Article Text

Research paper
Automated volumetry of the mesiotemporal structures in antibody-associated limbic encephalitis
  1. Jan Wagner1,2,
  2. Juri-Alexander Witt1,
  3. Christoph Helmstaedter1,
  4. Michael P Malter1,3,
  5. Bernd Weber1,2,4,
  6. Christian E Elger1,2,4
  1. 1Department of Epileptology, University of Bonn, Bonn, Germany
  2. 2Department of NeuroCognition/Imaging, Life & Brain Center, Bonn, Germany
  3. 3Department of Neurology, Marien-Krankenhaus, Bergisch Gladbach, Germany
  4. 4Center for Economics and Neuroscience, University of Bonn, Bonn, Germany
  1. Correspondence to Dr Jan Wagner, Department of Epileptology, University of Bonn, Sigmund-Freud-Str. 25, Bonn D-53127, Germany; jan.wagner{at}ukb.uni-bonn.de

Abstract

Objective Limbic encephalitis (LE) is an autoimmune mediated disease leading to temporal lobe epilepsy, mnestic and psychiatric symptoms. In recent years, several LE subforms defined by serum antibody findings have been described. MRI usually shows volume changes of the amygdala and hippocampus. However, studies quantifying longitudinal volume changes in the acute disease stage are lacking.

Methods The aim of this retrospective observational study was to evaluate and quantify these volume changes by applying a fully automated volumetric approach to serial MRIs of 28 patients with antibody-associated LE. The results were compared with those of 28 age-matched and gender-matched healthy controls and analysed separately for the different antibody profiles and correlated with clinical parameters. Antibody profile analyses were exploratory due to the relatively small sample sizes.

Results We found distinct volumetric and clinical courses depending on the associated antibody. While LE associated with voltage-gated potassium channel-complex antibodies (VGKC-LE) showed highly significant larger volumes of both the amygdala and the hippocampus within the first 12 months after disease onset, LE associated with glutamic acid decarboxylase antibodies (GAD-LE) only displayed greater amygdala volumes at this disease stage. Both subgroups showed a reduction of the amygdala and hippocampus volumes during follow-up with higher volume changes in VGKC-LE.

Conclusions These differences in the volumetric evolution corresponded to distinct clinical courses in terms of a more severe initial symptomatology regarding seizure, mnestic and psychiatric disturbances in VGKC-LE, which improved rapidly, corresponding to the evolution of the volumetric changes. In contrast to this, patients with GAD-LE were less severely affected at disease onset, showing a more unmodulated and chronic disease course during follow-up.

  • EPILEPSY
  • MRI
  • IMAGE ANALYSIS
  • LIMBIC SYSTEM

Statistics from Altmetric.com

Introduction

Limbic encephalitis (LE) is an autoimmune-mediated disease leading to temporal lobe epilepsy, mnestic deficits and psychiatric symptoms.1 ,2 It was first described in the 1960s as a paraneoplastic syndrome caused by inflammation in limbic structures in adults.3 ,4 Various onconeural antibodies, which are biomarkers for these paraneoplastic syndromes, have been discovered in the following decades.5 However, in recent years, non-paraneoplastic forms of LE have been increasingly recognised, associated with antibodies against glutamic acid decarboxylase (GAD) and the voltage-gated potassium channel (VGKC)-complex.6–9 These non-paraneoplastic forms have a better prognosis, but the outcome seems to differ between the two subgroups depending on the associated antibody. While patients with VGKC-complex-associated LE (VGKC-LE) mostly become seizure-free and show normalised antibody concentrations after immunotherapy, GAD-associated LE (GAD-LE) often displays a non-remitting course with antibody and seizure persistence.6 ,8 ,10

MRI in LE shows signal and volume changes of the mesiotemporal structures. However, studies quantifying volume changes in LE are scarce. In a recently published study, Irani et al11 found smaller volumes of the whole brain and hippocampus (HC) in VGKC-LE compared to healthy controls. However, MRI was performed in the convalescent phase of the disease and neither amygdala (AM) volumes nor longitudinal volume changes were investigated.

