Characterizing Presumed Displaced Retinal Ganglion Cells in the Living Human Retina of Healthy and Glaucomatous Eyes

Abstract

Purpose: The purpose of this study was to investigate the large somas presumed to be displaced retinal ganglion cells (dRGCs) located in the inner nuclear layer (INL) of the living human retina. Whereas dRGCs have previously been studied in mammals and human donor tissue, they have never been investigated in the living human retina.

Methods: Five young, healthy subjects and three subjects with varying types of glaucoma were imaged at multiple locations in the macula using adaptive optics optical coherence tomography. In the acquired volumes, bright large somas at the INL border with the inner plexiform layer were identified, and the morphometric biomarkers of soma density, en face diameter, and spatial distribution were measured at up to 13 degrees retinal eccentricity. Susceptibility to glaucoma was assessed.

Results: In the young, healthy individuals, mean density of the bright, large somas was greatest foveally (550 and 543 cells/mm² at 2 degrees temporal and nasal, respectively) and decreased with increasing retinal eccentricity (38 cells/mm² at 13 degrees temporal, the farthest we measured). Soma size distribution showed the opposite trend with diameters and size variation increasing with retinal eccentricity, from 12.7 ± 1.8 µm at 2 degrees to 15.7 ± 3.5 µm at 13 degrees temporal, and showed evidence of a bimodal distribution in more peripheral locations. Within and adjacent to the arcuate defects of the subjects with glaucoma, density of the bright large somas was significantly lower than found in the young, healthy individuals.

Conclusions: Our results suggest that the bright, large somas at the INL border are likely comprised of dRGCs but amacrine cells may contribute too. These somas appear highly susceptible to glaucomatous damage.

Recent advances in adaptive optics (AO) retinal imaging, in particular with optical coherence tomography (AO-OCT), have made it possible to directly observe individual retinal ganglion cells (RGCs) and other transparent neurons in the living human retina.^1–10 These cells had been extremely challenging to image owing to their intrinsic high translucency and low image contrast.^11–13 The ability to now quantitatively analyze and track these cells over time in human subjects has garnered significant scientific and clinical interest, as these cells are fundamental to retinal neural circuitry — processing visual signals generated by the capture of light — and are vulnerable to numerous diseases that lead to blindness. RGCs in particular are lost in neurodegenerative disorders, such as glaucoma, Alzheimer's disease, Parkinson's disease, and multiple sclerosis. 

High-resolution retinal imaging studies of inner retinal neurons have focused on the ganglion cell layer (GCL), where the somas of RGCs and displaced amacrine cells are located. However, as we and other groups have reported, similarly bright, large soma-like bodies also appear deeper at the inner edge of the inner nuclear layer (INL). Because a small fraction of RGCs are known to be located at this INL inner edge,^14,15 it has been hypothesized that the soma-like bodies are displaced RGCs (dRGCs),^1,4 but this speculation is confounded by the fact that somas of several other major cell types are located in the INL, including horizontal, bipolar, Müller, and amacrine cells. Much less is also known about dRGCs than RGCs in the GCL. In particular, little is known about their size and distribution across the retina in humans with the exception of intrinsically photosensitive (ip)dRGCs that have been well studied with about half reported displaced to the INL.¹⁶ The dRGCs have also been studied mostly in other species where their distribution across the retina has been found to vary notably (in the mouse,^17,18 rat,^18–20 hamster,²¹ guinea pig,¹⁸ rabbit,¹⁸ and monkey^18,22) and is affected by the level of melanin pigmentation,^19,23,24 both confounding what to expect in humans. 

To facilitate attribution, we characterize these previously reported bright, large soma-like bodies at the inner edge of the INL by performing a prospective pilot study, obtaining AO-OCT images in human subjects. We characterize key morphological properties of the somas, namely their density, size, and distribution across the horizontal meridian of the macula in healthy young subjects and in the arcuate defects of glaucomatous subjects. We compare these morphological properties to INL cell types reported in the literature. These results were first reported in Marte et al. (IOVS 2021;62:ARVO E-Abstract 25). 

Methods

Subjects and Clinical Testing

Eight subjects participated in the study (see the Table). Five were healthy, young subjects (subjects N1–N5) whose ages ranged from 24 to 28 (26.6 ± 1.3 years, average ± SD; 3 women and 2 men) and 3 were older subjects (subjects G1–G3) with open-angle glaucoma (64.7 ± 4.7 years; 1 woman and 2 men). Assuming an age-related loss rate of 0.1% to 0.5% of RGCs per year,^25–32 we expect age alone to reduce the RGC population in the subjects with glaucoma by 3.8% to 19.1% relative to the younger subjects, who are, on average, 38.1 years younger. 

