Abstract
Background:
We performed exploratory analyses of retinal thickness data from a clinical trial of the AβPP cleaving enzyme (BACE) inhibitor verubecestat in patients with Alzheimer’s disease (AD).
Objective:
To evaluate: 1) possible retinal thickness changes following BACE inhibition; and 2) possible association between retinal thickness and brain atrophy.
Methods:
Retinal thickness was measured using spectral-domain optical coherence tomography in a 78-week randomized placebo-controlled trial of verubecestat in 1,785 patients with mild-to-moderate AD. Changes from baseline in retinal pigment epithelium, macular grid retinal nerve fiber layer, central subfield retinal thickness, and macular grid volume were evaluated for verubecestat versus placebo. Correlation analyses were performed to investigate the potential association between macular grid retinal nerve fiber layer and central subfield retinal thickness with brain volumetric magnetic resonance imaging (vMRI) data at baseline, as well as correlations for changes from baseline at Week 78 in patients receiving placebo.
Results:
Verubecestat did not significantly alter retinal thickness during the trial compared with placebo. At baseline, mean macular grid retinal nerve fiber layer and central subfield retinal thickness were weakly but significantly correlated (Pearson’s r values≤0.23, p-values < 0.01) with vMRI of several brain regions including whole brain, hippocampus, and thalamus. At Week 78, correlations between retinal thickness and brain vMRI changes from baseline in the placebo group were small and mostly not statistically significant.
Conclusion:
BACE inhibition by verubecestat was not associated with adverse effects on retinal thickness in patients with mild-to-moderate AD. Correlations between retinal thickness and brain volume were observed at baseline.
Trial registration:
Clinicaltrials.gov NCT01739348 (registered December 3, 2012; https://clinicaltrials.gov/ct2/show/NCT01739348).
Keywords
INTRODUCTION
Alzheimer’s disease (AD), the most common form of dementia [1], is characterized by specific histo-pathological features in the brain, including amyloid-β (Aβ) peptide aggregates, intraneuronal neurofibrillary tangles, and neuronal degeneration [2]. The amyloid cascade hypothesis proposes that the progressive deposition of Aβ aggregates into complexes, such as plaques, triggers AD-related tau protein hyperphosphorylation, and subsequently leads to neuronal degeneration [3, 4]. Aβ aggregates are produced when the extracellular domain of amyloid-β protein precursor (AβPP) is cleaved by AβPP cleaving enzyme 1 (BACE1; also known as β-secretase) to create a membrane-bound fragment. This fragment is then cleaved and released by γ-secretase [5]. Depletion of Aβ via BACE1 inhibition has therefore been evaluated as a potential therapeutic strategy for slowing the progression of AD.
Preclinical studies have suggested that BACE inhibition may lead to retinal toxicity. Administration of the BACE1 inhibitor LY2811376 to rats resulted in changes to the retina, including cytoplasmic accumulation of granular autofluorescent material within the retinal epithelium, enlarged epithelial cells, and photoreceptor degeneration [6]. These toxicology findings led to the discontinuation of the concurrent LY2811376 Phase I clinical trial, though no trial participants showed clinically significant observations during safety follow-up examinations [6]. Autofluorescent granules were also observed in cultured human retinal pigment epithelium (RPE) cells exposed to BACE1 inhibitor or BACE1 small interfering RNA [7]. Additionally, BACE1–/– knockout mice were shown to have changes to the retina compared with wild-type mice, including thinning of the inner and outer nuclear retinal layers, shrunken and atrophic retinal ganglion cells, increase in lipofuscin autofluorescence, RPE thinning and atrophy, and overall reduction in retinal thickness [7]. In preclinical toxicology studies with the BACE1 inhibitor verubecestat, increased RPE thickness was observed in rats but not in monkeys. As a result of these preclinical findings, RPE and other retinal thickness measures were assessed using spectral-domain optical coherence tomography (SD-OCT) in the Phase II/III EPOCH trial of verubecestat in patients with mild-to-moderate AD [8]. The primary focus was to see if there was evidence of RPE hypertrophy for verubecestat versus placebo; secondary goals were to assess possible treatment-related changes in either direction across a range of retinal measures. We report on the results of those evaluations here.
