Cellular-Level Visualization of Retinal Pathology in Multiple Sclerosis With Adaptive Optics

Abstract

Purpose: To apply adaptive optics–optical coherence tomography (AO-OCT) to quantify multiple sclerosis (MS)–induced changes in axonal bundles in the macular nerve fiber layer, ganglion cell somas, and macrophage-like cells at the vitreomacular interface.

Methods: We used AO-OCT imaging in a pilot study of MS participants (n = 10), including those without and with a history of optic neuritis (ON, n = 4), and healthy volunteers (HV, n = 9) to reveal pathologic changes to inner retinal cells and structures affected by MS.

Results: We found that nerve fiber layer axonal bundles had 38% lower volume in MS participants (1.5 × 10⁻³ mm³) compared to HVs (2.4 × 10⁻³ mm³; P < 0.001). Retinal ganglion cell (RGC) density was 51% lower in MS participants (12.3 cells/mm² × 1000) compared to HVs (25.0 cells/mm² × 1000; P < 0.001). Spatial differences across the macula were observed in RGC density. RGC diameter was 15% higher in MS participants (11.7 µm) compared to HVs (10.1 µm; P < 0.001). A nonsignificant trend of higher density of macrophage-like cells in MS eyes was also observed. For all AO-OCT measures, outcomes were worse for MS participants with a history of ON compared to MS participants without a history of ON. AO-OCT measures were associated with key visual and physical disabilities in the MS cohort.

Conclusions: Our findings demonstrate the utility of AO-OCT for highly sensitive and specific detection of neurodegenerative changes in MS. Moreover, the results shed light on the mechanisms that underpin specific neuronal pathology that occurs when MS attacks the retina. The new findings support the further development of AO-based biomarkers for MS.

Multiple sclerosis (MS) is an autoimmune disease characterized by inflammatory demyelination, axon loss, and neurodegeneration in central nervous system (CNS) tissues such as the brain, spinal cord, and optic nerves.¹ MS pathology can also manifest in the retina, because of both the indirect effects of optic neuritis/neuropathy and direct inflammation and neurodegeneration in retinal tissues.² 

The impact of MS on the retina has been extensively studied using optical coherence tomography (OCT), which provides noninvasive high axial resolution cross-sectional imaging of the peripapillary region and macular retinal layers.³ As quantified by OCT, MS leads to thinning of the peripapillary nerve fiber layer (NFL), macular NFL, and macular ganglion cell–inner plexiform layers (GCIPL), with a small increase in the macular inner nuclear layer (INL).⁴ OCT also reveals microcystic macular edema (MME) associated with more aggressive MS phenotypes.⁵ NFL and ganglion cell layer (GCL) thinning are associated with higher relapse rates, contrast-enhancing MRI lesion formation, brain atrophy, and worsening of visual and physical disability.^6–8 The exact cause of retinal layer thinning in MS, however, is unclear. Direct optic nerve injury from either symptomatic optic neuritis (ON) or asymptomatic optic nerve lesions is a significant cause of retinal thinning.⁹ However, imaging studies in MS patients suggest trans-synaptic neurodegeneration caused by pathology in brain optic pathways may also contribute.^10,11 Postmortem histopathology and animal models such as experimental autoimmune encephalomyelitis (EAE) also provide evidence of direct retinal injury due to inflammatory activity from anti-retinal antibodies, complement deposition, and neurotoxic interactions between activated microglia, astrocytes, and neurons.^2,12,13 Without in vivo techniques for visualization of retinal structures at cellular-level resolution it will be difficult to better elucidate this process in people with MS. 

Adaptive optics (AO) augments traditional retinal imaging techniques like OCT to provide fine structural and functional detail of the retina on a cellular scale.^14–17 AO works by correcting ocular aberrations using real-time feedback from a wavefront sensor to a deformable mirror—significantly improving lateral image resolution. Because OCT axial resolution and optical sectioning are already decoupled from optical diffraction and instead are inversely proportional to the source bandwidth, integration of AO into OCT imagers (AO-OCT) can achieve isotropic micron-level resolution (1–3 µm), resulting in visualization of cells in inner retinal layers critical to MS disease pathology. We have previously demonstrated that AO-OCT provides in vivo visualization of the human retinal ganglion cell (RGC) mosaic¹⁸ and inflammatory cells at the vitreomacular interface.¹⁹ Application of our technique to glaucoma revealed that while the GCL thins as glaucoma progresses, RGC density is reduced and individual RGC somas enlarge,¹⁸ confirming that AO-OCT can be used to probe retinal biology in patients with ocular pathology. 

The aim of this study was to evaluate whether AO-OCT can similarly characterize the effect of MS on the inner retina at a cellular level. We sought to quantify MS-induced changes in axonal bundles in the macular NFL, RGC somas, and macrophage-like cells at the vitreomacular interface. This study provides new information on MS disease processes in the retina, which could potentially inform more sensitive disease biomarkers and novel treatment avenues for neurological diseases with retinal manifestations. 

Methods

Approvals, Participants, and Consent

The study protocol was approved by the Institutional Review Boards of the University of Maryland Baltimore and the U. S. Food and Drug Administration after an IRB authorization agreement was executed between the institutes. Participants aged 18 to 65 with a diagnosis of MS per 2017 revised criteria²⁰ were recruited from the University of Maryland Center for Multiple Sclerosis Treatment and Research. Healthy volunteers (HV) were recruited through on-campus advertisements at the University of Maryland, Baltimore (UMB), and HVs were chosen to approximately match the MS cohort by age and sex. Written informed consent was obtained after the potential risks were explained to each participant. Participants were examined between January 2021 and May 2022. 