Table.

 

Study Participants

 

All subjects were required to have had an eye examination within the past year and a best-corrected visual acuity of 20/20 or better. Clear optics were required, and therefore subjects with cataracts worse than 1+ or significant ocular surface disease were excluded. Exclusion criteria included a refractive error greater than 3 diopters of hyperopia, or 6 diopters of myopia; subjects were excluded if they had retinal pathology or dystrophies affecting the posterior pole. For the older, glaucomatous group, subjects were excluded if they had other retinal pathology, such as macular degeneration or diabetic retinopathy. 

Eye lengths of the 8 subjects ranged from 21.76 mm to 27.07 mm, as measured with the IOLMaster (Carl Zeiss Meditec, Jena, Germany) and were used to scale the AO-OCT retinal images from degrees to millimeters.³³ Clinical OCT and scanning laser ophthalmoscope (SLO) images (Spectralis, Heidelberg, Germany) were obtained on all subjects and used to assess the retinal locations imaged with the AO-OCT system and the retinal damage in the glaucomatous subjects. 

The three patients with glaucoma (subjects G1–G3) had open-angle glaucoma with arcuate retinal nerve fiber layer defects as determined by Spectralis OCT images (Heidelberg Engineering, Heidelberg, Germany) and Humphrey visual fields (Zeiss, Dublin, CA, USA). Glaucoma varied in type and stage of disease, with subject G1 having primary open-angle glaucoma (POAG), subject G2 having normal tension glaucoma (NTG), and subject G3 having pigment dispersion glaucoma (PDG; see Fig. 1). All three subjects had an inferior retinal nerve fiber layer defect and corresponding superior arcuate visual field defect. Two of these subjects (G1 and G3) also had glaucomatous damage in the superior hemifield with corresponding inferior visual field defects. Although we could have imaged these superior defects for this study, we selected the inferior ones as they were more centrally located and better defined, thus providing a clearer transition from relatively healthy to severely diseased retina. 

Figure 1.

 

En face attenuation maps from Spectralis images and pattern deviation maps from Humphrey visual fields reveal structural and functional defects in the glaucoma subjects. (Left column) Using Spectralis OCT images, en face attenuation coefficient maps constructed at the depth of the RNFL show the characteristic bright pattern of retinal nerve fibers. Images have been montaged for subjects G1 and G2 and cropped to an approximate dimension of 30 × 40 degrees.³⁴ A single posterior pole image was obtained for subject G3 with an approximate dimension of 20 × 30 degrees. (Right column) Pattern deviation probability maps from the 24-2 static perimetry cover 30 degrees nasal and 24 degrees temporal area. The black squares, shaded squares, lightly shaded squares, four dots, and one dot denote probabilities of <0.5%, <1%, <2%, <5%, and >5%, respectively.

All procedures on the subjects adhered to the tenets of the Declaration of Helsinki and were approved by the Institutional Review Board of Indiana University. Subject participation followed written informed consent after discussing risks, benefits, and alternatives to participation. 

Description of Indiana AO-OCT System

The AO-OCT system has been described previously.^1,5,35,36 Briefly, the system consists of a high-speed point-scanning spectral-domain OCT subsystem that uses a superluminescent diode with a center wavelength of 790 nm and bandwidth of 42 nm, providing a nominal axial resolution of 4.7 µm in retinal tissue (refractive index of 1.38). The diffraction-limited lateral resolution of 2.4 µm is realized by an AO subsystem, which dynamically measures and corrects the ocular aberration across a 6.7-mm pupil of the subject's eye. Speed of the AO system was increased during the study. Although this improvement had little impact on our ability to detect and measure the bright, large somas in our images, it did improve the execution of the experiments. The AO system converged faster after an eye blink, was more tolerant of eye motion, and was easier to operate. For 6 of the 8 subjects, the AO subsystem operated at a control loop rate of 158 measurements and corrections per second, and achieving a closed-loop bandwidth of 16 hertz (Hz). For the other 2 subjects, the AO-OCT system operated at 123 measurements and corrections per second and achieved a closed-loop bandwidth of 4.5 Hz.³⁷ Both AO operations used the same 20 × 20 microlenses of a Shack-Hartmann wavefront sensor to sample the eye pupil, and the same deformable mirror (DM97, ALPAO, France) to correct the ocular aberrations. 

