Toward Organelle Motility as a Functional Biomarker of Retinal Pigment Epithelium Health

Abstract

Purpose

The purpose of this study was to quantify characteristics of retinal pigment epithelium (RPE) organelle motility in human subjects across the temporal (T) macula.

Methods

In 11 healthy volunteers (HVs), we used high-speed adaptive optics-optical coherence tomography (AO-OCT) to image RPE cells from the fovea to 12°T. Organelle motility was calculated in the apical, middle, and basal layers as the time constant (τ) of the exponentially decaying speckle decorrelation function normalized to the photoreceptor layer. We imaged 7 HVs on 3 different days at the fovea and 7.5°T to characterize motility repeatability. We imaged 4 HVs on 2 systems with 830 nm and 1060 nm source wavelengths at 7.5°T to explore motility wavelength dependence.

Results

Across the temporal macula, τ decreased from 15.3 (±2.4) to 10.9 (±2.7) seconds (−0.33 seconds/degrees) for the apical RPE, decreased from 9.4 (±1.8) to 8.5 (±1.9) seconds (−0.11 seconds/degrees) for the middle RPE, and was nearly constant (−0.05 seconds/degrees) at 6.7 (±1.0) seconds for the basal RPE at 1060 nm. Motility repeatability was moderate (coefficient of variance [CV] = 10.7% for all subjects and locations). At all RPE layers, τ measured shorter at 830 nm (apical = 6.8 ± 2.2, middle = 4.8 ± 1.0, and basal = 3.8 ± 0.7 seconds) compared with 1060 nm (apical = 10.4 ± 2.8, middle = 6.5 ± 1.6, and basal = 5.0 ± 1.1 seconds, P < 0.005).

Conclusions

RPE organelle motility varies across the macula and cell depth. Our normative dataset will aid future studies of RPE-related retinal pathology.

Translational Relevance

AO-OCT measured organelle motility is a potential functional biomarker of RPE cell health and ocular disease pathophysiology.

Introduction

The retinal pigment epithelium (RPE) is composed of polarized cells arranged in a monolayer between the photoreceptors (PRs) and Bruch's membrane. These cells are crucial to maintain the health of PRs and overall vision by performing several specialized activities such as interchange of metabolites, ions, and water; protection against light radiation; and ingestion of shed PR outer segment (OS) discs.^1–3 RPE cells perform their diverse set of functions through their apical to basolateral polarity and the diversity of their various organelles.¹ Among the organelles are melanosomes, lipofuscin, melanolipofuscin, and mitochondria.^4,5 These organelles are spatially arranged in and move within the RPE cell body to carry out different functionalities.^4–7 Animal studies suggest that defects in the transport, motility, and localization of organelles could contribute to RPE dysfunction and retinal pathologies that impair vision, such as age-related macular degeneration (AMD),^8,9 choroideremia,^10,11 Stargardt macular dystrophy,¹² and Usher syndrome 1B.^13–15 RPE organelle characteristics (relative molecular constituents, density, motility, location, etc.) also change with aging (absent disease).^6,16,17 Thus, organelle movement can potentially serve as a sensitive indicator of cell health.

However, current clinical ophthalmic imaging standard-of-care cannot resolve RPE cells or organelles or assess their function. The high axial resolution of current clinical optical coherence tomography (OCT) devices allows discrimination of the RPE layer but not the RPE cell mosaic. Fundus autofluorescence uses signatures from melanin, lipofuscin, and melanolipofuscin fluorophores in the RPE to detect pigment distribution and loss across the macula. Neither method has the precision and sensitivity to detect normal intracellular distribution, sense very early disease-related or normal age-related changes in human RPE cells, or to provide insights into the mechanisms of RPE organelle movements. To adopt RPE organelle motility as a clinical biomarker for the living human eye, high resolution in vivo imaging techniques are needed to sense this activity.

In vivo imaging of RPE cells is possible using adaptive optics (AO)-enabled technologies such as AO scanning laser ophthalmoscopy (AO-SLO), AO transscleral optical phase imaging (AO-TOPI), and AO-OCT through various contrast mechanisms.¹⁸ AO-SLO devices visualize individual RPE cells by detecting the autofluorescence (AF) of lipofuscin^19–22 and melanin^21–24 granules, extrinsic fluorescence of an indocyanine green dye,^25,26 or multiply-scattered light.^27,28 AO-TOPI^29,30 uses transscleral illumination to provide diffuse wide-field illumination while by-passing strong PR waveguiding signal to preferentially visualize RPE cells with an AO full-field camera. Different from these passive contrast mechanisms, AO-OCT takes advantage of the cells’ intrinsic organelle motion to sense activity and increase overall cell contrast.^31,32 AO-OCT has the additional advantage of better RPE layer signal localization by its micron-scale depth sectioning capabilities inherent to its low coherence interferometric methodology. In a pilot study by Liu et al.,³² organelle motility-induced speckle field dynamics were characterized in terms of an exponential decay time constant, measured in healthy individuals, and confirmed on two separate AO-OCT imaging systems with similar optical coherence volumes. The motility dynamics were also characterized across the apical to basal slices of the RPE layer. One important takeaway from the study was that acquiring volumes with a time interval close to or greater than the decorrelation time of the RPE organelles yielded optimum cellular contrast. Another important clinical implication of that study was to propose speckle decorrelation (as characterized by its decay time constant) as a potential measure of RPE function as it relates to the interactions of the organelles within the imaging beam. That study provided the first step toward establishing a normative baseline for RPE organelle motility by imaging a small healthy cohort (7 participants) at a single location of the macula. Characterization of RPE motility is in a nascent stage and additional detailed and comprehensive studies are needed.

