Retinal Damage and Visual Network Reconfiguration Defines Visual Function Recovery in Optic Neuritis

Abstract

Background and Objectives

Recovery of vision after acute optic neuritis (AON) is critical to improving the quality of life of people with demyelinating diseases. The objective of the study was to prospectively assess the changes in visual acuity, retinal layer thickness, and cortical visual network in patients with AON to identify the predictors of permanent visual disability.

Methods

We studied a prospective cohort of 88 consecutive patients with AON with 6-month follow-up using high and low-contrast (2.5%) visual acuity, color vision, retinal thickness from optical coherence tomography, latencies and amplitudes of multifocal visual evoked potentials, mean deviation of visual fields, and diffusion-based structural (n = 53) and functional (n = 19) brain MRI to analyze the cortical visual network. The primary outcome was 2.5% low-contrast vision, and data were analyzed with mixed-effects and multivariate regression models.

Results

We found that after 6 months, low-contrast vision and quality of vision remained moderately impaired. The thickness of the ganglion cell layer at baseline was a predictor of low-contrast vision 6 months later (ß = 0.49 [CI 0.11–0.88], p = 0.012). The structural cortical visual network at baseline predicted low-contrast vision, the best predictors being the betweenness of the right parahippocampal cortex (ß = −036 [CI −0.66 to 0.06], p = 0.021), the node strength of the right V3 (ß = 1.72 [CI 0.29–3.15], p = 0.02), and the clustering coefficient of the left intraparietal sulcus (ß = 57.8 [CI 12.3–103.4], p = 0.015). The functional cortical visual network at baseline also predicted low-contrast vision, the best predictors being the betweenness of the left ventral occipital cortex (ß = 8.6 [CI: 4.03–13.3], p = 0.009), the node strength of the right intraparietal sulcus (ß = −2.79 [CI: −5.1–0.4], p = 0.03), and the clustering coefficient of the left superior parietal lobule (ß = 501.5 [CI 50.8–952.2], p = 0.03).

Discussion

The assessment of the visual pathway at baseline predicts permanent vision disability after AON, indicating that damage is produced early after disease onset and that it can be used for defining vision impairment and guiding therapy.

Introduction

Acute optic neuritis (AON) of demyelinating origin is commonly observed as a relapse of multiple sclerosis (MS), MOG-associated disease (MOGAD), AQP-4 neuromyelitis optica spectrum disorder (AQP4-NMOSD), or idiopathic.^1-5 AON is commonly seen at the clinic, with a mean annual prevalence of 8 cases per 100,000 person-year and AON mean annual incidence rate of 5 cases per 100,000 person-years (approximately 130,000 patients yearly).⁶ In AON, the autoimmune process destroys the myelin and sever axons, inducing retrograde degeneration of the retinal ganglion cells, with subsequent inner retina atrophy, as well as anterograde functional impairment to the posterior visual pathway, leading to visual impairment.⁷ The thinning of the ganglion cell plus inner plexiform layer (GCIPL) or peripapillary retinal nerve fiber layer (pRNFL) takes place in the first 3 to 4 months, but mainly in the first 4 weeks after onset and remains almost stable since that.^8,9 After inflammation resolution, remyelination may occur approximately 1 month after onset, but pathologic and molecular abnormalities remain, contributing to permanent visual disability.^10,11 The cortical visual network changes over time in response to the damage and may contribute to clinical recovery.^12-21 The standard of care for AON is high-dose corticosteroid therapy, and in cases with suboptimal response, plasmapheresis is used.^22,23 However, whether the use of corticosteroid therapy influences the residual disability after AON, based on the results of the Optic Neuritis Treatment Trial, is under discussion.²⁴ For this reason, it has been proposed that early treatment, in the first 7–10 days, should be pursued,^25,26 defining AON as an emergency,²⁷ the “Optic Neuritis Code.”

Recovery from AON is often incomplete and impairs patients' quality of life.^2,3 There is the misperception that recovery is good because of the lack of sensitivity of high-contrast visual acuity in this condition.²⁸ However, when using the most sensitive visual outcomes, such as 2.5% low-contrast vision or the visual quality-of-life tool Visual Functioning Questionnaire-25 (VFQ-25) plus 10 neuro-ophthalmologic items, it is observed that most of the cases suffer moderate vision impairment in the long term.²⁹ Previous studies have identified that visual acuity and color vision loss, retinal atrophy, length of the optic nerve damage on MRI, or latencies of the visual evoked potentials (VEPs) may predict visual recovery after AON,^30-35 but not all studies have found such predictors.³⁶ However, a comprehensive prospective analysis of the anterior and posterior visual pathway damage and its role in predicting long-term visual disability is still lacking.