The AO-OCT detection arm contains 4 high-speed spectrometers that are connected to a 1 × 4 fiber-based optical switch. By synchronizing the optical switch and spectrometer exposure durations, the system is scalable to A-line speeds of 250 kHz (one-camera mode), 500 kHz (two-camera mode), and 1 MHz (four-camera mode). In this study, we used the 500 kHz 2-camera mode operation. The energy entering the eye of the AO-OCT beam (<430 µW) was within safe limits established by the American National Standards Institute.³⁸ 

Experimental Design for AO-OCT Imaging

The subject's eye was dilated and partially cyclopleged with one drop of 1% tropicamide and aligned to the Indiana AO-OCT system using a bite bar apparatus, which also served to stabilize the head and eye during image acquisition. For the young, healthy subjects, system focus was placed at the border between the INL and the inner plexiform layer (IPL) in order to maximize sharpness of the targeted large somas that we had observed at this retinal depth. This was achieved by systematically varying focus in 7 to 11 µm steps (0.02 to 0.03D) over the IPL and inner portion of the outer nuclear layer (ONL) while simultaneously assessing image sharpness of the acquired images on the real-time display of the AO-OCT control computer. The 1.4 degrees × 1.5 degrees volume images were acquired with the AO-OCT system at 8 locations across the horizontal meridian of the retina: 2 degrees, 3 degrees, 6 degrees, 8 degrees, and 13 degrees temporal (T) and 2 degrees, 3 degrees, and 6 degrees nasal (N) to the fovea (see Fig. 2). Images were cropped to 1.4 degrees × 1.2 degrees in post processing to facilitate analysis. 

Figure 2.

 

The eight retinal locations imaged shown overlaid on a Spectralis SLO image from subject N1. The small, red-bordered boxes denote the eight locations imaged with the Indiana AO-OCT system.

For the three subjects with glaucoma, the system focus was shifted anteriorly to the GCL, as these subjects were imaged for an ongoing study looking at RGCs. The 1.5 degrees × 1.4 degrees volume images were acquired (cropped to 1.4 degrees × 1.2 degrees) in 3 to 4 adjacent locations that traversed into the arcuate defects 3 degrees to 4 degrees nasal and 1 degree to 6 degrees inferior to the fovea, corresponding to a radial eccentricity of 3.1 degrees to 7.2 degrees. The AO-OCT images were also acquired in the opposing hemifield at locations that mirrored those acquired in the arcuate defects. Thus, these additional measurements randomly sampled any arcuate defects that happened to be present in the opposing hemifield and were used only for additional GCL thickness measurements. Because the AO-OCT system focus for these subjects was in the GCL, cellular-level details in the INL were less sharp but the bright, large somas could still be identified and counted. 

For both subject groups at each retinal location, volume images were acquired at a rate of 2.4 Hz over a 5-second duration (12 volumes/video) with a lateral sampling of 1 µm/pixel and an A-line rate of 500,000 lines/second. Fifteen to 20 videos were acquired per location. These volume videos were then processed (described below) to create a single volume image at the location. This process was repeated for each retinal location. Imaging location was controlled by changing the position of a fixation target, which was generated on a liquid crystal display. 

To assess repeatability of our AO-OCT measurements, one of the healthy, young subjects was randomly selected and imaged a second time 6 months later. We returned to the same seven locations by comparing the real-time en face view displayed by our AO-OCT control software with the patterns of vessels and nerve fiber bundles in images captured during the baseline session. 

Data Processing and Analysis

Soma Density and Diameter Measurements

After data collection, we registered and averaged volumes (typically approximately 150 per retinal location) in order to increase signal-to-noise ratio and image contrast of the somas.^1,5,7 ImageJ software was used to view, and then manually count and measure the en face area of all bright, large round reflections at the INL edge. Due to their size and shape we interpreted these as somas. We defined the INL inner edge as being immediately below the intermediate vascular plexus^39,40 but within the INL. To be counted, the bright, round reflections typically needed to be at least 9 microns in diameter, a criterion we found useful as it separated them from the tightly packed smaller-sized reflections that permeated the INL and matched the size of bipolar somas (see Results and Discussion). In addition, to be counted, the bright, round reflections could not be part of any overlying or underlying vasculature, thus minimizing the possibility of vasculature confounding our analysis. Further refinements were needed to determine the center of each soma. Brightness and contrast of the images was regularly adjusted within ImageJ to aid in the visualization and identification of somas. Vertical and horizontal cross-sectional images were utilized to ensure the potential somas were visible in all dimensions. En face area was measured with an oval selector tool from which the diameter was calculated. Morphometric biomarkers of soma density, diameter, and their spatial distributions were measured. Once all somas were counted and measured, image dimensions in degrees were converted to microns by accounting for the axial length of each eye.³³ 

