Multiple Bioenergy-Linked OCT Biomarkers Suggest Greater-Than-Normal Rod Mitochondria Activity Early in Experimental Alzheimer's Disease

Abstract

Purpose: In Alzheimer's disease, central brain neurons show evidence for early hyperactivity. It is unclear if this occurs in the retina, another disease target. Here, we tested for imaging biomarker manifestation of prodromal hyperactivity in rod mitochondria in vivo in experimental Alzheimer's disease.

Methods: Light- and dark-adapted 4-month-old 5xFAD and wild-type (WT) mice, both on a C57BL/6J background, were studied with optical coherence tomography (OCT). We measured the reflectivity profile shape of the inner segment ellipsoid zone (EZ) as a proxy for mitochondria distribution. Two additional indices responsive to mitochondria activity were also measured: the thickness of the external limiting membrane–retinal pigment epithelium (ELM-RPE) region and the signal magnitude of a hyporeflective band (HB) between photoreceptor tips and apical RPE. Retinal laminar thickness and visual performance were evaluated.

Results: In response to low energy demand (light), WT mice showed the expected elongation in EZ reflectivity profile shape, relatively thicker ELM-RPE, and greater HB signal. Under high energy demand (dark), the EZ reflectivity profile shape was rounder, the ELM-RPE was thinner, and the HB was reduced. These OCT biomarker patterns for light-adapted 5xFAD mice did not match those of light-adapted WT mice but rather that of dark-adapted WT mice. Dark-adapted 5xFAD and WT mice showed the same biomarker pattern. The 5xFAD mice exhibited modest nuclear layer thinning and lower-than-normal contrast sensitivity.

Conclusions: Results from three OCT bioenergy biomarkers raise the novel possibility of early rod hyperactivity in vivo in a common Alzheimer's disease model.

Deposition of β-amyloid (Aβ) plaques in the brain is a histopathologic hallmark of Alzheimer's disease (AD) commonly considered to contribute to the devastating cognitive losses. However, results from many clinical trials that pharmaceutically reduced plaque build-up have failed to mitigate AD dementia.¹ The lack of treatment to prevent or delay the worsening cognitive trajectory of AD is driving the search for new insights into the pathogenesis of AD in order to develop effective prodromal treatments. 

Accumulating results indicate that cognitive loss (linked to circuit/synaptic dysfunction) and Aβ plaque deposition can occur independently of each other, with both driven by a cross-linked soluble amyloid β-peptide oligomer—the neuronal hyperactivity “viscous cycle.”^2–9 In this study, neuronal hyperactivity refers to greater than normal mitochondria activity, such as an increase in oxygen consumption rates. Remarkably, the prediction that cognitive dysfunction but not plaque deposition can be restored by treating neuronal hyperactivity has been confirmed in several AD models—for example, by drugs that prolong the opening time of endoplasmic reticulum ryanodine receptor type 2 (RyR2) calcium channels and suppress neuronal hyperactivity.^{2–5,10–18} These results raise the possibly that measuring biomarkers of neuronal hyperactivity in vivo may be a useful target for improving early diagnosis of AD and in the evaluation of treatment efficacy. However, conventional functional imaging methods, such as magnetic resonance imaging or positron emission tomography, have relatively low spatial resolution and are limited in their ability to evaluate mitochondria activity in vivo.^22–25 

One approach to addressing this problem may be to examine the retina. The retina develops AD pathology similar to that found in the brain of patients but at earlier time points.^6,17,19–23 In the 5xFAD mouse AD model, the retina develops soluble amyloid β-peptide oligomers and plaque deposition before their appearance in the brain, as well as phosphorylated tau and neurofibrillary tangles.^6,17,19–23 The 5xFAD transgenic mouse is a widely used model that overexpresses three mutant human amyloid precursor protein (APP) forms and human presenilin 1 (PS1) harboring two mutations.²⁴ Also, in humans, cognitive impairment is correlated with the volume of the region between the ellipsoid zone (EZ) and retinal pigment epithelium, a biomarker similar to the external limiting membrane–retinal pigment epithelium (ELM-RPE) thickness that has been linked to mitochondria activity as described below.^25,26 These functional results indirectly suggest early AD-linked alterations in photoreceptor mitochondria function, but further evidence in vivo is needed to test whether these changes represent hyperactivity. 

