1. Introduction

Adaptive optics scanning light ophthalmoscopy (AOSLO) has provided clinicians and scientists a tool for visualizing the living human retina with single-cell resolution and investigating retinal disease mechanisms [1–3]. Adaptive optics (AO) imaging technology, which is a system for measuring and compensating for optical imperfections in an imaging medium, has been transformative in providing improved resolution and contrast for imaging in the living human eye with cellular level detail [4,5]. Confocal reflectance imaging, a technique analogous to confocal microscopy, involves scanning a focused illumination beam of light and collecting the backscattered light, using a pinhole detector to select light from within the focus. In non-confocal split detection imaging (“split detection”), non-confocal light is divided onto multiple detectors [6]. Increasingly, confocal reflectance and split detection are implemented together and simultaneously collected [7].

Visualization of the different retinal cells in human eyes using AOSLO depends not only on the optical properties of the imaging system but also on the size and light scattering properties of the cell type being imaged. While confocal reflectance AOSLO can readily visualize structures such as the nerve fiber layer [8], vasculature [9–11], and photoreceptors [12], other cells such as retinal ganglion cells [13], bipolar cells, and photoreceptor cell nuclei are nearly transparent and have not yet been revealed using confocal reflectance AOSLO imaging. Non-confocal reflectance imaging allows for visualization of structures such as the retinal ganglion cells [14], vasculature [15] and photoreceptor inner segments [6,16]. In addition, many in vivo adaptive optics imaging techniques utilizing intrinsic or administered fluorophores have provided additional insight into the structure of the photoreceptor, retinal pigment epithelium (RPE), and nerve fiber layers [17–20]. While these in vivo techniques provide insight into cellular status within living retina, factors such as eye motion, numerical aperture (NA) of the eye, optical properties of the cornea, lens, and vitreous, as well as light safety considerations make anatomical investigation of living retina with sub-micron resolution challenging.

Histological studies using microscopy techniques such as Nomarski differential interference contrast microscopy (NDIC) have enabled high resolution investigation of structures in the ex vivo human and non-human primate retina. For example, these techniques helped characterize photoreceptor size and spacing [21], relative size and distribution of different cone populations [22], and cone inner segment structure and composition [23]. However, comparisons between histology and AOSLO are confounded by differences between the individual imaging modalities, as split detection is not typically used for histology, and conversely, NDIC is not currently used for AOSLO imaging. In a few studies, the retinal tissue was first imaged in vivo using AOSLO and then subsequently compared to histological images of the same tissue ex vivo [24–26]. These studies have been valuable for validating our interpretation of in vivo AOSLO imaging.

In this work, we present a translational imaging framework, an AO microscopy module integrated with an existing AOSLO, which enables protein specific fluorescence microscopy alongside simultaneous AOSLO imaging modalities. This work builds off previous work [19] examining fluorescent labeling in vivo and helps provide additional functionality for examining ex vivo samples. The simultaneous examination of ex vivo samples using AO microscopy and native AOSLO modalities facilitates comparisons by allowing the same cells to be directly compared across modalities. We build upon an earlier version of the translational imaging framework (introduced as a system for imaging in vivo and ex vivo tissues using shared illumination and detection optics [19]) with an improved microscope design and dedicated optical path. The addition of objective mounting and sample holding optomechanics enable more flexibility in sample imaging with greater stability and higher resolution. The new AO microscope visualizes retinal layers at sub-micron resolution using AOSLO imaging modalities while also enabling simultaneous visualization of fluorescently labeled structures.

