Ultrastructural alterations in the retinal pigment epithelium and photoreceptors of a Stargardt patient and three Stargardt mouse models: indication for the central role of RPE melanin in oxidative stress

Animals

Pigmented Abca4^−∕− mice (129S4/SvJae-Abca4^tm1Ght) were kindly donated by P. Charbel Issa (University of Oxford, Oxford, England). Albino Abca4^−∕− mice (BALB/c-Abca4 ^tm1Ght) were kindly donated by G. Travis and R. Radu (University of California, Los Angeles, CA). Double knock-out Abca4^−∕−.Rdh8^−∕− (129S4/SvJae- Abca4^tm1Ght*C57BL/6J-Rdh8^tm1Kpal) mice were kindly donated by K. Palczewski (Case Western Reserve University, Cleveland, Ohio). Control albino WT mice (Balb/c) were purchased from Charles River (Sulzfeld, Germany) and control pigmented WT mice (129S2) were purchased from Harlan Laboratories (Hillcrest, UK). The knock-out mouse strains were bred in our in-house facility. Light cycling was 12 h light (approximately 50 lux in cages)/12 h dark, food and water were available ad libitum. All procedures involving animals were in accordance with the German laws governing the use of experimental animals and were previously approved by the local agency for animal welfare (Einrichtung für Tierschutz, Tierärztlichen Dienst und Labortierkunde der Eberhard Karls Universität Tübingen, Tuebingen, Germany) and the local authorities (Regierungspräsidium Tübingen, Tuebingen, Germany).

Sample preparation and light, fluorescence and electron microscopy

Animals were sacrificed by carbon dioxide inhalation and subsequent cervical dislocation and the eyes were immediately enucleated.

For light and electron microscopy, eyes were fixed overnight at 4 °C in 5% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.4). After washing in cacodylate buffer, the cornea and lens were removed and eye cups were hemisected. Halves were post-fixed in 1% osmium tetroxide in 0.1 M cacodylate buffer and bloc-stained with uranyl acetate. Samples were dehydrated in a graded series of ethanol and propylene oxide and embedded in Epon. Reagents were purchased from AppliChem (Darmstadt, Germany), Merck (Darmstadt, Germany) and Serva (Heidelberg, Germany). Light microscopy on toluidine blue stained semi-thin sections (500 nm) was performed with a Zeiss Axioskop (Zeiss, Jena, Germany). Ultrathin sections (70 nm) were mounted on copper slot grids (Plano, Wetzlar, Germany) and stained with lead citrate and examined with a Zeiss 900 electron microscope (Zeiss, Jena, Germany).

For fluorescence microscopy and immunohistochemistry, eyes were fixed in 4.5% formaldehyde (Carl Roth, Karlsruhe, Germany) and embedded in paraffin wax, cut into 5 µm thick sections for fluorescence microscopy and 4 µm thick sections for immunohistochemistry. Samples for fluorescence microscopy were deparaffinized according to standard procedures, cover-slipped with FluorSave (Calbiochem, La Jolla, CA, USA) and were investigated with a Zeiss Axioplan2 imaging microscope (Zeiss, Jena, Germany). Lipofuscin autofluorescence was visualized with a custom lipofuscin filter set (excitation 360 nm, emission 540 nm).

Immunohistochemistry

Sections were deparaffinized, rehydrated and subjected to heat induced antigen retrieval in a pressure cooker for 2 min in either citrate buffer at pH 6 (anti-HNE, anti-MDA) or Tris buffer at pH 9 (anti-NITT). Sections were incubated overnight at 4 °C with either rabbit anti-HNE (4-hydroxy-2-noneal, HNE11-S, alpha diagnostic international; major product of endogenous lipid peroxidation, 1:3,500), rabbit anti-MDA (malondialdehyde, MDA11-S, alpha diagnostic international; byproduct of endogenous lipid peroxidation, 1:2,000) or rabbit anti-NITT (nitrotyrosine, NITT12-A, alpha diagnostic international; peroxynitrite-induced nitration of tyrosine residues in proteins, 1:1,000). Dilution of primary antibodies was done with antibody diluent with background-reducing components (Dako S3022; Agilent, Santa Clara, CA, USA). The primary antibodies were detected with the Dako REAL™ Detection System, Alkaline Phosphatase/RED, Rabbit/Mouse according to the manufacturer’s instructions. In brief, sections were incubated with biotinylated goat anti-mouse and anti-rabbit immunoglobulins for 15 min at room temperature with subsequent incubation with streptavidin conjugated to alkaline phosphatase for 15 min at room temperature. Reaction products were visualized with freshly prepared substrate working solution supplemented with levamisole to block endogenous peroxidases. Sections were counterstained with hematoxylin solution modified according to Gill III (Merck, Darmstadt, Germany). To control for non-specific binding, control sections were prepared without incubation with primary antibodies. Sections were investigated using a Zeiss Axioskop (Zeiss, Jena, Germany).

