MRI characteristics of periaqueductal lesions in multiple sclerosis

Download PDF
66 downloads 0 Views 1MB Size Report
Comment
MS and 14 a progressive disease course. ... et al., 2002) and the PAG region is thought to be associated ... individually selected disease modifying treatments.
Multiple Sclerosis and Related Disorders (2014) 3, 542–551

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/msard

MRI characteristics of periaqueductal lesions in multiple sclerosis Athina Papadopouloua,b, Yvonne Naegelina, Katrin Weiera, Michael Amanna,b,c, Jochen Hirschc, Stefanie von Feltend, Oezguer Yaldizlia, Till Sprengera,b,c, Ernst Wilhelm Radueb, Ludwig Kapposa, Achim Gassa,n a

Department of Neurology, University Hospital Basel, Petersgraben 4, CH-4031 Basel, Switzerland Medical Image Analysis Center, Schanzenstrasse 55, CH-4056 Basel, Switzerland c Department of Radiology and Nuclear Medicine, Division of Diagnostic and Interventional Neuroradiology, University Hospital Basel, Petersgraben 4, CH-4031 Basel, Switzerland d Clinical Trial Unit, University Hospital Basel, Schanzenstrasse 55, CH-4056 Basel, Switzerland b

Received 19 December 2013; accepted 8 January 2014

KEYWORDS Multiple sclerosis; Periaqueductal lesions; Brainstem symptoms

Abstract Background: In multiple sclerosis (MS), periaqueductal lesions (PAL) have been described histopathologically.

Objectives: We sought to investigate the frequency and characteristics of PAL on magnetic resonance images (MRIs) in patients with MS or clinically isolated syndrome (CIS). Methods: We analyzed proton density (PD)-weighted MRIs of 247 MS and 10 CIS patients. PAL were identified based on their abnormal hyperintensity and lesion shape on at least two consecutive slices. Patients with and without PAL were compared for clinical characteristics in a propensity score weighted analysis. Results: We identified PAL in 48/257 patients (18.7%), 34 of which had CIS or relapsing-remitting MS and 14 a progressive disease course. The shape of PAL was often circular (65%), or/and wedgelike (42%). Multi-planar image analysis in a subgroup of patients with double inversion recovery sequences revealed that 36% of PAL were periventricular lesions of the third ventricle extending towards the aqueduct. We found an association of PAL and brainstem functional system.

Abbreviations: CIS, clinically isolated syndrome; DIR, double inversion recovery; EDSS, expanded disability status scale; FS, functional system; MRI, magnetic resonance image; MS, multiple sclerosis; NMO, neuromyelitis optica; NRS, numerical rating scale; PAG, periaqueductal gray matter; PAL, periaqueductal lesions; PD, proton density n Corresponding author. Present address: Department of Neurology, University Hospital Mannheim, Theodor-Kutzer-Ufer 1-3, DE-68167 Mannheim, Germany. Tel.: +49 621 3833553; fax: +49 621 3833807. E-mail addresses: [email protected] (A. Papadopoulou), [email protected] (Y. Naegelin), [email protected] (K. Weier), [email protected] (M. Amann), [email protected] (J. Hirsch), [email protected] (S. von Felten), [email protected] (O. Yaldizli), [email protected] (T. Sprenger), [email protected] (E.W. Radue), [email protected] (L. Kappos), [email protected] (A. Gass). http://dx.doi.org/10.1016/j.msard.2014.01.001 2211-0348/& 2014 Elsevier B.V. All rights reserved.

MRI characteristics of PAL in MS

543 Conclusions: Although PAL may be underreported in MS, they are relatively frequent and found at all clinical stages and in CIS. They could be considered as a variant of periventricular lesions in the supratentorial midbrain and thus be useful in the diagnosis of MS. & 2014 Elsevier B.V. All rights reserved.

1.