2. Methods

2.1 Instrumentation

A custom-built multimodal AOSLO imaging device [16,27] was modified to include a custom AO microscopy module (Fig. 1). This existing device, and by extension the imaging framework, uses a 790 nm superluminescent diode (SLD) (S-790-G-I-15-M, Superlum, Carrigtwohill, Co. Cork, Ireland) for reflectance and fluorescence imaging. The full width at half maximum (FWHM) bandwidth of the source was 16 nm, reduced to 15 nm with a clean-up spectral filter (ET775/50x, Chroma, Bellow Falls, VT, USA). For wavefront sensing, an 880 nm SLD was used (SLD-mCS-341-HP1-SM-880, Superlum, Carrigtwohill, Co. Cork, Ireland). This light source has a FWHM bandwidth of 46 nm, reduced to 20 nm with a clean-up spectral filter (FF01-900/32, Semrock, Rochester, NY, USA). The optical power of these sources at the cornea and sample was kept below the maximum permissible exposure limit established by the American National Standards Institute standard Z136.1-2014. Beam raster scanning was achieved using a ∼15 kHz resonant scanner (SC-30, Electro-Optical Products Corp, Fresh Meadows, NY, USA) and a vertical tip-tilt scanner (S-334, PI-USA, Auburn, MA, USA). Spherical mirror telescope design and adaptive optics hardware were based on previously published work [28,29].

 

Fig. 1. The translational imaging framework consists of a multimodal adaptive optics scanning light ophthalmoscope (AOSLO) integrated with an adaptive optics (AO) microscopy module. (a) Simplified system diagram of multimodal AOSLO and AO microscopy module. This module can be integrated with an existing ophthalmoscope to enable protein specific fluorescence microscopy alongside AOSLO reflectance imaging modalities. Abbreviations: AM: reflective annular mask; BS: 80/20 (transmission/reflection) beamsplitter; D: dichroic beamsplitter; DM: deformable mirror; F: optical filter; P: pinhole; RS: resonant scanning mirror (horizontal); SHWS: Shack-Hartmann wavefront sensor; SM: spherical mirror; TT: tip/tilt scanning mirror (vertical). The DM, TT, RS, and SHWS are all conjugate to the pupil plane. Darkfield and split detection images are generated through summation and normalized subtraction of the two split-detection PMTs (Split PMT 1 and 2). (b) and (c) show a diagram and photo of the AO microscopy module, respectively. (a) is adapted from Bower [27].

Download Full Size | PDF

Backscattered 790 nm light returning from the eye or sample was separated using a dichroic filter (T810lpxr, Chroma, Bellows Falls, VT, USA). An annular reflective mask was used to reflect the confocal reflectance light and transmit non-confocal split detection light as previously described [6,16]. Confocal reflectance light was further spatially filtered by a pinhole and collected with a photomultiplier tube (PMT) (H7422 series, Hamamatsu Corporation, Hamamatsu, Japan). Split detection light was divided using a D-shaped mirror and collected with two additional PMTs (Split PMT 1 and 2 in Fig. 1(a)). Fluorescence emission light between 810-850 nm was separated with a custom dichroic filter (Semrock, Rochester, NY, USA). This light was passed through an emission filter (FF01-832/37, Semrock, Rochester, NY, USA), spatially filtered by a pinhole, and collected using a PMT.

A custom optical path was constructed to connect the AO microcopy module to the existing AOSLO device. A removable fold mirror was inserted into the AOSLO beam path between SM7 and SM8 to switch between the two devices. A second fold mirror steered the beam to a duplicate SM8 and the AO microscopy module. To allow for direct sample access without use of a coverslip, this system utilized a 40X/0.80 NA water dipping objective (N40X-NIR, Nikon, Melville, NY, USA) for light collection instead of a water immersion objective [30,31]. This objective was chosen to maximize the usable NA, given the beam diameter at the pupil plane, to minimize changes to the optical system, and to avoid back reflection from the use of a coverslip. The system structure was constructed using Thorlabs CERNA microscopy components: this includes the microscope body (CFB15000, Thorlabs, Newton, NJ, USA), fast XY scanning stage (MLS203-1, Thorlabs, Newton, NJ, USA), motorized objective z-stage (ZFM2020, Thorlabs, Newton, NJ, USA), and slide holder (MLS203P2, Thorlabs, Newton, NJ, USA). Objective and sample stage scanning was managed digitally using corresponding controllers (MCM3001, BBD202, Thorlabs, Newton, NJ, USA).