Owing to the retina–brain connection, the development of an ophthalmic biomarker for early detection of AD has gained considerable interest among the scientific and medical communities [9]. Thinning of the retinal nerve fiber layer (RNFL) has been reported in several meta-analyses of OCT studies in AD patients [10, 11], including an analysis restricted to studies which used SD-OCT, the second generation of OCT [12]. Two large community-based longitudinal studies in healthy participants also found a relationship between a thinner RNFL and poorer cognition [13] or risk of dementia [14]. A number of small studies have looked at the relation between retinal thickness and measures of brain atrophy. A study in patients with early-onset AD and healthy controls found a correlation between total macular thickness and posterior cortical atrophy (assessed by magnetic resonance imaging, MRI) in both groups, but no difference between groups [15]. In cognitively normal and cognitively impaired individuals participating in a dementia epidemiology study, ganglion cell-inner plexiform layer thinning, but not RNFL thinning, was associated with reduced brain volume in the occipital and temporal lobes [16]. In neurologically normal adults, RNFL thinning was associated with reduced medial temporal volume but not with the volume of other brain regions [17].
As the relationship between retinal degeneration and brain degeneration in AD is not well understood, and because an ophthalmic biomarker for early AD or prodromal symptoms may offer a potential tool for the diagnosis of AD [9], we took advantage of ophthalmic and brain imaging data from the EPOCH study to evaluate whether baseline macular grid RNFL thickness and central subfield retina thickness correlated with baseline measures of brain volume as assessed by volumetric MRI (vMRI); i.e., we sought to examine the possibility that lower retinal thickness at a single time point (baseline) might be associated with lower brain volume at that time point. In the EPOCH study, patients with mild-to-moderate AD showed a decline in measures of brain volume as assessed by vMRI, amounting to a mean reduction from baseline of –3% for whole brain volume and –5% for hippocampal volume at Week 78 in the placebo group [18]. Therefore, we also evaluated whether changes from baseline in macular grid RNFL thickness and central subfield retina thickness in the placebo group over 78 weeks correlated with AD-associated changes in brain atrophy/vMRI in the placebo group. That is, we sought to explore the possibility that reductions in retinal thickness over time might be associated with decreases in brain volume over time. Additionally, we assessed the relationship between these retinal thickness measures and cognitive function. In the EPOCH study, patients with mild-to-moderate AD showed a mean decline in cognition, as assessed by ADAS-Cog11 score, of –35% at Week 78 in the placebo group [8].
METHODS
Study design and patients
The full study methodology has been published previously [8]; an overview of the methodology is provided here, and features relevant to this explora-tory analysis are described below.
Study MK-8931-017 (EPOCH; Clinicaltrials.gov identifier, NCT01739348) was a randomized, double-blind, placebo-controlled, parallel-group, multisite Phase II/III study designed to assess the safety and efficacy of verubecestat for the treatment of mild-to-moderate AD in elderly patients over a period of 78 weeks. The study was conducted from December 2012 through April 2017, with ophthalmology data collected at ophthalmology centers associated with the primary investigative sites. The ophthalmology sites were qualified using an independent reading center to ensure that they met the imaging requirements of the study.
Men and women aged ≥55 to ≤85 years at first visit were eligible to enroll if they met clinical criteria for a diagnosis of probable AD dementia based on NINCDS-ADRDA and DSM-IV criteria [19, 20], and had a Mini-Mental State Examination (MMSE) score (not corrected for education) between 15 and 26 (mild AD, 21–26; moderate AD, 15–20) at screening. Concomitant AD medications were permitted. MRI or computed tomography (CT) was required within 12 months prior to, or at, screening to exclude those with dementia of non-AD etiologies. Patients with the following ophthalmological conditions were excluded from enrollment prior to a protocol amendment made in April 2015 in which ophthalmological screening was dropped: exudative (wet) age-related macular degeneration, active proliferative diabetic retinopathy, history of myopia or hyperopia >8 diopters, pigment dispersion syndrome, pseudo-exfoliation syndrome, pigmentary glaucoma, glaucoma requiring more than two classes of medications, intraocular pressure >21 mmHg (at the screening visit), clinically significant macular edema with diabetic retinopathy or advanced cataract to the degree that did not allow SD-OCT measurement, cataract surgery within 6 weeks prior to screening, nystagmus, other significant retinal diseases causing such significant distortion that baseline measurements would be too greatly abnormal to allow reasonable detection of possible change (e.g., retinal detachment).
