Near-Infrared Autofluorescence Imaging of the Fundus: Visualization of Ocular Melanin

Abstract

purpose. To evaluate the origin of the near-infrared autofluorescence (AF) of the fundus detected by scanning laser ophthalmoscopy and compare the distribution of this AF with that of lipofuscin.

methods. AF [787] fundus images (excitation [Exc.] 787 nm; emission [Emi.] >800 nm) were recorded with a confocal scanning laser ophthalmoscope, in 85 normal subjects (ages: 11–77 years) and in 25 patients with AMD and other retinal diseases. Standard AF [488] images (Exc. 488 nm; Emi. >500 nm) were recorded in a subset of the population.

results. The fovea exhibits higher AF[787] than the perifovea in an area ∼8° in diameter, roughly equivalent to the area of higher RPE melanin seen in AF[488] and color images. The ratio of foveal to perifoveal AF[787] decreases with age (P < 0.0001) and is higher in subjects with light irides (P = 0.04). Higher AF[787] emanates from hyperpigmentation, from the choroidal pigment (nevi, outer layers) and from the pigment epithelium and stroma of the iris. Low AF[787] is observed in geographic atrophy particularly in subjects with light irides.

conclusions. AF[787] originates from the RPE and to a varying degree from the choroid. Oxidized melanin, or compounds closely associated with melanin, contributes substantially to this AF, but other fluorophores cannot be excluded at this stage. Confocal AF[787] imaging may provide a new modality to visualize pathologic features of the RPE and the choroid, and, together with AF[488] imaging, offers a new tool to study biological changes associated with aging of the RPE and pathology.

Autofluorescence (AF) imaging is playing an increasingly important role in the diagnosis of age-related macular degeneration (AMD) and retinal dystrophies. Fundus AF generated with short-wavelength excitation is dominated by RPE lipofuscin, ¹ a complex mixture of fluorophores that are byproducts of the visual cycle, ^{2 3} and accumulate in the RPE after phagocytosis. Confocal scanning laser ophthalmoscopy (SLO) ⁴ has made this imaging modality accessible as a clinical tool. ^{5 6 7} An important advantage of AF imaging has been that the signal originates principally from the RPE, resulting in a relatively simple interpretation of the images. Fundus autofluorescence excited in the near-infrared (NIR) at 805 nm was first reported by Piccolino et al. ⁸ using a nonconfocal video-imaging system as part of an investigation of possible pseudofluorescence in ICG angiography. They demonstrated that pseudofluorescence did not substantially affect their images, but that a faint AF emanated from several pathologic structures (the normal eye was not investigated). They suggested that degradation products of blood, lipofuscin deposits, and/or melanin contributed to this AF. We have further investigated near-infrared AF imaging using the superior imaging modality provided by the confocal SLO, in conjunction with the same excitation wavelength (787 nm) and detection system that is normally used for ICG angiography (Keilhauer CN, et al. IOVS 2005;46:ARVO E-Abstract 1394; Weinberger AWA, et al. IOVS 2005;46:ARVO E-Abstract 2585). In this study, we demonstrate that normal eyes exhibit a characteristic near-infrared AF distribution and have investigated the origin of this AF, in a population of subjects with no retinal disease and in selected clinical cases. 

Methods

Population

NIR AF images were obtained in 85 subjects with normal retinal status (49 women and 36 men; mean age: 47 ± 18 years; range, 11–77 years). Iris colors were light (blue, gray) in 34 subjects and dark (green, hazel, brown) in 51 subjects. Two subjects were black Africans. Images were acquired from one eye in 49 subjects and from both eyes in 36 subjects; interocular correspondence was assessed, and one eye was selected randomly for other analyses. Images of fundus diseases have also been analyzed to differentiate AF contributions from fundus layers or to illustrate the AF of various pigments. These diseases were peripapillary atrophy (n = 3), hyperpigmentation (n = 8), geographic atrophy (n = 11, ages: 69–81 years) in AMD, and macular hole (n = 3). 

The tenets of the Declaration of Helsinki were observed. The Institutional Review Board of the Eye Clinic (University of Würzburg) granted approval for this project. Informed consent was obtained from all subjects. 

Retinal Imaging

NIR AF-images—AF[787]—were recorded with a confocal scanning laser ophthalmoscope (HRA, Retinal Angiograph; Heidelberg Engineering, Heidelberg, Germany). Laser diode excitation was at 787 ± 2 nm (power at the pupil: 1.9 mW) and the detection filter transmitted light above 800 nm (filter rejection: described later). The field was 30° × 30° (512 × 512 pixels), and always included the optic disc and the macula (the foveola was at least 100 pixels from the image edge). The confocal depth of the camera was ∼1100 μm (measured by moving an NIR-fluorescent retina in an artificial eye through the focal plane and measuring the locations were the returned signal is at half maximum). This field size is large enough to collect light simultaneously from the retina and choroid, while rejecting light originating from the lens and from a large part of the vitreous. 