The aim of this study was to extend the previous work by imaging a larger cohort of healthy individuals and to measure organelle motility over a larger area of the macula to further characterize regional variations. We assessed the repeatability of the AO-OCT–based motility measurements in a subset of the cohort and measured the wavelength dependence of the speckle-based motility dynamics for two AO-OCT systems. Our results answer questions about the 3D variability in human RPE organelle motility that furthers our ability to extract this important functional signal from AO-OCT volumes. Our validation steps pave the way toward the establishment of a normative baseline for RPE organelle motility that can be used as a biomarker of retinal health and dysfunction, once longitudinal sensitivity and further validation in diseased cohorts are demonstrated.

Materials and Methods

Participants

Fourteen individuals (6 women and 8 men with mean age of 37.7 ± 6.9 years) free of ocular disease participated in this study. Details on the participants are listed in Table 1. Study procedures and potential risks were explained to each participant and written informed consent was obtained afterward. The study protocol was approved by the Institutional Review Board of the US Food and Drug Administration (FDA). All participants underwent screening and a comprehensive eye examination by a board-certified ophthalmologist (authors O.J.S. or C.C.) prior to the AO imaging session at the FDA. Participants were screened for fixational ability, ocular pathology, or contraindication to pupil dilation. The eye examination included the documentation of visual acuity, fundus photography, biometry (IOLMaster 700; Carl Zeiss Meditec Inc., Dublin CA), and OCT imaging (Spectralis, Heidelberg Engineering GmbH, Heidelberg, Germany).

Table 1.

Summary of Participants

Subject ID	Age	Eye	Sex	Race	Axial Length, mm
0201	32.9	OS	F	A	22.14
0420	27.7	OD	F	B	25.29
0571	42.2	OS	M	A	26.35
1610	32.0	OD	M	W	25.02
2875	32.8	OS	M	W	24.13
3339	36.6	OS	M	W	22.14
4136	53.6	OS	M	W	25.76
5291	36.6	OD	M	W	25.21
5810	36.5	OS	F	W	22.29
6581	41.9	OD	F	A	23.73
7473	36.6	OS	M	W	23.96
7743	42.9	OS	F	A	25.02
8009	40.0	OD	F	W	24.30
8195	33.3	OD	M	W	24.11

A, Asian; B, Black; W, White.

AO-OCT Imaging System and Imaging Protocol

Participants were imaged with the custom-built FDA Fourier domain mode-locked (FDML)-based AO imaging system,³³ which collects AO-OCT volumes at 13.0 hertz (Hz) using a 3.4-MHz FDML swept source laser (λ_c = 1060 nm, Δλ = 76 nm). The lateral and axial resolution of the system in the eye are estimated to be 2.9 µm and 8.4 µm, respectively.

One eye of each participant was dilated for imaging with 1% tropicamide. For 11 participants, 2 degrees × 2 degrees field of view (FOV) AO-OCT volumes with the system focus set to the PR-RPE complex were collected from the fovea to 12 degrees temporal (T) retina in steps of 1.5 degrees using a fixation target (Figs. 1A, 1B). At each retinal location, three 10-second videos (20 volumes/video; 0.5 seconds sampling interval) were acquired to capture the organelle motility speckle decorrelation.³² Participants were instructed to blink naturally during the <1-minute interval between the video recordings. Imaging from 7 subjects was repeated on 2 additional days at the fovea and 7.5°T to assess the repeatability of the organelle motility measurements.

Figure 1.

AO-OCT imaging and RPE cell labeling for assessing motility dynamics. (A) Spectralis SLO image of a subject eye with nine AO-OCT imaging regions overlaid (black border). (B) Montage of AO-OCT en face images at the RPE layer for the same eye. (C) Data processing steps to calculate the correlation coefficient (CC). (D) (left) An en face image of the RPE mosaic at 12°T, (middle) the automatically identified RPE cell centers marked with magenta dots, and (right) Voronoi map of RPE cells in the central region marked by the yellow edge lines. (E) Example image series of an RPE cell over time. Pixels within the RPE Voronoi region are used to calculate the correlation coefficient between the reference (at time 0) and each consecutive image.

