Fundus autofluorescence (FAF) imaging is a valuable tool for imaging eyes with retinal diseases1 and serves as a quantifiable endpoint in clinical trials related to age-related macular degeneration (AMD).2 The autofluorescence method takes advantage of the intrinsic properties of lipofuscin that is normally present in the human retinal pigment epithelium (RPE) cells.3 Lipofuscin consists of lipid-containing granules with fluorescent properties, which accumulate in the RPE as a metabolic by-product of the visual cycle,4 more specifically, from the ingestion of photoreceptor outer segments. Thus, FAF measured noninvasively is dominated by the autofluorescence of lipofuscin.5
Lipofuscin accumulates as the RPE ages, however these deposits are also found irregularly distributed in patients with retinal disease, such as AMD. One of the earliest detectable disease markers in AMD is an abnormal pattern of FAF. These imaging changes are attributed to differences in the amount or distribution of bisretinoid fluorophores in the RPE lipofuscin. In vitro laboratory observations have suggested that lipofuscin and one of its major components, the fluorophore A2E, may exert toxic effects on the RPE6,7; however, in vivo evidence is against a high lipofuscin accumulation in patients with early or intermediate stages of AMD,8 and the debate is ongoing.9,10
Numerous studies have suggested that FAF imaging could assist in the detection and/or monitoring of, for example, geographic atrophy (GA)11 or choroidal neovascularization12 in AMD. However, it is unclear how much robust quantitative evidence exists to support the accuracy of FAF imaging.13 Quantitative fundus autofluorescence (qAF), introduced in 2011,14 is performed by calibrating the FAF image to an embedded reference of known fluorescence efficiency, making it possible to reproducibly quantify and compare the FAF intensity between patients and across time.
Gliem et al. published the first application of qAF to study non-neovascular AMD.8 Nevertheless, the authors recognized several limitations of their study. For example, measurements were not performed in patients older than 65 years, and patients with late AMD were not investigated either.
The aim of this study was to use qAF to objectively measure FAF intensities and compare them with normal age-matched controls. By examining FAF differences in a quantitative manner, via qAF imaging, the present study may shed light on the role of lipofuscin fluorophores in various AMD phenotypes and to re-examine the relationship of lipofuscin to progressive AMD in vivo.
Patients and Methods
The study protocol adhered to the tenets of human research as presented in the Declaration of Helsinki and was approved by both the Institutional Review Board (IRB) at New York University (NYU) School of Medicine, Department of Ophthalmology and Western IRB (Puyallup, WA) at the Vitreous-Retina-Macula Consultants of New York, a private practice associated with NYU. Data were collected, stored, and managed in compliance with the Health Insurance Portability and Accountability Act of 1996. From October 2016 to October 2017, patients with AMD were prospectively enrolled at the Brooklyn and Manhattan offices of the Vitreous-Retina-Macula Consultants of New York. A multicenter collaboration with NYU and Bellevue Hospital also allowed us to recruit controls. A written informed consent was obtained from each patient before the initiation of any study related activities.
Participants included pseudophakic patients aged 60 to 90 years and diagnosed with non-neovascular AMD in at least one eye, along with normal age-matched controls. The inclusion criteria for patients and controls were the clinical diagnosis of non-neovascular AMD based on the Beckman Initiative for Macular Research (renamed the Stephen J. Ryan Initiative for Macular Research) Classification Committee system.15 When available, eyes underwent additional multimodal imaging, including color fundus photography (Topcon TRC 50IX fundus camera; Topcon Medical Systems, Tokyo, Japan), ultra-widefield fundus photography (Optos 200Tx, Optos, Dunfermline, United Kingdom), conventional FAF using the scanning laser ophthalmoscope short-wave blue autofluorescence (Spectralis HRA + OCT; Heidelberg Engineering, Heidelberg, Germany) and/or green ultra-widefield autofluorescence (Optos 200Tx), fluorescein and/or indocyanine-green angiography (Optos 200Tx and/or Topcon TRC 50IX and/or Spectralis HRA + OCT), and imaging with optical coherence tomography angiography (RTVue XR Avanti; Optovue, Fremont, CA or Plex Elite 9000; Carl Zeiss Meditec, Dublin, CA) to characterize drusen phenotype.16
Exclusion criteria were considered if any of the following findings were present: diabetes mellitus, choroidal neovascularization in both eyes, anterior segment pathology or any other condition that would significantly interfere with ocular imaging, and diagnosis of any concomitant retinopathy other than AMD such as retinal dystrophies, vein occlusions, myopic degeneration, or central serous chorioretinopathy.
Referencing a normative dataset,17,18 we measured qAF levels in three non-neovascular AMD phenotypes: soft drusen (SD)/cuticular drusen (CD), which are located between the RPE and Bruch membrane,16 reticular macular disease (RMD)-subretinal drusenoid deposits (SDDs) internal to the RPE,16 and GA. Participants with normal aging and healthy maculas were also recruited and measured during this study.