Quantification of photoreceptor nuclei and outer segment length by light microscopy

Semi-thin sections of whole eye cups (up to 400 µm from the optic nerve head) were photographed using a 63×  oil objective. Areas adjacent to the optic nerve and to the ora serrata were excluded from analysis due to the physiological thinning of the retina.

Photoreceptor nuclei were counted semi-automatically in an average of 11 digital images per section in on average three eyes per age group using Fiji software (Schindelin et al., 2012). In each image, a region of interest with a width of 100 µm along the outer nuclear layer was defined and total photoreceptor nuclei within the region of interest were counted as number of photoreceptor nuclei per 100 µm width of retina. The total numbers of eyes investigated by light microscopy and used for photoreceptor nuclei quantification per strain were 34 eyes (albino Abca4^−∕− mice), 30 eyes (pigmented Abca4^−∕− mice) and 47 eyes (Abca4^−∕−.Rdh8^−∕− mice).

Outer segment layer length was measured manually in 11–16 positions per section in an average of two eyes per age group using Fiji software.

Electron microscopic investigation

Whole sections of mouse eyes were thoroughly investigated by electron microscopy. Total numbers of eyes investigated by electron microscopy for the different strains were 12 eyes (pigmented Abca4^−∕−), 16 eyes (albino Abca4^−∕−), 16 eyes (Abca4^−∕−.Rdh8^−∕−), six eyes (albino WT) and three eyes (pigmented WT). Sections were examined for changes in photoreceptor, RPE and Bruch’s membrane structure. Lipid droplets in RPE cells were counted and their diameter measured in whole sections. Thickness of Bruch’s membrane was measured in 15 consecutive images. All measurements were performed with iTEM 5.0 Software (Olympus Soft Imaging Solutions, Muenster; Germany).

Quantification of the area in RPE cytoplasm occupied by lipofuscin-like material

Lipofuscin-like material (derived from both lipofuscin and melanolipofuscin granules, hereafter referred to as lipofuscin) was quantified in young (aged 4 to 6 months) and old (aged 12 months) Stargardt mice in an average of 5 eyes per group. Thirty electron micrographs of the RPE (magnification× 20.000) per eye were taken in a way so only cytoplasm and no nuclei, microvilli and basal labyrinth were visible. To compensate for vignetting, only an oval region of interest in the image center was analyzed. The Trainable Weka Segmentation (Arganda-Carreras et al., 2017) plugin for Fiji software, a machine learning tool, was used to determine the area of lipofuscin per image. A representative subset of images was used to train the tool to specifically recognize lipofuscin, so melanin was excluded from the analysis. A combination of the mean, median, maximum, minimum, variance and the anisotropic diffusion-, entropy- and neighbor algorithms was used. Segmentation calculations were performed on a 16 virtual central processing units system with 64 GB working memory (de.NBI Cloud Tübingen, https://denbi.uni-tuebingen.de).

Human tissue

Glutaraldehyde-fixed perimacular tissue of a 72-year-old donor with clinically diagnosed Stargardt disease was obtained through Foundation Fighting Blindness (Columbia, MD, USA). Post mortem time until fixation was 48 h. Written informed consent of the donor for use in medical research and approval of the Institutional Review Board of the University of Tuebingen (Ethik-Kommission an der Medizinischen Fakultät der Eberhard-Karls-Universität und am Universitätsklinikum Tübingen, Tuebingen, Germany, approval number 462/2009BO2) were obtained. The experiments were performed in adherence to the tenets of the Declaration of Helsinki. Human tissue was embedded for standard electron microscopy as described above. For investigation of lipofuscin autofluorescence in semi-thin sections, heavy-metal treatment during embedding (osmium tetroxide, uranyl acetate) was omitted.