All images were acquired by the same operator (CNK) for eyes with dilated pupils. Focusing was achieved at 815 nm, and reflectance images were acquired. After a switch to the 787-nm excitation (ICG mode), the sensitivity was increased until the vessels and the disc appeared as faint features and ∼40 images were acquired. In 38 study subjects (mean age: 45 ± 21 years), we also acquired AF[488] images (excitation: 488 nm; power: 270 μW). 

Image Analysis

Eighteen AF[787] images were selected for highest exposure and absence of severe eye movements, aligned, and averaged (nine images were used for AF[488] and one for NIR reflectance). All images shown herein have been histogram stretched. For quantitative analysis of the AF distribution, we used nonstretched images in conjunction with IGOR image analysis software (WaveMetrics, Lake Oswego, OR). Mean gray levels (GL) were measured at the center of the fovea and at three perifoveal sites (Fig. 1A) . GLs from different images cannot be compared with each other (different sensitivities, laser powers, and pupil diameters), but ratios of GL from the same image can be used to assess AF distribution. We defined the zero GL as the mean GL of the 2500 least-exposed pixels of each image or ∼1% of the image area (histogram analysis). These pixels were generally found near the edge of the optic disc. Tests performed on eight subjects showed that our approximation was adequate; it was 0.6 ± 0.9 GL (range: −1.7 to +1.7) higher than the mean GL in images acquired with the camera pointing at a black screen in the dark (zero GL varied from 9 to 23, because of electronic variations). Only images with mean exposures at the three peripheral sites >2 GLs above zero were included for analysis. 

Filter Rejection

Because AF[787] signals were very low, it was important to assess whether filter rejection was adequate. We first determined the filters’ rejection by acquiring images of the reflection of a coverslip (placed at ∼20 cm from the camera) obtained (1) in the AF[787]-mode (power: 1.9 mW; sensitivity adjusted to show the leak), (2) in the 815 nm-reflectance-mode (power: 63 μW; same electronic sensitivity) with insertion of a glass neutral-density filter (NDF) placed at 45° to the camera axis (to avoid reflections from the NDF) between the coverslip and the camera, and (3) in no-light conditions to obtain the zero. The NDF filter was calibrated for 45° transmission: the optical density was 3.2 density units (for double pass at 780–820 nm). GLs measured at the reflection in both modes, differences in power and detector’s spectral sensitivity, and the attenuation of the neutral filter allowed us to calculate a rejection of (6.2 ± 2.2) × 10⁵ (three tests). Second, we compared, in three subjects, GLs in same fundus areas of AF[787] and 815-nm reflectance images. For equal retinal irradiance and sensitivity, AF[787] was 4.4 ± 1.0 × 10⁴ times less intense than reflectance. Thus, reflected excitation light leaking through the filters should be 14 ± 6 times lower than the AF[787] signal, small enough to cause no substantial interference with the AF signal. 

Results

The most striking feature of AF[787] images was an area of high AF that was roughly centered on the fovea (Fig. 1)and corresponded to the higher RPE melanin in color images and to the lower AF[488] surrounding the area of densest macular pigment (MP; Figs. 1A 1E ). Retinal vessels were seen against a bright background, but their contrast was always lower than in AF[488] images, and choroidal vessels were occasionally delineated. The optic disc was always dark, with its edges often being the darkest area of the AF[787] images. 

AF[787] images were always of lower contrast and exposure than those obtained with 488-nm excitation (AF[488]). Although AF[787] images were recorded using radiant powers approximately seven times higher and detection sensitivity three to five times higher than those used for AF[488] images. Mean GLs for the three perifoveal sites were ∼4.6 and ∼13.4 for the AF[787] and AF[488] images, respectively. Thus, AF[787] was 60 to 100 times less efficient than AF[488] when excited from outside the eye. 

RPE Contribution

Evidence that the RPE contributes to AF[787], in addition to the bright foveal area described earlier, is found from the combined interpretation of the AF-distribution in peripapillary atrophy (Fig. 2A)and in a full-thickness macular hole (Fig. 2B) . ^{9 10} Whereas the former suggests contribution from the neurosensory retina and/or the RPE, the latter implies contributions from the RPE and deeper layers. Taken together, these observations indicate that the RPE-choriocapillaris complex contributes to AF[787]. This conclusion is supported by the high AF[787] emanating from hyperpigmentation (Figs. 2A 2C)associated with clumps of stacked RPE cells (brownish foci in color images). ¹¹ Furthermore, imaging of the iris (Fig. 2D)reveals a continuous ring of high AF[787] at the pigment ruff. This ring corresponds to the two heavily pigmented epithelial cell layers of the iris that terminate at the pupil margin. ¹⁰  