The aim of the present study was to evaluate and quantify longitudinal volume changes of AM and HC in the acute and subacute disease stages of antibody-associated LE compared to healthy controls using a fully automated volumetric approach. Furthermore, results were analysed depending on the associated antibody and correlated with clinical parameters (eg, seizure outcome, mnestic and psychiatric symptoms). In doing so, we aimed at addressing the following questions: (1) Is there a difference in AM and HC volumes between patients with LE and healthy participants? (2) How is the temporal evolution of the volumetric changes? (3) Do volumetric changes correlate with clinical parameters? (4) Can we find distinct volumetric and clinical results depending on the associated antibody?

Materials and methods

Study groups and MRI examinations

We retrospectively studied all patients diagnosed with antibody-associated LE presenting at the Department of Epileptology, University of Bonn, from 1 January 2006 to 31 August 2012. LE was diagnosed based on the features of (1) limbic signs and symptoms, which manifested themselves in adolescence or adulthood (≥one of the following: seizures of temporal semiology, disturbance of episodic memory, psychiatric symptoms with affective and/or anxiety disturbances) and (2) presence of serum antibodies associated with LE (ie, onconeural, VGKC-complex, GAD).

Serum antibody testing for GAD, onconeural and VGKC-complex antibodies was performed as described previously.12 Antibodies against leucine-rich, glioma inactivated 1 protein (LGI1) and contactin-associated protein-2 (CASPR2) were detected by indirect immunofluorescence using formalin fixed HEK293 cells containing membrane bound LGI1 or CASPR2.13

To allow for an evaluation of the volume changes in the acute disease stage and to assess all included patients at the same disease period, we only included patients whose first MRI including a three-dimensional (3D) T1-weighted volume data set was performed within the first 12 months after LE onset. LE onset was defined as the time point of the first symptoms suggestive of LE (seizures and/or psychiatric and/or mnestic disturbances). If available, we furthermore analysed up to two follow-up MRIs each six to 12 months after the preceding MRI scan to assess the course of the volume changes within the first 36 months of the disease.

The control group was assembled from a pre-existing in-house database consisting of 1262 healthy participants with no neurological disorder. Age-matching and gender-matching with the LE group was achieved by building matched pairs for each individual patient. The MRIs of the selected controls were thoroughly inspected and were free of lesions especially in the mesiotemporal region. We did not exclude controls with unspecific extratemporal lesions on MRI (eg, microangiopathy). As follow-up MRI examinations with resembling time intervals were not available for the control group, the control group for the patients’ follow-up MRIs consisted of the corresponding matched participants (ie, for each missing patient, the corresponding matched participant was excluded).

All included patients and healthy participants underwent at least one MRI examination including a 3D T1-weighted volume data set using the same 3T MRI scanner (Magnetom Trio, Siemens, Erlangen, Germany) with the same sequence parameters in all participants (MP-RAGE, voxel size 1×1×1 mm, repetition time 1570 ms, echo time 3.42 ms, flip angle 15°, field of view 256 mm×256 mm).

MRI volumetry and volumetric measures

Fully automated volumetry of AM, HC and the whole brain was performed with the FreeSurfer image analysis suite (V.5.1.0, Martinos Center, Harvard University, Boston, Massachusetts, USA),14 ,15 which is documented and freely available for download online (http://surfer.nmr.mgh.harvard.edu/). All AM and HC segmentations were thoroughly visually inspected by an expert reader (JW) and found to be appropriate in all cases so that no manual corrections or even exclusion of particular study participants was necessary.

Volumes of AM and HC are given and were analysed in mm3 for comparing the entire LE group with the control group as these two groups were age-matched and gender-matched as described above. For comparisons of the different LE subgroups with one another and the corresponding control groups, normalised volumes of AM and HC as a percentage of the whole brain volume were calculated. This was carried out because of significant differences concerning age at MRI between the LE subgroups (see results) to adjust for brain size and age. A detailed description is available in the online supplementary material (text and figure S1).