In this study, a common experimental AD model (5xFAD; The Jackson Laboratory, Bar Harbor, ME, USA) was examined by optical coherence tomography (OCT) at 4 months of age, a time point showing sparse plaque deposition, as has been reported in the retina of this Alzheimer's model.^27,28 This model has been well characterized and exhibits neurodegeneration as a component of late-stage pathology.^24,29 We looked for agreement among the three distinct OCT biomarkers in order to minimize the impact of confounders that might influence any one of the measurements. First, we interrogated the shape of the hyperreflective band in the outer retina immediately posterior to the ELM (also referred to as “band 2”). A consensus clinical lexicon, based on a body of supporting literature, suggests that band 2 measured with spectral-domain OCT (SD-OCT) represents the inner segment EZ, whose volume is normally 75% filled with mitochondria.^30–35 Our functional SD-OCT studies strongly support this consensus interpretation that mitochondria are the major contributor to band 2.^36,37 For example, we have shown that the EZ reflectivity profile shape measured with SD-OCT agrees with mitochondria redistribution within the inner segment ellipsoid measured from electron microscopy in mice with different mitochondria activity modulated by either strain or light level.³⁶ We also find agreement between EZ reflectivity profile shape and oxygen consumption rate measurements.³⁷ Thus, in this study, we consider the EZ reflectivity profile shape measured with SD-OCT as a functional biomarker of rod mitochondria spatial arrangements, an important aspect of how the energy needs of the rod cell are met.^31,38–40 

In addition to the EZ reflectivity profile shape, we measured the thickness of the ELM-RPE region. This index measures mitochondrial-driven/pH-triggered/RPE water removal from the subretinal space, a signaling pathway that has been well charaterized.^41–43 Briefly, with dark adaptation, rod cyclic guanosine monophosphate (cGMP) accumulates in the outer segment to maintain persistently open, cyclic nucleotide-gated channels, an event that depolarizes the rod membrane and increases ion pumping/mitochondrial energy utilization for dark adaptation compared to light adaptation.^42–48 Increased mitochondrial activity in the dark also produces more lactate, CO₂, and wastewater, which acidifies the subretinal space and triggers an increase in apical RPE cotransporter-based water removal with concomitant shrinkage of the ELM-RPE region.⁴¹ Light–dark changes in ELM-RPE thickness have been reported in animal models and in humans for both rods and cones.^{42,43,47–56} 

For the third index, we used the magnitude of the relative decrease in reflectivity/intensity (i.e., hyporeflective band [HB]) between the tips of the photoreceptors and apical RPE.^54,57 The HB signal becomes smaller in the dark than in the light and is associated with changes in the ELM-RPE thickness in a light duration-dependent but pH-insensitive manner, suggesting a link to mitochondria activity.^54,57 We consider all three biomarkers showing higher than normal energy demand to be evidence that rod mitochondria activity has moved beyond its physiologic regulatory range into the “hyper” range. As we recently reported, the above biomarkers occur over a relatively slow time frame and measure mitochondria-driven physiology in the outer retina.⁵⁷ In contrast, other functional OCT approaches, such as optoretinography, measure faster and passive physiology, including the rapid, nanometer-scale electromechanical deformation in individual human cone photoreceptors at the onset of phototransduction.^57–59 

Finally, we note that visual contrast sensitivity may be an additional vision-based biomarker of interest. In patients with AD, reduced contrast sensitivity occurs before overt pathology and cognitive decline and is a major risk factor for falls and decreased survival; contrast sensitivity deficits are also linked to cerebral amyloid and tau deposits.^{6,19–23,60–63} Although visual performance studies in 5xFAD mice have found reduced visual acuity at 6 months of age but not at 4 months of age, it is unclear if contrast sensitivity declines at early time points in 5xFAD mice.²⁷ Thus, in this study, we also tested contrast sensitivity as a non-imaging index of early changes linked to the visual system in experimental AD. 

Methods

All mice were treated in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals, with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research, and with specific authorization by the Wayne State University Division of Laboratory Animal Resources Institutional Animal and Care Use Committee (IACUC). For these studies, non-littermate 4-month-old male wild-type (WT) and 5xFAD mice, both on a C57BL/6J background (#000664 and #034848, respectively; The Jackson Laboratory), were housed and maintained in 12-hour light/12-hour dark cycle laboratory lighting. After they were scanned, the mice were humanely euthanized by an overdose of ketamine/xylazine followed by cervical dislocation, per our Wayne State University IACUC-approved protocol. 