Isolation of porcine RPE melanin

RPE melanin from pig eyes was isolated as described by (Boulton & Marshall, 1985) with modifications (Zareba et al., 2006). In brief, pig eyes were opened close to the ora serrata and the anterior segment and vitreous were discarded. The retina was removed with forceps and the eye cup washed with PBS (Gibco, Carlsbad, CA, USA). Eye cups were filled with trypsin/EDTA (Gibco, Carlsbad, CA, USA) and incubated for 10 min at 37 °C. RPE cells were isolated by pipetting jet streams onto the cell layer and collected in DMEM supplemented with 10% fetal bovine serum (Gibco, Carlsbad, CA, USA) to stop the enzymatic reaction. RPE cell pellets were homogenized using a tissue grinder. Cell debris were removed by centrifugation for 7 min at 60 g and the obtained supernatant was centrifuged for 10 min at 6,000 g to pellet all pigment granules. The pellet was collected in 0.3 M sucrose, loaded on a two-step sucrose gradient (1 M and 2 M) and centrifuged for 1 h at 103.000 g. Isolated RPE melanin granules were washed in PBS and the final granule concentration was determined using a hemocytometer.

Quenching of superoxide in an NBT assay

Quenching capability of melanin was investigated with the photochemical nitroblue tetrazolium (NBT) assay (Cheng et al., 2015). In this assay, superoxide generated by illumination of riboflavin reduces pale-yellow NBT to blue-violet formazan. The reaction mixture had a final concentration of 0.1 mM EDTA, 13 µM methionine, 75 µM NBT and 4 mM riboflavin in PBS pH 7.4. EDTA, methionine, NBT and riboflavin were purchased from Sigma-Aldrich (St. Louis, MO, USA). The reaction mixture was illuminated for 15 min with blue light from an LED lamp at 2.5 mW/cm² (450 nm, SunaEco 1500 Ocean Blue XP, Tropic Marin, Wartenberg, Germany). Uniform illumination of the samples was ensured by placing the samples on a horizontally aligned rotator. The formazan was quantified at 560 nm (Synergy HT, BioTek, Winooski, VT, USA).

Controls included positive controls (all reagents, no melanin, illuminated), negative controls (all reagents, no melanin, kept in the dark) and blanks (no riboflavin, melanin, illuminated) to compensate for melanin absorption at readout.

Statistical analysis

Statistical analysis was performed with JMP 13 (SAS, Cary, NC, USA). All data sets were tested for normal distribution to decide on using parametric or non-parametric testing. Since all data sets were not normally distributed and contained several groups, we used the Steel-Dwass All Pairs test, a multiple non-parametrical test with alpha correction. The null hypothesis was that the groups were not significantly different. Values are given as mean ±standard deviation, p < 0.05 was considered statistically significant.

Photoreceptors

Light microscopic investigation of tissue of 12-month-old animals (Fig. 1A) showed overall typical histology in pigmented Abca4^−∕− mice compared to pigmented WT as already published in (Charbel Issa et al., 2013) (not shown). Albino Abca4^−∕− mice had shortened outer segments whereas Abca4^−∕−.Rdh8^−∕− showed a hypertrophy of the RPE (Fig. 1A, Fig. S1).

Outer segment length was measured in light microscopy. Compared to age-matched albino WT animals, albino Abca4^−∕− mice aged 4 months showed a significant reduction in outer segment length (albino Abca4^−∕− 4 months 22.3 ± 3.2 µm, albino WT 4 months 26.2 ± 2.5 µm, p < 0.001, Fig. 1B). In Abca4^−∕−.Rdh8^−∕− mice, outer segment shortening had later onset: no difference was found for 3-month-old animals, whereas 6-month-old animals had shortened outer segments (Abca4^−∕−.Rdh8^−∕− 3-month-old 32.1 ± 2.7 µm, pigmented WT 3-month-old 32.6 ± 1.6 µm, not significant; Abca4^−∕−.Rdh8^−∕− 6-month-old 27.7 ± 2.0 µm, pigmented WT 12-month-old 33.9 ± 2.0 µm, p < 0.0001, Fig. 1B). Pigmented Abca4^−∕− had no outer segment shortening compared to pigmented WT.

A detailed analysis of the age-dependent reduction of photoreceptor nuclei numbers of the three different Stargardt mouse strains revealed similar retinal degeneration courses for both albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice (Fig. 1C) as compared to the pigmented Abca4^−∕− mice for which it is known that they do not present signs of retinal degeneration (Charbel Issa et al., 2013; Weng et al., 1999). We quantified total photoreceptor nuclei numbers, as opposed to the more commonly used outer nuclear layer thickness, to account for nuclei disorganization and gaps between nuclei that occurred in albino Abca4^−∕− and, to a lesser extent, in Abca4^−∕−.Rdh8^−∕− mice.