Screening and Examinations

MS participants underwent a neurological evaluation, which included collection of demographic and clinical data and a neurological examination for the Expanded Disability Status Scale (EDSS) score.²¹ The component tests of the MS Functional Composite (MSFC) were also collected (Timed 25 Foot Walk (T25W), 9-hole Peg Test (9HPT), and Paced Serial Auditory Addition Test (PASAT)). The Symbol Digit Modalities Test (SDMT) and low-contrast letter acuity (LCLA) testing were also administered. 

All participants underwent a comprehensive ophthalmic exam, including screening for ability to maintain gaze fixation, other ocular pathology, or contraindication to pupil dilation, all of which were exclusionary factors. For all participants who passed screening, the ophthalmic examination included documentation of visual acuity (VA), fundus photography, biometry (IOLMaster 700; Carl Zeiss Meditec Inc., Dublin CA, USA), and standard automated perimetry (10-2 protocol, Humphrey Visual Field Analyzer; Carl Zeiss Meditec Inc.). 

OCT Imaging

Standard spectral domain OCT was obtained in both eyes using a Spectralis OCT2 device with Glaucoma Module Premium Edition software (Heidelberg Engineering GmbH, Heidelberg Germany). See Supplement for more information on the scan protocol used. Images underwent automated retinal layer segmentation using vendor provided software (Heidelberg Eye Explorer version 2.5.1, Heidelberg Engineering GmbH) followed by manual correction for segmentation errors. Segmentation and image quality was assessed using the OSCAR-IB criteria.²² Thickness measurements were derived from layer segmentation outputs, interpolated to the exact macular locations imaged with AO-OCT. 

AO-OCT Imaging

AO-OCT imaging occurred on a custom-built multi-modal AO (mAO) device. Detailed specifications of this system have been previously published.¹⁴ Briefly, the AO-OCT channel of the system operates at 830 nm (Δλ = 60 nm) and acquires volumes at a rate of 2.33 Hz. Aberration sensing uses a portion of the OCT imaging light as a beacon collected to a custom Shack-Hartmann wavefront sensor and aberration correction is achieved with a 97-actuator deformable mirror (Alpao, Montbonnot-Saint-Martin, France). The AO scanning laser ophthalmoscopy channel of the mAO system was used for initial subject alignment and AO correction assessment. 

The AO imaging session took place no more than one week after the ophthalmic examination. Because of the length of a typical AO-OCT scanning session (approximately two hours per eye), only one eye for each participant was chosen for imaging. This was determined by review of the Spectralis OCT segmentation results and selection of the eye with the thinnest GCL thickness, unless eye fixation stability dictated the more stable eye. For this reason, all visual testing outcomes and other ocular testing are reported in this paper for the eye imaged by AO only. The imaged eye was dilated and cyclopleged with Tropicamide 1%. Given previous work suggesting regional thinning of the GCIPL in the nasal macula of MS patients between 3° to 6° eccentricity,^8,23 we chose six imaging locations in a grid pattern 4° eccentric from the fovea (Fig. 1A). Each imaging location will be referred to using a naming convention providing the degree of eccentricity and directions (T – temporal, N – nasal, S – superior, I – inferior). At each location except the fovea, 300 AO-OCT volumes (Fig. 1B, lateral FOV: 1.5° × 1.5°, 300 × 300 pixels; pixel sampling density: ∼1.5 µm/pixel) were collected with the focus set to the GCL. An additional 30 AO-OCT volumes were collected at the fovea to examine this region only for the presence of hyperreflective structures, which may be associated with MS.²⁴ The volumes were de-warped, registered in three dimensions, averaged, and flattened to the inner limiting membrane (ILM) in preparation for further quantification. Macrophage-like cells, nerve fiber bundles, and ganglion cells were examined in the ILM, NFL, and GCL, respectively (Figs. 1C–K). Individual axial lengths measured with biometry were used to correct the lateral pixel size in all AO image analyses below.²⁵ 

Figure 1.

 

AO-OCT imaging acquisition and analysis methodology. (A) Seven locations in the macula were examined. (B) Isometric view of an AO-OCT volume. (C) Macrophage-like immune cells located at the ILM and their processes are resolved with AO-OCT. (D) Cross-sectional (CS) view of inner retina (linear scaling) illustrating individual nerve bundles. Dashed line indicates en face (EF) section depth. (E) EF view of NFL (logarithmic scaling) illustrates individual nerve bundles. Dashed line indicates CS position. (F) CS view of NFL mask for nerve bundle volume quantification. (G) EF view NFL mask for nerve bundle volume quantification and bundle count (yellow crosses). (H) CS view of inner retina (logarithmic scaling) illustrating GCL layer imaging. Dashed box indicates slab from which the EF projection was collected. (I) EF projection showing RGCs. (J) CS view of GCL with RGC segmentation overlay color-coded to soma diameter. (K) EF projection with RGC segmentation overlay.