2.2 Tissue and sample preparation

Human Tissues. Human donor eyes were obtained from Lions Eye Bank (Calgary, Alberta, Canada). Globes were collected 3 hours postmortem and fixed by immersion in 10% neutral buffered formalin then transferred for shipment in balanced salt solution. Globes were dissected to isolate the retina and 4-5 mm diameter trephine punches were taken centered at 5.5 mm temporal to the fovea. This region was selected to examine rods and larger peripheral cones. The retina was gently detached from the RPE with fine forceps. Samples were stored in 1x PBS containing 0.05% sodium azide.

Animal Tissues. Eyes from rhesus monkeys (Macaca mulatta) were obtained courtesy of the Center for Biologics Research and Testing, U.S. Food and Drug Administration (Silver Spring, MD, USA). Globes were removed 30 minutes postmortem and fixed by immersion in 4% paraformaldehyde in 1x PBS. Globes were dissected to isolate the retina and 5 mm diameter trephine punches were taken with the punch centered at 5.5 mm temporal to the fovea. This region was selected to examine rods and larger peripheral cones. The retina was gently detached from the RPE with fine forceps. Samples were stored in 1x PBS containing 0.05% sodium azide.

Fluorescent labeling. Before fluorescent labeling, ex vivo retina samples were permeabilized and blocked with a solution of 0.1% Triton X-100 and 1% BSA in PBS for 90 minutes at 4° C and washed three times with PBS. For f-actin visualization, a 1000X phalloidin DMSO stock solution was prepared (Phalloidin-iFluor 790 Conjugate #20559, Cayman Chemical, Ann Arbor, MI, USA). From this stock solution a working solution was prepared (1:1000 dilution with 1% BSA in PBS). The sample was incubated in the working solution for 90 minutes at room temperature and washed three times with PBS. For S cone visualization, samples were incubated with a primary anti S-opsin rabbit polyclonal antibody (1:200) (Millipore Sigma AB5407, Lot 2775513, Darmstadt, Germany). Following incubation in this primary antibody solution for 90 minutes at room temperature, samples were washed 3 times with PBS, and then incubated with Alexa Fluor 790 Goat anti-rabbit polyclonal antibody (1:200) (Thermo Fisher Scientific Invitrogen A11369, Lot UG280689, Waltham, MA, USA) as the secondary antibody for 90 minutes at room temperature. Finally, samples were washed three times with PBS.

Empirical Measurements of Resolution. Polystyrene fluorescent beads with a diameter of 0.2 µm (NIRex 4002, Phosphorex, Hopkinton, MA, USA) were imaged to characterize system resolution [16]. Beads were diluted with distilled water (1:700), mixed thoroughly, and a few drops transferred to filter paper. After drying, the filter paper was imaged under a thin layer of distilled water. At each location, 150 frame acquisitions were captured of a 1.5° square field of view. After averaging the frames, the lateral FWHM of 15 beads was measured and averaged.

2.3 Imaging

Multimodal AOSLO. For in vivo imaging of a human subject, one 39-year-old female subject was recruited from the National Eye Institute eye clinic for this study (NCT02317328; https://clinicaltrials.gov). This study was approved by the Institutional Review Board of the National Institutes of Health. Research procedures adhered to the tenets of the Declaration of Helsinki. Written, informed consent was obtained after the nature of the research and possible consequences were explained. Ocular biometry was performed after dilation (IOL Master, Carl Zeiss Meditec, Dublin, CA, USA) and used to calculate the scaling factor to convert retinal degrees to mm, as described previously [32]. After dilation with 2.5% phenylephrine hydrochloride (Akorn Inc, Lake Forest, IL, USA) and 1% tropicamide (Sandoz, Novartis, Basel, Switzerland), sub-diffraction limited imaging was performed using a custom-built AOSLO instrument (Fig. 1) as described previously [16]. Illumination light power levels measured prior to imaging were below 135 µW for 790 nm and below 43 µW for 880 nm.