On the ultrastructural level, shrunken, electron dense nuclei, resembling pyknosis, were only found in albino Abca4^−∕− animals (Fig. 1D). Furthermore, isolated photoreceptor nuclei were present in the subretinal space next to unusually thinned RPE cells, even in young albino Abca4^−∕− animals (Fig. 1E). In our colony of Abca4^−∕−.Rdh8^−∕−mice, we could not find the previously described rosette formation at 3 weeks and the rather prominent retinal degeneration beginning at 3 months (Maeda et al., 2008). Rosette formation was only seen in a single 12-month-old animal. Instead, RPE hypertrophy was regularly seen in Abca4^−∕−.Rdh8^−∕−animals (Fig. 1A, Fig. S1).

Genotype and age-dependent alterations of the disc membrane stacking were apparent on the ultrastructural level (Fig. 2A, Table 1). Those alterations seemed to start at the outer segment tips (next to the RPE) and spread towards the base of the outer segment. Alterations in outer segment tip organization were never seen in pigmented WT and pigmented Abca4^−∕− mice. In albino WT and Abca4^−∕−.Rdh8^−∕− animals, it was not seen in the youngest animals investigated (3–4 months), but in animals aged 6–7 months and older. In albino Abca4^−∕− mice, outer segment tip disorganization was already present in 4-month-old animals, the youngest group investigated in this strain.

In albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− animals, alterations in overall disc membrane stacking were observed starting with 6 months but were lacking in pigmented and albino WT and pigmented Abca4^−∕− animals.

Inner segments of photoreceptors appeared swollen and/or ruptured in albino Abca4^−∕− mice (Fig. 2B). In pigmented Abca4^−∕− mice and Abca4^−∕−.Rdh8^−∕− mice, inner segment damages were only occasionally observed (Table 1).

In albino Abca4^−∕− mice, agglomerated electron dense material was occasionally present between inner segment and outer segment (Fig. 2C) and disk membrane renewal seemed to be somewhat disturbed as disk membrane swirls were sporadically present at the nascent part of the outer segment (Fig. 2D). These changes were not present in albino WT and pigmented Stargardt mice. Outer segment tips were in contact to RPE microvilli in all mouse strains and age groups.

Vacuole-like structures within the RPE

In semi-thin sections of albino Abca4^−∕− mice, vacuole-like structures within the RPE monolayer were apparent in animals as young as 6 months (Fig. 3A). Electron microscopic investigation revealed that the vacuole-like structures were an enlargement of the intercellular space between RPE cells and were apically limited by the junctional complex (Fig. 3B). Sometimes vacuole-like structures were also seen within the cytoplasm, usually in proximity to the lateral plasma membrane and probably also of intercellular space origin. The contents of the vacuole-like structures varied between empty (not shown), filled with membranous material (Fig. 3B) and filled with a variety of fibrillous, granular, lamellar and amorphous materials (Fig. 3C). Different types of contents were often seen within the same eye. Ultrastructurally, vacuole-like structures were apparent in albino Abca4^−∕− mice of all age groups, but were barely present in pigmented Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice (Fig. 3D). The number of vacuole-like structures increased with age, though there was a considerable variance between eyes of the same age. Vacuole-like structures were considerably fewer in albino WT mice.

Lipofuscin

To compare the lipofuscin autofluorescence between the different Stargardt mouse strains, paraffin sections of 12-month-old animals were investigated by fluorescence microscopy (Fig. 4A). In pigmented Abca4^−∕− sections, the lipofuscin autofluorescence was clearly defined, indicating the presence of granules with well-defined borders. In albino Abca4^−∕− sections, the lipofuscin autofluorescence was predominantly blurry. In Abca4^−∕−.Rdh8^−∕− sections, both clearly defined and blurry lipofuscin autofluorescence were present. Autofluorescence in albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− tissue was elevated compared to pigmented Abca4^−∕− tissue.

The morphological appearance of lipofuscin granules in the different Stargardt mouse strains was investigated by electron microscopy. Typical morphological appearances are illustrated in Fig. 4B and age-dependent quantification of the different identified lipofuscin subtypes can be found in Table 1.

In pigmented Abca4^−∕− mice, we found exclusively lipofuscin with homogenous electron density and clearly defined margins indicating an intact limiting membrane (Fig. 4B), as previously described (Charbel Issa et al., 2013). Most of the lipofuscin was fused to melanin and formed melanolipofuscin.

In albino Abca4^−∕− mice, lipofuscin with homogenous electron density was also seen. However, the majority of lipofuscin granules belonged to another type that had a flocculent electron density (Fig. 4B, Table 1). The margins of this type of lipofuscin were often not clearly defined, suggesting loss of the lysosomal membrane. As albino Abca4^−∕− mice accumulated lipofuscin with age, the amount of lipofuscin with flocculent electron density increased while the amount of lipofuscin with homogenous electron density seemed to stay constant (Table 1).