Four subjects (3339, 4136, 6581, and 8009) were additionally imaged on another FDA custom-built multimodal AO (mAO) system³⁴ to assess the dependency of the organelle motility results on imaging wavelength and coherence volume. The coherence volume depends on imaging wavelength,³⁵ and ultimately affects the number of organelles contributing to each speckle, as well as the speckle decorrelation function, and hence the overall decorrelation time.³⁶ This additional experiment was therefore conducted to link the results in this paper to our previous studies all conducted at shorter wavelengths and smaller coherence volumes. The second system uses a superluminescent diode (λ_c = 830 nm, Δλ = 60 nm) for AO-OCT imaging and was used in one of the previous RPE motility studies.³² The lateral and axial resolution of the mAO system are estimated to be 2.5 µm and 3.7 µm, respectively, which produces a smaller coherence volume (and smaller speckle size) than the FDML AO system. The system signal-to-noise ratio (SNR) for the FDML AO and mAO imaging systems was previously measured to be 82.9 ± 1.0 decibels (dB) and 83 dB, respectively.^33,34 The numerical aperture (NA) of both systems is also similar.

To match the dwell time of the beam on the cells and other characteristics of the FDML AO and mAO systems that could influence organelle decorrelation and our motility results, we configured the systems with the same acquisition and experimental parameters to every extent possible for the wavelength experiment. First, the lateral pixel size was set to the same value (approximately 1.45 µm) on both systems, which was achieved with a 2.5-degree × 2.5-degree (512 × 512 pixels) FOV on the FDML AO system and a 1.5-degree × 1.5-degree (300 × 300 pixels) FOV on the mAO system. Second, the volume sampling intervals were matched. The mAO volume rate (300 × 300-pixel volumes) was 2.3 vol/sec, which equates to a sampling interval of 0.43 seconds for continuous acquisition. The FDML AO system has a much higher volume rate, and so we configured our acquisition to collect volumes at the same 0.43 seconds sampling interval. Three 8.6-second AO-OCT videos (20 volumes/video; approximately 0.43-second interval) separated by approximately 1 minute with the system focus set to the PR-RPE complex were collected at 7.5°T on both the FDML AO and mAO systems. Finally, the subjects were imaged consecutively on both systems to control for physiological variabilities to every extent possible.

Data Processing and Motility Analysis

AO-OCT volumes were processed and dewarped to correct for sinusoidal motion from the resonant scanner (Fig. 1C). Eye and head motions were corrected by automatic sub-cellular 3D registration of the volumes from the same retinal location to a selected reference volume³⁷ and flattened to the PR-RPE complex. A high-quality averaged AO-OCT volume was created by averaging the registered volumes for marking RPE cells. The RPE layer center was automatically identified from the averaged AO-OCT volume as the peak location of the average axial intensity profile and verified by an expert grader. RPE cells were automatically identified from the averaged, registered RPE en face image using an automated deep learning approach.³⁸ We used Voronoi analysis with the automatically identified cell centers to create the spatial masks for individual cells. If needed, RPE cell markings were manually corrected by an expert grader to ensure that hypo-reflective regions corresponding to vessels were omitted from further analysis. Post hoc analysis found that markings for only 4.5 ± 0.9% of cells across all locations in 5 subjects (45% of the cohort) changed with manual correction of the deep learning results, indicating the automated machine learning algorithm was highly accurate. Additionally, motility measurements were limited to cells within the central one-third section of the B-scans to minimize effects from any residual rotational eye motion or aberrations that tended to be strongest toward the edges of the fast-axis (Fig. 1D).

We used the previously described decorrelation-based approach³² to quantify the temporal dynamics of RPE organelle motility. Given the spatial mask of an RPE cell Voronoi patch (V) at a reference time point with amplitude R₀(x,y), the correlation coefficient (CC) with consecutive images at times t₁, t₂, ... is computed as shown in Equation 1 (see Figs. 1C, 1E):

$$\begin{array}{ccc}& & CC\left(t_n\right)\hfill \\ & & =\frac{\mathbb{E}\left[\left(R_0\left(x,y\right)-{\overline{R}}_0\right)\left(R\left(x,y;t_n\right)-\overline{R}\left(t_n\right)\right)\right]}{\sqrt{\mathbb{E}\left[{\left(R_0\left(x,y\right)-{\overline{R}}_0\right)}^2\right]\mathbb{E}\left[{\left(R\left(x,y;t_n\right)-\overline{R}\left(t_n\right)\right)}^2\right]}}\hfill \end{array}$$ (1)

were $\mathbb{E}$ is the mean operator over pixels within mask V, R(x, y; t_n) denotes the OCT amplitude at pixel (x, y) at time point t_n, and ${\overline{R}}_0$ and $\overline{R}\left(t_n\right)$ are the mean amplitude of all pixels in V at time points t₀ and t_n, respectively. As eye movement can cause gaps in the registered images (where excessive eye motion occurs), the unimaged regions were masked out to avoid biases in the CC calculation. The CC for individual RPE cells were computed and then averaged to yield the correlation function (CF) that can be expressed as an exponentially decaying function, as shown in Equation 2³²:

$$\begin{array}{c}\hfill CF\left(t\right)=A\left(t\right)\times exp\left(-t/\tau \right)\end{array}$$ (2)