Optical Coherence Tomography

In this study, we used a cross-sectional design in which anesthetized mice—100 mg/kg ketamine (Covetrus, Portland, ME, USA) and 6 mg/kg xylazine (MWI Animal Health, Boise, ID, USA)—were examined by OCT (Envisu UHR2200; Bioptigen, Durham, NC, USA) in the morning (i.e., before noon). Morning OCT studies were performed to match the time of day required to obtain visual performance studies because, in our experience, mice stop cooperating for optokinetic tracking in the afternoon. Mice were either dark adapted overnight and the following day room light adapted for just 1 hour prior to imaging, or dark adapted overnight and studied in the dark. In other words, light and dark studies were performed on different groups of mice on different days. In our experience, these conditions have sufficiently low variability to robustly detect light and dark changes in EZ reflectivity profile shape, ELM-RPE thickness, and HB magnitude while minimizing, for example, handling stress and multiple rounds of anesthesia (e.g., see error bars in the figures).^{36,53,64–66} To dilate the iris, 1% atropine sulfate was used, and Systane Ultra (Alcon, Geneva, Switzerland) was used to lubricate the eyes. 

From central retina, we collected radial volume scans with following parameters: A-scans/B-scans = 1000 lines; B-scans/volume = 1000 scans; and frames/B-scan = 1 frame. One hundred images extracted from B-scan numbers 450 to 549 (representing inferior–superior retina) were registered (in-house script for R; R Foundation for Statistical Computing, Vienna, Austria). Briefly, first-pass rigid body registration with RNiftyReg (function in R) was used to rotate the image and interpolate signal at each pixel. Next, non-rotational rigid-body approaches (at the level of a given row or column of pixels) were applied three times. The 100 images were finally compared visually as a final step before averaging. A representative image of the outer retina and a set of axial reflectance profiles are shown in Figure 1. 

Figure 1.

 

Summary of qualitative light–dark changes in WT and 5xFAD mice. (A) Representative OCTs of WT (top) and 5xFAD (bottom) mice retina. ONL, outer nuclear layer; ELM, external limiting membrane; EZ, inner segment ellipsoid zone; RPE, retinal pigment epithelium. (B) Representative reflectivity (or signal intensity) profile, also known as the A-line (green arrow), in a light-adapted WT mouse (top) and a 5xFAD mouse (bottom). The 5xFAD mouse has visibly appreciable thinning of its ELM-RPE region (black double-headed arrows), a lower HB intensity (red), and rounder EZ (shown in pink); together, these changes imply a higher energy demand in 5xFAD rods. WT and 5xFAD A-line profiles are scaled the same; no y-axis is shown because units are arbitrary. No signal intensity normalization was performed for these indices.

Our analysis methods were refined over the 2-year time period in which the study data were collected so that laminar boundaries for segmentation were estimated manually, or with a machine learning model–based computer program. The machine learning model was a U-net convolutional neural network trained using the “dice loss” function and the Adam optimizer (learning rate = 0.001), with 665 previously labeled images for training and 166 and 356 images for validation and testing, respectively.⁶⁷ To improve the performance of the model, its predictions were postprocessed by applying a shortest-path algorithm.⁶⁸ From either the manually drawn or model-based estimates, the segmentation estimates were then processed with another R script to segment the image and extract the output indices described below. Please note that we have confirmed that both manual and model-based boundary estimates produced similar outputs (see Supplementary Fig. S1 in Ref. 37). The intensity values used to generate the EZ reflectivity profile shape are measured from a log-based image with a 16-bit depth (default in the Bioptigen system). Previously, we compared images before and after converting them from log to linear values using a simple, empirically derived equation and noted that EZ reflectivity profile shape differences would be detected with either output; more detailed work in this area is warrented.^36,69 