Introduction

The aqueduct of Sylvius is a small channel that traverses the midbrain connecting the third with the fourth ventricle. It is surrounded by densely layered neurons, forming the periaqueductal gray matter (PAG). The exact functions of the PAG neurons are not fully understood, but experimental data suggest an important role in analgesia (Bartsch et al., 2004; Knight et al., 2002) and the PAG region is thought to be associated with migraine (Rocca et al., 2006; Welch et al., 2001). Furthermore PAG-neurons seem to be involved in the regulation of the micturition reflex in animals (Marson, 2004) and humans (Athwal et al., 2001; Benarroch, 2010; Fowler, 2008; Sakakibara et al., 1997). In multiple sclerosis (MS) patients, periaqueductal lesions (PAL) are a typical finding according to neuropathological data (Prineas and McDonald, 1997). However, PAL do not receive much attention in daily clinical practice. In the scientific literature PAL have been reported in MS patients only in association with headache, especially migraine (Gee et al., 2005; Fragoso and Brooks, 2007; Haas et al., 1993; Tortorella et al., 2006). In this cross-sectional observational study, we aimed to determine the frequency of PAL and their magnetic resonance imaging (MRI) characteristics in a relatively large cohort of patients with MS and clinically isolated syndrome (CIS).

2. 2.1.

Material and methods Patients

We analyzed the MR Images of all 257 patients (247 MS and 10 CIS patients) participating in an ongoing prospective observational study on the phenotypic–genotypic characterization of all clinical subtypes of MS, obtained at study baseline (2005). At the time of the MRI the patients were treated with best individually selected disease modifying treatments. All patients were clinically stable. Patients with an acute relapse were not examined and the MRI scan was postponed at least 30 days after the last dose of steroid treatment. In addition, 11 healthy control subjects (4 women and 7 men with a mean age 33.5 years) were also included, in order to study the normal MRI characteristics of the periaqueductal area. Informed consent was obtained in writing from all subjects, in accordance with the local ethics committee approval.

2.2.

Analysis of the MR images

MRI was obtained with a 1.5 T system (Avanto, Siemens Medical systems, Erlangen, Germany). The brain MRI protocol consisted of axial, double-echo proton density (PD)/T2-weighted sequences (repetition time (TR): 3980 s, echo time (TE): TE1/ TE2=14 ms/108 ms, inplane resolution of 0.98  0.98) as well

as post contrast T1-weighted images (T1w spin-echo scan with TR: 550 ms, TE: 17 ms, inplane resolution of 0.98 mm  0.98 mm and 3 mm slice thickness), positioned parallel to the inferior borders of the corpus callosum. The T2- and T1hypointense lesions were marked on the axial PD-weighted (-w) and T1-w post contrast images respectively and then segmented by means of the commercially available semiautomatic thresholding contour software AMIRA 3.1.1 (Mercury Computer Systems Inc) to calculate their total volumes. The majority of patients was willing to also undergo spinal cord MR (n=201) and a biplanar review (sagittal and axial planes of PDw and T2-w images) was performed to include the number of spinal cord lesions in the analysis. The axial PD-w MR images of the 247 MS, 10 CIS patients and 11 healthy controls were analyzed using a structured reporting scheme. The brainstem/midbrain area was evaluated on 5 consecutive 3 mm slices. Two experienced readers (AP and AG), unaware of patient identity and clinical characteristics, evaluated the images by consensus. In order to avoid false positive results, we used the following conservative criteria for the detection of PAL: they had to be identified on a minimum of two consecutive slices, to be hyperintense, have an asymmetrical or wedge-like shape or/and clearly exceed the normal, circular PA gray matter (Fig. 1). In 22 patients with PAL, a further 3-dimensional Double Inversion Recovery (3D DIR) sequence (repetition time (TR): 7.5 s, echo time (TE): 311 ms, inversion time (TI): 3 s, slice thickness: 1.5 mm and inplane spatial resolution 1.33 mm  1.33 mm) was performed. PAL were then reviewed in particular for their topographical relationship to surrounding anatomical structures in a multiplanar reconstruction mode visualizing 3 anatomical planes simultaneously.

2.3. Acquisition of the patients0 clinical and imaging-characteristics The demographic and clinical characteristics of the patients (age, gender, disease duration, disease course, Expanded Disability Status Scale (EDSS) score and Functional System (FS) scores, presence of migraine, presence and severity of pain) were obtained at the time of the MRI (presented in Tables 1 and 2). On the day of the MRI all patients underwent a standardized neurological examination by certified physicians (http://www.neurostatus.net), including the assessment of the brainstem- and bowel/bladder FS scores. The brainstem FS score refers to the cranial nerve functions and varies from 0 (normal) to 5 (inability to swallow or speak). It includes the assessment of extraocular movements impairment and nystag mus. The bowel/bladder FS score is based on reported symptoms by the patients, regarding potential urinary hesi tancy and retention, urinary urgency and incontinence, blad der catheterization or/and bowel dysfunction. It varies from 0

544

A. Papadopoulou et al. scale (NRS) with a range from 0 (no pain) to 10 (worst imaginable pain).

2.4.