According to standard of care, patients had a comprehensive dilated eye exam, which included imaging of the posterior segment of the eye. The patient's medical, family, and social history were reviewed using a comprehensive medical questionnaire. In particular, the following elements were included: age, race, iris color, history of cataract surgery, any history of prolonged sun exposure to the eyes, smoking history (including duration and frequency), personal history of diabetes or kidney disease, personal history of any other medical conditions, and any family history of macular degeneration, any other retinal degeneration, or cardiac disease (Table).
Demographic and Clinical Overview of the Study Population
A standardized protocol, consisting of a near-infrared reflectance, spectral-domain optical coherence tomography (SD-OCT) images and qAF, was performed simultaneously, and in registration for combined analysis, with the Spectralis HRA + OCT, a qAF-ready scanning laser ophthalmoscope. Because the natural lens absorbs the blue autofluorescence excitation light and aging changes of the lens (such as cataract development) can interfere with quantitative measurements, only pseudophakic eyes with known intraocular lens absorbance spectrum and clear posterior capsule were considered in this study. The process of converting a qualitative image into a quantitative image is performed by integrating an internal fluorescence reference in the Spectralis, which is also scanned at the time of imaging. This allows individual images to be normalized for variations in laser power or detector sensitivity between patients. Details of the qAF technique are in Delori et al.14 For research purposes, qAF images were taken on the patient's eligible eye.
For qAF imaging, the camera was positioned centered on the fovea of the patient by using the near-infrared reflectance mode and the internal fixation light. After switching to the qAF mode (488-nm excitation and 500-nm to 680-nm detection), focus and alignment were readjusted according to refraction and to obtain a uniform signal. For optical bleaching of photopigment, the retina was exposed to blue excitation light for at least 25 to 30 seconds. For each participant, 12 qAF images were obtained in sequence, and a mean image was computed using customized software developed by Heidelberg Engineering for recording qAF images (30° field of view and 768 pixels × 768 pixels). Up to three of the 12 images could be excluded to obtain the clearest composite possible.
After the recording of each series, the quality of the acquired images was thoroughly evaluated for image artifacts. Exclusion criteria mainly included insufficient optical bleaching or dark images, shadows, or unstable fixation. The qAF values were calculated by measuring the mean gray levels within eight concentric segments (qAF-8) for the pericentral mid-ring of the Delori pattern (Figure 1).
The quantitative fundus autofluorescence (qAF)-8 pattern. The qAF-8 is the mean of the qAF values within the eight numbered zones. This overlay tool should be in the center of the fovea and the border on the temporal edge of the optic disc.
Retina vessels and GA itself (in the patients with GA) were excluded from analysis. The qAF-gray values were automatically exported to a spreadsheet analysis program (Microsoft Excel 2016 for Windows, version 15.14; Microsoft Corporation, Redmond, WA) Finally, the qAF-8 values were statistically compared with those control individuals without eye disease, for each AMD phenotype.
Statistical analysis was performed using IBM SPSS software for Windows version 20.0.0 (IBM Corporation, Armonk, NY).
An independent samples Welch t test was run to determine whether there were differences in qAF-8 between AMD and control participants. There were no outliers in the data, as assessed by inspection of a box plot. The qAF-8 for each group was normally distributed, as assessed by the Shapiro-Wilk test (P > .05). The assumption of homogeneity of variances was violated, as assessed by Levene's test for equality of variances (P = .035).
The Mann-Whitney U test and Kruskal-Wallis test were used to compare continuous variables between and/or among groups.
A Kruskal-Wallis H test was run to determine whether there were differences in qAF-8 between three groups of participants: controls, those with SD/CD or drusen phenotypes, and those with a predominantly SDD phenotype, which we considered as RMD. Distributions of qAF-8 were not similar for all groups, as assessed by visual inspection of a box plot.
Subsequently, pairwise comparisons were performed using Dunn's procedure with a Bonferroni correction for multiple comparisons.
In addition to the pairwise comparison, a separate comparison of qAF-8 in SDD and drusen phenotypes was performed. Because of the presence of outliers in the data as assessed by inspection of a box plot, a Mann-Whitney U test was run to determine whether there were differences in qAF-8 between drusen and SDD phenotypes. Distributions of qAF-8 were similar, as assessed by visual inspection. A P value of .05 or less was accepted as significant.
The patients included in this study had a mean (standard deviation [SD]) age of 83.1 (5.39) years, and 19.35% of patients were men. Clinical and demographic characteristics are summarized in the Table.
Cigarette smoking, hypertension, hypercholesterolemia were some modifiable risk factors for AMD, which we found in our cohort. Controls had a mean (SD) age of 74.4 (7.97) years, and 27.77% of participants were men.