Once segmented, inferior and superior retinas (350–624 µm from the optic nerve head on the inferior and superior sides) were each analyzed; starting at 350 µm ensured that our data were analyzed away from the optic nerve head, where the outer retina is relatively uniform in all OCT data. Our analysis generated the following outputs: (1) a spreadsheet of distances from the optic nerve head and layer thickness, and (2) a spreadsheet containing the HB magnitude. Two output images were generated. The first was spatially normalized (stretched) to align layers. The R code uses fiduciary points to do this. The program assigns 12 points from –30 to 0 µm from the choroid to the RPE, inclusive of lower and upper values. All others are inclusive of upper values only, so the next line starts one step beyond the RPE and lands on the ELM. The program assigns 38 points to the RPE–outer limiting membrane (OLM) span; 36 points to the OLM–outer nuclear layer (ONL)/outer plexiform layer (OPL) span; 36 points to the ONL/OPL–inner nuclear layer (INL)+inner plexiform layer (IPL) span; 50 points to the INL+IPL–retinal nerve fiber layer (RNFL)/ganglion cell layer (GCL) span; and assign 18 points to the RNFL/GCL–retina/vitreous border span; and 10 points from the retina/vitreous border to 30 µm into the vitreous. Together, there are 200 points, which slightly over-represent the outer retina span of the region of interest. The second output image was not stretched and only the basal aspect of the RPE had a fixed position without stretching the image; this non-stretched image was used to generate the A-line reflectivity profiles shown in Figure 1. 

The analysis software, written before the data were collected for this specific experiment, assumes a whole retina thickness of 269-µm thickness. Because a typical mouse retina is usually smaller than this (∼200 µm) the retina is oversampled by the software, and no or negligible information is lost. Because we are oversampling the image, the accuracy of segmentation will be limited by spatial resolution of the original image, rather than sampling strategy. Given an image axial resolution of 1.8 µm/pixel, retinal thickness for our system is ∼111 pixels. The software was written to sample the original image at 200 points in the axial direction (178 samples in the retina + 10 samples in the vitreous and 12 in the choroid). Also, the number of points sampled from each span of the retina is based roughly on the proportional real-world size of those spans: 18 points are assigned to the RNFL/GCL and 36 points to the ONL/OPL–INL+IPL span, because the latter was roughly twice as thick as the former (again, in prior experiments). By sampling this way, the original image is slightly oversampled in every span. If the point assignment were flipped (36 points to the RNFL/GLC and 18 to the ONL/OPL–INL+IPL), then the RNFL/GCL would be grossly oversampled and the latter undersampled, leading to information loss. In other words, our sampling strategy results in the final spatially normalized retina visually resembling the original, easing interpretation. 

An advantage of the stretched/spatially normalized format is that the EZ is positioned in the same place for each image. To generate the EZ reflectivity profile shape, the initial image is resized to a width of 100 pixels and a height of 259 pixels using a bilinear interpolation; signal intensity is also set to a fixed value (macro for ImageJ; National Institutes of Health, Bethesda, MD, USA). A fixed-size region of interest is then positioned to span 350 to 624 µm from the optic nerve head on the inferior or superior sides using two separate ImageJ macros. These steps are performed to take advantage of a built-in function within ImageJ (Plot Lanes) for drawing reflectivity profiles. Next, a baseline is hand drawn, and the wand feature is used to define the EZ region of interest. Because the examiner in this study was not masked during this process, we also analyzed the data using an unbiased approach in which MATLAB code (MathWorks, Natick, MA, USA) determines the baseline for EZ reflectivity profile shape analysis from the initial, spatially un-normalized images. The two approaches reached the same conclusion (Supplementary Fig. S1). Thus, the fixed region-of-interest approach is not a necessary criterion for measuring EZ aspect ratio. In all cases, the shape of this EZ reflectivity profile is summarized using the Fit Ellipse command in ImageJ (please note the word “fit” may be somewhat misleading; see below). In the results window, the value under the column marked “round” is the minor-to-major aspect ratio for the fitted ellipse. The EZ reflectivity profile shape (i.e., OCT signal magnitude along the length of the EZ, illustrated in Fig. 2B) was extracted using ImageJ macros.⁷⁰ The magnitude values used to generate the EZ reflectivity profile shape are from the Bioptigen system.³⁶ The EZ reflectivity profile shape was determined not by fitting to the EZ region but rather after converting/transforming the EZ profile into an ellipse with the same direction and area in order to determine its minor-to-major aspect ratio; this process is described at https://imagej.nih.gov/ij/source/ij/process/EllipseFitter.java and more formally in the literature.^70,71 In this study, we chose to use the ellipse aspect ratio, a commonly used shape descriptor; whether it is an optimal shape description for the EZ reflectivity profile shape is not yet clear.⁷¹ Also, our experience to date is that the EZ reflectivity profile shape is not a function of ELM-RPE thickness, an impression supported by the data in this report (see below). 