The objective of this project was to define the course of AON in terms of visual outcomes, changes in retina thickness, and visual network metrics to identify predictors of permanent visual disability. To this aim, we analyzed the AON-Vis prospective cohort from Barcelona that follows patients for 6 months with visual acuity assessment and quality of vision, optical coherence tomography (OCT), VEPs, and advanced brain MRI characteristics.^37,38 The specific aims were (1) to define the visual impairment by the end of follow-up (month 6) and the retinal and cortical visual network changes at structural and functional levels from baseline to month 6; (2) to identify predictors of low-contrast visual acuity (LCVA) as the primary end point, as well as high-contrast visual acuity (HCVA), color vision, and quality of vision by the end of follow-up (month 6) as secondary end points, using retina assessments and cortical visual network as predictors.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

The Institutional Review Board of the Hospital Clinic of Barcelona approved the study, and all participants provided written informed consent. The article follows the EQUATOR guidelines and Advised Protocol for SD-OCT Study Terminology and Elements (APOSTEL) recommendations.³⁹

Subjects

We analyzed the first 88 consecutive patients with AON recruited into the Barcelona AON-Vis prospective cohort at the Hospital Clinic of Barcelona enrolled from February 2011 to November 2022 as described before.^37,38 We included patients with unilateral AON, including idiopathic (AON-idiopathic), AON as MS onset with dissemination in space (AON-CIS), AON as a relapse in relapsing-remitting MS (AON-MS), and AON as a relapse of MOGAD (AON-MOGAD). We excluded bilateral AON and AON in the setting of AQP4-NMOSD (by detecting anti–aquaporin-4 antibody) or systemic immune-mediated disorders because of the most severe damage and the different topographic damage of the retina compared with AON in MS.

Procedures

All subjects underwent weekly OCT assessments during the first month and then monthly for 6 months. Month 6 was defined as the follow-up end point because no further changes are observed in visual disability and retinal thickness after such a time point.³⁸ Moreover, patients conducted brain MRI, perimetry, and multifocal visual evoked potential (mfVEP) evaluations at baseline and month 6 as described before.³⁸ A certified optometrist masked to the statistical plan of the study evaluated monocular 2.5% LCVA using EDTRS Sloan plates, HCVA using EDTRS plates with LogMAR transformation, color vision using the Hardy Rand and Rittler Pseudoisochromatic Plates with the best correction. We evaluated the quality of life related to visual impairment with the VFQ-25 with the 10 neuro-ophthalmologic items as described before.³⁰

Spectral-domain OCT was obtained using a Spectralis device to quantify the thickness of the pRNFL, macular RNFL (mRNFL), GCIPL, and inner nuclear layer (INL) as described before.³⁸ The mfVEPs were recorded using the monocular VisionSearch1 as described before.³⁸

Brain MRI

MRI was performed on 53 patients with a 3T Magnetom Trio scanner (Siemens, Erlangen, Germany) with a 32-channel phased-array head coil. The protocol included structural 3D-magnetization prepared rapid acquisition gradient echo (MPRAGE, TR = 2200 ms, TE = 3 ms, FA = 7°, 1.0-mm isotropic voxels),⁴⁰ 3D-T2 fluid-attenuated inversion recovery (TR = 3,780 ms, TE = 96 ms, FA = 120°, voxel size 0.8 × 0.6 × 3.0 mm, 3.0-mm thick, 0.3-mm gap between slices, 40 axial slices) and diffusion-weighted imaging (DWI) (TR = 14,800 ms; TE = 103 ms; 100 contiguous axial slices; 1.5 mm isotropic voxel size; 154 × 154 matrix size; b value = 1,000 s/mm²; 60 diffusion encoding directions and a single baseline image acquired at 0 s/mm²).

Resting-state functional MRI (rs-fMRI) was acquired in 19 patients using a gradient-echo echo-planar pulse sequence sensitive to blood oxygenation level-dependent contrast (TR = 2,000 ms, TE = 30 ms, FA = 85°, 3.0-mm isotropic voxels, 3.0-mm thick, no gap between slices), acquiring a total of 450 frames over an acquisition time of 15:14 minas described before.⁴¹

Visual Network Parcellation

The visual network framework was derived from T1w images and integrated 50 nodes based on reference 42, which corresponded to 25 distinct topographic areas in each hemisphere. This included 8 ventral-temporal (V1v, V2v, V3v, hV4, VO1, VO2, PHC1, and PCH2), 9 dorsal-lateral (V1d, V2d, V3d, V3A, V3B, LO1, LO2, TO1, and TO2), 7 parietal (IPS0, IPS1, IPS2, IPS3, IPS4, IPS5, and SPL1), and 1 frontal (hFEF) regions (eFigure 1 and eTable 1). After this, the lateral geniculate nucleus (LGN) from both hemispheres was segmented using the FSL-FIRST package, resulting in 52 nodes. The volumetry of these regions was computed after normalizing by the skull size.