GCL Thickness Measurement

Additionally, the GCL thickness was measured at each of the retinal locations imaged in the glaucomatous subjects. This was achieved by averaging the central 10 AO-OCT B-scans of each volume to obtain a single averaged B-scan. This average better delineated the borders of the inner retinal layers. The GCL was then segmented by manually tracing along its upper and lower edges in the averaged AO-OCT B-scan. Based on this segmentation, GCL thickness was measured at 10 uniformly spaced locations along the averaged B-scan and averaged. This thickness was compared to the expected thickness in age-matched, location-matched normals reported in the literature,⁴¹ and used as a surrogate of GCL soma density,^8,9 and then compared it to the density of the underlying bright, large somas. 

Statistical Analysis

A 1-way ANOVA was performed on the bright, large soma density and soma size variance as a function of retinal eccentricity in the young, healthy subjects. Linear regression was used to test for a correlation between the bright, large soma density and GCL thickness in the glaucoma subjects, and a Bland-Altman test was performed on the repeated soma counts measured in one of the healthy subjects. 

Results

Bright Large Somas in the Young, Healthy Subjects

AO-OCT volume images were successfully acquired in the five young, healthy subjects at all eight retinal locations, enabling counting and measuring of the bright, large somas at the INL inner edge. Figures 3 and 4 illustrate the detailed views obtained by our AO-OCT method. The transverse slices were dissected digitally from our AO-OCT volume images in the same healthy subject. Capillaries are readily apparent, depicted as bright fragmented sections of the intermediate vascular plexus that happen to extend across the slice. Shadows of overlying retinal vasculature are also prominent. The most numerous cell-like structure in the images is a predominately faint, small soma-like body, indicated by red arrowheads in Figures 3 and 4. These structures tightly pack and permeate the entire INL thickness. To quantify their size, we measured the en face diameter of 60 randomly selected small faint cell bodies, which were evenly distributed across the AO-OCT images at 2 degrees, 6 degrees, and 13 degrees temporal retinal eccentricities in one of the healthy subjects (N4). We measured diameters of 5.26 ± 0.65 µm (2 degrees), 6.26 ± 0.51 µm (6 degrees), and 6.30 ± 0.61 µm (13 degrees). We did not measure the size of the small soma-like bodies in the other four healthy subjects but expect similar sizes given their comparable appearance in the images. Of importance for this study are the bright, large soma-like bodies that are prominent in the Figures 3 and 4 images. These soma-like bodies reflect more light than surrounding cells, and their size appears to increase and their prevalence decreases with increased retinal eccentricity. These cells also vary in size and prevalence between subjects, as evident in the Figure 4 images that were acquired at the same retinal location (6 degrees nasal) in the five healthy subjects. 

Figure 3.

 

Bright, large somas are observed at all retinal eccentricities but vary in size and prevalence. AO-OCT en face images at the INL inner edge are shown at the eight retinal locations in the same young, healthy subject (subject N4). The yellow arrowheads point to examples of the bright, large somas. The red arrowheads point to examples of predominately faint, small soma-like bodies.

Figure 4.

 

Bright, large somas are observed in all the healthy subjects but vary in prevalence. AO-OCT en face images at the INL inner edge are shown at 6 degrees nasal retinal eccentricity in the five young, healthy subjects. The yellow arrowheads point to examples of the bright, large somas. The red arrowheads point to examples of predominately faint, small soma-like bodies.

We next quantified our bright, large soma observations, starting with Figure 5 that plots soma density as a function of retinal eccentricity for (left) the individual healthy subjects and (right) the average of all healthy subjects. For these subjects, soma density is greatest near the fovea and decreases monotonically with increasing retinal eccentricity. The maximum and minimum average densities are 550 cells/mm² at 2 degrees temporal and 38 at 13 degrees temporal, respectively; this decrease was statistically significant (P value < 0.001) using a 1-way ANOVA. For comparing trends, the overlying RGC soma densities reported by Curcio et al.⁴² are also plotted in Figure 5 (right) using the secondary y-axis. The RGC soma densities are considerably larger, 50 to 200 times more dense than the bright, large somas in our images. 

