In this study we analysed the clinical and paraclincal features associated with myelitis and optic neuritis in a large cohort of Caucasian patients with neuromyelitis optica spectrum disorders in a stratified fashion according to the patients' AQP4-Ab serostatus. Clinically, the disease course was characterized by severe attacks in many patients despite immunosuppressive and immunomodulatory treatment, often resulting in disability and impaired mobility within short time. More than two thirds of all myelitis attacks were associated with motor symptoms (with sensorimotor paraparesis as the most common single manifestation), and around half of all motor attacks resulted in an MRC grade of 2 or lower in one or more limbs. Every fourth motor attack was associated with tetraparesis, leading to respiratory failure in some cases, and in more than a quarter of patients, the disease was not restricted to the spinal cord but involved the brain stem as well. The rate of no or incomplete recovery from myelitis was high (around 75%) already from the beginning, and worsened with the number of subsequent relapses. Accordingly, around 40% of all patients with a history of myelitis had an EDSS of 6 or higher at the end of the observational period. Similarly, the visual deficiency present during acute ON attacks was often severe (visual acuity ≤ 0.1 in more than 60% of all relapses), and remission was incomplete in most cases. Motor symptoms at onset, tetraparesis early in the disease course, and more than one myelitis attack during the first disease year were identified as possible predictors of a worse clinical outcome in patients with long term follow-up. The median number of relapses per year (0.53 myelitis attacks and 0.38 ON attacks) was similar to that observed in patients with MS under standard treatment with IFN-beta (0.29-1.82) or glatiramer acetate (0.34-1.19), but the time to EDSS 6-6.5 (walking aid needed) was much shorter (7.8 years) than that noted even in untreated exacerbating-remitting MS patients (23 years) . The latter finding is likely to reflect both the preferential affection of the spinal cord in NMO as compared to MS and the differential modes of inflammatory tissue damage in the two conditions. While MS primarily causes demyelination, NMO attacks are often associated with severe necrosis . Seropositive and seronegative patients were found to differ with regard to attack severity and clinical presentation. Visual acuity of ≤ 0.1 during acute ON attacks was more frequent among seropositives, motor symptoms were more common in seropositive patients, the median MRC grade during acute myelitis attacks worse, and MRC grades ≤ 2 more frequent, in particular if patients met Wingerchuk's criteria . On the other hand, simultaneous myelitis and ON as well as bilateral ON at disease onset, and sensory symptoms were more frequent in the seronegative group. By contrast, the two groups did not differ significantly with regard to the median annualized total relapse rate, the median annualized myelitis specific relapse rate, the median annualized ON specific relapse rate, relapse outcome (no remission, partial or complete remission), and the frequency of brainstem involvement.
Given both the low remission rate (as compared to MS) found in our study already at disease onset and the fact that the first event was followed by a relapse after a median latency of just 9 months (and only 5 months in the seronegative group), early treatment and, therefore, an early diagnosis of NMO is crucial. However, this study revealed a marked delay in the diagnosis of NMO (16 months from onset if the disease started with myelitis, and even 55 months if the disease started with ON; p < 0.013). Most importantly, a substantial number of patients (42.5%) were initially wrongly diagnosed with MS, and 52.6% of these patients were treated at least once with IFN-beta, a drug which is safe and effective in MS but considered to be harmful in NMO [45–50]. In addition to incorrect differential diagnostic considerations on the part of treating physicians, other factors may have also contributed to this delay. In some patients the latency between the first ON attack and the first myelitis attack or vice versa was extraordinarily long. Also, the presence of supratentorial brain lesions, which has only recently been recognized as a common feature of NMO [36, 51, 52], as well as the fact that short MRI lesions not exceeding two vertebral segments and/or positive OCBs, a hallmark of MS, were present at onset in some patients, might have played a role. The median time to diagnosis was shorter among seronegative patients (11 months) than among seropositive patients (45 months), which is partly explained by the fact that NMO started more frequently with simultaneous myelitis and ON in the seronegative group.
From a diagnostic point of view, the following findings are of particular importance.