Diffusion-Weight Connnectome

The diffusion MRI preprocessing pipeline involved the following steps: DWI denoising, Gibbs ringing correction, motion-induced distortion correction, and phase unwrapping procedure to correct geometric distortion using gradient field maps and bias field correction.⁴³ After implementing the necessary corrections, a fractional anisotropy (FA) scalar map was derived from the diffusion tensor model (DTI) using FSL's DTIFIT. Based on FA-weighted indices, the matrices for structural visual connectivity were formed using outcomes from the advanced probabilistic streamline tractography. To validate the biological plausibility of the reconstructed streamlines, connectome reconstruction incorporated the Anatomical Constrained Tractography framework. From this, a subset of 6M of streamlines was chosen, followed by postprocessing based on anatomical exclusion criteria, as detailed in references 44 and 45. The parceling approach of the visual network, consisting of 52 nodes from the anatomical image, was aligned to the FA map. Finally, the average FA value along each connection between a pair of nodes was used to form the FA-weighted adjacency matrix for the visual network's structural connectome.

Functional MRI Connectome

For the functional connectivity rs-fMRI analysis, brain signal correlation or synchronization was captured based on the Duke resting state pipeline approach.⁴⁶ This method encompassed several preprocessing stages, including slice timing adjustment, motion correction, spatial normalization to a standard template, and band-pass filtering to isolate frequencies between 0.001 and 0.08 Hz, all achieved using FSL tools described elsewhere.⁴⁷ The predetermined parcellation scheme was used to extract the average time series for each of the 52 visual regions, resulting in a functional connectivity matrix.

Topologic Analysis of Brain Connectivity Networks

Both structural (DTI) and functional (resting-state) connectivity matrices were adjusted for age and gender influences using linear regression, reducing potential biases.⁴⁸ After these adjustments, the matrices underwent normalization to ensure uniformity in connectivity measures across all participants. Afterward, using the Brain Connectivity toolbox⁴⁹ to compute graph measures from weighted connectivity data, we analyzed the topologic characteristics of networks in both structural and functional domains. On a global scale, we assessed the following average graph measures: betweenness centrality, which reflects the fraction of all shortest paths passing through a node, highlighting nodes that play a crucial role in information flow. Global efficiency, calculated as average inverse shortest path length, provides insight into overall network efficiency in information transmission. Node strength, representing the sum of weights of links connected to a node, offers insights related to the influence of individual nodes. The clustering coefficient, indicating the proportion of interconnected neighbors of a node, provides valuable perspectives on network tendency to form subnetworks. Finally, assortativity reflects the network propensity to connect nodes of similar degree⁴¹ (see supplementary material eMethods for details on graph analysis). On a more regional level, for each of the 52 nodes defined by the parcellation visual scheme, we computed measures including betweenness centrality, node strength, and clustering coefficient.⁴¹

Statistical Analyses

We described qualitative variables using absolute and relative numbers and quantitative variables using medians and interquartile ranges (IQR = P₂₅-P₇₅). We plotted the absolute thickness change of each macular retinal layer, referred to as baseline value, from the first symptom in each visit up to the 6-month visit. For the pRNFL thickness, we estimated the intereye asymmetry as the baseline value in the ON eye. Group comparison was performed with the t-test. We analyzed the temporal changes of retinal layer thickness using mixed-effects models with 2 knots for the 25^th and 75^th percentiles. We did not impute missing values, assuming missing at random, because of the robustness of mixed-effect models to missing data. Predicting visual acuity by month 6 was conducted using multiple linear regression models with splines using the published cutoffs of 60 and 75 µm of the pRNFL^36,50-52 and the median of the pRNFL and GCIPL from our study. The primary end point (dependent variable) was the 2.5% LCVA by month 6, and the independent variables were GCIPL and pRNFL thickness, nodal volumes and network metrics, latencies and amplitudes of the mfVEP, sex, age, and use of corticosteroids at baseline (the EDSS was not included because it includes visual acuity in the subscore). Statistical analyses were performed using R. We used false discovery rate (FDR) to correct for multiple comparisons.

Results

Visual Disability After AON

Consecutive patients (n = 88) with demyelinating AON, either idiopathic or due to MS or MOGAD, were recruited to this prospective study and followed for up to 6 months. Demographics and clinical characteristics at baseline and by month 6 are summarized in Table 1. At baseline, eyes with AON showed a profound impairment on 2.5% LCVA and mild impairment on HCVA and color vision. By the end of the follow-up (month 6), prominent LCVA disability remained (Sloan 2.5% chart: 19 letters out of 70), whereas HCVA and color vision significantly improved to almost complete recovery. At baseline, eyes showed an intermediate impairment of the visual fields (mean deviation: −11.39), prolonged latencies (173 ms), and decreased amplitudes (121 mV) of the mfVEP, which partially recovered by the end of follow-up (month 6), except for the latencies that remained significantly prolonged (168 ms). Regarding the quality of vision, even if the global quality of vision (VFQ25) returned to normal, it remained impaired for the score combined with the neuro-ophthalmologic items, below the cutoff of 85 (Table 1). In summary, patients with AON showed a moderate-to-severe visual disability at presentation and moderate residual disability (low-contrast vision and quality of vision) at the end of follow-up.

