Retinal Optical Coherence Tomography in Neuromyelitis Optica

Discussion

Our study specifies a severe and functionally relevant decrease of GCIP layer and pRNFL in NMOSD-ON compared with NMOSD-NON and HCs in AQP4-IgG–seropositive patients. Neuroaxonal damage is particularly large from the first episode of ON, where contribution to retinal damage in subsequent episodes of ON is still considerable but smaller. In contrast to previous smaller studies, the current study ascertains GCIP layer but not pRNFL thinning in NMOSD-NON compared with HC.^6,19 This effect was driven by eyes with contralateral ON and not statistically significant in eyes without a history of ON. INL was thicker in NMOSD-ON and was inversely correlated with GCIP layer. Of note, 13.1% of NMOSD-ON eyes showed MME indicative of secondary inflammatory changes.^20,21 This investigation overcame limits of small samples and OCT data heterogeneity of earlier studies through use of an international consortium of NMOSD specialists. This framework substantiated the CROCTINO study—a large, multicenter, retrospective evaluation of retinal pathology in NMOSD using OCT.

The neuroaxonal degeneration in NMOSD-ON demonstrated here is substantially greater than reported changes in multiple sclerosis (MS), a common differential diagnosis.^2,22 In a meta-analysis, an average pRNFL loss of 20 μm was estimated in MS after ON, which is nearly 2-fold higher in our NMOSD-ON cohort (−38.4 μm).²² For GCIP layer, our data suggested approximately 1.5-fold higher loss in NMOSD (−24.0 μm) compared with MS.²² These differences not only result from a higher ON frequency, but may also be caused by more severe retinal damage in NMOSD after a singular ON.²³ It is intriguing that the damage is particularly large after the first ON episode with smaller losses after subsequent episodes, which might be due to less neuroaxonal content in subsequent episodes.^2,24 Alternatively, although the analysis of treatment effects exceeds the scope of this study, it is possible that the longer time to effective anti-inflammatory therapy and the typical choice of less effective therapies (e.g., steroids instead of plasma exchange) at the first attack compared with following attacks may significantly contribute to this difference. Independent of the number of ONs, the percent loss is smaller in GCIP layer compared with pRNFL–pointing toward either (1) stronger loss of retinal nerve fibers than retinal ganglion cells, (2) impairment of ganglion cells not leading to extinction but axonal loss; (3) a larger amount of non-neuronal tissue in GCIP layer, or (4) RNFL loss in the periphery beyond the macular area measured by GCIP layer. These hypotheses are not mutually exclusive, and each might contribute to the effect. Consistent with these concepts, excessive vision loss relative to neuroaxonal content and VEP latency in AQP4-IgG–seropositive patients compared with MOG-IgG–seropositive and MS patients implies damage of the peripheral retina and optic nerve tissues, which are not reflected in the macula and pattern VEP measurements.⁴

Whether attack-independent neurodegeneration in AQP4-IgG–seropositive NMOSD occurs has been controversial.^{2,6,7,25,-,28} The current study identifies decreased GCIP layer in NMOSD-NON compared with HC, but not pRNFL. Such subclinical changes could be caused by contralateral involvement after unilateral ON.^17,18 Indeed, our cohort suggests pronounced neurodegeneration in eyes with contralateral ON. However, eyes of patients without ON also exhibit a trend for thinner GCIP layer compared with HCs. Underscored by longitudinal studies showing ON-independent neurodegeneration and VEP latency prolongation,^7,29 such patterns suggest at least 2 mechanisms of injury: (1) a primary retinopathy or optic neuropathy in context of an astrocytopathy or caused by direct damage to AQP4-expressing cells such as astrocytes and Müller cells by either AQP4-IgG or AQP4-specific T cells or (2) a global or afferent visual system specific chronic or episodic neurodegenerative process. Because lesions often spare the brain, and most studies outside of predisposed areas such as optic nerve and spinal cord failed to detect effects,³⁰ a constant global involvement seems unlikely. Several studies described changes affecting AQP4 expression and astrocytic end feet,³¹ parafoveal changes in agreement with an involvement of AQP4-expressing Müller cell,^6,26,27,32 and attack-independent spinal cord atrophy.³³ These findings are consistent with tissue infiltration by AQP4-IgG–specific T cells³⁴ and the attack-independent loss of retinal ganglion cells⁷—further supporting the existence of an ON-independent pathology, which might be restricted to the main disease foci. The latter hypotheses could be addressed by region-specific pathology or advanced imaging studies.