Neuropsychological and psychiatric assessment

Verbal learning and memory was assessed by the verbal learning and memory test (VLMT)16 and figural learning and memory was assessed by the revised Diagnosticum fuer Cerebralschaedigung (DCS-R).17 ,18 The main parameters of the VLMT (learning performance, memory loss, delayed free recall, recognition) and the DCS-R (learning performance, learning capacity, recognition) were standardised according to age-dependent norms derived from 488 healthy controls (age range 16–93 years). This control sample was recruited independently from the imaging control pool. A verbal or figural memory deficit was defined as a score of more than one SD below the mean performance of the normative sample in at least one of the main parameters. Furthermore, we analysed the number of impaired parameters for all patients in both tests. At the final visit, one patient with GAD-LE underwent a computerised assessment of verbal and figural memory by applying the neurocognitive effects test (NeuroCog FX).19

Psychiatric symptoms (mainly affective and anxiety disorders) were assessed based on the patient records and, if available, by evaluation of the Beck Depression Inventory (BDI-I) with a cut-off value of ≥16.20 ,21

Statistical analysis

Statistical analyses were performed using SPSS Statistics V.21.0 for Mac OS X (IBM, Armonk, New York, USA). All values throughout this report are given as median unless otherwise stated. For statistical comparisons of independent categorical data, Fisher's exact test was performed; for comparisons of independent metrical data, a Mann-Whitney U test was performed; for comparisons of paired metrical data, a Wilcoxon signed-rank test was performed, and for correlation analyses, a Pearson correlation test was performed. For comparisons of LE subgroups with the controls, a Kruskal-Wallis test was additionally performed. Correlation analyses of patients’ volumes with clinical parameters and antibody concentrations were performed using all available absolute volumes (entire LE group) and normalised volumes (LE subgroups) at all available time points. A probability (p) value <0.05 was regarded as statistically significant using two-tailed tests.

Results

Study groups

According to the aforementioned inclusion and exclusion criteria, 28 patients with LE and 28 age-matched and gender-matched healthy participants were included in the study. Clinical and demographical data of the LE group and the control group are summarised in table 1. Seven out of the 15 patients with VGKC-complex antibodies were tested for LGI1 and CASPR2 antibodies, whereas the remaining eight were studied before 2010, and therefore appropriate tests were not available at that time. Tumour searches were negative in all patients with LE including the case with amphiphysin antibodies (repeated extensive diagnostics were performed in this patient).

Table 1

Clinical and demographical data of patients with LE and controls

Development of mesiotemporal volumes in LE over time

Patients with LE showed highly significant larger volumes of both right and left AM compared to the control group at MRI 1 (left and right AM p<0.001), whereas no significant differences in HC volumes were found at this time point. At MRI 2, we found no significant differences between the two groups, whereas at MRI 3 the LE group developed a significant bilateral atrophy of the HC (left HC p=0.022; right HC p=0.010). We found no significant differences in the whole brain volume between the LE group and the control group, and there was no significant reduction of the whole brain volume during follow-up in the LE group as well. Results are summarised and illustrated in table 1 and figure 1.

Figure 1

Absolute AM and HC volumes of the entire limbic encephalitis (LE) group and the control group at all three MRI time points. AML, left amygdala; AMR, right amygdala; HCL, left hippocampus; HCR,right hippocampus. ***p≤0.001, **p≤0.01, *p<0.05.

With regard to the 25 patients who had a second MRI, we found a significant volume reduction of both AM and both HC between MRI 1 and MRI 2 (left and right AM p=0.001; left and right HC p<0.001). Referring to the 14 patients in whom a third MRI was available, the volume reduction of AM and HC between MRI 2 and MRI 3 was smaller compared to the volume change between MRI 1 and MRI 2, but was still significant for both AM (left AM p=0.025; right AM p=0.010), whereas we found borderline p values for both HC (left HC p=0.049; right HC p=0.051). Absolute and percentage longitudinal volume changes of AM and HC are summarised in table 2.

Table 2

Absolute and percentage longitudinal volume changes in the limbic encephalitis group

Disease duration correlated significantly with all available AM and HC volumes while no significant correlation with whole brain volume was found (see online supplementary figure S2).

LE subgroups

Clinical data of the LE subgroups and the corresponding control groups are given in table 3. We found a significantly earlier disease onset in GAD-LE compared to VGKC-LE (p=0.001). When evaluating mnestic deficits based on a dichotomous analysis (ie, deficit defined as an impairment in at least one parameter in VLMT/DCS-R), we only found a significant difference in figural memory performance between the two groups at MRI 1 (p=0.024). However, when analysing the number of impaired test parameters, we found a significantly poorer performance in both VLMT (median 1.0 vs 3.0, p=0.045) and DCS-R (median 0.5 vs 2.0, p=0.037) in VGKC-LE at the first visit. No significant differences in mnestic deficits and the number of impaired test parameters were found during follow-up at MRI 2 and MRI 3. Detailed results of the performance in VLMT and DCS-R are summarised in the online supplementary figure S3. Patients with VGKC-LE significantly more often suffered from psychiatric disturbances at MRI 1 (p=0.022), and we found a significantly better seizure outcome in this group compared to GAD-LE. Furthermore, most VGKC-LE cases became antibody-negative during follow-up, whereas we did not find any relevant changes in antibody concentration in GAD-LE.