Figure 5.

 

Bright, large soma density with retinal eccentricity. (Left) Individual density traces are shown for the young, healthy subjects. (Right) Average density trace is shown for the five young, healthy subjects with individual traces for the glaucoma subjects. RGC soma density from histology in Curcio et al., 1990⁴² (based on 6 young healthy human retinas from 5 subjects) is shown for comparison and plotted against the right ordinate axis. Error bars are ± 1 standard deviation (SD).

Central tendency and variability of soma size are captured by the frequency distributions plotted in Figure 6. Both average soma size and standard deviation (SD) of size increase monotonically with increasing retinal eccentricity, from 12.7 ± 1.8 µm (average ± SD) at 2 degrees temporal to 15.7 ± 3.5 µm at 13 degrees temporal. This variation in average size with retinal eccentricity was statistically significant using a 1-way ANOVA with a P value < 0.001. Size variation also showed a propensity for a bimodal distribution at the larger retinal eccentricities. To assess whether the gap in soma diameter around 18 µm at 13 degrees temporal retinal eccentricity was real or an artifact of the histogram binning, we decreased (0.5 µm) and increased (2.0 µm) the bin size, and found that the gap remained. 

Figure 6.

 

Bright, large soma size distribution with retinal eccentricity in the young, healthy subjects. Distributions are pooled from all five subjects. Averages are denoted by the red arrowheads along the abscissa and also plotted with their respective standard deviations (error bars) in the bottom-right panel. Retinal eccentricities of 2 degrees, 3 degrees, and 6 degrees are plotted with averaged measurements from both temporal and nasal locations. Note that the more central retinal locations have larger numbers of somas compared to the more peripheral locations, thus necessitating different ordinate ranges from 0 to 6 somas to 0 to 100 somas. A bin size of 1 µm was used for all 8 histogram plots.

Bright, Large Somas in the Glaucomatous Subjects

AO-OCT images were successfully obtained inside the arcuate defects of the three glaucomatous subjects (G1-G3). As a representative example, Figure 7 shows the confirmed locations that were imaged in subject G3 and that sampled traversely into the arcuate defect with volume 1 located immediately superior to the defect and volume 4 located at the defect's center. Figure 8 shows a cross-section of the four-volume composite (denoted by the white line in Fig. 7), and Supplementary Video S1 shows a cross-sectional flythrough of the central portion of the same composite. The inner retinal layers appear most normal in the first section on the left (volume 1), which locates just outside the arcuate defect. Volumes 2 and 3 reveal dramatic thinning of the GCL and volume 4 shows essentially no GCL remaining, that is, approaching the axial resolution of our AO-OCT system (4.7 µm) and the small irregularities in the boundary between individual RNFL bundles and GCL. In contrast to this sharp loss, the RNFL, IPL, and INL appear relatively unscathed with their thicknesses preserved across the composite. Supplemental Video S2 shows a similar finding in subject G2, also a cross-sectional flythrough of the central portion of the AO-OCT composite volume image. 

Figure 7.

 

GCL thickness map from Spectralis OCT is superimposed on the reflectance image from Spectralis SLO of subject G3. The thickness map reveals the general location of the inferior arcuate defect, which is outlined in black to highlight its area. Superimposed on the Spectralis map and image composite are the four projected en face AO-OCT images, whose locations were determined by manually aligning to the retinal vasculature in the Spectralis SLO image. En face size of the cropped AO-OCT images is 1.4 degrees × 1.2 degrees. The blue circles are centered on the fovea with diameters of 1, 3, and 6 mm (3.3 degrees, 10 degrees, and 20 degrees). The white line indicates location of composite cross-sectional AO-OCT image displayed in Figure 8. The white numbers correspond to image locations of cross-sections in Figure 8.

Figure 8.

 

Composite cross-sectional view of the arcuate defect in subject G3. The cross section extends across the 4 AO-OCT volume images that are shown in Figure 7 and labeled 1 through 4 in the top right corner of each section. The green arrowheads denote the thickness of the GCL at example points. Supplemental Video S1 shows a cross-sectional flythrough of the central portion of the AO-OCT composite volume image.