Figure 2.

 

Summary of quantitative light–dark changes in bioenergy OCT biomarkers within controls (left two columns) and within 5xFAD (right two columns) groups. As expected, in controls, light produced a more elongated EZ reflectivity profile shape (A), a thicker ELM-RPE (C), and higher HB intensity (E) panretinally compared to those in the dark. In contrast, in light versus dark-adapted 5xFAD mice, EZ aspect ratios were not different (B), ELM-RPE was thinner (D), and HB intensity was only significantly lower on the inferior side (F). Black horizontal line indicates P < 0.05 (two-tailed, linear mixed-model analysis; mean ± 95% confidence interval [CI]); n = 5 WT, n = 5 5xFAD. Individual data points represent the measured value for each mouse.

We measured the ELM-RPE thickness (illustrated in Fig. 1B) from in-house R scripts that objectively extracted layer boundaries obtained after searching the space provided by a hand-drawn estimate (“seed boundaries”); for details, see Reference 41. The ELM and RPE are initially identified by local signal maxima, and the R script determines the ELM-RPE thickness by calculating the distance from the ELM to the basal side of the RPE at the level of Bruch's membrane.^72–74 

The magnitude of the relative HB signal intensity is calculated as previously described by analyzing profile contours spanning the reflectance peak of the RPE and outer segment tips: A straight line is drawn between the RPE and the outer segment tip portions of the profile (intersecting only one point on each side of HB), and the largest departure from that line is the magnitude of the relative HB signal intensity; the magnitude of this decrease in reflectivity/intensity is presented herein.^54,66,75 

Optokinetic Tracking

Two visual performance metrics were measured from awake and freely moving mice not used in the OCT arm of the studies: spatial frequency thresholds (“acuity,” in cyc/deg) and contrast sensitivity, measured at the peak of the nominal curve (0.06 cyc/deg), and inverse Michelson contrast (unitless) using the optokinetic tracking (OKT) reflex (OptoMotry; CerebralMechanics, Inc., Lethbridge, AB, Canada), as described previously.⁷⁶ In brief, a vertical sine wave grating is projected as a virtual cylinder in three-dimensional coordinate space on computer monitors arranged in a quadrangle around a testing arena. Unrestrained mice were placed on an elevated platform at the center of the arena. An experimenter used a video image of the arena from above to view the animal and follow the position of its head with the aid of a computer mouse and a crosshair superimposed on the mouse head. The x, y positional coordinates of the crosshair are centered on the hub of the virtual cylinder, enabling its wall to be maintained at a constant “distance” from the animal's eyes and thereby adjusting the spatial frequency of the stimulus to a fixed viewing position. When the cylinder was rotated in the clockwise (CW) or counterclockwise (CCW) direction and the animal followed with head and neck movements that tracked the rotation, it was judged whether the animal's visual system could distinguish the grating. CW and CCW tracking provides a measure of left and right eye acuity and contrast sensitivity.^77,78 One set of acuity and peak of contrast sensitivity measurements can reliably be obtained in 30 minutes. 

Statistical Analysis

Data are presented as mean and 95% confidence intervals. At the minimum, each mouse had two OKT measurements, one for each side, and laminal layer thickness had values spanning 350 to 624 µm from the optic nerve head. Thus, all outcomes (OKT, EZ aspect ratio, HB intensity, and OCT layer thickness) had repeated measures for each mouse. As such, we used mixed models to analyze all outcomes using the Kenward–Roger method for calculating degrees of freedom in PROC GLIMMIX for SAS 9.4 (SAS Institute, Cary, NC, USA). We used generalized linear mixed models with a gamma distribution and log link to analyze contrast sensitivity. Analyses for all other measures assumed a normal distribution. All models included the fixed effects of strain (WT vs. 5xFAD), side, and the strain/side interaction. Only a random intercept for mouse nested within strain was included for these models. We evaluated the potential for heterogeneity of both the residual variance and the variance between mice using the Akaike and Schwarz Bayesian information criteria, although no outcome showed evidence for such heterogeneity. Higher order interactions were removed if not statistically significant. We used linear contrasts for all comparisons based on the final model. We used a 10% significance level for all interactions because tests of interactions have lower power, and we used a 5% significance level for all other tests. 