In the above equation, A(t) captures the residual motion from the eye and system (e.g. uncorrected scanning artifacts), which are reduced after subcellular image registration. The exponential term captures the organelle motility, in which the time constant τ is the de-correlation time of the speckle pattern and the metric we use to quantify RPE motility. To remove the time-dependent amplitude term A(t) from the measurements, the CF of the RPE is normalized to the CF of the cone PR layer (extracted from the same AO-OCT volume) as it can be approximated as A(t).³² Thus, as shown in Equation 3, the normalized CF (CF_norm) of the RPE becomes:

$$\begin{array}{c}\hfill C{F}_{norm}\left(t\right)=A\times exp\left(-t/\tau \right)+B\end{array}$$ (3)

where A and B are the fitting parameters. Outlier images, defined as cases with low image quality (suboptimal aberration correction) or significant residual motion, were manually identified and removed prior to estimating the parameters of Equation 3 through curve fitting. Post hoc analysis revealed that removed images had lower image quality as measured by a cross-correlation figure-of-merit than images kept for further processing, indicating that although the removal was subjective, it was applied in a consistent manner, and yielded a dataset with high quality RPE images. The Matlab (Mathworks, Natick, MA) fit function was used for curve fitting and images from the subsequent successive acquired videos (limited to an 80-second time-delay with the reference frame) were used for estimating model parameters (e.g. τ). At any imaged location, we measured motility by selecting one reference volume from each of the three video recordings and repeating the described analysis steps whenever possible. We report the average of the repeated motility measurements for each retinal location unless only one high-quality reference image could be selected, in which case we report a single measurement.

Given that different RPE organelles are distributed differently across the apical to basal axis, we measured motility across three axial planes of the RPE layer. The three RPE depths were selected at the apical, middle, and basal portions of the RPE layer. The apical and basal planes were set to be approximately 3 µm above and below the peak intensity of the RPE layer, a separation small enough to reduce signal contamination from the adjacent layers (i.e. PR and Bruch's membrane) but large enough to distinguish from the RPE peak signal.

RPE Organelle Motility Mapping

To visualize the spatial variation of RPE organelle motility across the macula, we generated cross-sectional CC maps (based on Equation 1) at the three depth layers described above and normalized to the PR layer. The CC values for each RPE cell (as defined by the Voronoi analysis) were calculated between the reference frame and the average of 3 frames centered approximately 5 seconds after the reference frame (i.e. 4.5, 5.0, and 5.5 seconds) and normalized to the cone PR layer. We used CC rather than τ to generate cellular-level maps at a narrowly defined timepoint. This mapping approach differed from the τ estimation described previously, in which at each time point CC values were averaged across many cells within a region to enhance the motility signal by reducing background noise and minimizing cell-to-cell variability.

Statistical Analysis

R software and Microsoft Excel with the Real Statistics Resource Pack and Analysis Toolkit add-ins were used for statistical analyses. A linear mixed-effects (LMEs) model with a random intercept for subject was used to analyze the spatial dependency of the decorrelation time constant τ using all the repeated measurements. We used coefficient of variance (CV) to characterize the agreement between repeat session motility measurements, which we define as repeatability. Despite the relatively long time between repeat measurements, all other experimental factors (e.g. device, image location, image time, scanning parameters, patient procedures, device operator, etc.) were identical for the three sessions. Student's t-test (paired) was used to compare the motility time constant for different wavelengths for the three RPE layers. The threshold P value for statistical significance was set to 0.05.

Results

RPE Organelle Motility Maps

An example of the RPE cell mosaic across the macula and the respective organelle motility maps as a function of depth for one subject with excellent RPE contrast is shown in Figure 2. The overall trend of decreasing CC values across the macula is evident, although there is significant variability in individual cell-to-cell values. For this subject, there is also a plateau <3 to 5 degrees before a falloff in values outside this central region. The same trend (i.e. plateau in central macula) was observed in several but not all subjects imaged. The uniform appearance of RPE cells in the structural image (see Fig. 2A) likely rules out image quality or eye motion artifact as factors in this plateaued appearance. The decrease is highest for the apical regions (τ = 14.0 seconds at the fovea to 7.8 seconds at 9°T), followed by the middle region (τ = 10.2 seconds at the fovea to 6.0 seconds at 9°T). The basal region showed only a slight decrease (τ = 6.9 seconds at the fovea to 5.0 seconds at 9°T).