In Abca4^−∕−.Rdh8^−∕− mice, the first type of lipofuscin with homogenous electron density had a considerable variability between age groups (Table 1). After the age of 12 months, the second type of lipofuscin with flocculent electron density was also present (Table 1). As in pigmented Abca4^−∕− mice, most of the lipofuscin was fused to melanin and formed melanolipofuscin.

Membranes enclosing lipofuscin of the flocculent electron density type were often damaged or absent (Fig. 4C). In very old animals (aged 23 months and older), the RPE cytoplasm was abnormally electron dense (Fig. 4D). Clear demarcations between lipofuscin granules and the cytoplasm were hardly visible.

The area of RPE cytoplasm occupied by lipofuscin (both derived from lipofuscin and melanolipofuscin granules) was quantified in electron micrographs of young (4–6 months) and old (12 months) Stargardt mice (Fig. 5). In both age groups, Abca4^−∕−.Rdh8^−∕− mice had the highest levels of lipofuscin (young 16.9% ± 8.3%, old 26.9% ± 5.3%). In young animals, albino Abca4^−∕− mice had almost twice the levels of pigmented Abca4^−∕− mice (albino Abca4^−∕− 7.5% ± 3.9%, pigmented Abca4^−∕− 3.9% ± 2.7%, p < 0.0001). In 12-month-old animals, the difference of lipofuscin levels in albino and pigmented Abca4^−∕− mice was lower but nevertheless statistically significant (albino Abca4^−∕− 11.0% ± 5.5%, pigmented Abca4^−∕− 9.3% ± 5.5%, p < 0.05).

Lipid droplets

Lipid droplets were frequently seen in RPE cells of all investigated mouse strains, especially in albino Abca4^−∕− mice (Table 1). In pigmented and albino WT animals, lipid droplets were small (median diameter: albino WT 235 nm, pigmented WT 353 nm) and the majority of droplets was fused to lipofuscin (albino WT 59% and pigmented WT 82% of total lipid droplets, Fig. S2). They were found in both apical and basal parts of the RPE cells, with a tendency to be more often apically located (Table 1). In Stargardt mouse strains, lipid droplets were bigger (median diameter: pigmented Abca4^−∕− 555 nm, albino Abca4^−∕− 705 nm, Abca4^−∕−.Rdh8^−∕− 687 nm) and only sporadically seen in the apical part of the RPE (Table 1, Fig. S2). Fusion with lipofuscin was rarely seen in albino Abca4^−∕− mice (3% of total lipid droplets) and absent in pigmented Stargardt mice.

RPE integrity and Bruch’s membrane

Dead RPE cells were regularly encountered in eyes of albino Abca4^−∕− mice of all ages (Fig. 6A, Table 1). In contrast, dead RPE cells were only found in aged Abca4^−∕−.Rdh8^−∕− mice (12 months and older). In pigmented Abca4^−∕− and young Abca4^−∕−.Rdh8^−∕− mice, dead RPE cells were not found. Locally, RPE cells were unusually thin in albino Abca4^−∕− mice (Fig. 1E).

RPE cells regularly contained undegraded phagosomes in the basal area of the cytoplasm in all three Stargardt mouse strains (Table 1). RPE cells, whose cytoplasm was largely occupied by extensive undegraded phagosomes, were found in albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice (Fig. 6B). The highest numbers of undegraded phagosomes were found in Abca4^−∕−.Rdh8^−∕− mice (Table 1).

Irregularly shaped basal infoldings were frequently seen in albino Abca4^−∕− mice, but rarely in pigmented Stargardt mice (Table 1). The irregular shape was due to the basal infoldings expanding to the lateral side of the cells (Fig. 6C). In general, basal infoldings were reduced in number and expansion in all Stargardt mouse strains compared to wild types (Fig. 6D).

Bruch’s membrane became thicker with age in all Stargardt mouse strains (Fig. 7). There was no significant difference between the different strains.

Histopathology and pigment changes in late Stargardt disease

For comparison, we also investigated a single eye of a 72-year-old Stargardt patient (Fig. 8). In this late stage of the disease, photoreceptors were almost completely degenerated resulting in geographic atrophy as is typical for Stargardt patients (Fig. 8A). A reliable quantification of neuroretinal changes was not possible due to the long death-to-fixation time of the donor, but pigments can still be investigated under these conditions. RPE cells were nearly completely filled with pigment granules (Fig. 8B), mostly lipofuscin and melanolipofuscin, melanin granules were barely present. Microvilli and basal infoldings were absent, while Bruch’s membrane was unorganized and contained electron dense deposits (Fig. 8B). Melanolipofuscin formed clusters (Figs. 8C, and 8D), as they were also found in pigmented Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice.