Mean AF[787] level in the bright foveal area was significantly higher (paired t > 17, P < 0.0001) than that at the three perifoveal locations, in contrast with AF[488] images where the opposite was true (t < −12, P < 0.0001). We found no differences between the nasal and superior sites (P = 0.6), but levels at both sites were higher than at the inferior site in the AF[787] images (n = 85, paired t > 2.7, P < 0.008) and AF[488] images (n = 38, t > 2.7, P < 0.009). The ratio of foveal to perifoveal AF is defined as:  

\[K_{\mathrm{FP}}\ {=}\ \frac{\mathrm{GL\ (at\ fovea)}}{\mathrm{GL\ (average\ of\ 3\ perifoveal\ areas)}},\]

where K _FP was 1.7 ± 0.2 on average, decreased with age (P < 0.0001, Fig. 3 ), and was lower in subjects with dark rather than light irides, particularly at old age (P = 0.04). Interocular correspondence in K _FP was moderately good (n = 36): K _FP for fellow eyes was not significantly different from each other (paired t = −1.0, P = 0.3) and the correlation between eyes was significant (r = +0.55, P = 0.001). 

Because the position of the foveola could not be identified in the AF[787] images, we obtained its position in AF[488] images as the location with maximum MP absorption (using retinal vessels as landmarks for both images). From curve fits to the horizontal GL-profiles through the foveola (Fig. 4) , we calculated the mean horizontal width of the bright area (at half its maximum) to be 8.8° ± 2.2° (n = 37, 1 outlier). The profiles exhibited a wide range of shapes, most of them (except for 5/37) being flatter than a Gaussian profile (Fig. 4) . Surprisingly, the center of the fitted profiles was found to be nasal to the foveola at a mean eccentricity of 0.78° ± 0.86° (t-test from 0, t = 5.6, P < 0.0001). This can also be appreciated in Figure 1 . The nasal eccentricity was not significantly correlated with age (r = 0.2, P = 0.3) or K _FP (r = −0.2, P = 0.16), and was not affected by iris color (P = 0.7). The asymmetry in the distribution of the bright area, can also be expressed in AF intensity: the GLs at 5° nasal to the foveola were 13% ± 19% higher than at 5° temporal (t = 4.3, P = 0.0001). We found no significant difference in the vertical direction between the position of the foveola and the center of the bright area (P = 0.5). 

At the foveola, we observed a small localized depression in the AF[787] images of some subjects (Figs. 1D 4) . We determined the ratio of GL in a 0.9° × 0.9° area centered on the foveola to the mean GL along an annulus centered at 2.5° radius (width: 0.9°). This mean ratio was 0.99 ± 0.04 (n = 38) ranging from 0.86 (deepest depression, Fig 1D ) to 1.07 (maximum at the foveola), and deepened with age (r = −0.36, P = 0.03), but was not affected by iris color (P = 0.8) and K _FP (P = 0.2). 

Choroidal Contribution

In addition to a contribution from the RPE, fundus AF[787] also emanates from the choroid, as evidenced by the AF of nevi (Fig. 1E)and by the visibility of choroidal vessels (Fig. 1C) . The visibility of these vessels was assessed by two observers (blind scoring: Yes or No). There was interobserver agreement for 71 of 85 images (χ² = 41, P < 0.0001). For those 71 subjects, choroidal vessels were more likely seen in subjects with dark rather than light irides (χ² = 5.6, P = 0.02) and were more often seen in older eyes (ages >47 years, median age) than in young eyes (χ² = 4.1, P = 0.04). There was no difference in the distribution of iris colors between the young and old age groups (P = 0.7). 

AF-imaging of geographic atrophy (GA; Fig. 5 ) revealed low levels of AF[787], consistent with the degeneration of the RPE and similar to that of AF[488]. ⁷ High AF at the margin of the atrophy corresponded to hyperpigmentation in both imaging modes, but the relative intensities of these bright foci were different in both modes (Fig. 5 , arrows). In 11 patients with geographic atrophy (GA, ages: 69–81 years), we compared mean GLs measured in the GA with those at the perifovea for both AF[787] and AF[488] images (Fig. 5C) . The ratio K _GA = (mean GL in GA)/(mean GL at perifovea) was always higher (lower contrast) for AF[787] images than for AF[488] images. The contrast of the GA for AF[787] images was lower in subjects with dark rather than light irides, but the opposite was true of AF[488] images (Fig. 5C) . 