Figure 2.

RPE organelle motility varies across the macula. (A) The RPE mosaic from subject 2875 (OS) from the fovea (*, left) to 12°T (right). Color-coded cross-correlation coefficients (CC) for each RPE Voronoi patch corresponding to an individual RPE cell for the: (B) apical, (C) middle, and (D) basal layers. (E) The corresponding τ at 1.5-degree increments calculated for each layer from the same dataset. The monotonically decreasing CC from fovea to peripheral retina can be seen in both the maps and τ. CC shows significant variability from cell-to-cell.

Depth-Dependent RPE Organelle Motility Across the Retina

We quantified the organelle motility across the apical, middle, and basal sections of the RPE layer for all subjects and across nine regions of the temporal retina (see Figs. 3A–C). For each of the three axial planes, the three repeated measures of τ for all subjects as a function of retinal eccentricity are shown in Figure 3D. The result of the LME model with random intercept for subjects is summarized in Table 2. The model showed that for each 1-degree increase in eccentricity, the decorrelation time constant τ decreased by 0.333 (95% confidence interval [CI] −0.417 to −0.250), 0.112 (95% CI = −0.167 to −0.058), and 0.052 (95% CI = −0.090 to −0.015) seconds at the apical, middle, and basal sections, respectively (all P values < 0.05) after accounting for subject-to-subject variability. An LME model with variable slopes for subjects did not converge for the apical plane, thus the simpler model with only random intercepts was used.

Figure 3.

AO-OCT measurement of RPE organelle motility based on speckle field decorrelation. (A) Cropped average B-scan image from subject 3339 depicting different outer retinal layers, with 3 depths across the RPE layer marked on the averaged A-line profile. (B) The 200 × 200 µm² en face images of the average volume (top) and single frames from the refence volume (bottom) at the apical, middle, and basal sections of the RPE layer at 10.5°T. (C) The color-coded normalized correlation function showing different rates of decay at the apical, middle, and basal layers of the RPE. (D) Measured motility τ with retinal eccentricity at different sections of the RPE layer for all 11 study participants. Each participant’s data are shown with a different marker. The solid red lines represent the predicted linear fits to the data using the LME model and the dashed black lines are one SD above and below the mean values for all participant data.

Table 2.

Estimated Linear Mixed-Effects Model Parameters (Intercept and Slope) for the Three Axial Planes of the RPE Layer

Section	Fixed Effect Parameter	Estimate	Lower 95% CI	Upper 95% CI	P Value
Apical	Intercept	14.491	13.599	15.383	0.001
	Eccentricity	−0.333	−0.417	−0.250	0.001
Middle	Intercept	9.462	8.870	10.053	0.001
	Eccentricity	−0.112	−0.167	−0.058	0.001
Basal	Intercept	6.908	6.458	7.358	0.001
	Eccentricity	−0.052	−0.090	−0.015	0.006

CI, confidence interval.

RPE Organelle Motility Repeatability

The mean (± standard deviation [SD]) separation between the first and third AO-imaging sessions for the repeatability cohort was 41.6 ± 36.5 days (minimum = 12 and maximum = 119). The mean separation between all consecutive visits was 20.8 ± 28.6 days (minimum = 2 and maximum = 112). Whereas all experimental factors were controlled to every extent possible, measurements on repeat days separated by weeks to months was designed to isolate and determine the effect of subject variability factors in a real-world clinical scenario. The sessions occurred within 1 or 2 hours of the same time each day to provide reasonable control over any potential influence on RPE organelle motility from the diurnal variability of PR disc shedding.³⁹ The individual participant measurements for the three sessions and their values normalized to the mean are shown Figures 4A and 4B, respectively. The CV for the fovea and 7.5°T locations across all axial planes were 12.6 ± 8.9% and 8.7 ± 4.4%, respectively, and 10.7 ± 7.2% for all subjects and locations. CV was less than 15% for 81% of the subject locations where organelle motility was measured. Our previous study⁴⁰ measured very low CV for RPE density quantification for repeat sessions (2.3%). Whereas AO-OCT measured organelle motility is more variable than RPE cell identification, probably owing to daily changes in RPE cellular physiology for human subjects and cell-to-cell variability (see Fig. 2), both results suggest that the structural and functional RPE measures have sufficient repeatability to act as biomarkers of retinal cellular health, pending motility measurements in diseased eyes.

Figure 4.

Repeatability of the AO-OCT RPE organelle motility measurements. (A) Repeated measurements for three imaging sessions (light blue/orange markers) and their mean ± SD (darker colored markers) for seven participants at the apical, middle, and basal sections of the RPE layer. (B) Plot of the repeated measurements normalized to their corresponding mean, depicting the spread of the repeated motility measurements. Dashed lines denote ±SD of the overall samples.