Oxidative stress markers are present in Stargardt mice

Since ultrastructural changes in the RPE of Stargardt mice were indicative of oxidative stress, we tested for oxidative stress markers in 12-month-old mice by immunohistochemistry. We found HNE, MDA and NITT adducts in the RPE cells of pigmented and albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice with signals being generally lowest in pigmented Abca4^−∕− mice (Fig. 9). HNE, MDA and NITT signals were also present in the neuroretina, reaching from the inner limiting membrane to the inner segments, in all Stargardt strains and were highest in Abca4^−∕−.Rdh8^−∕− mice in the case of HNE and MDA or at similar levels in the three strains in case of NITT (Fig. S3).

RPE melanin can quench superoxide

To investigate the capability of RPE melanin to quench superoxide, we employed the colorimetric NBT assay (Cheng et al., 2015). We first tested melanin from S. officinalis as a melanin standard for its quenching capabilities and found a dose–response relationship (Fig. 10A). Subsequently, we investigated RPE melanin isolated from pig eyes and found an approximately 30% reduction of detectable superoxide for 3 ×10⁴ RPE melanin granules utilized in the assay (Fig. 10B).

Among Stargardt mice, albino Abca4^−∕− mice show the earliest onset and most varied retinal damages

It was previously described that albino Abca4^−∕− mice show a mild retinal degeneration that is not present in albino WT mice (Sparrow et al., 2013; Wu, Nagasaki & Sparrow, 2010). We found that absolute photoreceptor nuclei numbers decrease with age, resulting in a steeper decline of photoreceptor nuclei counts compared to pigmented Abca4^−∕− mice (Fig. 1C). By contrast, Abca4^−∕−.Rdh8^−∕− mice are described to have regional retinal degeneration and rosette formation as early as 6 weeks of age and advanced retinal degeneration by 3 months (Maeda et al., 2008). We were unable to reproduce the published results with our colony of Abca4^−∕−.Rdh8^−∕− mice. Our animals had a similar rate of retinal degeneration than the albino Abca4^−∕− mice (Fig. 1C) and rosette formation was only found in one 12-month-old animal from a total of 25 Abca4^−∕−.Rdh8^−∕− animals examined light-microscopically up to an age of 25 months.

In addition to the photoreceptor loss, albino Abca4^−∕− mice show outer segment shortening (Fig. 1B), disordered disk membranes (Fig. 2A), damaged inner segments (Fig. 2B), and morphological changes at the base of the outer segments (Figs. 2C and 3D). These damages were not seen in pigmented Abca4^−∕− and to a much lesser extent in Abca4^−∕−.Rdh8^−∕− mice (Table 1). Shortening and disorganization of outer and inner segments was also described after conditional knockout of superoxide dismutase in the RPE, resulting in oxidative stress (Mao et al., 2014).

In Stargardt patients, the junction between the inner and outer segments was described to be unorganized and/or lost on SD-OCT (Gomes et al., 2009). Furthermore, photoreceptor anomalies were found in patients that did not display equivalent fundus autofluorescence anomalies (Gomes et al., 2009). Damaged inner segments and defective outer segment synthesis in albino Abca4^−∕− mice (Figs. 2B–2D), but not in pigmented Stargardt mouse strains, might be histological representations of the SD-OCT findings in Stargardt patients and needs further investigation.

Outer segment renewal starts at the base of the outer segment, so the disks at the tips are the oldest disks in a photoreceptor and are affected the longest by any disturbance, e.g., accumulation of retinoids due to lack of ABCA4. Loss of fundamental outer segment proteins like rhodopsin (Lem et al., 1999) or GARP (Huttl et al., 2005) leads to a severe outer segment damage starting early in life (reviewed by (Goldberg, Moritz & Williams, 2016)). ABCA4 is located in the rims of disk membranes (Papermaster et al., 1978) and does not form stable interactions with the peripherin-2⋅GARP complex (Poetsch, Molday & Molday, 2001), which is essential for linking disk rims to the outer segment plasma membrane and disk-disk stacking (Kaupp & Seifert, 2002). These findings, together with the lack of outer segment disk pathology in pigmented Abca4^−∕− mice (Table 1) (Charbel Issa et al., 2013), indicate that ABCA4 is not needed for proper disk formation, orientation and stability. Whether RPE malfunction is responsible for the presented outer segment defects in albino Abca4^−∕− and to a lesser extent Abca4^−∕−.Rdh8^−∕− mice needs to be elucidated.