Table 1.

Demographics and Clinic Characteristics of Patients With AON at Baseline

	Patients with AON
n	84
Age (y)	34.8 (29.3–42.7)
Sex (female)	65 (76%)
Duration (d)	9.0 (6.0–15.0)
Etiology	Idiopathic: 20 (24%); MS: 26 (31%); CIS: 32 (38%); MOGAD: 6 (7%)
Affected eye (right)	50 (59%)
Corticosteroid therapy	IV high dose: 47 (55%); oral high dose: 19 (22%); oral low dose: 3 (4%); none: 16 (19%)
Disease-modifying therapies baselinea	16 (24%)
Visual function affected eye	Baseline	Month 6
HCVA (EDTRS, logMar)	0.34 (0.08 to 0.86)	0 (−0.06 to 0.07)
2.5% LCVA (Sloan, # letters)	0 (0 to 0)	19 (3.5 to 28.5)
Color vision (HRR, # letters)	21 (0 to 31.25)	36 (32 to 36)
Visual fields (mean deviation)	−11.39 (−21.35 to −5.08)	−4.82 (−4.82 to −1.82)
VEP latencies (msec)	173 (165 to 177)	168 (159 to 178)
VEP amplitudes (mV)	121 (61 to 149)	136 (102 to 172)
EDSS	2.0 (1.5 to 3.0)	1.0 (0 to 1.5)
VFQ25 global	—	92.3 (85.5 to 96)
VFQ25 + 10 items neuro-ophthalmology	—	75.1 (63.1 to 84.2)

Abbreviations: EDSS = Expanded Disability Status Scale; HCVA = high-contrast visual acuity using EDTRS charts; HRR = Hardy Rand and Rittler Pseudoisochromatic Plates; LCVA = low-contrast visual acuity using 2.5% contrast Sloan charts; VEP = visual evoked potential; VFQ25 = Visual Function Questionnaire.

Data are shown as mean and interquartile range (IQR) or number (percentage). Visual acuity refers to the affected eye.

^aDMDs: teriflunomide: 1 (6%), interferon-beta: 5 (32%), glatiramer acetate: 4 (25%), fingolimod: 1 (6%), dimethyl fumarate: 1 (6%), anti-CD20: 4 (25%).

Role of Corticosteroid Therapy on Vision Recovery

Given the nonrandomized nature of this study, we conducted an exploratory analysis to investigate the impact of corticosteroid treatment on AON resolution (Table 2). We compared the visual acuity (low contrast, high contrast, and color vision) and quality of vision (VFQ-25) between patients treated with high doses of corticosteroids (either IV or orally) and those who did not. Our analysis revealed a significantly higher recovery of the HCVA in high-dose corticosteroid–treated patients (p = 0.043). However, no significant differences were observed for other visual outcomes.

Table 2.

Association Between the Use and Timing of Corticosteroid Therapy With Visual Disability by Month 6

Use corticosteroids	No corticosteroids (n = 16)	High-dose corticosteroids (n = 58)	p Value
HCVA (logMar)	−0.04 (−0.07 to 0)	0 (0.055 to 0.1)	0.043
2.5% LCVA	24 (14.2 to 30.2)	19 (3 to 28)	0.167
Color vision	36 (35.5 to 36)	36 (31.2 to 36)	0.125
VFQ25 global	NA	92.7 (87.4 to 96.3)	—
VFQ25 + 10neuro	NA	75.1 (64.1 to 83.2)	—

Onset corticosteroids	≤10 d (n = 37)	>10 d (n = 21)	p Value
HCVA LogMar	0.01 (−0.06 to 0.1)	0 (−0.03, 0.09)	0.852
2.5% LCVA	19 (4.5 to 28)	12.5 (0, 26.8)	0.511
Color vision	36 (31.2 to 36)	35 (31.5, 36)	0.738
VFQ25 global	92.8 (87.6 to 97.2)	91.3 (87.4 to 95.1)	0.376
VFQ25 + 10neuro	76.4 (60.9 to 83.8)	75 (70.1 to 80.7)	0.734

Abbreviations: EDSS = Expanded Disability Status Scale; HCVA = high-contrast visual acuity using EDTRS charts; HRR = Hardy Rand and Rittler Pseudoisochromatic Plates; LCVA = low-contrast visual acuity using 2.5% contrast Sloan charts; VEP = visual evoked potential; VFQ25 = Visual function questionnaire.

Data are shown as mean and interquartile range (IQR).

In addition, we explored the potential influence of treatment timing by comparing patients who received corticosteroids within 10 days of AON onset (as suggested in previous studies) with those treated later.^24,26 No significant differences in any visual outcomes were found based on treatment initiation time.