The study was conducted in accordance with the principles of Good Clinical Practice and was ap-proved by the appropriate institutional review boards and regulatory agencies; all patients provided written informed consent prior to initiation of the study.
Randomization and treatment
An interactive voice response system was used to randomly assign patients according to a computer-generated assignment schedule. Patients were randomized to receive a once-daily oral dose of either verubecestat (12, 40, or 60 mg) or placebo (1 : 1 : 1 : 1) for the Phase II component of the trial. The 60 mg dose was included to provide safety information on a higher dose than planned for Phase III (maximum dose of 40 mg) and was planned to be dropped from the Phase III component of the trial. Those initially receiving verubecestat 60 mg were transferred to the 40 mg cohort for the remainder of the trial according to the prespecified plan. Patients were stratified by geographic region, severity of AD, and treatment regimen at screening. Verubecestat was administered in tablet form.
Assessments
SD-OCT retinal assessments
Retinal thickness measures were assessed using sponsor-approved SD-OCT systems. The majority of the images (>80%) were acquired by SPECTRALIS® (Heidelberg Engineering, Heidelberg, Germany) and CIRRUS® (Zeiss, Oberkochen, Germany) SD-OCT and the remaining by RTVue® (Optovue, Fremont, CA, USA) and TopCon 3D systems (Topcon Medical Systems, Oakland, NJ, USA). The same platform was used for an individual patient throughout the trial. SD-OCT measurements of central subfield RPE thickness (μM), macular grid RPE thickness (μM), macular grid RNFL thickness (μM), central subfield retinal thickness (all layers, μM), and macular grid volume (all layers, μM3) were performed under standardized conditions according to a manual of ophthalmic procedures. Briefly, SD-OCT images were collected, and quality controlled by Parexel and then analyzed by a team of experts at the University of Wisconsin Fundus Photograph Reading Center using proprietary segmentation and image analysis software [21]. Additional exploratory measures included fundus autofluorescence and fundus photography (data not shown).
Prior to an April 2015 protocol amendment, retinal assessments were conducted at screening, Weeks 13, 26, 52, and 78; and the early discontinuation visit. Following the amendment, patients who already had baseline retinal assessments underwent further retinal assessments at Week 78 only, while new patients enrolled under the amendment did not have any retinal assessments performed. These changes were implemented based on an external Data Monitoring Committee recommendation following their review of unblinded safety data from this trial [22].
Volumetric MRI assessments
Three-dimensional T1-weighted MRI images were acquired on 1.5 T or 3 T scanners and were quality controlled by a central laboratory. The MRI images were segmented using Freesurfer and the longitudinal analysis of each measure was computed using a change analysis algorithm with tensor-based morphometry developed at BioClinica. These techniques produced one measure of volume change calculated from registration of serial vMRI scans for each pair of follow-up time points relative to baseline. The vMRI measures assessed were total hippocampal volume, left hippocampal volume, right hippocampal volume, whole brain volume, ventricular volume, and Mayo cortical thickness index [18].
Visual acuity assessments
Visual acuity testing was performed using either Early Treatment Diabetic Retinopathy Study (ETDRS), Landolt, or Snellen charts under standardized conditions according to a manual of ophthalmic procedures. Sites were instructed that use of a wall mounted ETDRS chart was highly preferred and, in practice, the majority of assessments were performed using this method.
Cognition assessments
Cognition was assessed using the 11-item cognitive subscale of the Alzheimer’s Disease Assessment Scale (ADAS-Cog11), with higher scores indicating worse cognition [23].
Statistical analysis
Descriptive statistics (mean, standard deviation [SD], counts, percentages) are given for patient de-mographics and clinical characteristics by treatment group.