Discussion

AF[787] appeared to originate from melanin in the RPE and to a varying degree from melanin in the choroidal layers. An RPE contribution is supported by the pattern of high foveal AF[787] that corresponds with the distribution of RPE melanin, by the reduced AF[787] in geographic and peripapillary atrophy, and by the high AF from PE cells of the iris, which are anatomically similar to those of the RPE but contain no lipofuscin. ¹² Furthermore, high AF[787] and AF[488] emanate from hyperpigmentation, ⁶ which is clumps of stacked or hyperpigmented (lipofuscin, melanin, and melanolipofuscin) RPE cells in the subretinal and sub-RPE space. ¹¹ NIR-reflectance images acquired with confocal SLOs show high reflectivity from hyperpigmentation (Fig. 2A 2C)resulting from backscattered light from the highly refractile melanin granules. ¹³ We demonstrated that pseudofluorescence (filter leakage) did not substantially contribute to the signal of our AF[787] images. 

A contribution to AF[787] from melanin in the choroid is indicated by the visibility of choroidal vessels, nevi, and melanin deposits in the iris stroma. Melanin is two to three times more abundant in the outer than in the inner choroid, ¹⁴ and this contributes to the visibility of absorbing choroidal vessels seen against the bright AF of the outer choroid. Choroidal vessel visibility varied with iris color as did the ratio of foveal to perifoveal AF[787] (Fig. 3)and the extent to which geographic atrophy contrasted with its surround (Fig. 5) . Because iris color is in large part related to the amount of melanin in the iris stroma, ¹⁵ it also reflects the amount of melanin in the choroidal stroma (uvea). ^{16 17} This melanin, as well as that in hair and skin, is embryologically derived from the neurocrest and exhibits marked racial variations. In contrast, melanin in the RPE is derived from the neuroepithelium cells, and its concentration is independent of race and iris color. ^{14 18}  

Endogenous Fundus Fluorophores

The dominant fundus fluorophore for short-wavelength excitation is RPE lipofuscin, ¹ a mixture of several fluorophores, ² of which only a few have been characterized to date: A2E and minor cis-isomers of A2E ^{3 19} and an all-trans-retinal dimer conjugate. ²⁰ Little is known about the other lipofuscin fluorophores. Lipofuscin in vivo has a broad excitation spectrum that peaks at 490 to 510 nm and drops to ∼30% of maximum at 600 nm. AF-imaging has been achieved with excitations as long as 580 nm. ^{21 22} We cannot at this point rule out that the excitation spectrum of one of the lipofuscin fluorophores extends to the NIR and contributes to AF[787]. 

Melanin absorption decreases monotonically with increasing wavelength (Fig. 6) . ^{23 24 25 26 27} Its AF was mostly studied for short-wavelength excitations (300–500 nm) for synthetic melanin, ²⁸ hair and skin, ^{29 30} and ocular melanins. ^{23 27 31 32} Boulton et al. ²³ and Docchio et al. ³¹ demonstrated that AF of melanin granules (peak emission: 440–560 nm; peak excitation; ∼450 nm) was 6 to 10 times less efficient than that of lipofuscin granules and that the optical density and fluorescence of melanin granules increased with age. Unfortunately, few data were presented in these studies to assess the magnitude of these age-related changes. Kayatz et al. ²⁷ reported similar AF properties for bovine melanosomes and showed that melanin only fluoresced efficiently if it had undergone oxidation by hydrogen peroxide. Because the RPE generates this compound after light exposure, ^{33 34} they suggested that the AF of melanin granules in vivo may increase with age and that ex vivo samples of ocular melanin, used in fluorescence and absorption studies, may be partially oxidized. Sarna et al. ³² further showed that photo-oxidation increases the AF of samples of melanosomes from human RPE. Thus, the efficiency of melanin AF is increased by oxidation. 

Melanolipofuscin is a complex granule derived from melanin and lipofuscin organelles, involving degradation and remodeling of RPE melanin. ^{35 36} Little is known about its AF properties except that its AF is intermediate between that of melanin and lipofuscin. ³¹  

For NIR excitation at 785 nm, Huang et al. ^{37 38} described broad emission spectra from synthetic melanin with maxima at 870 to 900 nm, and superimposed Raman emission lines at 880 and 895 nm (contributing ∼50% of the total signal). AF from the skin ³⁸ increased with the degree of pigmentation and the emission spectra generally decreased with increasing wavelength with less obvious Raman signals; differential analysis demonstrated that the AF was composed of that of melanin (similar to synthetic melanin) and that from other tissues components (decreasing with wavelength). Pilot spectra obtained with our spectrofluorometer ³⁹ using a laser diode as excitation (783 nm) demonstrated weak emission spectra that decreased with increasing wavelength (not shown), but the signal was too variable to detect Raman lines. The NIR fluorescence properties of ocular melanin ex vivo and in vivo warrant further study. 

Porphyrins are not believed to contribute significantly to the AF[787] images shown herein (excitation: 500–700 nm; emission: 600–800 nm), ⁴⁰ because their concentrations in blood are low and no AF appears to emanate from blood vessels. However, they may play a role in AF imaging of melanomas and lesions that contain degradation products of hemoglobin. ^{8 41} Other fluorophores such as collagen and elastin, ⁴⁰ that contribute to the AF of Bruch’s membrane for short-wavelength excitations, ⁴² could also contribute to AF[787], but no information is available on their NIR properties. 