Results

Rod Light–Dark EZ Reflectivity Profile Shape Changes

In WT mice, a more elongated EZ aspect ratio was found in inferior and superior retina in the light versus in the dark; light is a nominally low energy demand condition (i.e., lower oxygen consumption rate), and dark is a high energy demand condition (Fig. 2A). This light–dark difference in EZ aspect ratio was not found in inferior or superior retina of 5xFAD mice (Fig. 2B), as a consequence of the EZ aspect ratio being greater in the inferior and superior sides of light-adapted 5xFAD mice versus light-adapted WT mice (Fig. 3A); no such difference was noted in dark-adapted groups (Fig. 3B). These results suggest a more dark-like/greater than normal mitochondria activity in 5xFAD mice, a hypothesis next tested by two other bioenergy biomarkers, the ELM-RPE thickness and HB magnitude. 

Figure 3.

 

Summary of quantitative light–dark changes in bioenergy OCT biomarkers between control and 5xFAD mice. In the light (left two columns), 5xFAD mice showed a rounder EZ reflectivity profile shape (A), a thinner ELM-RPE that reached significance only on the superior side (C), and lower HB intensity compared (E) to controls. In the dark, no significant differences between controls and 5xFAD mice were noted in EZ reflectivity profile shape (B), ELM-RPE thickness (D), or HB intensity (F). Black horizontal line indicates P < 0.05 (two-tailed, linear mixed-model analysis; mean ± 95% CI); n = 5 WT, n = 5 5xFAD. Individual data points represent the measured value for each mouse.

Rod Light–Dark ELM-RPE Thickness Changes

In WT mice, the inferior and superior retina ELM-RPEs were thicker in the light than in the dark, as expected (Fig. 2C).^36,54,57 In 5xFAD mice, a similar light–dark difference pattern in ELM-RPE was noted in inferior or superior retina (Fig. 2D). Comparing light-adapted-only 5xFAD and WT mice showed that the ELM-RPE thickness in the light became significantly smaller (i.e., more like dark WT mice) in superior retina, although this difference did not achieve significance in inferior retina (Fig. 3C); dark-adapted 5xFAD and WT groups were similar (Fig. 3D). 

Rod Light–Dark HB Magnitude Changes

In WT mice, inferior and superior retina HB magnitudes are larger in the light than in the dark, as reported previously (Fig. 2E).^36,54,57 In 5xFAD mice, a significant light–dark difference pattern in HB magnitude was noted in inferior but not superior retina (Fig. 2F). However, comparing light-adapted-only 5xFAD and WT mice showed significantly smaller (i.e., more like dark WT mice) differences in HB magnitude in both inferior and superior retina (Fig. 3E); dark-adapted 5xFAD and WT groups were similar (Fig. 3F). 

Retinal Laminar Thickness

As shown in Figure 4, the thicknesses of the two nuclear layers (outer, superior side only; inner, inferior and superior sides) showed modest atrophy. On the other hand, the inner plexiform (Fig. 4C) and retinal nerve fiber (Fig. 4D) layers were thicker than normal. The whole retina thickness was subnormal (Fig. 4E), reaching significance only on the superior side. 

Figure 4.

 

Summary of retinal layer thickness changes between controls and 5xFAD mice; results in light and dark were similar but only the results for light are shown for clarity. (A) Compared to controls, 5xFAD showed laminar thinning for ONL but only reached significance on the superior side. (B) INL+OPL, thinning was not significantly different between inferior and superior sides and results were averaged. (C) IPL did not differ significantly between groups. (D) RFNL was greater than normal in 5xFAD mice and did not differ significantly on inferior and superior sides; results are averaged. (E) The whole retina showed significant thinning limited to the superior side. Black horizontal line indicates P < 0.05 (two-tailed, linear mixed-model analysis; mean ± 95% CI); n = 5 WT, n = 5 5xFAD. Individual data points represent the measured value for each mouse.

OKT Evaluation

Contrast sensitivity was significantly decreased in 5xFAD mice compared to WT mice (Fig. 5A). However, acuity did not achieve a statistically significant difference at 4 months, in agreement with results from 5xFAD mice reported elsewhere (Fig. 5B).²⁸ 

Figure 5.