Table 3

Clinical and demographical data of the limbic encephalitis (LE) subgroups and the corresponding control groups

Owing to the relatively small patient and control groups, we did not differentiate between left and right AM and HC when comparing the LE subgroups with one another and the control groups (ie, two values per participant for AM and HC).

We found significantly larger AM volumes in both GAD-LE and VGKC-LE compared to the corresponding controls at MRI 1 (Kruskal-Wallis test p<0.001; GAD-LE vs controls p=0.004; VGKC-LE vs controls p<0.001; figure 2) while the AM volume was even higher in VGKC-LE compared to GAD-LE (p=0.016). Patients with VGKC-LE additionally showed significantly larger volumes of the HC compared to the control group at MRI 1 (Kruskal-Wallis test p=0.007; VGKC-LE vs controls p=0.001). At MRI 2, there was only a significant difference in AM volume between GAD-LE and the corresponding control group (Kruskal-Wallis test p=0.031; GAD-LE vs controls p=0.011). Referring to the patients who had a second MRI, we found a significant reduction of both AM and HC volume in VGKC-LE between MRI 1 and MRI 2 (both p<0.001), whereas no significant reduction was present in GAD-LE. At MRI 3, we found no significant differences in AM (Kruskal-Wallis test p=0.816) and HC volume (Kruskal-Wallis test p=0.141) between the LE subgroups and the control groups. Again, a significant reduction of AM and HC volume between MRI 2 and MRI 3 could be found in VGKC-LE (AM p=0.004; HC p=0.001) while no significant volume changes were found in GAD-LE (referring only to the patients who underwent a third MRI).

Figure 2

Normalised volumes of amygdala (AM) and hippocampus (HC) of the limbic encephalitis (LE) subgroups and the corresponding control groups. Note the more prominent volume changes of AM and HC in VGKC-LE at MRI 1, which declines rapidly during follow-up while volume changes in GAD-LE do not show such a highly dynamic course. ***p≤0.001, **p≤0.01, *p<0.05. GAD, glutamic acid decarboxylase; VGKC, voltage-gated potassium channel.

Percentage volume changes are summarised in table 4. Here, we also found higher changes in VGKC-LE compared to GAD-LE. The single patient with amphiphysin antibodies showed high volume changes, but a valid statistical evaluation was not possible in this single case.

Table 4

Percentage volume changes in limbic encephalitis subgroups

Correlation analysis of all available AM and HC volumes with disease duration revealed a significant negative correlation of both structures in VGKC-LE while no significant correlations were found in GAD-LE (figure 3). Furthermore, the VGKC-complex antibody concentration correlated negatively with disease duration (r=−0.352, p=0.041) and positively with AM volume (r=0.400, p=0.019; not shown in figure 3). No significant correlation of the GAD antibody concentration with volumetric or clinical parameters was found.

Figure 3

Correlation of disease duration with all available amygdala (A and B) and hippocampus (C and D) volumes in GAD-LE (left column) and VGKC-LE (right column). Significant negative correlations could be found in VGKC-LE for both structures while no significant volume changes were found in GAD-LE during the course of the disease. GAD, glutamic acid decarboxylase; LE, limbic encephalitis; VGKC, voltage-gated potassium channel.

Discussion

Development of mesiotemporal volumes in LE over time

To the best of our knowledge, this is the first study quantifying longitudinal volume changes of the mesiotemporal structures in the acute and subacute disease stages of LE. Our results indicate larger bilateral AM volumes within the first 12 months in LE when referring to all patients independent of the serum antibody type. Within a time period of 6–12 months, the AM volume decreased markedly and did no longer differ from healthy controls during further follow-up. In contrast to this, both HC shrank progressively and developed a significant atrophy within the same time period.