INL changes have been suggested as a marker of neuroinflammation and potential treatment response in MS.^21,35,36 MME may develop as a consequence of neurodegeneration or other—non–disease-specific—processes.³⁵ Patients with NMOSD were described to have INL thickening and MME.^20,37 In our cohort, 13.1% of NMOSD-ON eyes were affected by MME, which is higher than the 2%–5% described in MS but comparable to incidences in NMOSD described before by Gelfand and colleagues.^35,37 However, INL thickening itself remained comparable to changes reported in MS.³⁵ This disparity could reflect a disrupted fluid homeostasis due to Müller cell involvement or loss of content of the INL with a parallel inflammatory reaction and development of MME. The limited accessibility of MMEs, especially in severely affected eyes due to limited image quality, has hindered their detailed assessments and most likely leads to underestimation of their incidence.³⁸ We demonstrate that the INL thickness is inversely correlated with neuroaxonal content and could be a valuable marker of disease severity also in NMOSD.

Our study cohort was representative of patients with AQP4-IgG–seropositive NMOSD with respect to a female predominance and ON history.^11,39 This OCT study included multiple ethnicity backgrounds, although the distribution was shifted toward Caucasian/White patients and other ethnicities were underrepresented (e.g., Hispanic White and Black), limiting the generalizability of results.⁴⁰ Patients from different heritages presented similar findings.

The current study was based on source data instead of a meta-analysis. Using the Guthy-Jackson Charitable Foundation network, this multicentric study was conducted without investigator reimbursement and illustrated how collaboration integrating international perspectives can produce meaningful results. To overcome technical challenges of heterogeneous source data, we developed novel OCT postprocessing techniques allowing us to perform standardized analyses and enabling the uniform analysis of the largest OCT image data set in NMOSD to date.¹⁰ Thus, the strengths of CROCTINO include its established infrastructure, large international network of experts representing multiple ethnicities and geographic regions, and the use of state-of-the-art OCT postprocessing techniques.⁴¹

We recognize limitations of the current investigation: The retrospective and heterogeneous data acquisition might have led to biases and impreciseness beyond the ability of our quality control. We addressed this by excluding uncertain cases. HCs were only included from a limited number of centers. The unbalanced data set limited some analyses, such as the influence of ethnicities or acute and disease-modifying treatments. Case-control matching was impossible, particularly with respect to subclinical progression dependent or independent of ON and to ethnicity. Similarly, comparisons with other NMOSD subtypes or MS were beyond the scope of this study. Longitudinal data, acute ON data, and AQP4-IgG–seronegative and MOG-IgG–seropositive patient data are part of the CROCTINO data set and will be analyzed in the future. Also, OCT data processing was performed by multiple raters potentially introducing interrater variability. MRI data and posterior visual pathway involvement were not investigated in this study. However, the current study achieved an unprecedented worldwide assessment of retinal damage in AQP4-IgG–seropositive NMOSD.

To conclude, AQP4-IgG–seropositive NMOSD is characterized by a severe, functionally relevant retinal neurodegeneration as a consequence of ON. Although the majority of damage occurs during the first episode, there is cumulative loss with each succeeding relapse. The ON-associated damage is not limited to the neuroaxonal content but can also induce—likely inflammation-mediated—INL increase and occurrence of MME. Our data also suggest attack-independent retinal damage in AQP4-IgG–seropositive NMOSD. Our study supports that attack prevention is key in avoiding neuroaxonal damage and vision loss in patients with NMOSD. It further suggests that the first ON episode causes the most damage, where only some patients with then established diagnosis will be on immunosuppressive therapy. This highlights the need for effective therapies that can ameliorate an ongoing attack or regenerate attack-generated damage, which is an unmet clinical need. Last, the study emphasizes the utility of OCT as a sensitive structural metric and its potential for monitoring progression and even treatment response in AQP4-IgG–seropositive NMOSD. The international CROCTINO program provides an unprecedented opportunity to apply OCT in a standardized manner to assess pathophysiology, clinical course, and therapeutic efficacy in NMOSD.