Over 80% of patients who initially opted out of treatment had mild or moderate vision loss. This raises the concern that treatment decisions might be influenced by the perceived mild severity of vision loss. In addition, 1 patient with severe vision loss also declined treatment. Therefore, selection bias cannot be excluded in our assessment of corticosteroid efficacy in AON, and further confirmation is warranted.

Changes in Retina Thickness in AON From Onset to 6-Month Follow-Up

The retina of AON eyes showed an early and severe thinning of the inner layers followed by a smaller sustained thinning until the end of follow-up by month 6. The pRNFL, mRNFL, and GCIPL showed a progressive thinning, which was more severe in the first 26 days, moderate from day 26–71, mild thereafter, and showed no significant change after the first ∼160 days after AON (see results from the mixed models below for statistical analysis) (Figure 1, A–C). The INL showed a transitory mild early thickening during the first 3 months (Figure 1D). Mixed-effects models predicted a rate of GCIPL thinning of 0.20 μm per day, with a clear time dependence: from day 1–26 (percentile 0-25^th) β = −0.32 μm/d, 95% CI (−0.38 to −0.25), p value <0.001; from day 26–71 (percentile 25-75^th) β = −0.12 μm/d, 95% CI (−0.15 to −0.1), p value < 0.001; and without changes beyond day 71 (percentile 75-110^th) β = −0.001 μm/d, 95% CI (−0.004 to 0.001), p value = 0.3 (Figure 2A). The pRNFL also showed a clear time dependence: from day 1–26 (percentile 0–25th) β = −0.14 μm/d, 95% CI (−0.45 to 0.18), p value < 0.3; from day 27–71 (percentile 25–75th) β = −0.58 μm/d, 95% CI (−0.69 to −0.48), p value < 0.001; and from day 71 (percentile 75–110th) β = −0.02 μm/d, 95% CI (−0.03 to −0.001), p value = 0.037 (Figure 2B). In summary, the inner retina showed early and moderate atrophy in the first month, followed by a smaller decline that stabilized by the end of the follow-up.

Figure 1. Change in Retina Layer Thicknesses During the 6-Month Follow-Up Period.

Graphs show the longitudinal changes (from baseline to month 6) in each patient on the thickness of the inner retina layers for the affected eye (ON) and the non-affected eye (no ON). The y-axis represents absolute retina thickness. The x axis represents the time (days) from clinical onset. (A) Ganglion cells plus inner plexiform layer (GCIPL); (B) peripapillary retinal nerve fiber layer (pRNFL); (C) macular retinal nerve fiber layer (mRNFL); (D) inner nuclear layer (INL).

Figure 2. Linear Spline Mixed-Effect Models of Retina Layer Thicknesses.

Linear spline mixed-effect models with 2 knots at 26 and 71 days for the GCIPL (A) and pRNFL (B). The black points joined by a dashed line represent the individual trajectories of retinal thickness changes, the thicker curves represent the individual fit of the model, and the dark red line represents the population model.

Predictors of Permanent Visual Disability Based on Retina Atrophy at Onset

To define the assessments at baseline that will predict the permanent visual disability after the AON episode, defined at the month 6 visit, we conducted multiple linear spline regression analyses. The model's dependent variable was 2.5% LCVA as the primary outcome and HCVA and color vision at month 6 as secondary outcomes. The independent variables included retina layer thicknesses, latencies and amplitudes of the mfVEP, sex, age, and use of corticosteroids at baseline. We tested 3 cutoffs for retinal thickness: the median of the layer thickness (71 µm for GCIPL and 108 µm for pRNFL) and 2 previously reported cutoffs for the pRNFL (60 µm and 75 µm).^36,50-52 We found that the best model for predicting permanent LCVA impairment was the GCIPL thickness (ß = 0.49 [CI 0.11–0.88], p = 0.012) (Figure 3). Moreover, we found significant models for color vision using either the GCIPL (ß = 0.38 [CI 0.14–0.62], p = 0.003) or the pRNFL <75 µm (ß = 1.37 [CI 1.66–2.08], p = 0.0003). Regarding HCVA, we found no significant models.

Figure 3. Predictors of Permanent Visual Impairment After Acute Optic Neuritis Based on the Severity of Retina Damage at Presentation.

Graphs show the significant models from linear regression analysis for 2.5% low-contrast visual acuity (LCVA) and color vision (HRR plates) with the ganglion cell inner plexiform layer thickness as the independent variable.

Changes in the Structural Visual Network in AON Predict Visual Disability

The structural visual network⁴² was reconstructed in 53 patients with AON from the DTI sequences (Figure 4A, eTable 2, and eFigure 1). We obtained the edges' weight of the visual network (ventral-temporal, dorsoparietal, and parietal-frontal streams) at baseline and month 6. We computed the following network metrics: betweenness centrality (average), global efficiency, node strength (average), clustering coefficient (average), and assortativity. We found that the global statistical properties of the visual network remained stable from baseline to the end of the follow-up (eTable 3). Similar results were obtained when conducting the analysis in patients with new-onset disease (idiopathic and CIS) compared with patients with long-standing disease (MS) (eTables 4–6), although AON in the context of MS showed a trend for higher clustering coefficient of the structural visual network.