Dependence of RPE Organelle Speckle Decorrelation Time on Imaging Wavelength

The effect of OCT imaging source wavelength on RPE organelle motility has not previously been considered. Our pilot results³² measured an organelle speckle decorrelation time constant of approximately 5 seconds and demonstrated optimal RPE cell contrast when the time separation between acquired volumes was equal to or greater than τ. That study also examined axial differences in τ in 4 healthy volunteers using 2 AO-OCT systems at 785 and 830 nm (which had similar coherence volumes) and found τ = 6.5 ± 0.9, 4.4 ± 0.2, and 3.8 ± 0.8 seconds for the apical, middle, and basal RPE layers.

In the current study, we extend the previous results by imaging 4 healthy volunteers at 7.5°T with 2 different systems and OCT imaging wavelengths (Fig. 5). We first measured the SNR (mean peak signal in volume divided by SD noise measured in the vitreous region) of the AO-OCT volumes from which the RPE images were extracted from both systems in all subjects and found similar SNR values: 48.8 ± 1.8 dB for the AO FDML volumes (1060 nm source) and 44.2 ± 2.3 dB for the mAO volumes (830 nm source). Although the SNR difference is small, it may contribute to the overall difference in time-varying speckle decorrelation values between the two systems.

Figure 5.

RPE cell contrast and organelle motility as a function of wavelength. (A) Average en face RPE images (middle RPE) taken at 7.5°T in a healthy volunteer with the mAO (830 nm) and FDML AO (1060 nm) systems and corresponding 2D power spectra. The FDML image is cropped to match the mAO region. Arrows indicated shadowing from the same overlying retinal vessel at a branch point. (B) Radial power spectra calculated from the 2D power spectra with peak distinctiveness (inset) as a measure of RPE cell contrast. (C) The decay time constant (τ) measured at the apical, middle, and basal cell planes for 4 subjects at 1060 nm (solid circles) and 830 nm (open diamonds). Mean values for all subjects are shown next to the individual data in black symbols. Error bars denote the SD of values acquired on three consecutive trials for the four individual subjects in the current study and for a single trial on four individual subjects for the previous study (Liu et al.³²). **P < 0.005.

Next, we assessed RPE contrast for the fixed volume interval of 0.43 seconds. When all 60 volumes of the 3 acquired videos are averaged (each video is 8.6 seconds long), there are observable differences in RPE contrast (see Fig. 5A) between the systems. Whereas both systems resolved cells with sufficient fidelity for density quantification, the cells from the mAO system (830 nm) appear better delineated than those from the FDML AO system (1060 nm). We quantified the RPE cell contrast with the peak distinctiveness (PD) used previously by Das et al.⁴¹ (see Fig. 5B): PD = PS_max − 1/2(PS_{min_left} + PS_{min_right}), where PS_max is the local frequency peak associated with RPE cells and PS_{min_left} and PS_{min_right} are the base values on either side of the local peak. Both systems estimate the RPE cell row-to-row spacing to be 13.4 µm, which is consistent with our density values at the same location.⁴⁰ The FDML system yields 37% lower PD contrast compared to the mAO system (6.48 ± 1.50 vs. 8.89 ± 1.05 dB), consistent with visual inspection of the RPE contrast in Figure 5A.

Finally, we calculated the time constant for individual consecutive trials for all subjects. We observed consistently higher τ values for all RPE layers at 1060 nm compared to 830 nm (see Fig. 5C): 10.4 ± 2.8 vs. 6.8 ± 2.2 (P = 0.0024) for apical, 6.5 ± 1.6 vs. 4.8 ± 1.0 (P = 0.0013) for middle, and 5.0 ± 1.1 vs. 3.8 ± 0.7 (P = 0.0015) for basal. Furthermore, the values at 830 nm matched our previously results³² at similar wavelengths for all 3 layers. The lower RPE cell contrast for the 1060 nm image is thus explained by the decorrelation results and our pilot study results, whereby the short interval between acquired volumes (0.43 seconds) used in the current study to capture the decorrelation curve predicts poorer RPE cell contrast in the 1060 nm image compared to the 830 nm image. Cell contrast with 1060 nm OCT illumination is expected to improve using a longer volume interval time and more effective volume averaging to allow the speckle pattern to evolve.

