard ranges (SR) were 4–3000 pg/ml for Aβ38 and Aβ42, and 27.4–20 000 pg/ml for Aβ40. The lower limits of detection (LLOD) for Aβ42, Aβ40 and Aβ38 were 12.37 pg/ml, 1.28 pg/ml and 8.69 pg/ml, respectively. The limits of quantitation (LOQ) for Aβ42, Aβ40 and Aβ38 were ∼35 pg/ml, ∼50 pg/ml and <25 pg/ml, respectively. The intra-assay coefficients of variation (CV) were <10% for each of the analytes.
CSF levels of T-tau and P-tau were analysed in 143 subjects (109 PD, 14 NC, 20 AD) by one of the authors (EM) using a commercial sandwich ELISA (INNOTEST hTAU-Ag and INNOTEST PHOSPHO-TAU(181P), Innogenetics, Gent, Belgium). The T-tau assay utilises monoclonal antibody (AT120) for capture, and biotinylated monoclonal antibodies (HT7 and BT2) for detection. The assay had an SR of 75–1200 pg/ml, LLOD of 59 pg/ml, and intra- and interassay CVs of 1.2–5.9% and 1.7–6.0%. The P-tau assay utilises monoclonal antibody HT7 for capture, and monoclonal antibody AT720 as detector antibody. The P-tau assay had an SR of 15.6–500 pg/ml, LLOD of 15.6 pg/ml and intra- and interassay CVs of <5% and <10%.
All CSF samples were randomised and run in duplicates according to the manufacturer's instructions.
APOE ε genotyping
In 107 patients with PD and 19 subjects with AD, genomic DNA was extracted from 200 μl of EDTA-blood using the QIAamp 96 DNA Blood Kit (Qiagen, Hilden, Germany).APOE ε2, ε3 and ε4 genotypes were detected using the LightCycler APOE Mutation Detection Kit (Roche Diagnostics, Mannheim, Germany). The assay was performed according to the manufacturer's instructions.
Statistics
Differences in continuous clinical and demographic variables were analysed using Mann–Whitney and Kruskal–Wallis tests. Differences in proportions for categorical data were analysed by linear-by-linear association tests and χ2 tests. Primary comparisons were PD patients versus normal controls, and secondary comparisons were PD versus AD patients. Due to non-normal distribution, all CSF variables were corrected to normality by a logarithmic transform. Between-group differences in CSF protein levels were investigated using ANCOVA with age, sex and education as covariates, and, when significant, followed by simple planned contrasts.
Sequential multiple regression analyses were conducted to investigate the association of CSF markers and cognitive domain z-scores, entering as independent variables age, education and sex at step 1, disease duration, UPDRS motor score and MADRS score at step 2, and each log-transformed CSF parameter at step 3. Separate analyses were conducted with each cognitive domain z-score as dependent variable.
Results
A total of 165 subjects were included in this study: 109 untreated, non-demented patients with newly diagnosed PD, 36 age-matched normal controls and 20 patients with mild AD matched for age and disease duration. The 109 PD patients who provided CSF had lower UPDRS motor scores than those who did not but did not differ significantly in other demographic, clinical or neuropsychological variables (table 1).
Demographic, clinical, and neuropsychological characteristics in patients with early, untreated Parkinson's disease (PD)
Clinical and demographic characteristics of PD and AD patients and normal controls are presented in table 2. There were significant differences between the groups in MMSE score (p<0.001) and significantly more APOE ε4 carriers among AD patients than PD patients (p<0.001).
Demographic and clinical characteristics in normal controls, Parkinson's disease and Alzheimer's disease
CSF marker levels in PD, AD and NC
The CSF characteristics in normal controls and patients with PD and AD are shown in figure 1 and table 3. ANCOVA with age, sex and education entered as covariates demonstrated significant differences between the groups in all CSF marker levels (Aβ42: p<0.001, η2=0.121); Aβ40 (p=0.029, η2=0.044); Aβ38 (p=0.016, η2=0.051; T-tau: p<0.001, η2=0.189; P-tau: p=0.004, η2=0.078).
Box plots showing cerebrospinal fluid (CSF) values (pg/ml) for Aβ42 (A), Aβ40 (B), Aβ38 (C), T-tau (D) and P-tau (E) in normal controls (NC), Parkinson's disease (PD) and Alzheimer's disease (AD). Boxes indicate IQR, with bars representing medians. Circles represent outliers.
Cerebrospinal fluid marker concentrations in normal controls, Parkinson's disease and Alzheimer's disease
Simple planned contrast analyses revealed significant differences between PD patients and controls for CSF levels of Aβ42 (p=0.009, 19% reduction), Aβ40 (p=0.008, 15.5% reduction) and Aβ38 (p=0.004, 23% reduction), but not T-tau (p=0.816) or P-tau (p=0.531). Compared with AD subjects, patients with PD had significantly higher levels of Aβ42 (p=0.002) and lower levels of T-tau (p<0.001) and P-tau (p=0.002), whereas levels of Aβ40 (p=0.425) and Aβ38 (p=0.416) did not differ significantly between the two groups. The difference in Aβ42 between PD and AD remained significant (p=0.025) after adjustment for differences in APOE ε4 carrier frequency between these groups.