Figure 4. Structural and Functional Visual Network in Patients With AON Predicting Long-Term Visual Function.

(A) The visual network as described by Wang et al.,⁴² depicting the ventral-temporal nodes in blue, dorsoparietal nodes in green, and parietal-frontal nodes in orange; (B) the visual structural network predicting visual acuity; (C) the visual functional network nodes predicting visual acuity. Visual function is defined with 2.5% low-contrast visual acuity (2.5% LCVA), high-contrast vision using LogMAR (HVCA LogMAR), and color vision using the HRR plates 2.5% contrast. The network parameters are (a) betweenness centrality, (b) node strength, and (c) clustering coefficient.

We used the volume of the network regions of interest to conduct correlation and regression analysis because even if voxels are assigned to a given region, they are deformed to reflect the changes in volume for a given patient. We observed a correlation between 2.5% LCVA at the end of follow-up and the volume of the bilateral ventral-temporal and posterior-occipital streams at baseline. By contrast, the HCVA correlated with the left intraparietal sulcus volume, whereas color vision correlated with the left dorsal lateral, posterior-occipital, and right cortical dorsal lateral volume (eTable 7).

To assess whether the structural visual network from the DTI connectivity analysis predicts permanent visual disability at month 6, we conducted a multivariate regression analysis having 2.5% LCVA by month 6 as the dependent variable, and the independent variables were GCIPL and pRNFL thickness, nodal volumes and network metrics, latencies and amplitudes of the mfVEP, sex, age, and use of corticosteroids at baseline. By using the visual network obtained at baseline, we found significant models predicting 2.5% LCVA at month 6 using the betweenness centrality (ß = −0.92, p = 0.023) and clustering coefficient (ß = 320.55, p = 0.003) (eTable 8). Indeed, we found models predicting HCVA with the betweenness centrality (ß = 0.01, p = 0.002), clustering coefficient (ß = −2.05, p = 0.008), and assortativity (ß = −0.5, p = 0.027) (eTable 9).

Furthermore, by using the structural visual network obtained at the same time than the vision outcomes by month 6, we found significant models for 2.5% LCVA using the network clustering coefficient (ß = 166.99, p = 0.03) (eTable 8). Moreover, we found a significant model for HCVA using network assortativity (ß = 0.71, p = 0.036) (eTable 9). No significant models were found for color vision at month 6 (eTable 10).

Several nodal graph metrics of the DTI structural visual network at baseline contributed to predicting the visual disability at month 6 (Figure 4B). The nodes contributing to predicting 2.5% LCVA were the left V2 and V3, left intraparietal sulcus, right visual region V3 and V4, right parahippocampal cortex, left lateral occipital cortex, and left intraparietal sulcus (eTable 11). The nodes contributing to predicting the HCVA were the bilateral parahippocampal cortex, left intraparietal sulcus 1 and 4, right lateral occipital cortex 2, and right intraparietal sulcus 4 and 5 (eTable 12). Finally, the nodes contributing to predicting color vision were bilateral V2, left intraparietal sulcus 1 and 4, right lateral occipital cortex 2, left parahippocampal cortex 1, right lateral occipital cortex 2, and right intraparietal sulcus 5 (eTable 13).

Changes in the Functional Visual Network in AON

The functional visual network at baseline and month 6 was obtained from the resting-state fMRI sequences in a subgroup of 19 AON cases (eTable 14). The network metrics were calculated as described above. We found nonsignificant changes in network metrics from baseline to the end of follow-up (eTable 15) or by comparing new-onset vs chronic MS (eTables 16, 17). Then, we tested the ability of functional visual global and nodal network metrics at baseline for predicting visual disability by the end of follow-up using multivariate regression models (Figure 4C). First, we did not find models based on network metrics predicting visual disability at month 6. Second, the nodes that predicted the 2.5% LCVA were the left ventral occipital cortex 2, right lateral geniculate nucleus, bilateral superior parietal lobule 1, and right intraparietal sulcus 1 (eTable 18). Moreover, the nodes that predicted HCVA were the right lateral geniculate nucleus and right superior parietal lobule 1 (eTable 19), and the nodes that predicted color vision were right intraparietal sulcus 5, bilateral V4, V1, left lateral occipital cortex 2, right V3, right temporal, occipital cortex 1 and 2, left superior parietal lobule 1, and right intraparietal sulcus 5 (eTable 20).

Discussion

By studying a prospective cohort of patients with AON for 6 months, we confirmed that low-contrast vision and quality of vision remain impaired, jeopardizing the patient's quality of life. We identified predictors of low-contrast visual acuity 6 months after onset based on the thinning of the ganglion cell layer at baseline. The structural cortical visual network at baseline also predicted the low-contrast vision being associated with the secondary and tertiary visual cortex, intraparietal sulcus, parahippocampal cortex, and lateral occipital cortex as the main nodes involved. At the visual network level, the best predictors were always the betweenness and clustering coefficient of the network. The cortical nodes from the functional visual network that predicted residual visual disability were the ventral occipital cortex 2, lateral geniculate nucleus, bilateral superior parietal lobule, and intraparietal sulcus.