The population for the analysis evaluating differences between verubecestat and placebo in changes from baseline in retinal thickness and brain volume was the primary population described in the previous study report [8] and included all patients who underwent randomization except those randomly assigned to initially receive the 60 mg dose of verubecestat and the first 200 patients enrolled who were part of an initial safety cohort study. Differences between treatments in changes from baseline in retinal thickness measures were assessed using a longitudinal mixed model, with treatment, week, treatment-by-week, baseline, baseline-by-week, region, manufacturer model, and presence of background AD medication as categorical factors, and baseline MMSE score as a continuous covariate, using an unstructured covariance matrix. In addition, a greater than 100% increase from baseline in central subfield or macular grid RPE thickness in either eye was specified as an ophthalmological adverse event of clinical interest; the number and percentage of patients meeting these criteria were calculated.
The population for the analysis evaluating the relationship between macular grid RNFL or central subfield retinal thickness and vMRI or ADAS-Cog11 measures at baseline included all patients who had the relevant retinal thickness assessment and a vMRI/ADAS-Cog11 assessment at baseline. Only SD-OCT images acquired by SPECTRALIS® and CIRRUS® platforms (i.e., the two most commonly used platforms, accounting for >80% of images) were used for this analysis. We also evaluated the relationship between macular grid RNFL or central subfield retinal thickness and brain amyloid load, as assessed by positron emission tomography (PET) standardized uptake value ratio (SUVR), in the small subset of patients who underwent PET imaging [8].
The population for the analysis evaluating the relationship between change from baseline in macular grid RNFL or central subfield retinal thickness measures and vMRI or ADAS-Cog11 measures at Week 78 included those who had the relevant retinal thickness assessment and vMRI/ADAS-Cog11 assessment at baseline and Week 78; only data from the placebo group were included as verubecestat was previously shown to increase the rate of decline in brain vMRI relative to placebo in this trial [8]). Only SD-OCT images acquired by SPECTRALIS® and CIRRUS® platforms (i.e., the two most commonly used platforms, accounting for >80% of images) were used for the analysis. We also evaluated the relationship between change from baseline in macular grid RNFL or central subfield retinal thickness measures and PET amyloid SUVR at Week 78 in the small subset of patients in the placebo group who underwent PET imaging.
The inferential statistics related to the relationship between retinal thickness measures and vMRI, ADAS-Cog11, or PET amyloid measures at a given time point were calculated via linear regression (vMRI or ADAS-Cog11 measure regressed on the retinal measure) for each eye separately.
All analyses were exploratory and performed without adjustment for multiplicity. The analyses were conducted using SAS® v9.3 or SAS® v9.4.
RESULTS
Study population
Patient demographics and baseline clinical characteristics for those with an SD-OCT assessment are shown in Table 1. In brief, patients’ characteristics were distributed evenly across all treatment groups. The majority of patients were women (55%), white (80%), not Hispanic or Latino (86%), and had a mean (SD) age of 72 (8) years. Most patients were APOE4 positive (64%) and were receiving acetylcholinesterase inhibitor monotherapy at screening (56%).
Demographics and baseline clinical characteristics for all patients who received an SD-OCT assessment
AChEI, acetylcholinesterase inhibitor; AD, Alzheimer’s disease; APOE4, apolipoprotein E4; CSF, cerebrospinal fluid; MMSE, Mini-Mental State Examination; n, number of subjects; PET, positron emission tomography; SD, standard deviation. The verubecestat 60/40 mg group refers to those patients initially treated with 60 mg who were switched to 40 mg. a Based on ratio of tau/Aβ42 [26]
The percentages of patients who had a “cannot grade” SD-OCT assessment in at least one eye varied depending on the measure, being notably greater for macular grid RNFL thickness, but were similar across treatment groups. At baseline, the percentages were: central subfield RPE, 12 mg = 2%, 40 mg = 3%, placebo = 2%; macular grid RPE, 12 mg = 12%, 40 mg = 11%, placebo = 11%; central subfield retina, 12 mg = 7%, 40 mg = 8%, placebo = 6%; macular grid RNFL, 12 mg = 32%, 40 mg = 33%, placebo = 32%; macular grid volume, 12 mg = 15.5%, 40 mg = 15.4%, placebo = 15.1%. There did not appear to be a relationship between the patient’s level of cognitive impairment and “cannot grade” status for macular grid RNFL thickness at baseline; mean (SD) ADAS-Cog11 scores were similar among those who had no “cannot grade” assessments (21.4 [7.2], N = 1,040), those with a “cannot grade” assessment in a single eye (21.5 [7.9], N = 239), and those with a “cannot grade” assessment in both eyes (21.0 [7.8], N = 138).