Neuropsychological correlates of CSF markers in PD
Multiple sequential regression analyses with adjustment for age, sex, education at step 1, and disease duration, UPDRS motor score and MADRS score at step 2 revealed significant linear associations between CSF levels of Aβ42 (β=0.205, p=0.019, ΔR2=0.040, supplemental table 1), Aβ40 (β=0.378, p<0.001, ΔR2=0.132, supplemental table 2), and Aβ38 (β=0.288, p=0.001, ΔR2=0.077, supplemental table 3) and memory domain z-score at the final step. Aβ42, Aβ40 and Aβ38 did not correlate with attentional-executive or visuospatial domain scores. T-tau and P-tau did not correlate with any cognitive domain score (all p>0.05).
Four PD patients displayed T-tau protein, three patients P-tau and one patient both T-tau and P-tau protein levels above the expected range (marked as outliers in figure 1). Exclusion of these patients did not affect the results significantly (see supplemental tables 4–7).
Discussion
This study explored a range of established and more novel AD-related CSF biomarkers in a large community-based cohort of newly diagnosed and unmedicated patients with PD. The main finding of our study is that already at this early stage, all assessed amyloid-β peptide CSF levels but not Tau protein concentrations were significantly reduced in PD compared with normal controls. Furthermore, we found significant linear associations between CSF Aβ42, Aβ40 and Aβ38 levels and memory performance, but not visuospatial or executive-attentional dysfunction in PD. These findings suggest that already at the time of diagnosis, alterations in Aβ protein metabolism but not τ pathology are detectable and relate to pattern of cognitive performance in PD.
Few studies have investigated AD-related CSF markers in PD patients without dementia,16–18 28–32 only one of these studies32 has assessed CSF Aβ levels other than Aβ42, and none has explored CSF biomarkers at this early stage or in a population-based setting. The observed 19% reduction in CSF Aβ42 in our cohort, with similar decreases in Aβ40 (15.5%) and Aβ38 (23%), is comparable with the 13–30% decrease in CSF Aβ42 levels found in six16–18 28 29 32 out of eight16–18 28–32 clinic-based investigations in non-demented PD with longer disease duration. However, whereas our findings on CSF Aβ alterations in PD were significant, this was not the case in several previous studies.17 28–32 Given the similar extent in CSF Aβ reductions in most of these studies and our cohort, and the small effect sizes observed in our study, differences in sample characteristics and, in particular, sample size rather than technical variations may account for these inconsistencies. There is a remarkable interindividual heterogeneity in PD, both clinically and pathologically, which may covary with CSF measures. This was also reflected in the significant variation in CSF biomarker levels, with considerable overlap in the ranges of CSF Aβ values between PD patients and normal controls. As expected, many PD patients exhibited normal CSF Aβ values, and in those with decreased levels, reductions were often mild to modest.
In vitro and animal studies suggest that Aβ, and in particular its oligomeric and protofibrillar assemblies, promotes the accumulation of α-synuclein and accelerates cognitive decline in PD.13 A strong negative correlation between Aβ load and time to dementia in PD has been demonstrated in clinicopathological studies.8 9 Well-designed autopsy studies in PD demonstrate that of those who become demented within 5–10 years, 70–80% exhibit amyloid pathology at autopsy.9 33 Previous CSF studies report intermediate CSF Aβ42 reductions in non-demented PD compared with normal controls and PDD, suggesting an increase in the extent or severity of amyloid pathology with advance in cognitive deficits.16–18 Thus, the observed reduced Aβ peptide concentrations in a subset of our cohort may reflect early or ongoing Aβ fibrillisation or aggregation processes that will evolve into more severe amyloid pathology in those who are at highest risk of developing dementia.
Patients with PD show impairment in a range of neuropsychological tests at time of diagnosis.2–4 Although non-amnestic deficits with impaired executive, attentional and visuospatial dysfunction are considered the characteristic cognitive changes in PD,19 memory impairment is found in 20–34% of patients with newly diagnosed PD.2 3 Memory deficits in early PD have been related to atrophy in the hippocampus34 and surrounding limbic and paralimbic structures, which show significant progressive cell loss over time in these patients.35 Given the progressive nature of memory deficits in non-demented PD36 37 and their association with higher risk for dementia,5 our observation of significant associations between CSF Aβ levels and memory impairment further suggests that reduced levels of these markers may signal evolving PDD, at least in a subset of patients. Previous clinic-based studies in non-demented PD failed to demonstrate significant correlations between CSF and neuropsychological measures17 or found associations between Aβ42 and executive impairment,18 which also have been associated with increased risk for dementia.5 It remains unclear whether these inconsistencies relate to differences in sample size, or the fact that previous studies investigated patients without dementia with a longer disease duration who were on dopaminergic medication, which may affect the cognitive profile in PD. In addition, given that amyloid pathology is most pronounced in patients who develop dementia early during the course of their disease,8 9 the biological correlates, including CSF biomarkers, of early versus late cognitive impairment in PD probably differ. Therefore, prospective longitudinal CSF studies with detailed and repetitive neuropsychological and neuroradiological examinations are needed to further clarify the relation of lowered CSF marker levels, functional and structural alterations in specific brain regions, and pattern and rate of cognitive decline in early and late PD.