Discussion

RPE cells reside in a monolayer between Bruch's membrane and the rod and cone PRs. Their apical processes extend into the subretinal space forming a close interdigitation with the PR outer segments, which allows a close symbiotic functional relationship between the cells. During this functional activity various organelles within RPE cells are in constant motion, often along microtubules as part of the transcytosis structure of polarized RPE cells.^7,9 RPE organelles include melanosomes (M), lipofuscin (L), and melanolipofuscin (ML), as well as the organelles involved in disk phagocytosis (OS phagosome), autophagy (autophagosome), metabolism (peroxisomes), and the breakdown, recycling, and degradation of macromolecules in the endolysosomal system (endosomes and lysosomes).⁴² It is predominantly melanosomes and lipofuscin that give rise to contrast that is used by color fundus photography, AF imaging,⁴³ and OCT to distinguish the RPE layer, where melanosome absorption, scattering, and fluorescence are detected by near infrared stimulation and lipofuscin fluorescence by visible (blue) light stimulation. For OCT, organelle motion causes the speckle pattern to change, and the temporal decorrelation of this pattern can be measured to indirectly determine organelle speed. Other organelles including nuclei (N) and mitochondria (MC) may also contribute to the OCT speckle signal.

Before RPE organelle motility can be established as a functional biomarker of ocular disease, its characteristics must be documented and understood with respect to the broader RPE anatomy and physiology in both healthy and diseased eyes. The speckle field dynamics of the light-organelle interaction must also be fully explored. We observed that the measured decorrelation time constant that arises from RPE organelle motion is dependent on illumination wavelength with a longer time constant for longer wavelengths. Although the small SNR difference between the two imaging systems to some degree may have contributed to the overall difference in the measured decorrelation values, we attribute this observation mainly to the dependence on the coherence volume, where organelles must travel a greater distance to alter the speckle field for the longer wavelength system. Previous experimental and theoretical speckle decorrelation studies have shown that smaller speckles produce shorter decorrelation times compared to larger speckles,^36,44,45 which matches our observations. We note that the underlying RPE organelle motility velocity is the same regardless of illumination coherence volume (and wavelength), and the dependency is in the measured decorrelation time constant, not the organelle speed. Further studies will be necessary to directly link organelle speed and decorrelation time for a particular coherence volume.

RPE organelle motility time constant also depends on the axial layer, ordered for the apical, middle, and basal layers from longer to shorter values. This finding is consistent with our earlier pilot results.³² Across the temporal macula, the time constant decreases with increasing eccentricity for the apical layer and to a much lesser extent for the middle layer. The time constant changes very little with eccentricity for the basal layer. Two questions arise from our results: 1. Which organelles are the source of our motility signal axially? 2. What is the reason that the apical and, to a lesser extent, the middle layer time constants decrease with increasing eccentricity while the basal signal is relatively flat? Given the complexity of RPE cell biology and function, organelle motility as measured by the OCT decorrelation signal likely arises from multiple sources within RPE cells. However, we can draw some conclusions based upon what is known of the topographical arrangement of cells in the PR-RPE complex and also recent studies of RPE organelle distribution.

The topography of PRs, whose functional activities the RPE cells serve, is well known.^40,46 Cone PRs peak in the fovea at approximately 200,000 cells/mm² (mean) where our visual acuity is highest and decay with eccentricity (by a power function) to 5000 to 10,000 cells/mm² in the perifovea (10–15 degrees or 3 to 5 mm). Rod PRs are absent in the central fovea (1.25 degrees or 0.35 mm) and density increases for increasing eccentricity to an elliptical peak of approximately 140,000 to 160,000 cells/mm² at perifoveal locations, depending on direction (i.e. temporal-nasal or superior-inferior). RPE cell density is maximum in the fovea at 7335 cells/mm² and decreases across the macula by 123 cells/mm²/degrees to 5546 cells/mm² at 12°T.⁴⁰ RPE cells are therefore smallest in the fovea,²¹ increasing from approximately 13.2 to 15.2 µm. PR/RPE ratios are high in the central fovea where cone PR density is highest but also in the perifovea where rod density peaks. In terms of cell topography, the faster speckle decorrelation, that is, increase in RPE organelle motility speed – especially in the apical region – for increasing eccentricity seems more likely to do with the increased demand of rod PRs in the perifoveal region than the decreased size of RPE cells in the central fovea. It seems unlikely that the smaller area for foveal compared to perifoveal RPE cells would spatially confine the movement of approximately 1 µm-diameter organelles to cause a significant increase in speckle decorrelation time (and indirectly a decrease in organelle speed). However, the variation in RPE cell size cannot yet be ruled out as a cause for variations in organelle motility without further investigation.

Another important consideration to RPE function is the distribution of organelles, both axially within RPE cells, and also laterally across the macula. Several recent ex vivo studies have provided information on the topographical arrangement of RPE organelles in the macula.^4–6 Bermond et al. used structured illumination and confocal fluorescence microscopy to characterize the distribution of autofluorescent granules across the axial position within the cell and across the lateral macular location in flat mounted human retinas without ocular disease.⁶ Axially, they observed that L granules, found in the middle layer, and ML granules, found in both the apical and middle layers, both vastly outnumber M granules, which are localized predominantly in the apical interdigitation zone. Laterally, L granule density is higher in the perifovea (at 4 mm or 13.7 degrees superior to the fovea) compared to the fovea and ML granule density is flat or slightly elevated in the fovea compared to the perifovea. The authors attribute the higher ML granule proportion in the fovea to cone PRs and the higher L granule proportion in the perifovea to rod PRs.