Effect of verubecestat on retinal measures
Changes from baseline in central subfield RPE thickness over time in left and right eye were generally similar across treatment groups at each time point during the trial and showed no consistent evidence of RPE hypertrophy (potential toxicity signal of concern based on preclinical data) over 78 weeks (Table 2). At Week 78, the 95% confidence interval (CI) for the percentage difference between verubecestat versus placebo excluded zero for the left (12 mg and 40 mg groups) and right (12 mg group) eye, suggesting the possibility of increased RPE thickness (by approximately 2%) with verubecestat versus placebo at that time point. There were no patients who met the event of clinical interest criteria of a greater than 100% increase from baseline in central subfield RPE thickness in either eye.
Effects of verubecestat versus placebo on retinal thickness measures over 78 weeks
CFB, change from baseline; RNFL, retinal nerve fiber layer; RPE, retinal pigment epithelium; SD, standard deviation. aModel-based estimates based on a longitudinal mixed model, modeling % -CFB, with treatment, week, treatment-by-week, baseline, baseline-by-week, region, manufacturer model, and presence of background AD medication as categorical factors, and with baseline Mini-Mental State Examination score as a continuous covariate, using an unstructured covariance matrix.
Changes from baseline in macular grid RPE thickness over time in left and right eye were similar across treatment groups at all time points and showed no evidence of RPE hypertrophy over 78 weeks (Table 2). The 95% CI for the percentage difference between verubecestat versus placebo included zero at all time points (Table 2). There were no patients who met the event of clinical interest criteria of a greater than 100% increase from baseline in macular grid RPE thickness in either eye.
The other retinal measures (central subfield retina thickness, macular grid RNFL thickness, macular grid volume) showed some evidence of slight retinal thinning over 78 weeks for verubecestat relative to placebo in the order of –1% to –2%, with the 95% CIs at Week 78 excluding zero for all but one of the comparisons (Table 2). These small differences were not thought to be of clinical significance and no differences between treatments in visual acuity were observed in those patients who had visual acuity assessments. For example, most change from baseline ETDRS scores (logarithm of the minimum angle of resolution) at Week 78 were in the –0.2 to +0.2 range (left eye: 12 mg = 193/227 [85.0% ], 40 mg = 187/225 [83.1%], placebo = 199/230 [86.5%]; right eye: 12 mg = 176/224 [78.6%], 40 mg = 185/223 [83.0%], placebo = 186/230 [80.9%]). No treatment related differences were observed for fundus photography and autofluorescence measures (data not shown).
Correlations of baseline retinal thickness measures with baseline vMRI brain measures or ADAS-Cog11 score
Observed Pearson correlations and model-based slopes for regression for vMRI measures of a range of brain regions on mean macular grid RNFL thickness and mean central subfield retinal thickness for all patients at baseline are given in Table 3. Mean macular grid RNFL thickness was weakly (Pearson’s r values≤0.11) but statistically significantly positively correlated for both left and right eyes with several vMRI measures, including whole brain volume (p = 0.005 and p = 0.006 for left and right eye, respectively), total hippocampal volume (p = 0.002 and p < 0.001), and thalamus (p = 0.036 and p = 0.006). Mean central subfield retinal thickness was weakly (Pearson’s r values≤0.23) but statistically significantly positively correlated for left and right eyes with whole brain volume, total hippocampal volume, thalamus (all p < 0.001), parietal lobe (p = 0.003 and p = 0.001 for left and right eye, respectively), and occipital lobe (p = 0.015 and p = 0.002). We also looked at the subgroups of patients who were APOE4 carriers (a risk factor for AD) versus those who were non-carriers to see if there was evidence for a stronger correlation in the APOE4 carriers; the correlations were similar between carriers and non-carriers (Supplementary Tables 1, 2).