AO Microscope. For sample imaging, retinal samples were placed in 35 mm cell culture dishes and immersed in PBS. To reduce sample motion, tissue was restrained with a tissue slice anchor (Warner Instruments, Holliston, MA, USA). Immediately prior to sample imaging, an AO correction was obtained using a paper test sample to reduce system aberration and astigmatism. This static correction was used throughout imaging.

Image Acquisition and Processing. Images were acquired using custom acquisition software [6]. 5–13 second videos were acquired at a frame rate of 16.7 Hz. Videos were acquired at a field of view of either 56 × 53 µm or 84 × 80 µm (750 × 605 pixels). Videos were processed to adjust for the impact of sinusoidal illumination scanning, corrected for eye motion based on the eye motion correction computed from one of the simultaneously acquired nonfluorescent channels [33], and averaged to form a composite image. Image processing was completed with custom analysis software [6] and ImageJ (ImageJ, National Institutes of Health). Cone segmentation was performed with custom segmentation software [32,34] and cone boundaries were analyzed with custom analysis software.

3. Results

3.1 Translational imaging framework provides enhanced resolution

The translational imaging framework (Fig. 1) provides an opportunity to assess ex vivo retinal samples using a nearly identical optical system as that used for in vivo imaging of human subjects [6,28,35]. Empirical measurements using sub-diffraction near-infrared (NIR) fluorescent beads showed that the AO microscope achieved a resolution of 0.770 +/- 0.075 µm (mean +/- SD; n = 15 beads). This value can be compared with the expected theoretical resolution of 0.633 µm, calculated using the Rayleigh criterion (Eq. (1)) assuming a central wavelength of 830 nm for ICG fluorescence [19].

In comparison, empirical measurements of the resolution of the base AOSLO imaging system performed using NIR fluorescent beads (measured using an f = 19 mm lens) has been previously reported to be 2.36 µm [16]. Thus, the AO microscope has ∼3X improved resolution compared to the base AOSLO imaging system.

 

Fig. 2. Multimodal imaging of retina layers in an ex vivo fixed human sample using the AO microscope. Each row represents a set of simultaneously captured images from different planes of focus across a 63 µm range. Shown are examples of nerve fiber bundles (a-c), ganglion cells somata (d-f), outlines of cells near the inner nuclear layer (g-i), and rod and cone photoreceptors (j-l). Cyan arrows in (d-f) denote ganglion cell soma. Cyan arrows in (j-l) mark cone photoreceptor cells surrounded by the smaller rod photoreceptor mosaic. Images were acquired from the vitreal side of the retina, the same directionality corresponding to in vivo clinical imaging. Images are contrast adjusted for visualization purposes. Scale bar, 10 µm.

Download Full Size | PDF

3.2 Visualization of retinal layers in ex vivo samples using intrinsic contrast

The translational imaging framework enabled visualization of a variety of cell types using modalities native to AOSLO imaging. In the absence of eye motion, an inherently registered through-focus imaging stack could be readily acquired through different retinal layers from ex vivo fixed human tissue, with simultaneous acquisition of confocal reflectance, split detection, and darkfield modalities at each focal plane. Images in (Fig. 2) represent en face views of the same region of retina over an axial depth of 63 µm, including layers of the retina that are nearly transparent in vivo. Overall, the confocal reflectance images revealed axon bundles within the nerve fiber layer, capillaries, and photoreceptors, similar to previously-reported in vivo imaging [3], but with improved resolution.