Spatial Distribution

AF[787] is the combination of AF from the RPE and from the choroid (attenuated by absorption and scattering of the excitation light and the fluorescence in the RPE and choriocapillaris). Because only choroidal melanin is affected by iris color, it is understandable that the contrast of the bright “foveal” area (caused by RPE melanin) is higher in lightly pigmented eyes when less AF emanates from the choroid (Fig. 3) . 

The foveal area of high AF[787] corresponds roughly with the area of higher melanin pigmentation generally seen on color images and with an area of reduced AF[488] surrounding the area of strongest MP absorption (Figs. 1A 1E) . The latter results in part from attenuation of lipofuscin AF by RPE melanin, which is located more apically than lipofuscin in the cell. ^{14 43} Ex vivo studies ^{14 24} demonstrated a maximum in optical density of melanin centered on the fovea with a half-width of 6° to 10°, similar to the width of ∼8° found for our bright foveal area. In donor eyes, the foveal to perifoveal density ratio was ∼1.9 whereas cell heights were ∼1.13 taller at the fovea than at the perifovea. ¹⁴ Thus, it is likely that the increased foveal melanin is caused by an increase in both cell height and in melanin concentration in these cells. 

The center of the bright foveal area was found to be on average ∼0.8° nasal to the foveola (Figs. 1 4) . Although small compared to the distribution width (∼8°), this asymmetry is unexpected because no marked horizontal asymmetries have been reported in the distributions of rods and cones, ⁴⁴ MP, ⁴⁵ and lipofuscin. ⁴⁶ We considered whether spurious effects could cause this shift. The fact that a small depression was found at the foveola in some AF[787] images indicates that the effect was not due to image distortion (the depression would otherwise be at the center) or differences in magnification (the latter was 1% larger for AF[488] than for AF[787] images). If the excitation or detection were not uniform but higher at the center of the field, then a shift toward the center (and thus nasally) could occur particularly if the distribution is very flat-topped (Fig. 4) . However, we found no relationship between the nasal shift and the position of the bright spot in the field (P = 0.6, n = 38), nor could we detect a substantial change in the asymmetry when, in two subjects with nasally displaced bright areas, we located the fovea on the left and right sides of the center of the field (not shown). Although we cannot rule out other effects, we hypothesize that the distribution of RPE cell heights and/or of melanin concentration could be slightly asymmetrical (∼13% higher at 5° nasal than at 5° temporal) but we found no data in the literature to support or reject this possibility. The choroidal AF distribution could also contribute to the nasal shift if it were peaked at the posterior pole (nasal of the foveola), but the only existing data were not acquired with enough resolution to verify this possibility. ^{14 24}  

The localized depression in AF[787] at the foveola (Figs. 1D 4)remains largely unexplained. It is not likely to be the result of absorption by cone or MP, because NIR absorption of these pigments is very low. It may result from depigmentation, because the depression was found to be deeper (and perhaps larger) in older subjects. Alternatively, the lower AF[787] at the foveola may be the result of reduced photo-oxidation of RPE melanin, secondary to protection from short-wavelength light by the densest part of the MP distribution. Higher contrast AF[787] imaging may help clarify the significance of this feature and provide, in contrast to AF[488], a clinical tool to examine the fovea, without any masking by the MP (Fig. 2C) . 

An inverse relationship between the amounts of melanin and lipofuscin in RPE cells was demonstrated ex vivo. ¹⁴ This may be reflected by the higher lipofuscin levels found temporally than nasally ^{46 47} —the opposite of the melanin distribution suggested by the nasal displacement of the peak AF[787]. However, this inverse relationship was not confirmed in the vertical direction; both AF[787] and AF[488] were lower inferiorly than superiorly. Lower inferior AF[488] levels were observed in some, ⁴⁶ but not all, ⁴⁷ previous studies. Clarification of these issues awaits more detailed quantification of the AF[787] and AF[488] distributions in the same subjects. 

Contribution of Melanin Fluorescence to AF[488]

We estimated that AF[787] is 60 to 100 times less than AF of lipofuscin, AF[488], for equal instrumental sensitivity and excitation power outside the eye. Assuming that AF[787] is caused only by melanin and that its excitation and absorption spectra are essentially the same (Fig. 6) , melanin AF would be ∼9 times more efficient at 488 nm than at 787 nm. Crystalline lens double transmission is ∼1 at 787 to 820 nm ⁴⁸ but is 0.4 to 0.7 at 488 nm (ages: 75 and 25, respectively). ⁴⁹ Thus, AF from melanin for 488-nm excitation, would be only 3% to 10% of the AF from lipofuscin. It is unclear, however, that the secondary fluorophore detected in vivo during lipofuscin measurements ¹ was melanin. ²⁷ This minor fluorophore may be vitreous AF, because it was best detected at the fovea, where 488-nm excitation of any RPE fluorophore would be strongly attenuated by MP. 