Statistical analysis

To compare characteristics of patients with and without PAL (age, gender, disease duration, disease course, EDSS, T2- and T1-lesion volumes, presence and number of spinal cord lesions) we used a Wilcoxon rank sum test (Mann–Whitney test) for numerical variables and a chi-square test for the difference in proportions of categorical variables. To assess a potential effect of PAL on specific clinical variables (brainstem FS score, bowel/ bladder FS score, pain, migraine) we used propensity score weighted analyses. First, propensity scores were calculated using a logistic regression model predicting the occurrence of PAL (binary variable). All potential confounders, i.e., variables that may affect the occurrence of PAL as well as the outcome were used as predictors (i.e., all patient characteristics mentioned above). We used the R-package twang (Toolkit for Weighting and Analysis of Nonequivalent Groups [computer program], 2010) that implements propensity score estimation by boosted logistic regression, iteratively minimizing the covariate imbalance between the two groups (with and without PAL). Inverse probability weighting (IPW) was used to estimate the adjusted effect of PAL on the endpoints brainstem FS score, bowel/bladder FS score and pain, as assessed by a numerical rating scale with a linear model, and on the binary endpoint presence of migraine with a logistic regression model. All scores were log-transformed to meet the assumption of normality. In addition to the IPW estimate, an unadjusted (crude) estimate for the effect of PAL was derived for each outcome variable, using the models described above but without IPW. More details about the propensity score based analysis are provided in Appendix A. In addition to the model based analysis, yielding IPW and crude estimates, we performed an unadjusted analysis for the same end-points, using a non-parametric Wilcoxon rank sum test for the scores and a chi-square test for the difference in proportions of migraine.

Fig. 1 Proton density-weighted images demonstrating normal MRI features of the periaqueductal region and typical periaqueductal lesions. Panel A: MRI at the midbrain level demonstrating (a) normal periaqueductal gray matter, (b–d) variations of the appearance of the periaqueductal region not considered as lesions: (b) absence of the central hypointense (flow void) signal of the aqueduct, (c) thin, linear symmetrical hyperintensities in the periaqueductal white matter and (d) symmetrical faint hyperintensities in the periaqueductal white matter. Panel B: examples of PAL. (1) asymmetrical lesions; (2) circular lesions (abnormal, hyperintense signal and abnormal thickness of the whole PAG rim); (3) wedge-shaped lesions; (4) severe affection of the periaqueductal area with all abnormal characteristics (hyperintense, asymmetrical, wedge-shaped lesions with additional circular lesion).

(normal) to 6 (loss of bowel and bladder function). The presence of migraine and pain were reported by the patients on a questionnaire. The severity of pain was assessed through the widely used (Farrar et al., 2001) 11-point numerical rating

3. 3.1.

Results Frequency and characteristics of PAL

Proton density-weighted images of the healthy control subjects showed normal PA gray matter as a thin, symmetric, homogeneous rim surrounding the aqueduct of Sylvius. In contrast to the MS patients, none of the healthy control subjects had a wedge-shaped appearance of the PAG. On qualitative assessment, the healthy control PAG signal was slightly brighter than cortical gray matter (Fig. 1). Analysis of PDw images identified PAL in 48/257 (18.7%) patients. In 20/48 (42%) cases, the periaqueductal lesion had a wedge-like shape (Fig. 1). Moreover, 31/48 patients with PAL (65%) showed also an abnormally hyperintense, broad PAG rim, considered as a circular PAL (Fig. 1). A severe affection of the PAL, with all abnormal characteristics (hyperintense, asymmetrical, wedge-shaped lesion with additional circular PAL) was seen in 15 patients (15/48=31% of patients with PAL) (Fig. 1). None of the PAL showed contrast enhancement (Fig. 2).