NFL Axonal Bundle Analysis

NFL axonal bundle analysis was performed by two independent raters. Averaged AO-OCT volumes were visualized in Medical Imaging Processing and Visualization (MIPAV, version 11.0, https://mipav.cit.nih.gov/), and the NFL was identified in three orthogonal planes. The NFL axonal bundle and volume marking rules for the raters are described in the Supplement. Individual axonal bundles were marked in the en face view using a point voxel of interest (VOI) tool. Bundle counts were converted to density from the total en face area of the imaged volume. Semi-automated region-filling paint tools in MIPAV were used to mask the volume of all NFL axonal bundles on each image (Figs. 1D–G), with manual correction. The voxel count of this mask, converted to real world units (mm³), was used to calculate the total NFL axonal bundle volume and volume per bundle. Reported results are the mean values for the two raters. 

RGC Analysis

The GCL contains predominantly RGCs, as well as a small fraction of displaced amacrine cells (∼3% at 3°T).^26,27 We made no effort to distinguish RGCs from displaced amacrine cells during quantification, and we will use the term RGC throughout to describe all individual cells found in the GCL. RGC somas were segmented using an automated machine learning (ML) algorithm (Figs. 1H–K), which was previously shown to provide accurate and reproducible segmentation, even in the presence of disease.²⁸ Following automated analysis, cell soma identification was manually corrected by review of every averaged AO-OCT volume by a single rater. RGC soma density was calculated for the lateral field size (mm²) (i.e., flattened in depth, excluding pixels occupied by blood vessels). Soma diameter and ellipticity (vertical diameter/horizontal diameter) was extracted from the corrected segmentation masks. 

Macrophage-Like Cell Analysis

We examined the region within ∼10 µm of the ILM for the presence of macrophage-like cells, as previously described.¹⁹ The cells were confirmed by a bright hyperreflective soma and multiple processes (Fig. 1C). Cells were counted by two independent raters by examination of each volume in the en face plane using custom software, with the average count between the two raters at each location reported. Macrophage-like cell identification rules are described in the Supplement. Because of eye motion, blinks, and other factors, process motility could only be evaluated for a subset of cells counted. Process motility was measured by marking the process endpoint in each frame of the time-lapsed video by a single rater. 

Foveal Structures

We examined the foveal region for the presence of hyperreflective puncta (HRP), which are minute hyper-reflective dots that occasionally appeared as a beaded string, and scattering features (SF), which are larger structures that cast shadows on the underlying retinal layers in the previously reported AO-SLO images.²⁴ A structure was determined to exist only when both raters indicated its presence in the volume of interest. 

Statistical Analysis

Demographic and clinical characteristics were compared between groups using t-test or χ² testing, where appropriate. When analyzing data with multiple locations, such as NFL and RGC measures, we used a mixed-effects model with repeated measurements (MMRM) implemented through generalized least squares. This approach accounted for within-subject correlations and heterogeneous variances, while also considering other covariates such as age, gender, and prior history of ON (if ON significantly affects the outcome). We used model selection with Akaike's information criterion to determine the appropriate correlation and variance structures for each analysis.²⁹ Conversely, we used a multivariable generalized linear model³⁰ when outcomes were averaged over locations, such as macrophage-like cell density. To identify group differences in NFL axonal bundle, RGC, and macrophage-like cell measures among cohorts of HVs and MS participants with or without a prior history of ON, we conducted multiple comparisons with P values adjusted using the Holm-Bonferroni method. Moreover, we only considered MS participants with or without ON to detect associations with disability and vision measures, and we evaluated differences between HVs and MS participants in the model analyzing associations with RGC by either mean or interaction models. 

Only data from the eye chosen for AO scanning was used for all statistical analyses involving visual outcomes. For associations evaluated between AO-OCT and visual field (VF) measures (e.g., 10-2 VF total deviation), the VF values were interpolated to the AO-OCT location, accounting for the displacement in neural projections from the photoreceptors (i.e., VF location) to the ganglion cells (i.e., AO imaging location), similar to analysis performed in a previous glaucoma study.¹⁸ Similarly interpolated conventional OCT thickness measures at the same location as each AO-OCT location were used in comparisons between AO-OCT and conventional OCT. All statistical tests were performed using R software, version 4.1.0. Inter-rater variability was assessed by Lin's concordance coefficient (LCC)^31,32 for NFL axonal bundle counts and macrophage-like cell counts and Dice Similarity Coefficient³³ for overlapping NFL volume voxels. 

Results

Participants

Five of 24 (21%) participants failed initial screening for various reasons (excessive eye motion, steroid use, narrow angles, glaucoma suspect). AO data were extracted from all 19 participants enrolled, although several locations had excessive eye motion where GCL soma quantification could not be performed. Three participants had locations rescanned on successive visits to acquire higher quality images. 

Demographic and clinical characteristics of the participants who underwent AO-OCT imaging are shown in the Table. There were 10 participants with MS and 9 HVs. The cohorts were similar in age (42.1 ± 10.4 and 41.0 ± 13.0 years, respectively) and sex. Four (40%) of the MS participants had a prior episode of ON, three of which were in the distant (years) past, and for one participant was 5 months prior. The mean disease duration for the MS cohort was 14.0 ± 10.2 years. All MS participants were on disease modifying therapy. Median EDSS was 2.5 (1-6.5) in the MS cohort. Participants with a prior history of ON had significantly worse full contrast visual acuity in the eye imaged by AO, and the participants with ON failed to identify any letters on LCLA testing in the affected eye. 

Table.