Observed Pearson correlations and model-based slopes for regression of various vMRI measures, ADAS-Cog11 score, and PET amyloid SUVR on mean macular grid RNFL thickness and mean central subfield retinal thickness measures at baseline
ADAS-Cog11, Alzheimer’s Disease Assessment Scale –Cognitive subscale (11-item version); CI, confidence interval; n, the number of pairs contributing to the analyses; PET, positron emission tomography; RNFL, retinal nerve fiber layer; SUVR, standardized uptake value ratio; vMRI, volumetric magnetic resonance imaging
Observed Pearson correlations and model-based slopes for regression of ADAS-Cog11 score on mean macular grid RNFL thickness and mean central subfield retinal thickness for all patients at baseline are given in Table 3. There were no significant correlations between mean macular grid RNFL thickness or mean central subfield retinal thickness at baseline and ADAS-Cog11 score at baseline (Pearson’s r values < –0.1, all p values > 0.39).
Observed Pearson correlations and model-based slopes for regression of PET amyloid SUVR on mean macular grid RNFL thickness and mean central subfield retinal thickness at baseline in the subset of patients who underwent PET imaging are given in Table 3. There were weak positive correlations between retinal thickness measures and amyloid load at baseline (Pearson’s r values of 0.15 to 0.36; the correlation between right eye mean central subfield retinal thickness and amyloid load was statistically significant, p = 0.028).
Correlations of changes from baseline in retinal thickness measures and vMRI brain measures or ADAS-Cog11 score at Week 78 in the placebo group
Observed Pearson correlations and model-based slopes for regression of change from baseline for vMRI measures of a range of brain regions on change from baseline for mean macular grid RNFL thickness and mean central subfield retinal thickness at Week 78 for patients receiving placebo are given in Table 4. There were no significant correlations between vMRI measures and mean macular grid RNFL thickness. Change from baseline in right eye mean central subfield retinal thickness was weakly correlated (Pearson’s r≤0.17) with change from baseline in Mayo cortical thickness index (p = 0.033), parietal lobe (p = 0.008), and thalamus (p = 0.032). No other significant correlations of right or left eye Week 78 retinal thickness measurements with vMRI measurements were observed.
Observed Pearson correlations and model-based slopes for regression of percent change from baseline for various vMRI measures, ADAS-Cog11 score, and PET amyloid SUVR on percent change from baseline for mean macular grid RNFL thickness and mean central subfield retinal thickness measures at Week 78 in placebo subjects
ADAS-Cog11, Alzheimer’s Disease Assessment Scale –Cognitive subscale (11-item version); CI, confidence interval; n, number of pairs contributing to the analyses; PET, positron emission tomography; RNFL, retinal nerve fiber layer; SUVR, standardized uptake value ratio; vMRI, volumetric magnetic resonance imaging
Observed Pearson correlations and model-based slopes for regression of ADAS-Cog11 score on change from baseline for mean macular grid RNFL thickness and mean central subfield retinal thickness at Week 78 for patients receiving placebo are given in Table 4. There were no significant correlations between change from baseline in mean macular grid RNFL thickness or mean central subfield retinal thickness and change from baseline in ADAS-Cog11 score (Pearson’s r values < –0.1, all p values≥0.19).
Observed Pearson correlations and model-based slopes for regression of PET amyloid SUVR on change from baseline for mean macular grid RNFL thickness and mean central subfield retinal thickness at Week 78 in the subset of patients in the placebo group who underwent PET imaging are given in Table 4. There were no significant correlations (all p values≥0.18).
DISCUSSION
Retinal thickness was assessed in the EPOCH study due to previous preclinical findings that suggested retinal damage, including the possibility of RPE hypertrophy, following the use of BACE inhibitors [6, 7]. Our findings indicated that BACE inhibition by verubecestat doses of 12 mg and 40 mg, estimated to reduce amyloid by 60–75% in the cerebrospinal fluid [8], was not associated with adverse effects on SD-OCT measures of RPE thickness or other retinal measures over 78 weeks in patients with mild-to-moderate AD. This is in line with clinical findings for another BACE inhibitor [6] and suggests that, at doses investigated clinically in humans, BACE inhibition does not appear to adversely impact the retina or visual acuity.