Age Relationship of K _FP

AF depends on the amount of melanin in the RPE and choroid, its absorption, and its fluorescence efficiency. Each of these parameters undergoes age-related changes whose magnitudes are poorly known and may occur in opposition to each other. Nevertheless, we examine different mechanisms that could account for the age-related decrease in the ratio of foveal to perifoveal AF (K _FP, equation 1 ; Fig. 3 ). The ratio K _FP can be equated as  

\[K_{\mathrm{FP}}\ {=}\ \frac{R_{\mathrm{F}}\ {+}\ T_{\mathrm{F}}\ {\cdot}\ C_{\mathrm{F}}}{R_{\mathrm{P}}\ {+}\ T_{\mathrm{P}}\ {\cdot}\ C_{\mathrm{P}}}\ {\approx}\ \frac{{\alpha}\ {+}\ C/R_{\mathrm{P}}}{1\ {+}\ C/R_{\mathrm{P}}},\]

where R and C are the AF of the RPE and choroid, respectively; T is the double transmission (excitation and emission wavelengths) of the RPE and choriocapillaris, and the subscript F and P refer to the fovea and perifovea, respectively. The parameter α is R _F/R _P and is assumed to be larger than 1. On the right-hand side of the equation, we assumed that the choroidal AF is the same at both sites (C _F = C _P = C) and that T _F ≈ T _P ≈ 1, since the absorption by the RPE (Fig. 1)and the choriocapillaris is very small in the NIR. 

In addition to an age-related increase in lipofuscin, loss of RPE melanin granules has been observed in all regions after age 40 years, ⁵⁰ a time course similar to that of K _FP (Fig. 3) . In the macula, the number of melanin granules decreases by >50% over a lifetime. However, less marked age-related decreases have been found in the amount of macular RPE melanin in optical density ^{14 51} and concentration ¹⁸ measurements, probably a result of the increased absorption of the granules. ²³ If the AF efficiency of RPE melanin is independent of age and if the loss in melanin is proportional at all sites (α = constant, which is reasonable since the foveal-to-perifoveal optical density ratio was not found to be affected by age ¹⁴ ), then K _FP will decrease with age (equation 2 , α > 1, C constant, decrease in R _P). However, oxidation of melanin throughout life will increase the AF efficiency of melanin, ^{27 31 32} although this may be in part prevented by the presence of powerful antioxidants in the RPE. ^{52 53} Thus, a slower age-related decrease in R _P and in K _FP can be expected, as long as the increase in efficiency is offset by a greater loss of granules. 

The potential protective effect of the MP against melanin oxidation by short wavelengths light may result in a slower age-related increase in AF efficiency at the fovea than at the perifovea, causing c1 α to decrease with age. This, combined with the increase in C/R _P, results in a marked decrease in K _FP (equation 2) . In that case, one would expect a lower K _FP for high MP density. After accounting for age and iris color (as in Fig. 3 ), we found a weak negative correlation (n = 38; r = −0.23; P = 0.16) between K _FP and the foveal absorbance at 488 nm, −log [K _FP,488], which is in large part due to MP absorption and did not vary with age (P = 0.6). ⁵⁴  

The ratio K _FP will also decrease with age if the choroidal AF increases (equation 2 , c1 α and R_P constant). No age-related change in the amount of melanin was found for the choroid ¹⁴ and for the iris stroma ¹⁵ in donor eyes. However, choroidal melanin absorbs a large fraction of the white light transmitted by the RPE (including for λ > 600 nm where melanin, unlike blood, still absorbs) and may be more susceptible to photo-oxidation because of the absence of local antioxidants. Thus, an age-related increase in AF from choroidal melanin is likely, with a resultant decrease in K _FP. Other factors such as redistribution of melanin toward the outer choroidal layers ¹⁴ and decrease of the choroidal blood volume with age ⁵⁵ could also contribute to the increase in choroidal AF. It should be noted that computations using equation 2 , without neglecting the NIR absorption of the RPE and choriocapillaris (T _P < 1 and T _F < 1), demonstrated similar trends for changes in K _FP. 

It is uncertain, at this point, whether the combination of an age-related increase in choroidal AF[787] and the decreases in melanin AF from the RPE can account for the observed change in K _FP between the younger and older subjects, or if other factors are involved. Our observation that choroidal vessels were better detected at older age is also consistent with an increase in AF from the outer choroid and/or with a decrease in RPE fluorescence, since the latter would diminish veiling of the deeper layers by RPE fluorescence. 