However, despite little apparent loss in INL thickness, we find that in all three glaucomatous subjects the number of bright, large somas at the inner edge of the INL is dramatically reduced from that of the healthy subjects (see Figure 5 [right] soma density traces), obtained using the same protocol for identifying and counting the bright, large somas in the young, healthy subjects. In all cases, density of the somas decreased sharply near the arcuate edge (between volumes 1 and 2) and plateaued at exceedingly low numbers of somas (between the two volumes deepest into the arcuate defect). Average soma density at the two locations deepest into the arcuate defects of the three glaucomatous subjects (e.g. locations 3 and 4 shown in Figure 8 for subject G3) was 9.95 ± 7.98 cells/mm², 20 times lower than the average soma density at 6 degrees nasal in the healthy subjects. Average thickness of the overlying GCL at these same locations was 6.3 ± 1.7 µm. We found soma density decreased with decrease in the GCL thickness across the arcuate defects regardless of subject (Fig. 9 [left]) but the correlation was significant only when the measurements across the three subjects were pooled (R² = 0.46, P = 0.02). A similar correlation (R² = 0.43, P = 0.04) was obtained for the 10 measurements in the opposing hemifield that mirrored locations in the arcuate defects examined (Fig. 9 [right]). 

Figure 9.

 

Bright, large soma density as a function of the overlying GCL thickness of the three glaucomatous subjects. (Left) Measurements in the arcuate defects are separated by subject. (Right) For comparison, the 10 measurements in the opposing hemifield of the glaucomatous subjects are pooled. The dashed lines denote linear regression with fits specified by R² values. Regression slopes are significant only when the measurements across the three subjects are pooled: P values of 0.02 (left plot) and 0.038 (right plot). GCL thickness was measured from the AO-OCT images.

To test for repeatability, differences between soma counts taken 6 months apart were plotted against their mean following the Bland–Altman method (Fig. 10).⁴³ There was no proportional bias with mean soma count (Kendall's tau = 0.33, P value = 0.37). Bias and 95% limits of agreement were −0.79% ± 24.8%. 

Figure 10.

 

Bland–Altman plot showing difference against mean for repeated measurements of the bright, large somas. Somas were recounted at 7 locations in one of the healthy subjects in a second imaging session 6 months later. The solid and dashed lines indicate the mean difference (bias) and 95% confidence intervals (±1.96 standard deviations of the mean differences), respectively.

Discussion

In all the eyes we imaged, we observed bright, large somas within the INL that were sparsely distributed along the border of the IPL. We successfully characterized the densities, sizes, and spatial distributions of these somas across the horizontal meridian of the macula in five healthy young subjects and within the arcuate defects of three glaucomatous subjects. Whereas our study focused on a single meridian and specific locations within arcuate defects, our measurements should generalize well to other locations due to the radial symmetry of the macula about the fovea and the common neural layering of the retina across the posterior pole. 

The bright, large somas we observe must be attributed to cell types within the INL, with major types being horizontal, bipolar, Müller, amacrine, and dRGCs. From histology, we know that these cell types coarsely stratify the INL.¹⁴ Here, we summarize the key findings from our study and compare our measurements to the existing literature to assess potential attributions for these somas. Our evidence points most strongly toward a dRGC contribution, but subpopulations of amacrine cells may also contribute. 

Soma Density and Distribution

We observed that the density of the bright, large somas in our AO-OCT images peaks near the fovea (average of 550 cells/mm² at 2 degrees temporal, the closest location we measured) and decreases monotonically with retinal eccentricity (average of 38 cells/mm² at 13 degrees temporal, the furthest location we measured). Despite a much lower soma density, this trend follows that of the overlying GCL soma density reported by Curcio and Allen (1990; see Fig. 5), except near the fovea where the GCL soma density peaks at 3 degrees and decreases rapidly at smaller eccentricities. 

We believe this discrepancy is unlikely to reflect methodological differences between AO-OCT and histology. Previously, we used the same AO-OCT instrument to measure RGC density and obtained densities consistent with Curcio and Allen (1990; refer to fig. 3 in Liu et al., 2017).¹ Noise in our AO-OCT measurements is also an improbable cause. The intrasubject test-retest bias of our soma density measurements was only 0.79%, with an SD of 12.7% for measurements taken 6 months apart (see Fig. 10). This SD is three times smaller than our intersubject SD in soma density as a percentage of the mean soma density in Figure 5, which we measured at 38.1% (range = 28.6% to 62.5%) and averaged over the 8 retinal locations. This suggests that the measurement error is relatively small compared to the natural variability between subjects. 