 

Demographics and Clinical Data

 

Qualitative Comparison of AO Cellular-Level Images

Representative images from three participants at the same 4N location are shown in Figure 2. Qualitative comparison of the cross-sectional (CS) and en face (EF) images from the participants reveal considerable differences. In the CS images, we observed a large difference in NFL and GCL thickness between cohorts, while the plexiform layers retained their approximate thickness. In the HVs (top row), the NFL bundles are thick and the GCL soma density is five to six layers deep. Some hyper-reflective cells, thought to be displaced RGCs, are observed in the INL, as has been previously observed.²⁶ In the MS participant without ON (middle row), the NFL bundles are thinning, the GCL soma diameter is larger, the GCL soma density is lower, and the INL contains displaced RGCs similar to HVs. In the MS-ON participant (bottom row), the NFL bundles are nearly nonexistent, the GCL somas are even larger and less dense compared to the MS participant without ON, and the INL contains microcysts, which are visible in both CS and EF views. 

Figure 2.

 

Representative AO-OCT images from MS and HV participants. Representative AO-OCT images from: HV (top), MS-no ON (middle), and MS-ON (bottom) participants at 4N. Left to right: B-scan CS view of inner retina, and EF view of the NFL, GCL, and INL. Nerve fiber bundle thinning, enlarged RGCs, and lower RGC density were observed in the MS participants. INL microcysts were revealed as hyporeflective regions in the MS-ON subject. INL hyper-reflective cells, thought to be displaced RGCs, were observed in both HV and MS participants. The average horizontal RGC diameters for these three cases are 11.0, 12.4, and 16.6 µm. Yellow dashed lines in the CS view denote the axial section location of the EF views.

NFL Bundle Volume but Not Density Is Reduced in MS

There was generally good agreement between the two raters for NFL axonal bundles. LCC was 0.73 (95% confidence interval 0.65, 0.79) for NFL axonal bundle count and 0.96 (0.95, 0.97) for volume. The average Dice Similarity Coefficient for voxel overlap was 0.82 (± 0.12) across all locations. NFL axonal bundle volume was lower in MS participants (1.5 × 10⁻³ mm³ [± 1.1 × 10⁻³]) compared to HVs (2.4 × 10⁻³ mm³ [±1.3 × 10⁻³], P < 0.001). This difference was most pronounced on the nasal side of the fovea (Figs. 3A, 3B, Supplementary Tables S1, S2), where volume was higher at corresponding locations (i.e., 4N4S vs 4T4S) in all participants. Participants with a prior history of ON had particularly profound reductions in NFL volume, with an average NFL axonal bundle volume of 1.2 × 10⁻³ mm³ (±6.5 × 10⁻⁴, P < 0.001 compared to HVs). Although NFL axonal bundle density trended toward smaller values in all locations compared to HVs, this difference was not significant except for temporal locations (Fig. 3C, Supplementary Tables S1, S2). The NFL volume per bundle (Fig. 3D, Supplementary Table S1, S2) was significantly lower in all MS participants (6.08 × 10⁻⁵ mm³ [±4.18 × 10⁻⁵]) compared to HVs (9.21 × 10⁻⁵ mm³ [±5.43 × 10⁻⁵], P < 0.01), with the greatest difference observed in participants with ON (5.11 × 10⁻⁵ mm³ [±2.96 × 10⁻⁵], P < 0.01). 

Figure 3.

 

NFL axonal bundle volume is reduced in MS. (A) NFL axonal bundle volume measurements for the three cohorts and the combined MS group at each location. (B) NFL bundle volume for all groups at all T, all N, and all locations combined. (C) NFL axonal bundle density for all groups at all T, all N, and all locations combined showed less significant differences among the groups. (D) NFL bundle volume per bundle show similar trends as NFL volume. Box and whisker plot shows mean (X), median, minimum, maximum, lower and upper quartile ranges, and individual points. P values were calculated using MMRM and were adjusted for age, gender, ON, and multiple comparisons. NS, not significant; *P < 0.05; **P < 0.01; ***P < 0.001.

RGC Soma Density Is Reduced and Diameter Is Increased in MS

RGC soma density was found to be a highly sensitive cellular biomarker for MS pathology as visualized by AO-OCT (Figs. 4A, 4B, Supplementary Tables S1, S2). In all participants, RGC soma density was higher along the midline (4N and 4T locations) compared to the superior and inferior locations. RGC soma density was lower by approximately half in participants with MS (12.3 ± 9.9 cells/mm² × 1000) compared to HVs (25.0 ± 4.0 cells/mm² × 1000, P < 0.001). This difference was most profound in participants with a history of ON, where mean RGC density was 3.9 ± 3.5 cells/mm² × 1000, P < 0.001 for comparison to both HVs and MS without ON. 

Figure 4.

 

RGC soma density is reduced and diameter is increased in MS. (A) RGC soma density for the three cohorts and the combined MS group at each imaged location. (B) RGC soma density for the three cohorts at all N, all T, and all locations combined. (C) RGC soma diameter for the three cohorts at each imaged location. (D) RGC soma diameter for the three cohorts at all N, all T, and all locations combined. P values were calculated using MMRM and were adjusted for age, gender, ON, and multiple comparisons. NS, not significant; *P < 0.05; **P < 0.01; ***P < 0.001.

In addition to decreased cell density, we observed a significant enlargement of RGC somas in MS at all locations (Figs. 4C, 4D, Supplementary Tables S1, S2). Mean cell diameter was 10.1 ± 0.4 µm in HVs and 11.7 ± 1.5 µm in participants with MS (P < 0.001). The difference was largest in MS participants with a history of ON, in which mean cell diameter was 13.0 ± 1.3 µm (P < 0.001 compared to HVs and to MS without a history of ON). In contrast to soma density, there appear to be no regional differences in cell size for all participants and no discernible spatial pattern to the cell enlargement in MS. We observed no difference in cellular ellipticity for the GCL soma in MS participants compared to those found in HVs. 