While ganglion cell somata are nearly transparent in vivo when imaged using confocal reflectance or split detection, in the images from the AO microscopy module, ganglion cell somata are visible in the confocal reflectance channel (e.g. cyan arrows in Fig. 2(d)-(f)). The increased contrast of these cells may be due to the scattering properties being altered during the fixation process [36]. It is also possible that these somata are visible due to other factors such as differences in lateral resolution or image contrast due to the relative ease of averaging images at any arbitrary depth location without needing to take into consideration sample or eye motion. A negative mosaic of cells near the inner nuclear layers was observed in Fig. 2(g)-(i), where bipolar and horizontal cells are expected to reside. In Fig. 2(k), split detection imaging shows the rod inner segment mosaic, which is typically not visible in split detection without enhanced resolution [16]. In these images, cones can be seen in the darkfield image, but are more obscured in confocal reflectance and split detection images (e.g. cyan arrows in Fig. 2(j)-(l)).

 

Fig. 3. Comparison of multimodal imaging of photoreceptors acquired in the living human eye and in ex vivo samples using the translational imaging framework. (a-c) In vivo, sub-diffraction limit image of cone and rod photoreceptors from a human subject, obtained using annular pupil illumination and sub-Airy disk diameter confocal pinhole detection. (d-f) Images of ex vivo fixed human retina imaged from the scleral side. (g-i) Images of ex vivo fixed macaque retina imaged from the scleral side. Images in (d-i) were acquired using the AO microscopy module. In all cases, confocal reflectance, split detection, and darkfield modalities were simultaneous acquired. Cyan arrows denote examples of cone photoreceptor cells. Images are contrast adjusted for visualization purposes. Scale bar for all images, 20 µm.

Download Full Size | PDF

Whereas in vivo images from living human subjects are generally obtained with the illumination light approaching the retina from the anterior side of the retina (i.e. from the vitreal side), the AO microscopy module also enables the retina to be directly imaged from the posterior side (i.e. illumination from the scleral side accomplished by inverting the sample), which might be helpful for imaging photoreceptors in the absence of scattering from other layers. Images of the cone photoreceptors obtained using the AO microscopy module revealed subcellular features within cone photoreceptors (Fig. 3(d)-(i)), most visible from the scleral side, and similar to those observed in vivo using sub-diffraction limit resolution [16]. These features were also observed in the darkfield channels. Confocal reflectance and split detection imaging of the macaque sample (Fig. 3(g),(h)) provided an even clearer view of these features. The rod photoreceptor mosaic, which can be observed as small white dots in the confocal image and corresponding dots in the split detection image of Fig. 3(a),(b), is visible in Fig. 3(d),(g) confocal images but is much easier to discern in the split detection images (Fig. 3(e),(h)). In summary, the AO microscopy module enables direct comparison between histological images and in vivo human imaging using identical modalities, imaged with increased magnification and resolution to reveal additional structure and detail not readily visible in in vivo imaging.

3.3 Comparison of fluorescently labeled actin with photoreceptor structure

One key advantage of the translational imaging framework is the ability to image fluorescent labels alongside conventional AOSLO imaging modalities. This allows for direct comparison between intrinsic reflective contrast and fluorescently labeled structures. Here, we demonstrate the imaging of fluorescently labeled actin, acquired across different focal planes alongside the non-fluorescence modalities (confocal reflectance and split detection). The outer limiting membrane (OLM, or external limiting membrane, ELM), is a network of junctional complexes between Müller cells and the photoreceptors they encircle that contributes to barrier functions in the neural retina. The distribution of actin along the junctions of the OLM [38] makes actin a convenient surrogate for visualizing the OLM. To examine this structure we used phalloidin, which localizes to filamentous actin and has been shown to be useful in visualizing OLM in fixed rat and macaque retina samples [39].