 

Summary of visual performance in controls and 5xFAD. Impaired contrast sensitivity (A) but not acuity (B) was noted in male 4-month-old 5xFAD mice. Black horizontal line indicates P < 0.05 (two-tailed, linear mixed-model analysis; mean ± 95% CI); n = 5 WT, n = 5 5xFAD. Individual data points represent the measured value for each mouse.

Discussion

In this study, no light–dark changes were noted in the inferior or superior EZ aspect ratios of 5xFAD mice, unlike those seen in WT mice, whereas WT-like light–dark differences in inferior and superior ELM-RPE thickness and in superior HB magnitude were still evident in 5xFAD mice (Fig. 2). The lack of a light–dark change in EZ aspect ratios was explainable by higher than normal EZ aspect ratios in the light for 5xFAD mice; notably, both superior ELM-RPE and inferior and superior HB magnitude were more like those in dark-adapted WT mice. Based on extrapolation of the interpretation of biomarker changes from studies in light–dark controls, our impression from all three OCT biomarkers in light and dark is that rod mitochondria are hyperactive in vivo in a common experimental model before the appearance of AD histopathology (reported elsewhere in the same model and vendor).²⁷ 

The three OCT biomarkers examined herein have been previously shown to reflect upstream mitochondria activity as reflected in their distribution within the inner segment ellipsoid (EZ reflectivity profile shape), as well as the impact of this activity on downstream physiology (ELM-RPE thickness, HB signal magnitude).^{36,43,53,54,57,66} These interpretations are based on a body of work in WT mice that finds agreement between the EZ aspect ratio, ELM-RPE thickness, and HB magnitude and gold-standard methods, including adenosine triphosphate synthesis efficacy, oxygen consumption rate, and electron microscopy.^{36,43,53,54,57,66} This first-in-kind biomarker evidence for rod hyperactivity in AD mice (Figs. 2, 3) suggests a new, testable hypothesis for future studies using conventional assays of mitochondria activity in experimental AD models and in patients with, for example, mild cognitive impairment. 

It is interesting that we did not observe a difference between light and dark EZ aspect ratios in inferior or superior retina of 5xFAD mice (likely due to the higher values in the light) but did measure light–dark changes in ELM-RPE thickness (inferior and superior) and HB magnitude (inferior only) (Figs. 2D, 2F). The latter biomarkers showing an overall light–dark pattern consistent with that in controls (Figs. 2C, 2E). This apparent discrepancy between light–dark changes in EZ and ELM-RPE and HB changes might arise from the fact that mitochondria are able to alter their energy output without spatial rearrangements—for example, by modifying expression of optic atrophy type 1 on the inner mitochondria membrane, a major regulator of mitochondrial homeostasis and cristae remodeling.^38,79 Such molecular-level modifications would not be detectable with OCT. However, this notion remains to be formally tested. We cannot rule out the possibility that small changes in mitochondria distribution occur that were below our detection sensitivity. 

We also note that the outer retina biomarker changes in 5xFAD mice occurred in the presence of modest nuclear layer thinning. A review of the literature regarding retinal layer thickness in 5xFAD mice produces mixed results, with reports of no change or thicker or thinner retina layers compared to controls, inconsistencies that may be due to the various experimental conditions used such as fixation (histology) versus no fixation (OCT), age differences, and/or bias in manual segmentation or lack of side comparisons in OCT studies.^27,28,80 In this study, we used an unbiased segmentation approach and found in vivo evidence supporting a pattern of localized retinal nuclear thinning previously measured with OCT; evidence for RFNL thickening was also found but the etiology is currently unclear.⁸⁰ It is also not clear how rod mitochondria respond to, or possibly cause, modest degeneration in the outer retina; more investigations are needed to understand this relationship. We note that senescence-linked changes in retinal laminar thickness alone may be difficult to interpret as they occur with healthy aging and age-related disease.^39,81,82 