MRI characteristics of PAL in MS

Table 1

545

Comparison of clinical and MRI characteristics in patients with and without periaqueductal lesions. All patients (n =257)

Age (years) Mean7SD 44.4711 Median 44 Gender: male 78/257 (30.4%) Disease duration (years) Mean7SD 12.579.1 Median 10 Disease course CIS and RRMS 187/257 (72.8%) SPMS and PPMS 70/257 (27.2%) EDSS Mean7SD 3.271.7 Median 3.0 T2 hyperintense lesion volume (μl) Mean7SD 6082.676895.8 Median 3276.1 T1 hypointense lesion volume (μl) Mean7SD 2344.573673.8 Median 967.1 Patient with Z1 spinal cord lesion 150/201 (74.6%) Number of spinal cord lesions per patient Mean7SD 372.5 Median 3

Patients without PAL (n =209)

Patients with PAL (n =48)

P value

44.8711.2 44.2 62/209 (29.7%)

42.7710.4 42 16/48 (33.3%)

0.19

12.979.3 10

11.178.2 10

0.27

153/209 (73.2%) 56/209 (26.8%)

34/48 (70.8%) 14/48 (29.2%)

0.88

3.171.7 3.0

3.671.7 3.5

0.014*

5438.476346.6 2761.1

8887.978418.3 5426.3

o0.001*

2012.873012.9 786.8 119/166 (71.7%)

3788.975542.8 2042.9 31/35 (88.6%)

0.002*

2.672.4 2

4.572.4 5

0.75

0.061 o0.001*

P-values were calculated using a Wilcoxon rank sum test (Mann–Whitney test) for numeric variables and a chi-square test for categorical variables. Means are given together with standard deviations (SD). Note that only a subgroup of patients had available data regarding the presence of spinal cord lesions (n=201, 166 patients without and 35 with PAL). Abbreviations: PAL=periaqueductal lesions, CIS=clinically isolated syndrome, RRMS=relapsing remitting multiple sclerosis, SPMS=secondary progressive multiple sclerosis, PPMS=primary progressive multiple sclerosis, EDSS=expanded disability status scale. n Significant differences.

3.2.

Characteristics of PAL on DIR-Sequence

In 22 of the 48 patients with PAL, 3D DIR images were reviewed in 3 planes simultaneously by using a multi-planar reconstruction mode (MPR). In all 22 cases, the PAL observed on PD-w SE was confirmed on DIR. PAL contrast against the hypointense background of the aqueduct, 3rd ventricle and surrounding white matter tissue was stronger than on PDw images (Fig. 3). The majority of PAL (14/22, 64%) were isolated lesions involving the gray- and white-matter surrounding the aqueduct. However, 8/22 (36%) PAL were shown to be extensions of periventricular/subependymal lesions of the 3rd ventricle, extending caudally towards the aqueduct and involving the PA gray matter (Fig. 4).

3.3.

Comparison of patients with and without PAL

The imaging and clinical characteristics of patients with PAL and without PAL are presented in Table 1. Periaqueductal lesions were seen in CIS (2/10 patients) and in all clinical courses of MS. Patients with PAL had on average higher T2- and T1-lesion volumes as well as higher number of spinal cord lesions. They also had higher EDSS scores. There was no difference in age, gender, disease duration or disease course

(CIS/RRMS vs. progressive disease course) between patients with and without PAL (Table 1).

3.4.

Associations of PAL with clinical symptoms

The propensity score weighted analysis (Table 2, right part) showed an association between PAL and brainstem FS (effect size 1.17, 95% CI [1.02; 1.35]). This indicates, that the brainstem FS in patients with PAL was on average increased by 17% compared to patients without PAL (95% CI: 2–35.2%; P=0.031), probably independent of other factors like total T2 lesion load. This is consistent with the results of the unadjusted comparison (Table 2, left part), which showed a significantly higher mean brainstem FS in patients with PAL compared to patients without PAL. There was no significant effect of PAL on the incidence of migraine, severity of pain, or bowel/bladder FS score.

4.

Discussion

Periaqueductal lesions (PAL) may be underreported in MS for several reasons. They are usually small and located in a region of a busy MRI signal profile, due to the close vicinity of cerebrospinal fluid (CSF) flow, gray- and white-matter. Furthermore they are often difficult to distinguish from

546

A. Papadopoulou et al.

Fig. 2 MS patient with a periaqueductal lesion (PAL) and multiple other lesions, typical for MS. Note that none of the PAL in this cohort showed contrast enhancement. From left to right: PD-weighted, T2-weighted and T1-weighted sequence after i.v. administration of contrast agent.