The OLM was successfully visualized in the fluorescence imaging channel (Fig. 4(d)), within which a negative mosaic of cones (larger circles) and rods (smaller circles) could be readily observed. The phalloidin-labeled bands encircling cones colocalized with the darker rings that surround the cones in darkfield imaging (visible in Fig. 4(c)). In addition, we detected additional actin structures in layers posterior to the OLM (Fig. 4(h),(l)). These structures, which seem to be colocalized to individual photoreceptors, are similar to the findings of Omri [39]. At the plane of best photoreceptor focus for split detection and darkfield imaging (Fig. 4(j),(k)), there is fluorescence signal observed in the centers of both rod and cone photoreceptors. As we described earlier (Section 3.2), due to the illumination direction (through the inner retina), the confocal reflectance channel contrast is reduced. With the caveat that fixation has been associated with shrinkage of retinal thickness in the photoreceptor layer [36], these image stacks suggest an axial distance of approximately 7 µm between OLM and photoreceptor plane of best focus as determined using split detection. This application, enabled by protein specific fluorescent labeling, provides a valuable benchmark for assessing the focal planes at which cone photoreceptors are often imaged in relation to the OLM and may be useful for informing future in vivo imaging explorations.

 

Fig. 4. Multimodal imaging of f-actin and photoreceptors in an ex vivo fixed human sample. Fluorescent labeling with Alexa Fluor 790 phalloidin enables actin visualization using the NIR detector used for indocyanine green imaging [37]. Each row shows selected layers of a z-stack captured at the different planes throughout the photoreceptor layer. (a-d) Images acquired at the plane of best focus for OLM fluorescence. (e-h) Images acquired at an intermediate plane (4 µm deeper than a-d). (i-l) Images acquired at the plane of best photoreceptor focus split detection imaging (7 µm deeper than a-d). Images were acquired from the vitreal side of the retina, the same directionality corresponding to in vivo clinical imaging. Images are contrast adjusted for visualization purposes. Scale bar, 10 µm.

Download Full Size | PDF

3.4 Assessing S cone photoreceptor structure using immunofluorescence microscopy

Immunofluorescent staining has proven useful for specifically labeling opsins in canine and human retinal samples [22,40]. In this section, we demonstrate the combination of immunofluorescent labeling, using Alexa Fluor 790-conjugated antibodies against s-opsin to investigate the structure of blue cones (S cones) observed in reflectance imaging modes that are native to the base AOSLO.

Figure 5 shows multimodal imaging of a fixed macaque sample with immunofluorescent labeling of s-opsin. The S cones can be identified in the fluorescence imaging channel, which are colocalized with the simultaneously acquired reflectance channels. This approach reveals the appearance and characteristics of S cones at high resolution in the reflective imaging modalities. The focal planes in Fig. 5, which were also used for cone segmentation analysis, represent the plane of best photoreceptor focus in split detection. By using custom software to manually segment boundaries [32,34], we quantified both the outer diameter of cones and the inner core diameter (defined as the diameter of a circle with equivalent area) (Fig. 5(g)). On average, the diameter of the inner core was 46% of the outer cone diameter (inner: 3.76 µm, outer: 8.22 µm). To further explore whether these structures varied in size between different types of cones, we compared S cones to L/M cones. Measurements from 8 S cones and 54 L/M cones revealed that the outer diameter of S cones was 15.8% smaller (S cones: 7.06 +/- 0.51 µm, L/M cones: 8.38 +/- 0.47 µm, p < 0.001, two-tailed t-test). Similarly, when comparing the inner cores of the same S, L, and M cones, S cones were 19.2% smaller (S cones: 3.11 +/- 0.25 µm, L/M cones: 3.85 +/- 0.30 µm, p < 0.001, two-tailed t-test; mean +/- SD for all values). Overall, these results demonstrate an application for investigating the subcellular structure of cone photoreceptors alongside fluorescent labeling.

 

Fig. 5. Multimodal imaging of photoreceptors with immunofluorescent labels in an ex vivo fixed macaque sample. (a-f) Multimodal images of photoreceptors, including immunofluorescent labeling of S cones. Confocal reflectance and split detection images were simultaneously acquired alongside the fluorescence images. S cones are identified with blue arrows. (g) 2X zoom of green squares in d-f, with dotted lines to denote inner core diameters and solid lines to denote outer diameters. Measurements of cone diameters reveal differences in the distributions of S cone and L/M cone size in both cone inner core diameters (dotted curve) and outer diameters (solid curve) (h) (n= 8 S cones and 54 L/M cones). Images are contrast adjusted for visualization purposes. Scale bar, 10 µm.