First, mild symptoms or a benign long-term course do not rule out the diagnosis of NMO, but occur in a substantial number of (both seropositive and seronegative) NMO cases. While the concept of benign MS is now well recognized , only a few cases of benign NMO have been reported thus far [54–56]. It is therefore of interest that around 17% of patients with long-term follow-up (≥ 100 months) in the present cohort were fully ambulatory (EDSS < 4) at last examination, including one (seropositive) case with an EDSS of 1.5 (indicating minimal neurological signs, but no disability) after 102 months. Similarly, a considerable number of myelitis attacks were either purely sensory (around 30%) or associated with only mild paresis, especially in seronegatives.
Second, a subset of patients (irrespective of the AQP4-Ab serostatus) may present with rare manifestations not typically associated with NMO. While the broad majority of patients presented with tetraparesis and paraparesis in our cohort, in a minority of cases hemiparesis, Brown Sequard syndrome, or monoparesis may occur, even at disease onset. NMO attacks may also exclusively affect the brain stem, independent of myelitis or optic neuritis, as observed in 15 cases in our series. Isolated brain stem encephalitis may even be the presenting symptom, as was the case in 5 of our patients. One of these patients, a young girl, was initially falsely diagnosed with bulimia nervosa because of repeated vomiting, others with gastro-intestinal infection. Moreover, besides intractable vomiting and hiccups, which have been repeatedly described in NMO [23, 57, 58], other brain stem symptoms such as dysphagia, laryngeal spasm, and oculomotor dysfunction occurred in a subset of patients and do not rule out the diagnosis of NMO as demonstrated here. Accordingly, brainstem lesions as detected by MRI were not in all cases restricted to the medulla oblongata and to the diencephalon, which are known sites of predilection in NMO , but may also occur in the pons, the cerebral peduncles, and the cerebellar peduncles.
Third, most patients did not present with simultaneous myelitis and bilateral ON, the classical syndrome described by Devic. Instead, NMO started in most cases with unilateral ON and, more rarely, with isolated myelitis. Importantly, the first ON attack was followed by myelitis after a median of just 14 months in the seropositive group, and the first myelitis was followed by ON after a median of only 3 months. This corroborates findings from previous, smaller studies, which found that AQP4-Ab seropositivity in patients with isolated ON or myelitis confers a high risk of conversion to NMO within one year [18–21] and strongly underlines the need for early prophylactic treatment in patients presenting with seropositive isolated ON or myelitis.
We found a strong female preponderance in our cohort. Interestingly, the female/male ratio was considerably higher than that observed in MS (1:2)  in the seropositive group (10.4), which is in accordance with a smaller previous study , but was similar to MS in the seronegative group (1.9). This finding further supports the concept that seropositive NMO is a distinct entity, different from classical MS and seronegative NMO. In line with the latter presumption, the median age of onset (40 years) in the seropositive group was higher by a decade than that in classical MS (29 years) .
The mortality rate was much lower in the present cohort (6% after a median observation time of 57.5 months) than that in a previously published North American cohort, which reported a five-year survival rate of only 68% . This may partly reflect an increase in awareness of the condition and thus earlier treatment. While the median year of onset in our study was 2004 (range, 1977-2011), in that previous study disease had started in 1985 in the relapsing subgroup, and in 1977 in the monophasic subgroup. The difference in mortality rates might also be the result of improved treatment options. Fifty-seven patients in our cohort were treated with rituximab at least once over the course of disease, and 37 with mitoxantrone, drugs that are now considered to be effective in NMO, but which were not available in the 1980s and 1990s when most of the patients reported by Wingerchuk and colleagues were seen. Similarly, testing for NMO-IgG/AQP4-Ab, which largely contributes to the laboratory differentiation of NMO and classical MS and which can thus guide treatment decisions, only became available in the middle of the last decade. Finally, genetic factors might also play a role. While around 40% of the patients in the North American series were classified as 'non-white' by the authors, our cohort consisted of an exclusively Caucasian population. Importantly, only 5 of the 9 patients who died during the observation period in our study died from attack-related neurological complications.