Fig. 3 Multiplanar Double inversion recovery (DIR) images showing a periaqueductal lesion. From left to right: axial, coronal and sagittal sections. Note the strong contrast between the lesion and the hypointense background of the aqueduct and surrounding white matter tissue.

normal periaqueductal gray matter, particularly if no PD-w or DIR images are available. In addition, unlike in MS patients with well-known brainstem syndromes e.g., trigeminal sensory loss, trigeminal neuralgia or internuclear opthalmoplegia, where characteristic MRI lesions may be identified in the brainstem (Gass et al., 1997; Frohman et al., 2001; Gass and Hennerici, 1997), PAL rarely show a unique clinical correlate and may thus not receive the same focused attention as other brainstem lesions. However, according to neuropathological data (Prineas and McDonald, 1997) periaqueductal lesions are a typical finding in patients with MS. In our study, we identified PAL in almost 20% of the patients, at different clinical stages of MS, including patients with a CIS. Moreover, PAL were often extensions of periventricular 3rd ventricle lesions and could be therefore considered a subtype of supratentorial periventricular lesions at the midbrain level.

From this perspective PAL might be included in the location of current lesions accepted to fulfill the McDonald criteria (Polman et al., 2011). As demonstrated in Figs. 3 and 4, the 3D DIR data set was very useful to visualize the topographical relationship of PAL to the 3rd ventricle. Due to a suppression of the signal of the CSF and white matter, the typical DIR signal characteristics result in a strong contrast between PAL and surrounding tissue, which was particularly helpful in ascertaining the small periaqueductal lesions. Moreover, the reduction of CSF flow artifacts due to the 3D data acquisition is another important feature, particularly in areas near the aqueduct (Pouwels et al., 2006). Further studies are needed to also examine the specificity of PAL in the differentiation of MS from other conditions. Interestingly, the periaqueductal region and the region around

MRI characteristics of PAL in MS

547

Fig. 4 Multiplanar DIR images showing a lesion extending from the right periventricular area of the 3rd ventricle to the periaqueductal gray matter. Note the role of 3D DIR images to visualize the topographical relationship of PAL to the 3rd ventricle as extensions of periventricular 3rd ventricle lesions. Table 2 Outcome variables

Associations of PAL with brainstem and bowel/bladder functional system scores, migraine and pain. Unadjusted comparisons between patients with and without PAL

Estimate for the effect of PAL on the clinical variables

All patients (n =257) Brainstem FS score Mean7SD 0.870.8 Median 1 Bowel/bladder FS score Mean7SD 0.971 Median 1 Migraine 57/257 (22.2%) Pain Mean7SD 272.4 Median 1

Patients without PAL (n =209)

Patients with PAL (n =48)

P value

Crude estimate [95% CI]

IPW estimate [95% CI]

0.770.8 1

1.271 1

0.003

1.25 [1.09–1.44]

1.17 [1.02–1.35]

0.871 1 47/209 (22.5%)

1.170.9 1 10/48 (20.8%)

0.006

1.22 [1.05–1.43] 0.91 [0.4–1.9]

1.13 [0.95–1.34] 1.25 [0.5–3.15]

272.3 1

272.6 1

1 [0.79–1.27]

1.06 [0.83–1.35]

0.96

0.99

The first 4 columns on the left of the table show the results of the comparison between the two patient groups (without PAL vs. with PAL) using a Wilcoxon rank sum test (Mann–Whitney test) for numerical variables and a chi-square test for categorical variables. The last two columns on the right show the estimated effect size of PAL on the clinical variables before (crude estimates) and after inverse probability weighting by the propensity score (IPW estimates). Estimates 41 indicate an increase in the outcome variable due to PAL (e.g., in case of brainstem FS score). Abbreviations: FS =functional system, PAL=periaqueductal lesions, IPW=inverse probability weighting, 95% CI=95% confidence interval.

the 3rd ventricle have been recently suggested to be preferentially involved in patients with neuromyelitis optica (NMO) (Pittock et al., 2006). Although the presence of PAL in our patients was significantly associated with higher numbers of spinal cord lesions (Table 1) none of the patients fulfilled the

clinical or MRI criteria for NMO (Wingerchuk et al., 2006) and were therefore not tested for NMO-IgG antibodies. The pathogenetic mechanisms of lesion formation are different between NMO and MS. In NMO, the periaqueductal and the periventricular area around the 3rd ventricle are probably