A recently published study reported about a significant HC and whole brain volume reduction in eight patients with VGKC-complex and LGI1 serum antibodies compared to healthy controls.11 As the MRI was performed during the convalescent phase of LE (median 384 days after LE onset), these results are in agreement with those of our study concerning the HC, whereas we did not find significant abnormalities regarding the whole brain volume. Notably, the age matching between patients and controls in the cited study was not ideal as 50% of the patients were older than the oldest healthy participant, which may be a confounder of their results. Furthermore, neither AM volumes nor longitudinal volume changes were investigated so that our finding of a predominant pathology of the AM in the acute disease stage can neither be confirmed nor contradicted by the cited study.

In a recently published study, we quantified fluid-attenuated inversion recovery (FLAIR) signal intensities of AM and HC in LE and hippocampal sclerosis.22 Here, predominant signal abnormalities of the AM in LE were found while signal abnormalities of the HC were significantly less frequent in LE compared to hippocampal sclerosis (mean disease duration at MRI 2.4 years). These results, on the one hand, corroborate the findings of the present study indicating a predominant AM pathology in LE but, on the other hand, taking the disease duration into account, suggest that signal changes seem to evolve in a distinct manner compared to volume changes. At MRI 3 (which corresponds best to the time point after LE onset in our previous study), we found reduced volumes of both HC while the AM volume did no longer differ significantly from the control group in the present study. In a post hoc analysis, no correlation of FLAIR signal intensities with disease duration in the LE cohort investigated in our previous study was found, which further supports the hypothesis that volume changes evolve differently from signal changes in LE (unpublished in-house data). Thus, normalisation of AM volume does not necessarily mean normalisation of signal intensity or even of function. Furthermore, development of HC atrophy does not necessarily mean development of hippocampal sclerosis as FLAIR signal intensities differed significantly from those of patients with hippocampal sclerosis in our previous study.

Larger volume on MRI most likely results from parenchymal swelling caused by infiltration of inflammatory cells and complement activation in mesiotemporal structures, which could be shown in various histopathological studies.9 ,23 ,24 However, it remains unclear why these changes are limited to limbic structures in the majority of patients. A possible explanation for this is the antigen distribution as it could be shown that the VGKC-complex expression is particularly strong in the HC.8 However, this hypothesis is not applicable to GAD, which is expressed widely across the human brain without preferential localisation. Further studies are needed to clarify this issue.

LE subgroups

Both GAD-LE and VGKC-LE showed significantly larger AM volumes in the acute disease stage while the AM volume was even higher in VGKC-LE compared to GAD-LE. Furthermore, the HC volume was larger in VGKC-LE compared to the corresponding controls while we did not find a difference between GAD-LE and controls. These findings may be an indicator of a more severe initial disease course in VGKC-LE compared to GAD-LE. Compatible with this hypothesis, patients with VGKC-LE significantly more often suffered from psychiatric symptoms at this disease stage, which is most likely caused by a more severe pathology of the AM as it is known that the AM plays a key role in the spectrum of affective and anxiety disorders.25–29 Corresponding results were found by two previous studies, both reporting about a significantly higher prevalence of non-seizure limbic disturbances (eg, depression, anxiety, confusion, personality changes) in VGKC-LE compared to GAD-LE at disease onset.6 ,10 However, it must be pointed out that these two studies were also conducted at our department and there is an overlap of patients that were included in the present study and in the study of Malter et al. (32% overlap) and Frisch et al. (64% overlap).

In accordance with our finding of a more severe HC involvement in VGKC-LE at the initial disease stage, verbal and figural memory performance was poorer in this group compared to GAD-LE at MRI 1. Similar results regarding mnestic deficits in LE have been reported previously.6 ,10

During follow-up, VGKC-LE showed a rapid decline of both AM and HC volumes and did no longer differ from the controls 6–12 months after the initial MRI, corresponding to the clinical finding of a marked mnestic, seizure and psychiatric improvement in these patients within the same time period. The slight relative increase in memory deficits between MRI 2 and MRI 3 is most probably caused by a selection bias, as unimpaired patients more likely become lost to follow-up than impaired patients. In contrast to VGKC-LE, volume changes in the GAD-LE group were considerably smaller during follow-up. In line with this volumetric course, we did not find a similar clinical improvement in this group compared to VGKC-LE. A worse mnestic and seizure outcome in GAD-LE has been reported in three previous studies complementing our results.6 ,10 ,30 In contrast to the analysis of the entire LE group, we neither found a significant HC atrophy in GAD-LE nor in VGKC-LE at the last follow-up, which is most probably caused by the relatively small patient groups.