Macrophage-Like Cell Density and Activity Increase in MS

Macrophage counting was highly concordant between raters, with an LCC of 0.99 (0.98, 1.00). We observed an increase in the number of macrophage-like cells at the ILM for MS participants compared to HVs (Fig. 5). MS participants had an average density of 11.8 ± 9.3 cells/mm² counted in all six locations compared to 5.7 ± 9.4 cells/mm² for HVs. This difference was not significant, however (P = 0.230). A similar trend as NFL and RGC measures was observed where MS participants with a history of ON had the highest density (18.8 ± 10.1 cells/mm²), compared to 7.0 ± 5.2 cells/mm² in MS participants without a history of ON (P = 0.210). All participants had a higher density on the temporal side, but the difference between the nasal side was highest in MS participants with ON. In seven MS participants and three HVs in whom sufficient macrophage-like cells were seen with adequate visualization of cell processes, cell process motility was also measured (Supplementary Table S7, Videos SV1, SV2, and SV3). Mean process motility was slightly higher in MS participants (22.1 ± 8.7 µm/min) compared to HVs (20.5 ± 5.5 µm/min). The area probed by the processes was also slightly higher for MS participants (359 ± 248 µm²) compared to HVs (256 ± 187 µm²). Given the small number of participants and selectivity of this process, a statistical comparison was not performed. 

Figure 5.

 

Macrophage-like cellular density and dynamics. (A) Three example AO-OCT EF projections (∼7 µm slice thickness) showing macrophage-like cells in the three cohorts. Cells are denoted by yellow arrows. (B) The density of macrophage-like cells (total cells/mm² at all locations) was greater for all MS cohorts. P values were calculated using generalized linear model and were adjusted for age, gender, and multiple comparisons. *P < 0.05. (C) Average EF projection in one MS-ON subject with three macrophage-like cells. (D) In the zoomed region around one cell from the time-lapsed video, five processes were manually marked and tracked. (E) In the same zoomed regions, the probed area for the processes was measured.

Conventional OCT—AO-OCT Associations in MS

The group comparisons for retinal layer thicknesses segmented from conventional Spectralis OCT scans and 10-2 VF total deviation (TD) values obtained during the ophthalmology evaluation are shown in Supplementary Table S3. AO-OCT measures were evaluated for their relationship with these conventional OCT thicknesses and 10-2 VF TD values at the same interpolated locations (Fig. 6) and for grouped (nasal, temporal, and all) regions (Supplementary Table S4). AO-OCT–measured RGC density was most associated with GCL and IPL thicknesses, particularly in the nasal retina. This relationship was similar in MS and HVs. NFL axonal bundle volume measured by AO-OCT was most significantly associated with conventional NFL thickness measurement and with GCL thickness in the nasal retina, with some differences in these relationships seen between MS and HVs. 

Figure 6.

 

Association of AO-OCT metrics and layer thickness from conventional OCT. Correlation between RGC soma density measured from AO-OCT volumes and (A) NFL, (B) GCL, and (C) IPL thicknesses measured from OCT volume segmentation at all locations. Correlation between NFL axonal bundle thickness measured from AO-OCT volumes and (D) NFL, (E) GCL, and (F) IPL thicknesses measured from OCT volume segmentation at all locations (each data point represents an individual location, e.g., 4N). MMRM regressions were adjusted for age, sex, and ON history and accounted for inter- and intra-subject correlations. Coeff, MMRM regression coefficient; SE, standard error; *P < 0.05 for that group and association; † P < 0.05 for difference between coefficient in HV and MS. MS-ON are shown with filled symbols for illustration purpose only (i.e., parameters were analyzed together in a single “MS All” cohort in the model).

Visual Function—AO-OCT Associations in MS

The relationship between conventional OCT and AO-OCT imaging metrics and measurements of visual function in the eye scanned by AO was probed at the grouped regions. Results are found in Figures 7A–D for all locations and in Supplementary Table S5 for the grouped regions. Higher visual functional system scores on EDSS testing (thus worse visual function) were associated with increased RGC diameter (P = 0.029) and NFL axonal bundle density (P < 0.001), but not with any conventional OCT thickness measure. Increased RGC diameter was associated with decreased performance on 100% contrast VA (P = 0.007). Decreased AO-measured RGC density (P = 0.004) and conventional OCT GCL thickness (P = 0.009) were associated with poorer LCLA as measured on 2.5% contrast charts. Decreased RGC and NFL axonal bundle density and increased macrophage-like cell density were all associated with poorer LCLA as measured on 1.25% contrast charts (P = 0.006, P = 0.037, and P = 0.007, respectively). Conventional OCT GCL thickness was also associated with poorer LCLA 1.25% contrast measure (P = 0.029). Decreased GCC thickness, RGC ellipticity, and increased macrophage-like cell density were associated with greater mean deviation on visual field testing (P = 0.04, P < 0.001, and P < 0.001, respectively). 

Figure 7.