Recent studies have revealed a high frequency of brain lesions in Asian and mixed Asian/Caucasian NMO cohorts. Our data corroborate this finding in an exclusively Caucasian cohort. Supratentorial brain lesions were detectable in around 50% of all patients at first examination, irrespective of the AQP4-Ab antibody status, both in the total cohort and among those patients who met Wingerchuk's criteria, and at least once over the course of disease in 60.1%. The fact that around 17.7% of these MRIs were classified as "compatible with MS" by the reading radiologists underscores the relevance of AQP4-Ab, which was positive in 78.6% of these patients, in the differential diagnosis of MS and NMO.
Besides AQP4-Ab, the length of the spinal cord lesions is thought to discriminate between NMO and MS with high specificity . In fact, the first MRI showed at least one longitudinally extensive lesion in the vast majority of our patients. Importantly, however, 7.3% had a lesion shorter than three segments at first examination. Such cases can pose a serious differential diagnostic challenge (even more so as 60% had brain lesions on their first MRI). In 8 of these patients, the diagnosis of NMO could be established based on a positive AQP4-Ab test result. In the seronegative ones, the diagnosis of NMO could only be made based on the finding of lesions spanning over ≥ 3 segments on a follow-up spinal MRI later in the disease course and a negative brain MRI at onset. This underlines the importance of early AQP4-Ab testing also in patients presenting with myelitis and short spinal cord lesions as well as a need for repeated spinal cord imaging in these patients.
In the three largest previous studies [1, 27, 36], spinal cord lesions were reported to have spanned over 3 or more vertebral segments in most patients, but the exact length was not stated. In our cohort, spinal lesions had a median length of 6 segments, which is normally not found in MS . In addition, we evaluated whether more than one spinal cord lesion were present, and if so, calculated the total lesion load. In fact, 21 patients had an additional second or even a third lesion already at first examination, some of which also extended over three or more segments, corresponding to a median total lesion load of 8 vertebral segments. Again, such high lesion load is very unusual in classical MS. In line with the finding of more severe myelitis in the seropositive group, the total spinal cord lesion load was also higher among the seropositives in our cohort, lesions > = 6 segments were more frequent, and involvement of the entire spinal cord was observed only among seropositives.
At the time of first MRI examination, most patients had lesions located both in the cervical and thoracic portions of the spine. Among those with lesions restricted to either of the two sites, cervical involvement was slightly more common. However, 15 patients had additional lesions of the lumbar portion and/or the spinal conus. Together with our finding of multiple lesions in a considerable subset of patients, these data demonstrate that MRI examinations should not be restricted to the cervical spine but should ideally include the entire spinal cord.
The diagnosis of NMO according to Wingerchuk et al. (2006)  could be made based on the presence of two index events and brain and spinal cord MRI findings alone in 79% of all NMO patients and in 74.4% of the seropositive ones (not shown). Conversely, either brain MRI or spinal cord MRI was not formally required in 25.6% of the AQP4-Ab positive NMO patients to establish the diagnosis. However, we do not recommend dismissing brain and spinal cord imaging in seropositive cases, all the more as fulfilment of all three supportive criteria has been demonstrated to improve the specificity of the 2006 diagnostic criteria .
CSF findings in NMO are known to differ significantly from those in classical MS [62–64]. In accordance with previous studies, OCBs were present in only one third of our patients (compared to > 95% in MS) [65, 66]. Importantly, more than half of patients positive for OCBs at first LP became negative at follow-up. This is in sharp contrast to MS, where OCBs were shown to remain detectable over decades. Accordingly, repeat LP performed during remission may improve the CSF based laboratory differential diagnosis of NMO and MS substantially. In a previous study, we have shown that the polyspecific, intrathecal humoral immune response to measles, rubella, and varicella zoster (the so called 'MRZ reaction'), a highly sensitive (~90%) and possibly specific marker of classical MS, is mostly negative in NMO [39, 40, 63, 67]. Corroborating these findings, we found a negative MRZ reaction in 11 additional patients in the present cohort. It is of note that OCBs were rare and tended to disappear over the course of disease also in seronegative patients; also the MRZ reaction was negative in all five seronegatives tested. This provides further support for the hypothesis that seronegative NMO is not simply a clinical variant of MS.
Infections have been discussed in the past both as a possible trigger of disease relapse in NMO and as a possible aetiological factor [1, 68]. In our cohort, ON and myelitis were reported to have been preceded by acute infection at least once in 29.3% of the seropositive patients and in 17.9% of the seronegative cases. This is in good accordance with a study by Wingerchuk and colleagues who found a frequency of 30% in a North American mixed Caucasian/Asian cohort .