In summary, the results presented here indicate a highly dynamic and acute disease course in VGKC-LE within the first 30 months after onset. The marked volumetric changes of AM and HC during this disease stage mirror the clinical course of a severe symptomatology regarding seizure, mnestic and psychiatric disturbances at the beginning of the disease, which improve rapidly corresponding to the evolution of the volumetric changes. In contrast to this, GAD-LE shows a more chronic disease course with less prominent mesiotemporal volume changes and less severe clinical symptomatology at disease onset not evolving with a comparable dynamic during follow-up. A possible explanation of this finding may be the different pathogenic mechanisms in these two disease entities. While VGKC-LE appears to be primarily antibody-mediated and complement-mediated, T cell-mediated neuronal cytotoxicity seems to be the destructive factor in GAD-LE.23

In our previous study, we found higher HC signal intensities in GAD-LE compared to VGKC-LE after a mean disease duration of 2.4 years.22 Owing to the relatively long interval after onset, one can hypothesise that the signal intensity in VGKC-LE was already declined, whereas it remained high in GAD-LE, corroborating the hypothesis of a more chronic disease course. Another explanation could be that GAD-LE primarily leads to signal changes while VGKC-LE primarily leads to volume changes.

Moreover, our results raise the question as to what extent temporal volumetry may be helpful in diagnosing and monitoring the disease course in LE. Owing to its time efficiency and good practicability without the need for additional examinations or special MRI sequences, temporal volumetry may be a powerful tool that may increase the diagnostic sensitivity of MRI in LE. Furthermore, it may lead to an earlier identification of LE by prompting antibody determination in patients showing greater volumes of AM and/or HC, possibly leading to a better outcome due to earlier initiation of anti-inflammatory therapy. Moreover, it may be particularly helpful in patients with no detectable antibody, in whom diagnosis, treatment and prognosis are especially challenging for both epileptologists and neuroradiologists. It is tempting to speculate that antibody-negative patients with an acute and dynamic disease and volumetric course respond better to therapy and thus have a better prognosis, whereas less marked volumetric changes at disease onset may be an indicator of a poorer prognosis despite less severe initial clinical symptomatology.

Limitations

Owing to the small sample sizes especially at MRI 3, it should be explicitly stated that our results and analyses of follow-up MRI data are exploratory and not hypothesis testing. These small sample sizes may also lead to a selection bias at MRI 3, which is most likely the reason for the higher median seizure frequency in GAD-LE and the higher relative share of mnestic deficits in VGKC-LE at MRI 3 compared to MRI 2, as unimpaired/less impaired patients more likely get lost to follow-up than severely impaired patients.

Furthermore, follow-up MRI data with resembling time intervals was not available for the control group, thereby making a quantification of the physiological volume changes of the mesiotemporal structures impossible. We are aware of the fact that this may be a confounder of our results, especially when comparing follow-up MRI studies of the patients with the corresponding controls. However, the following points argue against a substantial influence of this issue on the results presented here: The time interval between the patients’ follow-up MRIs was rather short (≤24 months between MRI 1 and MRI 3) and previous studies report about HC volume changes of 0–2% in healthy adults over a time period of 3.5 years31 and 2 years, respectively.32 In contrast, we found a considerably higher median percentage HC volume loss of 12.5% between MRI 1 and MRI 3 in LE. We did not find any studies quantifying longitudinal physiological volume changes of the AM. Our results argue against a relevant AM volume change during normal ageing, as we did not find a correlation between age and AM volume in our control group.

Finally, our results are limited to a specific subset of antibody-associated LE as no patients with antibodies to AMPA, GABA-B and metabotropic glutamate receptors were identified in our institution during the time of this study.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

Footnotes

  • Funding The study was partly supported by the SFB TR3 projects A1 and A8 of the DFG.

  • Competing interests JW was supported by the Gerok Program of the BONFOR commission, University of Bonn. BW was supported by the Deutsche Forschungsgemeinschaft (DFG) with a Heisenberg grant (BW: WE 4427/3-1).

  • Ethics approval Local Ethics Committee University of Bonn.

  • 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.