Contribution of Choroidal Melanin

The contribution of choroidal AF to the total AF at the perifovea can be estimated from the measurements in geographic atrophy (K _GA; Fig. 5C ). From equation 2 , with R _F = 0 and T _F = 1, we can equate K _GA = C/(R _P + T _P × C). The double-transmission T_P of perifoveal RPE and choriocapillaris (25 c1 μm blood layer) for the excitation and emission wavelengths (T _P = 0.87 ± 0.03 for AF[787], and 0.30 ± 0.06 for AF[488] ^{14 56} ). The proportion of choroidal AF (T _P × C) relative to the total AF at the perifovea (R _P + T _P × C) is then K _GA × T _P which is 52% ± 5% and 62% ± 9% for 70- to 80-year-old subjects with light and dark irides, respectively. In contrast, for AF[488] images, the choroid contributes 12% ± 3% and 9% ± 3%, respectively (the fluorophore in this case may be stromal collagen or the sclera, and the difference reflects the difference in absorption by melanin). Thus, choroidal melanin contributes a large fraction of the total AF at older ages; this predominance may be a limitation of AF[787] imaging as a clinical modality in AMD because it will reduce the contrast of AF features in the RPE. Choroidal contribution at younger ages could not be directly estimated, since cases of frank atrophy were not found among young patients. It would be smaller for essentially the same reasons as proposed for the higher K _FP at young age. Lower contribution of the choroid in young subjects is consistent with the weaker dependence of K _FP on degree of pigmentation (Fig. 3 , the two young black African subjects have K _FPs similar to those of the other subjects) and with the decreased visibility of choroidal vessels in young subjects. 

In regard to the amount of melanin in the choroid, we derived C/R _P = K _GA/[1 − T _P × K _GA] or 1.9 ± 0.3 and 1.3 ± 0.1 in subjects with dark and light irides, respectively. Thus, the amount of choroidal melanin is 1.5 ± 0.3 times higher in subjects with dark rather than light irides. This difference is in reasonable agreement with an estimate of a ratio of 1.2 for the iris stroma and PE (necessarily an underestimate since the PE, which does not greatly affect iris color, was included), ¹⁵ and with a ratio of 2.0 for the amount of choroidal melanin in black and white individuals. ¹⁴  

In summary, the ocular fundus exhibits a faint AF under NIR-excitation that largely contrasts, in intensity and distribution, to the AF under blue light excitation which is known to be dominated by RPE lipofuscin. Topographic distribution, dependence on iris color, and observations in selected clinical cases suggest that melanin and/or compounds closely related to melanin (oxidized melanin, melanolipofuscin) is in large part responsible for the observed AF. Small contributions from other fluorophores, such as lipofuscin fluorophores, cannot be excluded at this point. In the future, quantitative measurements of the NIR autofluorescence may give information on the spectral properties and age relationship of this AF. Such measurements as well as long-term follow-up by imaging of melanin and lipofuscin are important for a better understanding of the unresolved biophysical questions and particularly of the biological changes associated with aging and disease. 

 Supported in part by National Eye Institute Grant EY08511 (FCD).

 Submitted for publication February 3, 2006; revised April 12, 2006; accepted June 12, 2006.

 Disclosure: C.N. Keilhauer, None; F.C. Delori, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

AF[787] images in five subjects with normal retinal status. Ages and iris color (L: light; D: dark) as indicated. AF[488] images are shown for comparison in (A) and (E). Bars and plus signs indicate the position of the fovea as defined by the darkest point of MP distribution in the AF[488] image. The square (1.25° × 1.25°) and rectangles (2.50° × 1.25°) in (A) show the areas in which mean GLs were measured on both AF[488] and AF[787] images; the three perifoveal sites are at the same distance from the fovea, and the nasal site is midway between the fovea and the disc center. A temporal site was not used because it often was partially outside the image. All AF[787] images showed an area of high IR fluorescence which corresponded to the area of higher melanin pigmentation in the AF[488] images, extending outside the densest MP distribution (A, E). Images (B) and (C) are from subjects with the largest and smallest contrast of the bright area in our population. The fovea location in (D) is marked by a local reduction in AF. Choroidal vessels in (C) and (D) are delineated against the brighter AF originating from the outer choroid. A nevus (E, arrows) exhibits bright AF[787] but without marked increase of lipofuscin in the RPE, as demonstrated by the AF[488]. The GL on the nevus was 1.9 times higher than the neighboring area.

Figure 2.