Both well defined co-existing autoimmune disorders as well as serological signs of co-existing autoimmunity were significantly more common in seropositive patients. Besides lupus erythematosus, Sjögren's syndrome, thyroid diseases, myasthenia gravis, or celiac disease, which have been all repeatedly observed in association with NMO spectrum disorders [54, 69–75], a broad variety of other autoimmune conditions was present in our patients, including pyoderma gangrenosum, psoriasis, sarcoidosis, ankylosing spondylitis, collagenous colitis, atopic dermatitis, rheumatoid arthritis, Sharp syndrome, and vitiligo. The frequency of co-existing autoimmunity found in our seropositive NMO patients is higher than that observed in classical MS .
Four per cent of our patients had a history of cancer, including breast cancer, cervical cancer, rectal cancer, nasopharyngeal cancer, and skin cancer, all of whom were positive for AQP4-Ab. This is of interest because AQP4 was reported to be expressed in a variety of tumour tissues . However, so far it is unknown whether seropositive NMO may occur as a paraneoplastic disorder, or whether the co-occurrence of the two conditions is a simple coincidence. Interestingly, our findings are in exact accordance with a recent North American study that had reported a frequency of cancer in NMO spectrum disorders in 5% . In three additional patients, hysterectomy and/or ovarectomy was performed, though it remained unclear from the patients' records whether a tumour was present.
We recognize that there are some obvious limitations of our study. First, the study design was retrospective, as in all previous studies, and a multitude of neurological centres was involved. However, due to the low prevalence of NMO in Caucasians, prospective monocentre studies including sufficiently large patient numbers are impracticable. Moreover, the multicentre design of this study, which included more than 25 academic centres, reduces the risk of referral bias, which was acknowledged as a possible limitation by the authors of the two largest previous monocentre studies [1, 36]. Second, similar to previous studies, analysis of MRI results was based upon patient records. However, this disadvantage is partly compensated for by the very high number of brain (N = 326) and spinal cord (N = 326) MRI results available for analysis. Third, as in previous studies, most patients were treated at least once with immunomodulatory or immunosuppressive agents. However, given the much more aggressive course of disease in NMO compared to MS, which requires early treatment, a longitudinal study on untreated patients with NMO cannot be performed. Moreover, AQP4-Ab has been previously shown to remain detectable in NMOSD even with strong immunosuppression. In a recent study, we found AQP4-Ab in 95/96 samples obtained from patients treated with a broad variety of immunosuppressive or immunomodulatory drugs such as azathioprine, rituximab, cyclophosphamide, IVIG, interferon beta, glatiramer acetate, steroids, or mitoxantrone . Therefore, we do not believe that treatment effects played an important role in the present study with regard to stratification, though we can, of course, not completely exclude such effects. Fourth, assay sensitivity could be a potential limitation. However, 98.3% of our patients were tested using recombinant assays. Recombinant assays employing human AQP4-Ab have been repeatedly shown in independent studies to be significantly more sensitive compared to indirect immunofluorescence on rodent brain tissue [4, 5, 38, 79], which was used in many of the previous studies. 19/19 (100%) samples from the seronegative group had remained negative at repeat testing (up to three times) or when tested in another assay (up to four methods); from the remaining patients no data on repeat testing and no samples were available due to the retrospective setting of this study. Fifth, patients with a benign long-term course are less likely to be admitted to hospital and might thus be under-represented in the present cohort. However, this type of potential bias is inherent in hospital-based studies and cannot be completely avoided. It is important in this context that all centres involved in the present study also have specialized neuroinflammatory outpatient departments and that patients were recruited among both inpatients and outpatients. Finally, the threshold for admission is low in Germany, where public healthcare is free. Accordingly, 7.4% of our patients had an EDSS of 0-1.5, i.e. no disability, at first admission.
In summary, our study provides a comprehensive overview of the clinical, MRI, and laboratory features of NMO in Caucasians and extends our knowledge of the clinical, diagnostic and prognostic impact of AQP4 antibody positive serostatus in this rare yet often devastating condition.