Our intracellular (axial) results could be explained by higher MC organelle speed arising from rapid metabolic coupling activity of peroxisomes in the basal layer compared to M, ML, or L granule transport speeds, collectively, in the apical and middle layers. Within the latter, our results indicate higher L granule speed in the middle layer compared to the speed of M or ML granules, which tend to localize toward the apical side of the cells. The axial differences in motility may also be related to differences in the complex recycling and degradation routes of the endolysosomal system,⁴² although it is not clear if this activity changes the image contrast of the OCT decorrelation signal. The lateral differences across the macula could be explained by higher L granule speeds in the perifovea compared with ML granules in the fovea. This explanation is consistent with the topographical explanation above. Unfortunately, there does not appear to be a study that directly compares RPE organelle velocities, although this is a consideration in the extensive RPE biology literature.⁴²

There is other evidence from known RPE biology that may be helpful in determining the source of RPE organelle decorrelation signals.⁴² One is the arrangement of microtubules in RPE cells, which act as active highways along which melanosomes and phagosomes are transported. These organelles exhibit bidirectional movements along the microtubules within the overall directional migration of macromolecules across polarized RPE cells. Vertical and horizontal microtubules exist in the cells, with the horizontal ones predominantly in the apical layer.⁹ This arrangement may be reflected in the difference in OCT decorrelation time constant across the RPE cell layers that we observed. There is also evidence that melanosome motion is not uniformly constant but occurs in short bursts along the actin filaments,⁴⁷ further confounding our ability to localize the source of our motility signals and possibly the cause of the significant cell-to-cell variability observed in our data. Melanosome localization in the actin-rich apical region of RPE cells has been shown to be depending on the presence of myosin-7a, the mutation of which plays a central role in Usher's Syndrome 1B.⁹

Our measurements represent an important first step in feasibility and initial measurement of normative values and repeatability for a potential AO-OCT based RPE motility biomarker. However, characterization of longitudinal changes in healthy eyes with aging and in diseased cohorts are necessary additional steps before RPE motility can be fully validated as a clinical biomarker. AO-OCT is not a current clinical standard-of-care method and RPE motility temporal decorrelation measurements are not rapid or straightforward, although some progress has been made streamlining the methodology in the decade since AO-OCT resolution of the RPE mosaic was demonstrated.³² Moreover, although our repeatability results indicate relatively low variance for repeat measurements – a mean change of approximately 10% in τ when measured over weeks to months – the level of variance is probably not low enough for a clinical biomarker, where values <5% are ideal,⁴⁸ particularly if disease-related effect sizes are found to be in a similar range.

The experimental factors that contribute to the variability can be further identified and mitigated for further development of RPE organelle motility as a clinical biomarker. Our measurement is currently sensitive to image quality, which is impacted by eye motion, variable aberration correction, registration accuracy, volume flattening, and subcellular layer identification, as well as other factors. There were also some relatively uncontrolled factors in our study, including a small difference in image session start time with respect to the diurnal cycle and inconsistent subject (light or dark) adaptation, although the extent to which these factors affect RPE organelle motility as measured by OCT decorrelation is unknown. Our relatively large system optical resolution, comparable to the RPE sub-layer distance, may limit our current ability to find a definitive source for the organelle motility signal observed in the AO-OCT volumes. However, the finer axial pixel sampling, coupled with knowledge of organelle topography, allows some interpretation of potential signal sources from our measurements. Another limitation of our study is the relatively low number of subjects. Once our AO-OCT organelle motility measurement method is tested in a larger cohort of healthy subjects, there is ample evidence that organelle motility impairment associated with many retinal diseases will be sensitively detected with our technique.^3,12,13 The next step in biomarker development, therefore, will be to characterize motility values from patients with various retinal diseases known to initiate in or affect RPE cells (Usher syndrome 1B, AMD, etc.).

An interesting question is whether RPE organelle motility can be sensed by other AO-based technologies that can resolve RPE cells, including the various AO-SLO AF techniques⁴⁹ and AO-TOPI.⁵⁰ It is possible that the imaging methods that use intrinsic organelle contrast – via melanosome or lipofuscin fluorescence or phase-based approaches – may be able to detect organelle motion by acquisition of organelle time-resolved signatures. Given the importance of the RPE in outer retinal disease initiation and progression, the invention and clinical translation of additional methodology to measure the functional interplay between PR and RPE cells will surely benefit patients with ocular disease.

In summary, we have identified key characteristics of AO-OCT–based RPE organelle motility time constants in human eyes, including variability across the macula, within the layers of the cells, and with respect to OCT illumination wavelength. This functional cellular measure may prove to be an effective means for early detection, progression tracking, and treatment monitoring for many retinal diseases that affect the RPE.