Of the 38 AMD patients recruited, 7 were excluded from analysis because of image defects. The mean qAF-8 was higher in control participants (132.2 ± 42.8) than in AMD patients (75.9 ± 23.3, P < .003), a statistically significant difference (Figure 2). We also observed that qAF levels were significantly lower in patients with SDD/RMD (95.5 ± 47.2) than in participants without AMD (131.2 ± 38.7, P = .0162) (Figure 3). Patients with SD/CD had intermediate qAF values (97.2 ± 35.9 au), but were not statistically significantly different from either healthy eyes (131.2 au ± 38.7 au, P = .0684) or SDD/RMD eyes at this sample size (Figure 4).
Quantitative fundus autofluorescence (qAF)-8 for control versus patients with age-related macular degeneration (without geographic atrophy). Lower qAF-8 values were significantly different (P = .0026).
Quantitative fundus autofluorescence (qAF)-8 for control versus patients with reticular macular disease / subretinal drusenoid deposits (without geographic atrophy). Lower qAF-8 values were significantly different (P = .0162).
Quantitative fundus autofluorescence (qAF)-8 for control versus patients with soft drusen/cuticular drusen (without geographic atrophy). Lower qAF-8 values were not significantly different (P = .0684)
A further analysis revealed statistically significant differences in mean qAF-8 between the SDD eyes (95.5 au ± 47.2 au) and control eyes (131.2 au ± 38.7 au) for a P value of .0279, but not between the SD/CD eyes (97.2 au ± 35.9 au) and control eyes, nor between the SD/CD eyes and the SDD/RMD eyes (Figure 5).
Quantitative fundus autofluorescence (qAF)-8 for control versus non-neovascular age-related macular degeneration phenotypes (without geographic atrophy). A statistically significant difference was only found for comparison between control versus patients with reticular macular disease/subretinal drusenoid deposits (P = .0279).
The lowest qAF-8 au values were in the patients with GA, especially those associated with RMD/SDD. The mean qAF-8 in nine participants was 80 au ± 48.6 au, with a significant statistical difference (P = .0112).
qAF in Pseudophakic Patients
In 2015, Armenti et al. studied 31 eyes in 27 healthy pseudophakic patients. They concluded that qAF displays a linear overall increase until age 75 to 80 years, as previously suggested,19,20 but then appears to decline thereafter. It was hypothesized that decreasing qAF after age 75 may be due to changes in lipofuscin turnover. We found comparable results in both pseudophakic eyes with AMD and normal aging femal controls (Figure 6). Our study participants also demonstrated an upward trend of qAF-8 up to 75 years of age, but there was also a decline in qAF-8 with decreasing RPE health and increasing severity of non-neovascular AMD stage (soft drusen to RMD to GA) in this sample.
Quantitative fundus autofluorescence (qAF)-8 for control versus patients with age-related macular degeneration (AMD). A downward trend is also noticeable after the age of 80 in patients with AMD (red arrow).
Gliem et al. did not report the lens status of their study patients,8 so it might have changed their results. It is also unknown how much qAF values might vary after a cataract extraction in individual patients. We propose to study this further.
qAF in Different AMD Phenotypes
Reports of quantitative measurements from FAF imaging are sparse. Frampton et al. identified only eight primary research studies, from 2,240 unique references, which report FAF imaging accuracy, for diagnosis and monitoring disease, quantitatively.13 Furthermore, among three studies for obtaining FAF images in reticular pseudodrusen in AMD,21–23 only one of these studies reported an objective (quantitative) approach for determining how abnormal (hypo or hyper) autofluorescence was defined.23
Although the composition of SD, CD, and SDD may share similarities, they are differentiable by multimodal imaging when the RPE's position and light absorption characteristics are taken into account.16
Both soft drusen and cuticular drusen are subject to variable amount of light attenuation because of their location under the RPE. In contrast, SDDs are internal to the RPE. Moreover, the natural history and sequelae of CD and SD exhibit characteristics of coalescence, resorption, and RPE disturbances that lead us to consider them within the same AMD phenotype for our study.24
The qAF-8 in SD/CD eyes presented lower levels than age-matched control eyes. The FAF patterns of SD themselves ranged from hypo to hyperautofluorescent. Possible explanations for an increased drusen-associated autofluorescence compared with background encompass greater accumulation of fluorophores within the overlying RPE or intrinsic fluorescence of drusen.8
Reduced qAF-8 values also obtained in CD were in line with a previous report using conventional FAF imaging and compatible with a proposed thinning of the overlying RPE.25 Balaratnasingam et al. also described the CD phenotype on FAF, characterized by a hypofluorescent center with a hyperfluorescent margin, without quantification.24 We found variable qAF levels in eyes with CD phenotypes, ranging from near-normal for age (Figure 7) to the lowest levels associated with GA (Figure 8).