The CSF marker pattern in PD differed from that in AD patients who displayed more severe Aβ42 reductions (mean 53%) than PD patients, as well as increased CSF T-tau and P-tau levels, which were in the expected range for AD.10 The observed significant reductions in Aβ40 (14%) and Aβ38 (18%) in AD compared with controls were somewhat unexpected, since most previous studies have shown normal CSF levels of these Aβ species in AD,38 although similar39 or even higher32 absolute reductions of these markers in AD have been reported. In PD without dementia, CSF Aβ40 and Aβ38 levels have been explored in only one previous, small study,32 demonstrating slightly (6% and 15%, respectively) but non-significantly decreased values in patients (n=11) compared with controls (n=19). We found significant reductions in both Aβ40 and Aβ38 with strong relations to memory performance, highlighting the need for further studies into the role of these markers in PD.
The APOE ε4 allele is associated with a higher risk and earlier onset of AD and has been found to be associated with low Aβ42 in both AD patients and cognitively normal controls.40 In line with previous studies,41 about a third of our PD patients were APOE ε4 carriers. This figure was significantly lower than that in patients with AD. However, the lower prevalence of APOE ε4 carriers in PD than in AD did not explain the less severe Aβ reductions in the PD group. Associations of the ε4 allele type with Lewy body pathology have been reported in PD,42 and relations between APOE genotype and age at motor onset and risk for PD and PDD have been hypothesised,42 although this was not confirmed by others.41 43 A current meta-analysis concluded that there is over-representation of APOE ε4 carriers in PDD compared with PD, but that effect sizes are at best modest.43 Further, large studies are needed to clarify the relation between APOE polymorphisms, AD-related CSF markers and cognitive decline in PD.
Our study has a few limitations. First, our results are cross-sectional and should therefore be interpreted cautiously until longitudinal data from our cohort or other studies are available. Second, neuropsychological assessments were not conducted in CSF-control subjects, and so it is unclear whether correlations between CSF Aβ levels and memory impairment might be observable in the general population, although two recent CSF studies in PD found no associations between AD-related CSF measures and neuropsychological function in normal control subjects.17 18 Finally, as other studies in early PD, we face the unavoidable problem that pathological confirmation of the clinical diagnosis is not yet available. However, given the prospective and comprehensive diagnostic procedures, in which the diagnosis was based on accepted research criteria and validated at several time points, we consider it unlikely that our results are significantly biased by erroneous inclusion of subjects with atypical parkinsonian disorders or dementias such as AD, DLB and PDD. Still, it is possible that further follow-up of our cohort may reveal that some of our patients with presumed PD may have been misdiagnosed.
Strengths of our study include the standardised clinical, neuropsychological and CSF assessments which were conducted in close temporal relationship to each other. In addition, our cohort was considerably larger than previous PD samples without dementia, it was population-based, and patients were assessed unmedicated, eliminating potential confounding by sample selection and dopaminergic agents, which may influence cognition. Finally, we included subjects with AD as positive controls, which was helpful to gauge the extent of CSF marker alterations in the PD cohort.
In conclusion, we demonstrate that alterations in Aβ protein metabolism, as measured in vivo by CSF Aβ peptide levels, are present in a subset of patients with PD at time of diagnosis and relate to pattern of cognitive impairment. Our study lends further support to the already strong evidence from clinicopathological, animal and in vitro studies that amyloid pathology may play a role in the pathogenesis of cognitive decline and dementia associated with PD. Longitudinal CSF studies are warranted to clarify whether reduced CSF Aβ peptide levels may be useful prognostic markers of early cognitive decline and PDD.
Acknowledgments
We are very grateful to the patients for their donation of cerebrospinal fluid and to families and care givers for their support. The authors thank J Selvaag and HL Nakkestad, for help with preparation of CSF samples, K Simonsen, for expert technical support, and OB Nilsen, for help with management of databases. We are very grateful to all research nurses and study clinicians for their contribution to the collection of CSF and clinical data.
References
Footnotes
Funding The Norwegian ParkWest study is funded by the Western Norway Regional Health Authority (grant no 911218), the Research Council of Norway (grant no 177966) and the Norwegian Parkinson's Disease Association.
Competing interests None.
Patient consent Obtained.
Ethics approval Ethics approval was provided by the Regional Committee for Medical Research Ethics, University of Bergen, Norway.
Provenance and peer review Not commissioned; externally peer reviewed.