 

AO-OCT measures are associated with MS visual and neurological disability. Significant associations were found between AO-OCT measures and visual function including (A) RGC density with 2.5% LCLA, (B) RGC diameter and 100% VA, (C) NFL axonal bundle density with 1.25% LCLA, and (D) macrophage density and 1.25% LCLA. Significant associations were also found between AO-OCT measures and MS disability scales including (E) RGC density and EDSS, (F) RGC density and PASAT, (G) RGC diameter and SDMT, (H) NFL axonal bundle volume and PASAT, and (I) macrophage-like cell density and EDSS. All associations are P < 0.05. Regression coefficient (Coeff) and standard error (SE) values are shown on each panel. MS-ON are shown with filled symbols for illustration purpose only (i.e., parameters were analyzed together in a single “MS All” cohort in the model). Results represent MMRM regression evaluating the relationship between AO-OCT measurements and measurements of visual or physical function. Models included all data from the locations shown, and were adjusted for age, sex, and ON history and accounted for inter- and intra-subject correlations. LCLA, low-contrast letter acuity; VA, visual acuity; EDSS, Expanded Disability Status Scale; PASAT, Paced Auditory Serial Addition Test; SDMT, Symbol Digit Modalities Test.

Neurologic Function—AO-OCT Associations in MS

The relationship between conventional OCT and AO-OCT imaging metrics and scores on neurological disability evaluations were probed at the grouped regions. Results are found in Figures 7E–I for all locations and Supplementary Table S6 for the grouped regions. Decreased RGC density and increased macrophage-like cell density were associated with greater overall disability as measured by EDSS (P = 0.017 and P = 0.032, respectively) and worse performance on hand dexterity testing (9HPT, P = 0.001). Walking speed measured by the T25W test was slower with reduced RGC ellipticity (P = 0.002) and NFL axonal bundle density (P = 0.032). Poorer performance on cognitive tests (PASAT and SDMT) were associated with reduced RGC density and NFL axonal bundle volume, increased RGC diameter, and reduced conventional OCT inner retinal thickness measures. 

Foveal Hyper-Reflective Structures Exist in Both Healthy and MS Eyes

The two raters generally agreed on the presence of HRP and SF in the fovea scans of the participants (Supplementary Fig. S1), with 31 of 38 (82%) locations in agreement. HRP were found in 40% of MS participants and 33% of HVs and SF were found in 50% of MS participants and 44% of HVs. Both HRP and SF were simultaneously present in 10% of MS participants and 0% of HVs. 

Visualization of Microcystic Macular Edema in MS by AO-OCT

Small hyporeflective regions in the INL, consistent with microcysts, were observed in all four (40%) participants with MS with prior ON, and none were seen in other MS participants or HVs. Microcysts were more readily observed on AO-OCT images than on Spectralis OCT images. In those with microcysts visible on both, a greater number of cystic structures were seen with better clarity on AO-OCT images. In many cases, smaller microcysts were observed in the AO-OCT images but not resolved with Spectralis OCT (Supplementary Fig. S2). 

Discussion

This pilot study demonstrates that AO-OCT can provide quantifiable images of the direct impact of MS on individual cellular structures in the retina in addition to the recruitment and activity of inflammatory cells. Images of RGCs revealed a 51% reduction in RGC density in eyes of participants with MS compared to HVs, with a 15% increase in the diameter of the remaining cells. Images of NFL axonal bundles revealed a 38% reduction in volume in participants with MS and images of macrophage-like cells at the ILM revealed increased cell density, especially in those with prior ON. Despite the small, pilot nature of this study, we were able to demonstrate that AO-OCT is sensitive to clinically significant differences in the cellular composition of the retina in MS compared to HVs. The evidence for this was best observed in relationships between retinal cellular changes and measures of visual function (such as LCLA and VF mean/total deviation) and neurologic disability (such as EDSS, PASAT, and SDMT). The ability of cellular measures in the retina by AO-OCT to both discriminate MS from non-MS and provide clinically relevant measures of pathology in such a small cohort suggests that AO-OCT holds strong promise as an exquisitely sensitive means by which to visualize, evaluate, and track the underlying disease processes in people with MS. 

The ability of AO-OCT to directly visualize and quantify neuronal cell structures in living patients clearly holds promise as a surrogate biomarker for use in future research into the mechanisms of neurodegeneration in MS. Although other imaging methods for neurodegenerative changes in MS exist, none is as cellularly specific as AO-OCT. Longitudinal atrophy of brain structures captured by MRI has long been used as a surrogate of MS-related neurodegeneration and slowing the rate of brain atrophy has repeatedly proven beneficial in trials of disease modifying therapies.^34–36 More recently, monitoring retinal layer thicknesses by OCT over time has been integrated into therapeutic trials as an additional metric of neurodegeneration.³⁷ Both techniques, however, are limited in their specificity to the neuronal aspects of neurodegeneration. Many pathologic processes can contribute to the atrophy of macroscopic scale CNS tissue, including alterations to glial cells, vasculature, and noncellular support structures.^35,38 Extracellular water also contributes significant noise to macroscopic scale measurement of neurodegeneration, as evidenced by the effect of inflammatory edema on brain atrophy and retinal layer thicknesses during acute inflammatory activity.^39,40 Direct measurement of cellular integrity by AO-OCT allows assessment of neurodegenerative changes in MS with limited additional measurement noise from secondary effects. 