The availability of both retinal thickness and brain atrophy (vMRI) measures in EPOCH allowed us to explore the possible relation between the two in AD patients. Results from several meta-analyses suggest that retinal thickness is significantly decreased in patients with AD compared with healthy controls, although the differences generally appear small, thereby potentially limiting its use as a biomarker in a clinical setting [10–12]. Indeed, a cross-sectional study that included SD-OCT, as well as other established biomarkers for AD, found that retinal thickness measures did not discriminate well-characterized cases of AD from control participants [24]. In our exploratory analysis we found statistically significant but weak correlations of baseline SD-OCT measures of mean macular grid RNFL thickness and mean central subfield retinal thickness with baseline vMRI data for whole brain and several brain regions, notably hippocampus and thalamus, in patients with mild-to-moderate AD. The observed correlations support the possibility of an association between retinal thinning and brain atrophy, but suggest that the relationship may not be of sufficient strength to support use of retinal thinning as a potential biomarker for diagnosing or detecting AD. Furthermore, when we looked at changes in retinal thickness and brain volume measures over 78 weeks in the placebo group there was little consistent evidence of a correlation. Thus, while whole brain and hippocampal volume shrank by 3% and 5% respectively over 78 weeks in the placebo group [18], retinal thickness measures changed by <1% in the placebo group over the same period (Table 2). This suggests that ongoing brain atrophy resulting from AD progression is not reflected in retinal changes over the time period evaluated.
We also did not find an association between mean macular grid RNFL thickness or mean central subfield retinal thickness with cognition (ADAS-Cog11 score), either at baseline or with regard to change from baseline at 78 weeks in the placebo group. Previous longitudinal studies that reported a relationship between retinal thinning and cognitive decline had follow-up periods of 3 to 4.5 years, so it is possible that 78 weeks was too short a duration to detect a relationship [13, 14].
In the subset of patients who underwent PET amyloid imaging, there was some evidence for a possible association between mean macular grid RNFL thickness or mean central subfield retinal thickness with brain amyloid load at baseline. The directionality of the correlations was positive whereas, based on some prior reports, a negative correlation might be expected (i.e., higher PET amyloid SUVR scores would be associated with lower retinal thickness values), although a recent study also found a positive association in people with preclinical AD [25]. However, due to our relatively small sample sizes and multiple comparisons, a type 1 error may be more likely.
A strength of our analysis is that the EPOCH study provided a large sample size compared with previous studies of the retina in AD patients, as well as assessing brain vMRI data, and longitudinal data. Patients were free of significant ophthalmological abnormalities at baseline, which should have had the effect of reducing noise to allow detection of any retinal changes. The analysis had a number of limitations. Most notably, the trial did not include a control group of people without AD for comparison. Other factors may also impact the results, such as the use of several different SD-OCT instruments, although the inter-instrument variability was not thought to impact on the analyses given that the same individual was consistently assessed by the same instrument throughout the study. In addition, Instrument was included as a co-variate and visual inspection of separate data for each instrument showed consistency for the results in Table 2. The percentage of patients who had a “cannot grade” assessment in at least one eye was ≤15% for most retinal measures but notably higher for the macular grid RNFL measure (33%). It is possible that the “cannot grade” assessments may have been influenced by dementia severity which could have influenced the results, but we did not find a relationship between “cannot grade” assessments and level of cognitive impairment for the macular grid RNFL measure at baseline. It is also possible that the study of other specific retinal layers or brain regions may produce different results. Our analysis evaluated a limited number of measures of retinal thickness and the study only examined two individual retinal layers (RPE and RNFL) since reliable algorithms for other metrics were not available at the time of study initiation. The exploratory nature of the analyses without adjustment for multiplicity further limits the conclusions that can be drawn. Finally, our conclusions are limited to patients at the mild-moderate stage of AD and findings could differ in those at an earlier or later stage of the disease process.
Footnotes
ACKNOWLEDGMENTS
Funding for this research was provided by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA. Wen Li is thanked for statistical support. Medical writing assistance, under the direction of the authors, was provided by Kirsty Muirhead, PhD, of CMC AFFINITY, McCann Health Medical Communications, and Christopher Lines, PhD, of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA, in accordance with Good Publication Practice (GPP3) guidelines. This assistance was funded by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ, USA. The authors would like to thank the study patients and investigators.