Although a lack of acuity changes in 4-month-old 5xFAD mice has been reported,^83–85 visual contrast sensitivity per se has not been studied to our knowledge. This is important to address, because an early reduction in visual contrast sensitivity is also found in humans with AD.^60,62,86,87 A prodromal contrast sensitivity decline has been reported in the transgenic APP_SWE/PS1_∆E9 murine model of AD.^63,88 In this study, we observed for the first time a decline in visual contrast sensitivity without visual acuity problems in 5xFAD mice at 4 months of age. As previously reported, the two visual performance indices studied here do not always change together in other experimental models.^89,90 Also, the contrast sensitivity deficit shown in this study may not be due to modest rod death, based on results from other studies in different experimental models.^91–93 Nonetheless, AD affects neurons in brain and retina, so caution is needed to avoid assigning a change in contrast sensitivity completely to changes in either retina or brain. Finally, we note a caveat that, unlike testing in humans which involves V1 cortex, the visual contrast sensitivity measured in mice is done with visuomotor behavior testing, which involves subcortical central pathways.^77,89,94 

It is outside the scope of this study to test for a link between biomarker indications of greater than normal mitochondria activity and decline in contrast sensitivity. Nonetheless, this possibility is suggested by literature that reports a potential signaling pathway linking the OCT and OKT biomarkers involving dopamine. In the retina, dopamine regulates many aspects of light-adapted vision via different dopaminergic receptor signaling pathways.⁸⁹ Furthermore, rods modulate light-evoked release of dopamine, an important neurotransmitter regulating cone-based vision.^89,95,96 Evidence in patients with AD and in 5xFAD mice suggests a role for abnormal dopamine signaling in patients with AD.^97,98 Dopamine signaling, however, may not be the sole contributor to the observed impairment in visual performance reported herein.^{89,95 99} For example, rods can also regulate cone function via connexin 36 gap junctions.^100–104 It is also somewhat underappreciated that rods (the most numerous cell in the retina) are a major regulator of the volume of the subretinal space, which increases in the light and causes redistribution within the subretinal space of vision-critical components for both rods and cones.^{42–44,48,49,89,95,103,105–108} Testing these various hypotheses will require further study. 

In patients with mild cognitive impairment, Bissig et al.²⁵ found that photoreceptors have an early, exaggerated OCT reflectance light response. It has been suggested that reflectance responses might not represent mitochondria activity, although this remains an area of active investigation.^25,57 Also, between-subject reflectance measurements can be difficult to interpret, as they require some type of normalization that can vary between groups.⁶⁹ In addition, the OCT data from the Bissig et al.²⁵ study were not collected under normal light conditions and thus differed from the data used in this study. Thus, it is difficult to perform comparisons between the results of the Bissig et al.²⁵ study and the present experiments. 

In this study, we examined mice on the same C57BL/6J background strain for WT controls and 5xFAD groups. One potential limitation of this study is that littermate controls were not studied; these littermate controls would also be on a C57BL/6J background. Littermates are thought to be “clones” of one another, in that all of the alleles are identical except those being manipulated. Also, there are environmental considerations—maternal care, housing differences, and such. These factors are of importance in behavioral studies. On the other hand, our OCT biomarkers measured under the same set of conditions (light vs. dark) found reproducibility in the outcomes in over 100 batches of C57BL/6J mice from The Jackson Laboratory. Further, The Jackson Laboratory indicated that they could not guarantee that the control mice sent would be littermates (email correspondence 01/20/23). Thus, it seems unlikely that the present results were an experimental design artifact of not studying littermate controls. Nonetheless, more studies would be helpful to confirm the findings of this work between littermate controls and 5xFAD mice. 

In summary, to offset the indirect nature of our biomarkers, we looked for a consistent pattern among our three indirect OCT measurements in order to triangulate the status of mitochondria activity in vivo with accessible technology that is clinically relevant. The present results raise the possibility that multiple OCT biomarkers together may be useful in the early diagnosis and testing of therapeutic efficacy for correcting mitochondria hyperactivity in AD.^{2–5,10–17} 

Acknowledgments

The authors thank Geoffry Murphy, PhD, and HaoHua Qian, PhD, for their comments and insight; Ryan Katz and Cole Goodman for their help during the early stages of this project and manuscript preparation; and David Bissig, PhD, for ongoing R scripts and development. 

Supported by grants from the National Institutes of Health (RO1 EY026584 and R01 AG058171 to BAB), by National Eye Institute Core Grant P30 EY04068, and by an unrestricted grant from Research to Prevent Blindness (Kresge Eye Institute to BAB). 

Disclosure: B.A. Berkowitz, None; R.H. Podolsky, None; K.L. Childers, None; R. Roberts, None; R. Waseem, None