Download Full Size | PDF

4. Discussion

We introduce a translational imaging framework that enables the visualization of retinal cells with the same imaging modalities used in many in vivo AOSLO imaging systems, but with resolution that exceeds the current limits achievable in ophthalmic imaging systems [16]. These images may aid in identifying subcellular features that are good targets for in vivo imaging. We demonstrate the value of using protein-specific fluorescence imaging to help improve interpretation of in vivo images. The overall approach can be implemented in most existing AOSLO imaging systems. However, there are some important considerations.

Depending on the imaging applications, selection of an appropriate microscope objective may differ. While even better resolution could be achieved through selection of a higher NA objective, the similarity in cellular-level features between in vivo and ex vivo images may start to diverge at higher resolutions. For our instrument, the choice of microscope objective was constrained by the beam diameter at the pupil plane; to maximize the usable NA and minimize changes to the optical system, we selected an objective with a matching back focal plane diameter. While this increased NA provides additional imaging resolution, light from the same Airy disk diameter is captured in the split detection channel regardless of the objective NA. However, it is possible that an increase in illumination NA changes the split detection contrast. In addition, the use of a water dipping objective was convenient, as it reduced back reflections (e.g. from index of refraction mismatches) which could interfere with non-fluorescent imaging as well as with our current implementation of Shack-Hartmann wavefront sensing. Despite these design constraints, this implementation of the translational imaging framework provided three-fold improved resolution over what is currently achievable using in vivo imaging, as well as precise and stable selection of imaging focal planes which can be visualized with shorter acquisition times due to the absence of factors related to imaging in the living human eye (e.g. eye movement, cataract, or other optical characteristics of the eye).

The potential for artifacts to be introduced by the fixation process is common to most histological studies. Fixation can cause decreased transparency of retinal layers (increase scattering), as well as tissue shrinkage or distortion. In general, the integrity of structures in the retina is best maintained by minimizing the postmortem to fixation interval of samples used for study, notable in the improved sample quality in macaque vs. human eyes in our data. While fixation may introduce artifacts, this is common to conventional histological studies. In addition to fixation, there is a potential for artifacts due to physical disturbance by sample dissection and handling. For example, bacillary layer detachment is sometimes observed in histologic studies of postmortem retinal samples [41] and may affect photoreceptor imaging. Furthermore, the contrast and noise characteristics for both confocal and non-confocal reflective modalities may be altered by the presence or absence of other retinal layers, such as the RPE or choroid. One key advantage of our approach is the direct use of AOSLO imaging modalities for histological samples, which helps to maximize the fidelity between histology and AOSLO images. Our evaluation of S cones alongside AOSLO imaging modalities provides insight into the overall diameter of the S cone inner segment and may be useful in identifying subtle distinguishing features between cone photoreceptor populations. This approach complements the growing list of techniques for classifying cone photoreceptor types in the living human eye [42–45].

We introduce a translational imaging framework for visualizing the retina with the same imaging modalities as those used for in vivo AOSLO imaging, but with improved resolution. As a further extension of this approach, simultaneous imaging of fluorescently labeled structures alongside reflectance imaging facilitates the exploration of cells across different layers of the retina, including those that are nearly transparent in vivo.

5. Conclusions

While the current state-of-the-art AOSLO imaging already provides single-cell resolution in the living human retina, the translational imaging framework provides improved subcellular resolution to ex vivo retinal imaging. As advances in optical engineering and instrumentation continue to result in the development of new modalities, the translational imaging framework can be leveraged to help validate these new technological advances in a more controlled imaging setting free of patient-related factors (e.g. eye motion). At the same time, the ability to validate the interpretation of new imaging findings observed in both healthy and diseased eyes by replicating imaging conditions from existing histologic samples could prove transformative in retinal disease research by helping to establish how distressed cells appear on AOSLO imaging, both in vivo and ex vivo.