Other in vivo cell-specific visualization methods of retinal pathology exist, such as AO-SLO and fluorescent labeling techniques combined with confocal SLO. However, AO-SLO does not have the depth-sectioning capabilities of AO-OCT and consequently has been unable to demonstrate resolution of the full mosaic of transparent inner cells like RGCs with sufficient fidelity for density or diameter quantification.⁴¹ The effects of retinal pathology on RGCs can also be directly measured in living patients with the Detection of Apoptosing Retinal Cells (DARC) technique, which uses fluorescently labeled Annexin A5 to highlight RGCs undergoing apoptosis when viewed with confocal SLO.⁴² Although this technology holds promise to measure neurodegeneration in MS, it is limited in its narrow focus on cells currently undergoing apoptosis without the ability to visualize adjacent healthy cells or other three-dimensional cellular changes and requires systemic injection of fluorescent dye into patients. 

The perifoveal radial pattern of reduction in RGC density in this cohort with more severe reductions on the nasal and superior regions of the macula is consistent with the previously described “horseshoe-like” pattern of regions of greatest GCIPL thickness reductions seen in MS eyes.^8,23 Indeed, the difference in RGC density between MS and HVs was smallest at the 4T location. This finding, along with the strong associations found between regionally-specific RGC density reductions and OCT-measured GCL and IPL thinning provide support for the hypothesis that a large portion of the thickness reductions seen on OCT in the GCL and IPL are directly due to RGC loss. The RGC density reduction seen in this study is also consistent with prior histopathologic findings showing significant reductions in the number of RGCs in eyes with and without prior ON. Although the most profound reductions in RGC density were seen in those with prior ON (85% compared to HVs), even MS eyes without prior ON had a 25% reduction in RGC density, suggesting indirect neurodegenerative changes. 

The exact mechanisms by which RGC loss occurs in MS is informed by work conducted in rodent models of MS, particularly acute and chronic forms of EAE, in addition to animal models of traumatic optic nerve injury. This work suggests both indirect neurodegenerative mechanisms, such as retrograde degeneration, and direct neuronal toxicity due to inflammatory changes in the retina.^43–45 In both optic nerve crush injury and optic neuritis in EAE, RGC loss occurs progressively after initial insult to the optic nerve, with induction of apoptotic changes in RGCs confirmed by caspase-1 expression and TUNEL staining.^44,46 Although little to no lymphocytic infiltrations are seen in the retina itself in EAE, other inflammatory changes are seen, such as increased TNF-α, MCP-1, and C3 expression in astrocytes and other glial cells, which triggers microglial activation, monocyte chemoattraction, and possibly direct C3-induced neuronal toxicity mediated by astrocytes.^12,46 Animal models of optic nerve injury, glaucoma, and diabetic retinopathy also suggest that a small percentage of RGCs are injury resistant, particularly α-RGCs and some subtypes of melanopsin-containing intrinsically photosensitive RGCs.^47,48 These injury resistant RGCs appear to undergo significant remodeling after retinal insult, including increasing axonal regrowth, cell soma hypertrophy, and dendritic remodeling. These data suggest that the reduced RGC density and increased RGC diameter noted in MS in this study are the result of apoptosis of injury-sensitive RGCs and hypertrophic remodeling of injury-resistant RGC subtypes. 

Given that the NFL is composed of the axons of RGCs, NFL axonal bundles would be susceptible to the same pathologic processes resulting in RGC loss. Histopathologic evaluation of both the optic nerve and the retinal NFL shows axonal loss in both eyes that have experienced ON and those that have not, with more severe changes in the former.⁴⁴ Our results demonstrate that AO-OCT provides a very sensitive measure of this axonal pathology. The NFL axonal bundles visualized by AO-OCT conform to histopathologic observations that axons in the NFL are grouped together into bundles within glial tunnels formed by Müller cell processes.⁴⁹ Individual bundles are observable by AO-OCT because of the broadened foot endings of individual Müller cells, which have different light reflectance properties than axonal bundles observed both by light microscopy and previous work with AO-OCT.^49,50 In animal models of optic nerve injury, there is delayed drop out of damaged axons and resultant thinning of individual fiber bundles.⁴⁹ This is consistent with what was observed in this study, as NFL axonal bundle volume and volume per bundle were both reduced, but bundle density was not. The significant associations of bundle volume loss with thinning of both the NFL and GCL on OCT in this cohort suggest that the processes of RGC loss and both retrograde and anterograde axonal degeneration contribute towards thickness loss in retinal layers on OCT. 

Axonal pathology measured by bundle volume loss was less clinically significant, however, as it was not associated with any of the visual outcomes measured, whereas RGC loss was associated with LCLA and diameter alterations were associated with visual functional system scores and visual acuity. This finding is evidence of the relative inadequacy of conventional OCT NFL thickness measurements for MS-mediated axonal pathology, further reinforced by our results showing poor association between macular NFL thickness and visual deficits. Further associations between RGC alterations and physical disability measures (EDSS, T25W, 9HPT) and cognitive performance (SDMT, PASAT), the former of which are weaker with conventional OCT thickness measures, confirm that quantification of RGC loss not only allows probing visual structure-function relationships but also provides surrogate metrics for global neurodegenerative changes occurring elsewhere in the CNS. 

The ability of AO-OCT to visualize neurodegenerative changes is also confirmed by our observations regarding INL microcysts. Microcystic macular edema, which occurs in areas of retinal neuronal loss, is thought to represent transsynaptic neurodegeneration and fluid accumulation by impaired Müller cells.⁵¹ Although this can occur in optic neuropathy of various etiologies and is thus not specific to MS, patients with MS who have microcystic macular edema noted on OCT appear to have a more severe MS phenotype.⁵ If future study confirms our findings that INL microcysts can be viewed with more sensitivity by AO-OCT than standard OCT, this technique could contribute to better prognostication of patients with early MS. 