 

AF[787] images in four patients compared to IR reflectance (IR-R) and AF[488] images. Ages and iris color (L: light; D: dark) as indicated. (A) Peripapillary atrophy (AMD). Ring of hyperpigmentation (arrow) scattered intensely in IR-R and fluoresces in both AF[787] and AF[488]. Zone (a) may represent Bruch’s membrane denuded of RPE cells. ⁹ Zone (b) is exposed sclera. AF in both zones was lower than at the intact retina, indicating contribution from the RPE or the neurosensory retina. (B) Full-thickness macular hole. AF[787] in the hole was similar to that of the perifoveal region, indicating that AF[787] emanates from the RPE and deeper layers. In the AF[488] image, the RPE is partially masked by the operculum. (C) Hyperpigmentation (HP) in AMD showed high AF[787] and IR-R, but not in AF[488] because of MP absorption. GL profiles (along line between bars) showed that the reflected light from the nerve fibers was brighter than the HP, but this difference was not observed in AF[787], confirming that filter leakage does not play a large role in these fundus images. (D) AF[787] image of an iris, obtained with a 22-mm focal length lens in front of the HRA, is compared to a black/white reflectance image (B/W). AF[787] showed a continuous ring of high AF corresponding to the extension of pigmented epithelium (PE) of the iris around the pupil margin. Several iris freckles (solid arrows) located in the anterior stroma, exhibited low reflectance but variable AF[787]. Fuch’s crypts (interrupted arrow) match loose spaces in the stroma ¹⁰ ; high AF may represent the PE of the iris. The small spot in the pupil is filter leakage from the corneal reflection contrasted against the dark background.

Figure 3.

 

Variation of the ratio of foveal-to-perifoveal AF as a function of age. The GLs measured at the fovea and at 3 sites in the perifovea as shown in Figure 1A . (•) and (▪) subjects with light and dark irides, respectively; (○) black African subjects. Arrows: the two subjects in Figures 1B and 1C . The ratio K _FP decreased with age particularly for ages >40 years. Stepwise regression analysis including age and iris color (IC) yielded K _FP = 1.85–0.080 × (age/47)³ − 0.049 × IC × (age/47)³, with IC = 1 for dark or 0 for light irides (adjusted r ² = 0.33, P < 0.0001). Dotted lines: regression results, for light (L) and dark (D) irides.

Figure 4.

 

Horizontal AF[787] profiles (thick lines) through the foveola in six subjects (ages as indicated) using a 0.7-deg² sampling spot. The profiles were displaced by five GLs to avoid overlap; dashed horizontal lines: zero GLs for the respective profiles. Thin traces: symmetrical function fitted to each profile to determine its width and peak (eccentric) position. The function was Gaussian-like in which the eccentricity exponent varied from 1 (exponential), through 2 (Gaussian), to 7 (very flattened top). The profile’s maxima (bars) were often located on the nasal side of the foveola. Several profiles (24, 35, 58) exhibited a slight depression at the site of the fovea (58-year-old subject: see also Fig. 1D ). The top profile is from a black African subject and had substantially larger GLs than the others, but this finding should be confirmed by absolute measurements.

Figure 5.

 

AF[787] and AF[488] images in two patients with AMD with GA. Reduced AF in the GA is observed in all images, because the RPE cells were absent, but the contrast of the GA in the AF[787] images (top) was higher when the choroidal pigmentation was low (A) than when it was high (B). High AF at the margin of the atrophy corresponds to hyperpigmentation in both imaging modes; the relative intensities of these bright foci may be different in both modes, and some foci are only seen in the IR mode (arrows). (C) Ratio K _GA = GL(atrophy)/GL(perifovea) measured in 11 patients with AMD with GA for both AF[787] images (•) and AF[488] images (○). Measurements in the GA involved the entire zone of atrophy, avoiding the foveola and any islands of remaining RPE. The perifoveal measurements were made at the same three sites as shown in Figure 1A . Bars: standard deviations obtained by propagation of errors for the measurements at the various sites. The mean (±SD) for K _GA is given for each subgroup. K _GA was significantly higher (lower contrast) in AF[787] than in AF[488] images (paired, t = 5.0, P = 0.0004). In AF[787] images, K _GA was larger in dark than light irides (t = 2.5, P = 0.03), whereas the opposite was true of AF[488] images (P = 0.009).

Figure 6.

 

Absorption spectra of melanin, normalized at 500 nm. The spectral data are (B) from Boulton et al. ²³ for human RPE melanin in solution in three age groups; arrows: increasing age. (G) Data from Gabel et al. ²⁴ for human RPE; (M) data from Menon et al. ²⁵ for iris melanin in solution, and (J) a c1 λ^−3.48 wavelength-dependence suggested by Jacques and McAuliffe ²⁶ based on measurements of internal absorption for RPE melanin (bovine). Average relative absorptions (inset) were calculated from these data, with extrapolations to 800 nm by exponential fits to the data. Melanin in these ex vivo samples was probably partially oxidized ²⁷ ; the excitation spectrum of melanin AF would, in the first approximation, be similar to the absorption spectrum.

The authors thank Jörg Fischer (Heidelberg Engineering, Heidelberg, Germany) for support with issues concerning instrumentation, Janet Sparrow (Columbia University, New York, NY) and Haishan Zeng (British Columbia Cancer Research Center, Vancouver, Canada) for useful discussions and critical suggestions, and Doug Goger for expert technical assistance.