Indications for lysosomal dysfunction and oxidative stress in albino Abca4^−∕− mice

Formation of vacuole-like structures in the RPE as seen in albino Abca4^−∕− mice (Fig. 3) has been described after lysosomal integrity has been compromised and after oxidative stress. Increasing the lysosomal pH with chloroquine (Peters et al., 2006), inhibiting lysosomal cysteine proteases (Okubo et al., 2000) and a lack of functional βA3/A1-crystallin (Valapala et al., 2014; Zigler Jr et al., 2011) all lead to formation of vacuole-like structures within the RPE cell layer. Lysosomal pH was shown to be increased after treatment of RPE cells with the lipofuscin compound A2E in vitro (Holz et al., 1999; Liu et al., 2008) and in primary RPE cells isolated from pigmented Abca4^−∕− mice (Liu et al., 2008). Since the melanosome is a specialized lysosome (Orlow, 1995) that contains many lysosomal enzymes (Diment et al., 1995) and can fuse with phagosomes (Schraermeyer & Heimann, 1999), it is possible that albinism itself adds to the lysosomal dysfunction seen in albino Abca4^−∕− mice due to lack of melanosomes.

Similar pathologies were also described after oxidative stress. Changes in RPE morphology, such as vacuolization and loss of basal labyrinth, and deposit formation were seen after a lack of superoxide dismutase (Mao et al., 2014; Seo et al., 2012), knockout of Nrf2, a transcription factor that regulates genes involved in the cellular oxidative stress response (Zhao et al., 2011), and after chronic exposure to cigarette smoke (Fujihara et al., 2008). In some models, thinning of the outer nuclear layer (Mao et al., 2014; Seo et al., 2012) and shortening and disorganization of inner and outer segments (Mao et al., 2014) or RPE hyper- and hypopigmentation and choroidal neovascularization (Zhao et al., 2011) were also described. Besides the high oxygen partial pressure and the presence of light, other sources for oxidative stress in the RPE are phagocytosis of photoreceptor outer segments (Miceli, Liles & Newsome, 1994) and lipofuscin (Rozanowska et al., 1995).

Previous studies investigating oxidative stress in Stargardt mice found that albino Abca4^−∕− mice have higher levels of MDA and HNE adducts than age-matched albino WT and pigmented Abca4^−∕− mice (Radu et al., 2011) and that 6-month-old Abca4^−∕−.Rdh8^−∕− mice have higher levels of 8-OHdG than 4-week-old Abca4^−∕−.Rdh8^−∕− mice and age-matched WT mice (Sawada et al., 2014). We found similar results indicating that pigmented Abca4^−∕− mice have the lowest oxidative stress levels (Fig. 9).

Interestingly, Abca4^−∕−.Rdh8^−∕− mice had higher numbers of undegraded phagocytosed outer segments than albino Abca4^−∕−mice (Table 1), but albino Abca4^−∕− mice had higher numbers of vacuole-like structures (Fig. 3D). One reason might be higher oxidative stress levels in albino Abca4^−∕− mice. Melanin was shown to be a scavenger of superoxide anion and singlet oxygen (Bustamante et al., 1993; Tada, Kohno & Niwano, 2010) and protects human RPE cells from oxidative stress in vitro (Burke et al., 2011). Intravitreal application of paraquat, a superoxide anion generator, leads to higher superoxide levels and more pronounced lipid oxidation and retinal degeneration in albino compared to pigmented mice (Cingolani et al., 2006).

Additionally, changes of RPE morphology in albino animals might play a role in RPE vacuolization: embryonic RPE from albino mice was found to have an irregular morphology, compared to pigmented littermates (Iwai-Takekoshi et al., 2016). Furthermore, an altered distribution of the junctional protein P-cadherin was found in albino RPE and it was hypothesized that this might lead to defects in cell–cell adhesion (Iwai-Takekoshi et al., 2016). It is reasonable that a combination of weakened cell–cell contacts, impaired lysosomal function and elevated oxidative stress as described above might lead to the defects found in albino Abca4^−∕− mice.