548

A. Papadopoulou et al.

affected due to their high concentrations of aquaporin-4 channels (Pittock et al., 2006). In MS, lesion formation in the periaqueductal area is probably mainly related to inflammation around the subependymal veins, similar to the areas around the lateral ventricles. Moreover, the close vicinity to the CSF and potential direct gliotoxic effects from the CSF might be an additional mechanism for the formation of PAL in MS (Cid et al., 2002; Menard et al., 1998). We could not find a relation between PAL and the incidence of migraine in our cohort. This is in line with the earlier report by Tortorella et al. (2006) but not with the report of Gee et al. (2005), who described a four-fold increase in migraine-like headaches in patients with a lesion near to or in the periaqueductal gray matter. It remains open if acute, contrast-enhancing PAL would be closer associated with “focal” symptoms. Our results suggest an association between PAL and brainstem FS scores, which were determined mostly by oculomotor signs in our patients. However, the clinical associations or the lack of associations have to be interpreted with caution, due to the observational nature of the data, which typically involve the problem of confounding. For example, an association of PAL with oculomotor sings could be due to the co-incidence of other brainstem lesions. The propensity weighted analysis reduces such confounding bias, but cannot ensure its complete elimination. Another methodological limitation of this study is that the assessment of migraine was based on the patients0 correspondence and not on a questionnaire with specific questions, according to the diagnostic criteria of migraine. Moreover, pain was only defined on the basis of a numeric rating scale, without further information about the characteristics, localization of pain etc.

5.

Conclusions

Our study clearly shows that PAL are a relatively frequent MRI finding in MS patients, found at all stages of the disease and even in CIS. Since they may share pathogenetic mechanisms with periventricular lesions and sometimes have a direct anatomical relationship with lesions around the 3rd ventricle, they could be considered a subtype of supratentorial periventricular lesions in the midbrain. They could therefore be useful in the diagnosis of MS and CIS. Thus, neurologists and neuroradiologists should be familiar with the normal periaqueductal appearance on MRI and the typical periaqueductal lesions as described on PD-w and DIR images.

Conflict of interest All authors state that they have no conflict of interest regarding this manuscript.

Disclosures of authors0 financial relationships Athina Papadopoulou has received consultation fee from TEVA. Yvonne Naegelin, Katrin Weier, Stefanie von Felten, Michael Amann and Jochen Hirsch have nothing to disclose. Oezguer Yaldizli received lecture fees from Teva and Bayer Schering that were exclusively used for funding of research and continuing education for residents in the Department of Neurology at the University Hospital Basel.

Till Sprenger has served on advisory boards for Eli Lilly, Genzyme, Mitsubishi Pharma, and Biogen and received reimbursement for travel expenses from Pfizer, Bayer Schering, Eli Lilly, and Biogen. The University Hospital Basel as employer of Ludwig Kappos has received and dedicated to research support fees for board membership, consultancy or speaking, or grants, in the last 3 years from Actelion, Advancell, Allozyne, Bayer, Bayhill, Biogen Idec, BioMarin, CSL Behring, Eli Lilly, European Union, Genmab, GeNeuro SA, Gianni Rubatto Foundation, Glenmark, Merck Serono, MediciNova, Mitsubishi Pharma, Novartis, Novartis Research Foundation, Novonordisk, Peptimmune, Roche, Roche Research Foundation, Santhera, Sanofi Aventis, Swiss MS Society, Swiss National Research Foundation, Teva, UCB, Wyeth. Ernst Wilhelm Radue has received research support (mainly for MS projects) and lecture fees from Actelion, Basilea, Bayer Schering, Biogen Idec, Merck-Serono and Novartis. Achim Gass has received honoraria for lecturing, travel expenses for attending meetings, and financial support for research from Bayer Schering, Biogen Idec, Merck Serono, Novartis and TEVA Neurosciences.

Authors0 contribution A.P. read the MRI data, drafted the manuscript and contributed to the study design and interpretation of the results. Y.N. was responsible for all the clinical data, contributed to the study design and the interpretation of the results. K.W. analyzed part of the MRI data and contributed to the interpretation of the results. M.A. and J.H. provided important technical assistance in the sequence design and data analysis. S.v.F. performed the statistical analysis. O.Y., T.S., E.W.R. and L.K. contributed to the study design and interpretation of the results. A.G. designed the study, read the MRI data and contributed to the interpretation of the results and the manuscript preparation. All authors also commented on the different drafts of the manuscript. They all have reviewed and approved the final submitted version of the manuscript.

Appendix A. weighting 1.

Details on propensity score

Calculation of the propensity score

The propensity score was estimated by boosted logistic regression, as implemented in package twang (Ridgeway et al., 2010) of the statistical software package R (version 2.13.1; R Development Core Team, 2011). The method is iterative, minimizing the imbalance of covariates between groups. The average effect size difference across all covariates was used as imbalance criterion. The strength of boosted logistic regression is to deliver stable estimates even with correlated predictors, and that interactions between predictors can be taken into account (McCaffrey et al., 2004).