Soma Size and Distribution

The central tendency and variability of the size of bright, large somas in our AO-OCT images increased monotonically with retinal eccentricity, following a similar trend to the overlying GCL somas. This trend is evident in the frequency distributions shown in Figure 6, where the average is marked by red arrowheads and plotted in the bottom-right panel. Figure 11 provides a comparison of these same average soma diameters with our previously reported GCL soma diameter measurements (green diamonds) in different subjects using the same AO-OCT instrument.¹ Across the five retinal eccentricities we imaged, our bright, large somas (see the black circles in Fig. 11) were, on average, 9.9% larger in diameter than the GCL somas in Liu et al.¹ (green diamonds) with a difference of 1.39 ± 0.47 µm (average ± SD). 

Figure 11.

 

Comparison of ganglion cell soma size and its range in human and non-human primates. Average diameter of the bright, large somas in the young, healthy individuals are compared with (1) in vivo measurements of GCL somas in human¹ and (2) histological measurements of dRGC and outer-stratifying ipRGC (os-ipRGC) somas in human and primate donor tissue from the literature. Four of the six histological studies reported a single average soma size for somas measured over a wide range of retinal eccentricities. Bunt and Minckler²² reported two size clusters of dRGCs for the same retinal eccentricity range. Chandra et al.⁵³ reported sizes up to 47 degrees retinal eccentricity for os-ipRGCs with somas in the INL, but because their measurements were sparse and roughly clustered, we analyzed them as 3 groups (5 degrees, 20 degrees, and 39 degrees) to simplify our plot. Sizes for Nasir-Ahmad et al.,¹⁶ Liao et al.,⁵⁴ and Jusuf et al.⁵⁵ are for os-ipRGC somas, roughly half of which are expected to be displaced in the INL. Vertical error bars denote standard deviation, except for Bunt and Minckler who reported a range.²² Horizontal error bars denote retinal eccentricity range where somas were measured.^16,22,54,55 Plotted soma diameters are based on 1713 bright, large somas (5 human subjects) in this study, tens of thousands of GCL somas (4 human subjects) in Liu et al. (2017), 74 cells (5 human retinas) in Nasir-Ahmad et al. (2019), 76 cells (1 human retina) and 281 cells (11 retinas from 10 macaque monkeys) in Liao et al. (2016), 35 cells (2 human retinas) in Chandra et al. (2019), 8 cells (8 eyes from 7 marmosets) in Jusuf et al (2007), and an unspecified cell number (4 macaque monkeys) in Bunt and Minckler (1977).

We also noted evidence of bimodality in the size distribution of our bright, large somas for retinal eccentricities at and greater than 6 degrees (see Fig. 6), a trend that is consistent with the findings in Liu et al.¹ for GCL somas (as seen in their figure 4A). 

Bipolar, Müller, and Horizontal Cells Not Likely to Contribute to the Bright, Large Somas

In our AO-OCT images, the most numerous cell-like structures in the INL are the predominately faint, small soma-like bodies whose diameters we measured at 5.26 ± 0.65 µm (2 degrees), 6.26 ± 0.51 µm (6 degrees), and 6.30 ± 0.61 µm (13 degrees; see the red arrowheads in Figs. 3, 4) and that densely populate the entire INL. Their size and spatial distribution are suggestive of bipolar somas, which are the most numerous cells in the INL and found in all its sublayers. Literature reports of bipolar soma sizes in human and macaque monkeys vary but typically range from 6 to 8 µm,^44–47 which is consistent with our 5.26 to 6.30 µm range. Their density, ranging from 40,000 to 60,000 somas/mm², is higher than that of RGCs in the GCL, even where RGCs concentrate in a ring around the fovea.⁴⁸ These characteristics led us to conclude that these small soma-like bodies in the INL in our AO-OCT images are most likely bipolar somas, ruling out bipolar cells as potential contributors to the bright, large somas at the IPL border. 

We can also probably rule out Müller and horizontal cell somas as candidates because these cell types stratify the middle and outer (adjacent to the OPL) sublayers of the INL, respectively. However, it should be noted that we are unable to detect any large somas in our AO-OCT images in the middle and outer sublayers, where Müller and horizontal somas are expected to be. This prevents us from confirming their presence and assessing their appearance relative to the bright large somas. One possible explanation for their absence is that Müller and horizontal somas backscatter less light and are more transparent due to better refractive index matching with their internal content and the surrounding INL composition, compared to, for example, RGC somas. This reduced signal and contrast would hinder their detection in the AO-OCT images. Supporting this explanation is the observation that the reflectance of the INL is six times dimmer than that of the GCL in clinical (non-AO) OCT images. This difference is derived from the normalized retinal reflectance profile in clinical OCT images⁴⁹ and the typical OCT system sensitivity (−95 decibel [dB]) and dynamic range in retinal tissue (35 dB). The decreased reflectance suggests reduced backscatter and improved refractive index matching of neurons and glial cells in the INL compared to those in the GCL. This could also explain why any RGC somas displaced to the INL would appear brighter than the surrounding neural tissue in that layer. 