Lipofuscin varies considerably between Stargardt mouse strains

Morphological appearance of lipofuscin granules differs in the three different Stargardt mouse strains. It has been described that the chemical composition of lipofuscin in pigmented and albino WT mice differs in terms of the amounts of the different types of bisretinoids that can be found (Ueda et al., 2016). It is hypothesized that the higher intraocular light levels in albino animals lead to photooxidation and photodegradation of A2E and other bisretinoids (Ueda et al., 2016), which is in accordance with earlier publications that demonstrated lower A2E and higher A2E oxirane levels in albino Abca4^−∕− mice compared to pigmented Abca4^−∕− mice (Radu et al., 2004). In Abca4^−∕−.Rdh8^−∕− mice, higher amounts of all-trans retinal dimer can be found than in Abca4^−∕− mice (Maeda et al., 2008). Our results show that in addition to the difference in the chemical composition of lipofuscin between the three strains, there is also a considerable histological difference.

Strikingly, lipofuscin levels in 12-month-old animals seem to be elevated in albino Abca4^−∕−compared to pigmented Abca4^−∕−mice, when investigated by lipofuscin autofluorescence (Fig. 4A), while quantification of RPE cytoplasm occupied by lipofuscin granules shows similar lipofuscin levels in 12-month-old pigmented and albino Abca4^−∕−mice (Fig. 5). Both quenching effects of the melanin and differences in nature and amount of present fluorophores, as described above, are possible explanations for this seeming contradiction.

The Met450 variant of RPE65 was described to be associated with lower levels of the lipofuscin fluorophores A2E and iso-A2E in Abca4^−∕−.Rdh8^−∕− mice (Kim et al., 2004). Since pigmented and albino Abca4^−∕− mice carry the Leu450 variant of RPE65 and we did not genotype the Abca4^−∕−.Rdh8^−∕− mice for RPE65, lipofuscin levels in Figs. 4 and 5 in Abca4^−∕−.Rdh8^−∕− mice might be underreported.

The bisretinoid A2E was shown to have surfactant-like properties and to be able to compromise lysosomal and cellular membranes (Schutt et al., 2002; Sparrow et al., 1999). In fact, we found that lipofuscin granules often lacked membranes (Fig. 4C). Furthermore, in very old animals (aged 23 months and older), we found abnormally electron dense RPE cytoplasm (Fig. 4D), making distinguishing between lipofuscin granules and cytoplasm very difficult. Whether the RPE cytoplasm gained electron density due to free lipofuscin components that leaked from lipofuscin granules after losing their membranes is unclear but appears likely.

Bruch’s membrane thickness does not vary between Stargardt mouse strains

For all three Stargardt mouse strains, a thickening of Bruch’s membrane compared to WT was described (Maeda et al., 2008; Radu et al., 2011; Weng et al., 1999). Compared to each other, the three strains show comparable thickening of Bruch’s membrane with age (Fig. 7). An increase in the thickness of Bruch’s membrane is also typical in Stargardt patients (Bonilha et al., 2016).

The significance of RPE melanin and its role as a scavenger of oxidative stress

The role of reactive oxygen species in retinal health was extensively investigated (Becquet et al., 1994; Cingolani et al., 2006; Mao et al., 2014; Seo et al., 2012; Zhao et al., 2011) and melanin was identified as a potential scavenger of reactive oxygen species (Bustamante et al., 1993; Tada, Kohno & Niwano, 2010). However, RPE melanin differs from other melanin species in several aspects. For instance, while melanocytes continuously synthesize new melanin, it is still under debate whether melanin turnover in the RPE occurs and if so, to what extent (Schraermeyer, 1993). Furthermore, RPE melanin can react differently than melanin in the melanocytes of the choroid (Schraermeyer & Heimann, 1999). Therefore, we analyzed isolated RPE melanin granules to corroborate that also this melanin species is capable to quench superoxide radicals and in fact can play a role in protection against oxidative stress in the RPE (Fig. 10). Anti-oxidative properties of RPE melanin have also been confirmed in cell culture experiments (Burke et al., 2011).

Certain photoreceptor and RPE-damages typical for oxidative stress and/or lysosomal impairment were exclusively found in albino Abca4^−∕− mice (Figs. 1D, 2B–2D, 3, 6C), while others were found in albino Abca4^−∕− and Abca4^−∕−.Rdh8^−∕− mice, but onset was earlier in albino Abca4^−∕− mice (Figs. 1A–1B, 2A, 6A & 6D). Notably, even though Abca4^−∕−.Rdh8^−∕− mice have a more severe disease-causing genotype compared to albino Abca4^−∕− mice, they only show a more pathologic phenotype regarding RPE hypertrophy (Fig. 1A, Fig. S1), the number of undegraded phagosomes (Table 1) and the percentage of RPE cytoplasm occupied by lipofuscin (Fig. 5).