MRI characteristics of PAL in MS The propensity score model was a (boosted) logistic regression model predicting the occurrence of periaqueductal lesions (PAL) (binary dependent variable, 1=yes, 0=no). The predictors used were all potential confounders, i.e., variables that may affect the occurrence of PAL as well as the outcome (brainstem FS, bowel/bladder FS, pain): patient age, sex, disease course (with factor levels CIS, RRMS, SPMS and PPMS), EDSS, disease duration, T2-lesion volume, T1-lesion volume, and the number of spinal cord lesions (Fig. A1). The estimated propensity score for each patient is a number between 0 and 1,

549 with larger scores indicating a higher probability for the occurrence of PAL (Fig. A2). Typically, the group with PAL has higher propensity scores than the group without PAL.

2.

We used Inverse Probability Weighting (IPW) to estimate the effect of PAL on outcome variables (Hernan and Robins, 2006). The method is also implemented in the R-package twang. IPW estimates the average effect on the population. Patients with PAL received the weight 1/p, i.e., small p results in large weight, patients without PAL received the weight 1/(1 p), i.e., large p results in large weight. The resulting pseudo-population shows the same distribution of covariates as the original population (all patients in the data set). Figs. A3 and A4 show the improvement of covariate balance by propensity score weighting.

3.

Fig. A1 Relative influence (%) of covariates on the estimation of propensity scores (SC.lesions.no: number of spinal cord lesions, age Years: age (y), T1BhVol: T1-lesion volume, DISCRS: disease course, disease Years: disease duration (y), ACTUAL. EDSS: expanded disability status scale (EDSS), T2Vol: T2-lesion volume, SEX: gender).

Propensity score weighting

Estimating the effect on the outcome

The effect of PAL on several outcome variables was estimated by using a treatment model with brainstem FS, bowel/bladder FS, pain or the occurrence of migraine as dependent variable, the occurrence of PAL as explanatory factor (analogous to treatment in a clinical trial). In the treatment model, patients were weighted as described above. A linear treatment model was used for the endpoints brainstem FS, bowel/bladder FS and pain, which were log-transformed after addition of 1 (since the logarithm of zero is undefined) to meet the normality assumption. For the

Fig. A2 Propensity scores of patients with (top) and without (bottom) periaqueductal lesions (PAL), estimated by boosted logistic regression (n =257).

550

A. Papadopoulou et al.

Fig. A3 Improvement of covariate balance by propensity score weighting shown as increasing P-values of the Kolmogorov–Smirnov statistic due to weighting. P-values before and after weighting are shown as solid and hollow circles, respectively. Small P-values indicate imbalance; P-values above the 451-line are larger than expected by chance (i.e., in a randomized trial).

Fig. A4 Improvement of covariate balance by propensity score weighting shown as decrease in the standardised difference for most covariates (blue lines). Filled red circles represent significant differences.

binary endpoint occurrence of migraine a generalized linear treatment model was used. In addition to the propensity score weighted estimation (using IPW), the “crude” effect was estimated using a “naïve” model, without propensity score weighting. The propensity score weighed analysis was chosen to adjust for differences between the two patient groups, regarding characteristics other than the occurrence of PAL, thereby minimizing confounding bias. In order to fit a statistical model, we treated outcome variables as continuous, although they are scores and not truly continuous.

References Athwal BS, Berkley KJ, Hussain I, Brennan A, Craggs M, Sakakibara R, et al. Brain responses to changes in bladder volume and urge to void in healthy men. Brain 2001;124:369–77.

Bartsch T, Knight YE, Goadsby PJ. Activation of 5-HT(1B/1D) receptor in the periaqueductal gray inhibits nociception. Ann Neurol 2004;56:371–81. Benarroch EE. Neural control of the bladder: recent advances and neurologic implications. Neurology 2010;75:1839–46. Cid C, Alcazar A, Regidor I, Masjuan J, Salinas M, Alvarez-Cermeno JC. Neuronal apoptosis induced by cerebrospinal fluid from multiple sclerosis patients correlates with hypointense lesions on T1 magnetic resonance imaging. J Neurol Sci 2002;193:103–9. Farrar JT, Young Jr. JP, LaMoreaux L, Werth JL, Poole RM. Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain 2001;94:149–58. Fowler CJ, Griffiths D, de Groat WC. The neural control of micturition. Nat Rev Neurosci 2008;9:453–66. Fragoso YD, Brooks JB. Two cases of lesions in brainstem in multiple sclerosis and refractory migraine. Headache 2007;47:852–4. Frohman EM, Zhang H, Kramer PD, Fleckenstein J, Hawker K, Racke MK, et al. MRI characteristics of the MLF in MS patients with