Previous studies have identified predictors of visual dysfunction. However, most of the studies were either retrospective or cross-sectional; patients were included several weeks after disease onset (>10–14 days) or used small sample sizes (<50 subjects).^30-35 Despite such limitations, it has been consistently reported that the GCIPL and pRNFL can predict visual acuity several months after AON onset. In an earlier analysis of this cohort, we found that the change from baseline to month 1 of follow-up for the GCIPL and pRNFL were predictors of low-contrast vision.³⁰ Regarding VEP, the Danish team elegantly proved in a prospective and well-powered cohort that the amplitudes and latencies correlated with the GCIPL thickness and the amplitudes correlated with the LCVA.³⁴ Indeed, the optic nerve damage length using double-inversion recovery sequences also predicted the visual dysfunction at follow-up.³⁵ It will be highly relevant having available biomarkers of the visual pathway damage in AON; new treatments are being tested for decreasing the visual pathway damage in this disease.^53-56

The cortical visual network has been previously analyzed in patients with AON.^12-20 Backner et al. described a higher degree and lower density of the visual network in AON cases than controls, with reduced information transfer efficiency and modularity in the non-NMO cases compared with controls.¹⁷ The same team reported that in patients who had suffered AON 1 to 28 months before, the functional connectivity within the visual network was higher in the AON group, even after controlling by the lesional level of the optic radiations.¹⁸ In the case of NMOSD, the visual network during the chronic phase of cases with previous AON showed an increase in functional connectivity in the primary and secondary visual networks that correlates with low and high-contrast vision and retina thickness.¹⁹ Similarly, the functional visual network was reduced in NMOSD or recurrent optic neuritis.¹⁵ In this study, we demonstrate that early structural and functional changes in the connectivity of the visual network were associated with the 6-month visual outcomes after AON, pointing out the regional specificity of network modifications and their predictive power. One message from our analysis based on the lack of changes from baseline to month 6 is that damage is already stablished in the first few days or weeks after disease onset, and there is not significant changes at the network level based on the technologies we used. This might suggest that recovery is mainly due to functional restoration of nerve conduction, more than structural changes that will be reflected on DTI and resting-state analysis. Alternatively, the current state of the brain network analysis may lack resolution to capture minute changes on the networks or that the noise of the analysis or lack of power may reduce the ability to capture network changes. Overall, AON alters the structure and function of the cortical visual network, reflecting the anterograde damage of the pathway and attempting to adapt and eventually compensate for such damage.

Our study points to several nodes of the visual network as predictors of vision impairment. We should keep in mind that the damage in AON is done in the anterior pathway, and both functional (e.g., nerve conduction block, desynchronization of afferent inputs to the visual cortex) and structural (e.g., anterograde and trans-synaptic degeneration) damage would spread from focal damage in the optic nerve to the rest of the visual network. Then, visual network changes would imply adaptations to the first (lateral geniculate nucleus), secondary (calcarine cortex V1 and V2), and tertiary (dorsal and ventral visual streams) networks. As expected, one of the most altered node was the lateral geniculate nucleus, confirming the spread of damage and dysfunction from the site of damage to the next neuron in the posterior visual pathway. Indeed, the bilateral primary visual cortex V1 and V2 regions were also identified as predictors of visual impairment, most likely being affected by the trans-synaptic degeneration induced by axonal transection in the optic nerve. Beyond the first and second neuron in the visual pathway, other nodes of the visual network identified included the intraparietal sulcus, parahippocampal cortex, lateral and ventral occipital cortex, and superior parietal lobule. In a recent study by Levin's team in patients with MS, they found that the damage of the visual network was similar at the functional level in patients with and without previous AON, but patients with previous AON showed degree and global efficiency of the anatomical visual network.⁵⁷ Such results support the role of trans-synaptic degeneration damaging the visual network after AON and the ability of the functional visual network to compensate for such damage. In our study, we did not find a preference for functional changes in the dorsolateral stream of the visual network in patients with AON¹⁷ or a preferential damage of the lateral middle occipital gyrus or the inferior peristriate cortex as described in patients with MS,⁵⁸ but a widespread damage of the overall network as observed in patients with MS as well. Because our study did not compare the visual network between AON and controls or other CIS relapses, we could not confirm the presence of increased functional connectivity of the visual network as was described before.¹⁸ Whether the involvement of such high-order visual regions are related with brain plasticity to compensate visual dysfunction or represent dysfunctional areas because of desynchronization of visual afferences will require further studies.