An illustrative case of the right eye of an 81-year-old man with cuticular drusen (CD) phenotype as seen on (A) color photography, (B) fluorescein angiography, (C) indocyanine green angiography, and (D) conventional fundus autofluorescence. Large drusen interspersed with CD. (F) A cross-sectional enhanced depth imaging optical coherence tomography (OCT) B-scan enables a combined analysis with quantitative fundus autofluorescence (qAF)-8 values obtained or (G) a region-of-interest (ROI)-qAF in CD cluster. (H) The OCT volume and (I) magnified inset of the area of the line scan were registered to the ROI-qAF to permit precise spatial correlation between the two methods.
A representative example of geographic atrophy (GA) in a cuticular drusen (CD) phenotype in a 77-year-old woman as seen on (A) color photography, and (B) an ultra-widefield fluorescein angiography with similar patterns of peripheral drusen and macular CD. (C) Conventional fundus autofluorescence did not permit to quantify low quantitative fundus autofluorescence (qAF) levels in (D) segment of Delori pattern or in (E) a region-of-interest. (F) Near-infrared reflectance in combination with (G) an enhanced depth imaging optical coherence tomography B-scan and (H) magnified inset of the area of the line scan showed greater qAF values in GA lobules associated with focal retinal pigment epithelium preservation.
The lowest qAF-8 levels were obtained for the RMD/SDD phenotype, possibly because of masking of autofluorescence by SDD above the RPE. It has also been suggested for RMD that its associated choroidal insufficiency and presence of SDD might be linked with a slowing of the visual cycle,26 reduced rod and cone densities due to outer retinal atrophy,27 or a different composition of lipofuscin, resulting in reduced lipofuscin accumulation and/or lower qAF measures.8
One of the strengths of this study was our cohort of participants older than 60 years of age, in comparison to the younger cohort studied by Gliem et al.8 This feature allowed us to use qAF in patients most likely to have late stages in AMD. For instance, macular complications of RPE-related changes, such as GA, often appear more frequently in patients with AMD who are older than 60 years.
We also evaluated thoroughly the non-neovascular AMD stage using state-of-the-art multimodal imaging techniques for precise drusen phenotyping, especially late non-neovascular AMD patients who have more progressed fundus changes requiring different examination strategies to acquire reliable data. For instance, the use of optical coherence tomography angiography can reveal whether choroid capillaries are present, allowing clear differentiation between GA and atrophy secondary to another disease or for the detection of subclinical choroidal neovascularization.28
Finally, we only studied pseudophakic patients with clear posterior capsule to avoid the increasing variability of lens opacities with age, which makes the resulting qAF values more inaccurate. Likewise, a small capsulotomy in an opacified capsule might be visually adequate for the patient, but not for the qAF method.
Also, the study had several limitations. The sample size was small partly because of our strict inclusion criteria aiming to include only patients with high image quality and certain distinct phenotypes. There were also periods of time at which the internal reference needed recalibration in both facilities. The goal was to keep the measurements as consistent as possible, but small errors may have accumulated between calibrations. Some patients declined to be assessed by this investigational device, possibly biasing the results.
The qAF imaging technique itself has intrinsic biologic limitations to its capacity for directly measuring lipofuscin concentration in the RPE. Differential distribution of lipofuscin within RPE cells29 or composition of lipofuscin fluorophores30 could affect intensity of autofluorescence; therefore, these phenomena need to be addressed in future studies. However, the FAF spectra of macular fluorophores also have high potential for noninvasive dissection into spatially and molecularly precise identification through hyperspectral imaging.31,32 Finally, the qAF measures may be reduced by absorbers of the excitation or emitted fluorescence light such as melanin granules30 or subretinal drusenoid deposits.33
In summary, our study complemented that of Gliem et al.,8with similar findings on an older cohort. We concluded, for our small study groups, that qAF-8 is higher in healthy eyes than in eyes with AMD, suggesting that the source of autofluorescence, lipofuscin, may not be as harmful as it was thought. This is consistent with alternate theories that lipofuscin sequesters oxidative products to prevent damage, or that it is a neutral bystander.34 Thus, qAF may be a useful positive outcome variable for treatment trials of AMD. Prospective studies with larger groups of patients are indicated to validate these conclusions.
It would be useful to evaluate qAF in the junctional zone35 and/or lobules of GA36 to follow recommendations from the Classification of Atrophy Consensus Meetings37 with regards to qAF levels outside atrophic areas38 and within the grayish rather than black lobules of atrophy in some GA subtypes.39 We will present our findings using qAF in GA due to RMD.40 Since the Spectralis combines qAF and SD-OCT in precise registration, enabling a spatiotemporal tableau of RPE health and retinal structure in combination, the qAF method provides a useful tool to study this non-neovascular late AMD stage further.