MRI characteristics of PAL in MS chronic internuclear ophthalmoparesis. Neurology 2001;57: 762–768. Gass A, Hennerici MG. Neurological picture. Bilateral internuclear ophthalmoplegia in multiple sclerosis. J Neurol Neurosurg Psychiatry 1997;63:564. Gass A, Kitchen N, MacManus DG, Moseley IF, Hennerici MG, Miller DH. Trigeminal neuralgia in patients with multiple sclerosis: lesion localization with magnetic resonance imaging. Neurology 1997;49:1142–4. Gee JR, Chang J, Dublin AB, Vijayan N. The association of brainstem lesions with migraine-like headache: an imaging study of multiple sclerosis. Headache 2005;45:670–7. Haas DC, Kent PF, Friedman DI. Headache caused by a single lesion of multiple sclerosis in the periaqueductal gray area. Headache 1993;33:452–5. Hernan MA, Robins JM. Estimating causal effects from epidemiological data. J Epidemiol Community Health 2006;60:578–86. Knight YE, Bartsch T, Kaube H, Goadsby PJ. P/Q-type calciumchannel blockade in the periaqueductal gray facilitates trigeminal nociception: a functional genetic link for migraine? J Neurosci 2002;22:RC213. Marson L. Lesions of the periaqueductal gray block the medial preoptic area-induced activation of the urethrogenital reflex in male rats. Neurosci Lett 2004;367:278–82. Menard A, Pierig R, Pelletier J, Bensa P, Belliveau J, Mandrand B, et al. Detection of a gliotoxic activity in the cerebrospinal fluid from multiple sclerosis patients. Neurosci Lett 1998;245:49–52. McCaffrey DF, Ridgeway G, Morral A. Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychol Methods 2004;9:403–25. Pittock SJ, Weinshenker BG, Lucchinetti CF, Wingerchuk DM, Corboy JR, Lennon VA. Neuromyelitis optica brain lesions localized at sites of high aquaporin 4 expression. Arch Neurol 2006;63:964–8.

551 Polman CH, Reingold SC, Banwell B, Clanet M, Cohen JA, Filippi M, et al. Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann Neurol 2011;69:292–302. Pouwels PJ, Kuijer JP, Mugler III JP, Guttmann CR, Barkhof F. Human gray matter: feasibility of single-slab 3D double inversion-recovery high-spatial-resolution MR imaging. Radiology 2006;241:873–9. Prineas JW, McDonald WI. Demyelinating diseases. In: Graham DI, Lantos PL, editors. Greenfield0 s neuropathology. 6th ed. London: Arnold; 1997. p. 821–4. Rocca MA, Ceccarelli A, Falini A, Colombo B, Tortorella P, Bernasconi L, et al. Brain gray matter changes in migraine patients with T2-visible lesions: a 3-T MRI study. Stroke 2006;37:1765–70. Ridgeway G, McCaffrey D, Morral A. Toolkit for weighting and analysis of nonequivalent groups. R package version 1.0-3; 2010. R Development Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2011. 〈http://www.R-project.org〉. Sakakibara R, Hattori T, Yasuda K, Yamanishi T, Tojo M, Mori M. Micturitional disturbance in Wernicke0 s encephalopathy. Neurourol Urodyn 1997;16:111–5. Toolkit for Weighting and Analysis of Nonequivalent Groups [computer program]. Version R package version 1.0-3; 2010. Tortorella P, Rocca MA, Colombo B, Annovazzi P, Comi G, Filippi M. Assessment of MRI abnormalities of the brainstem from patients with migraine and multiple sclerosis. J Neurol Sci 2006;244:137–41. Welch KM, Nagesh V, Aurora SK, Gelman N. Periaqueductal gray matter dysfunction in migraine: cause or the burden of illness? Headache 2001;41:629–37. Wingerchuk DM, Lennon VA, Pittock SJ, Lucchinetti CF, Weinshenker BG. Revised diagnostic criteria for neuromyelitis optica. Neurology 2006;66:1485–9.