Several limitations apply to our study. The single-center design may limit the generalization of the results. As such, the ongoing ACON trial will provide multicentric prospective data regarding vision outcomes in AON.²⁶ Moreover, we could not assess the influence of ethnicity, which seems to influence the severity of retinal damage,² because our cohort was composed almost only of White patients, reflecting the population our center serves. Indeed, our study included disease-naive cases (idiopathic and CIS) and cases with long-standing MS, which may create additional noise. For this reason, we have conducted a sensitive analysis comparing both populations observing similar trends, although sample sizes were smaller for subgroup analysis. Regarding the role of corticosteroid therapy in the recovery from AON, our analysis was exploratory because only powered, randomized clinical trials could provide class I evidence for its efficacy. Indeed, the fact that milder cases were not treated with corticosteroids may suggest the presence of selection bias. Our study did not include the analysis of the optic nerve damage by MRI using specific sequences such as double inversion recovery³⁵ because we should make a compromise between the goal of our study of studying the structural and functional networks using DWI and fMRI and the scanning time for the patient. Indeed, we have not explicitly analyzed the severity of the optic radiation damage¹⁸ because most of our cases were first episodes of MS without lesions in the optic radiations. The visual network analysis was restrained to the evolution of cases after disease onset, and therefore, no comparison with the healthy network is available. The visual network from healthy controls has been used in the past,¹⁷ but this will introduce the noise using different individuals for different time points. In any case, our goal was mainly to test the ability of the visual network at baseline and end of follow-up to explain the visual disability. Only 53 patients had DWI images and 19 patients had fMRI images, because of timing for such techniques implementation in the cohort or the presence of artifacts or modifications in the sequences over the years that precluded their comparability. We have not assessed functional connectivity as in other fMRI studies of the visual network because by using the previously described visual network,⁴² higher specificity and reproducibility were expected, as well as for decreasing the multiple testing derived noise. For this reason, we have not attempted to validate such previous results.

Vision recovery is incomplete after AON and limits the quality of life, and such incomplete recovery is predicted by damage in the anterior and posterior (cortical) visual pathways. The assessment of the visual pathway at baseline predicts permanent vision disability after AON, indicating that damage is produced early after disease onset and that it can be used for defining vision impairment severity and guiding therapy.

Data Availability Statement

PV and BSD had full access to all the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis. ES conducted the statistical analysis. Anonymized data will be shared by request from any qualified investigator to replicate results provided data transfer follows the EU legislation on the general data protection regulation.

Glossary

AON

acute optic neuritis

AQP4-NMOSD

AQP-4 neuromyelitis optica spectrum disorder

DTI

diffusion tensor model

DWI

diffusion-weighted imaging

FA

fractional anisotropy

GCIPL

ganglion cell plus inner plexiform layer

HCVA

high-contrast visual acuity

INL

inner nuclear layer

LCVA

low-contrast visual acuity

MOGAD

MOG-associated disease

mRNFL

macular RNFL

MS

multiple sclerosis

OCT

optical coherence tomography

pRNFL

peripapillary retinal nerve fiber layer

rs-fMRI

resting-state functional MRI

VEP

visual evoked potentials

VFQ-25

Visual Functioning Questionnaire-25

Appendix. Authors

Name	Location	Contribution
Pablo Villoslada, MD	Department of Neurology, Hospital Del Mar Research Institute	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Elisabeth Solana, PhD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Salut Alba-Arbalat, MSc	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona; Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Eloy Martinez-Heras, PhD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Francesc Vivo, MD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content
Elisabet Lopez-Soley, PhD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content
Alberto Calvi, MD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content
Anna Camos-Carreras, MD	Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Marina Dotti-Boada, MD	Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data
Rafel Alcubierre Bailac, MD, FEBO	Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content
Elena H. Martinez-Lapiscina, MD, PhD, MPH	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design
Yolanda Blanco, MD, PhD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content
Sara Llufriu, MD, PhD	Neurology Service, Laboratory of Advanced Imaging in Neuroimmunological Diseases, Hospital Clinic of Barcelona; Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Bernardo F. Sanchez Dalmau, MD, PhD	Ophthalmology Service, Hospital Clinic of Barcelona	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data

Study Funding

This study was supported by the Instituto de Salud Carlos III, Spain, and Fondo Europeo de Desarrollo Regional (FEDER): PI15/0061, JR16/00006, FI16/00,168 and PI20/01236.

Disclosure

P. Villoslada has received consultancy fees and holds stocks in Bionure Investment, Accure Therapeutics, Attune Neurosciences, QMENTA, CLight, NeuroPrex, Spiral Therapeutics, and Adhera Health, none related to this study. P. Villoslada holds patent rights and has received royalties and consultancy fees from Oculis Holding AG for using OCS-05 (aka BN201) to treat optic neuritis (NCT04762017). S. Llufriu received compensation for consulting services and speaker honoraria from Biogen Idec, Novartis, Janssen, Merck and Bristol-Myers Squibb, and holds grants from the Instituto de Salud Carlos III. Go to Neurology.org/NN for full disclosures.