Visualization and quantification of immune cells by AO-OCT provides an opportunity to gain a greater understanding of the interrelationship between chronic inflammation and neurodegeneration in MS. While it is possible that the macrophage-like cells seen at the ILM in this study are systemically derived macrophages, their thin, distinct cytoplasm and highly ramified processes are far more consistent with the morphology of ramified microglia or pyramidal hyalocytes. Similar cells are seen in healthy eyes at lower resolution by both AO-SLO and optimized conventional OCT.^52,53 Histopathology and accompanying immunohistochemistry and electron microscopy of ILMs removed during vitrectomy show that most non-fibroblast cells at the vitreoretinal surface are of glial origin – mostly microglia and a small number of hyalocytes.^54,55 This is supported by OCT and confocal immunofluorescence in rodents, where it was found that at the resting state 82% of cells 5 to 10 µm from the retinal surface are microglia, 9% are hyalocytes, and 9% are perivascular macrophages.⁵⁶ Cell populations at the ILM can be altered in response to retinal inflammation, however. After chemokine ligand 2 (CCL2) administration, the same rodent models show infiltration of the ILM with neutrophils and macrophages. Other data shows migration of activated microglia to sites of injury and conversion of hyalocytes to an immune cell-like phenotype.^57–59 Although the ILM creates a physical barrier between the vitreous and the retinal neuronal layers, it may act as a recruitment site for cells during neuroinflammation. It is also possible that cells at the ILM could respond to chemoattractant cytokines being released by retinal glia, even if they cannot traverse the ILM barrier. Either response would explain the increased number of macrophage-like cells seen in MS participants, particularly those with prior ON. This finding, along with the significant inverse relationship between the number of macrophage-like cells and retinal layer thicknesses and increased process motility (similar to our prior work in glaucoma¹⁹) further supports the hypothesis of migration of immune cells to areas of greatest injury or possibly a direct role of the interaction between glia and neurons in propagation of neurodegeneration in MS.^60,61 

The relative presence of FAZ hyperreflective structures (HRP and SF) in both MS and HV cohorts suggests that they are not exclusively associated with MS. Previous AO studies have identified HRP in eyes with and without neurological and ophthalmic disease.⁶² Hargrave et al.²⁴ found HRP in a large percentage (74%) of MS eyes and found associations between SF and multiple MS parameters (MS duration, EDSS, various treatments). The proportion in our study was lower but likely represents a lower limit because our criteria only counted structures identified by both raters. These results point to foveal hyper-reflective structures as a secondary marker of retinal pathology not specific to MS. 

Although significant and sensitive associations with visual and physical disability in such a small sample size attest to the suitability of AO cellular measures in an investigation of MS, the number of participants is the primary limitation of the current study. Further studies in larger cohorts are needed to confirm our preliminary findings. Additionally, although we demonstrated significant associations between macrophage-like cells at the vitreomacular interface and other aspects of clinically relevant MS pathology, we were unable to directly visualize the activated microglia known to reside within retinal layers.² This may require further optimization of AO-OCT or other visualization methods. Also, our study was focused on cellular changes in the macula. Further investigation of the peripapillary region with AO-OCT is warranted. There were several limitations related to AO methodology itself. The amount of averaging required to resolve RGCs and the number of locations chosen precluded imaging more than one eye per subject in an AO imaging session, which both limited the number of eyes examined, but also forestalled the potentially insightful comparison of ON in the same subject (i.e., ON eye vs. non-ON eye). AO also has a limited field of view, which hindered the ability to finely sample spatial variations across the macula. 

Despite these limitations, cellular-level AO-OCT imaging provides several potential advantages over conventional OCT as a source of MS disease biomarkers, including finer scale and direct delineation of changes to cellular integrity with disease, a richer set of metrics from which to characterize subtle disease initiation and progression, particularly for often slowly progressing or episodic diseases like MS, access to retinal architecture pathways and connectivity, higher sensitivity and specificity, better and more direct correlation with functional measures, more targeted information capture, and access to dynamic activity and functional information to bridge the structure-function correlation. Overall, AO-OCT represents a promising emerging technology with which to investigate MS. 

Acknowledgments

The authors thank Donald Miller (Indiana University School of Optometry) for use of OCT 3-D registration software. We thank Anant Agrawal for OCT system characterization and Achyut Raghavendra for technical contributions and helpful comments. We also thank Kerry Naunton, our research nurse, for coordination of this study. 

Supported by grants from FDA Medical Countermeasures Initiative, Critical Path Initiative, and the Office of the Chief Scientist and internal funds from the University of Maryland Department of Neurology. 

Disclaimer: The mention of commercial products, their sources, or their use in connection with material reported herein is not to be construed as either an actual or implied endorsement of such products by the U.S. Department of Health and Human Services. 

Disclosure: D.X. Hammer, None; K. Kovalick, None; Z. Liu, Indiana University (P); C. Chen, None; O.J. Saeedi, Heidelberg Engineering (F, R), Aerie Pharmaceuticals (R), Topcon Healthcare (C), Broadcast Medical (R). D.M. Harrison, EMD-Serono (F, R), Roche-Genentech (F), Horizon Therapeutics (R), TG Therapeutics (R), American College of Physicians (R), Up To Date, Inc. (R)