Taken together, this evidence and line of reasoning suggest that bipolar, Müller, and horizontal cells are unlikely candidates for attribution. This leads us to consider dRGCs and amacrine cells as the primary candidates. 

dRGCs Likely Contribute to the Bright, Large Somas

We found that the density of the bright, large somas corresponds to 0.5% to 2% of the expected overlying RGC density⁴² (compare the black and red curves in Fig. 5 plot), a fraction that is similar to the 2% to 3% of the total RGC population reported to be displaced to the INL in mouse^23,50 and rat⁴⁴ retina. However, we are not aware of any literature that reports the density distribution of dRGCs in humans, except for ipdRGCs (discussed below), and thus we cannot confirm that the small fraction of dRGCs observed in the mouse and rat also occurs in humans. Indeed, the distribution of dRGCs is known to vary among species (mouse,^17,18 rat,^18–20 hamster,²¹ guinea pig,¹⁸ rabbit,¹⁸ and monkey,^18,22) and is affected by other factors, such as the level of melanin pigmentation in the choroid and retinal pigment epithelium.^19,23,24 In the macaque monkey retina, for example, the density of dRGCs has been reported to be greatest in the peripapillary region and decreased in the peripheral retina.²² In contrast, in mice, dRGCs are present throughout the retina but are more numerous in the midperipheral and peripheral retina.^17,18 Guinea pigs also show a different distribution, with more numerous dRGCs in the central retina compared to the peripheral retina.¹⁸ 

A similar challenge occurs when attempting to compare our soma size distribution, as we are not aware of any literature on the size of dRGCs in humans, except for ipdRGCs (discussed below). However, dRGC soma sizes have been measured in the macaque monkey,²² as shown in Figure 11 alongside our measurements. Bunt and Minckler reported only two size clusters (10–13 µm and 15–25 µm) for the entire retinal eccentricity range of 0 degrees to 27.4 degrees (converted from millimeters using 0.223 mm/degrees) with most somas reportedly found in the peripapillary region. The coarseness of their measurements and the range of soma sizes that we measured over a fraction of their retinal eccentricity range, that is, from 2 degrees to 13 degrees, limit the comparison of the 2 studies. However, it is noteworthy that our bimodal size distributions in Figure 6, for example, at 6 degrees nasal retinal eccentricity, have modes (sizes of 12 µm and 18 µm) that are consistent with the two size clusters (11.5 µm and 20 µm) that they report for dRGCs. 

Interestingly in Liu et al.,¹ the lower- and higher-diameter modes of the bimodal distribution for GCL somas were interpreted as distinguishing the two primary subtypes of RGCs in the macula (midget and parasol). This was because the size distributions of the modes fell within the range reported for midget and parasol somas in the human literature. Furthermore, 86% to 91% fell into the lower-diameter mode for the GCL somas, consistent with fractional estimates of midget RGCs by Dacey⁵¹ and Drasdo et al.⁵² Although we did not quantify the fraction of bright, large somas falling into the lower- and higher-diameter modes in Figure 6, it appears that they are roughly equal. These similar distributions could suggest the presence of two or more subtypes of cells, such as dRGCs or a contingent of amacrine cells (as discussed below). If attributed to RGCs, this could imply that RGC types with larger somas are more likely to be displaced than those with smaller somas, given the greater prevalence of smaller somas in the overlying GCL. More work is needed to better characterize this bimodal distribution, which is limited in this study due to the small number of somas measured at higher retinal eccentricities. 

Unlike the general population of dRGCs, ipRGCs have been studied extensively in humans and nonhuman primates.^16,53–56 Three recent reports measured the density of outer-stratifying ipRGCs (os-ipRGCs) in humans as a function of retinal eccentricity,^16,54,56 roughly half of which are expected to be displaced in the INL. One reported 8 to 10 cells/mm² in the central retina tapering off to 2 to 3 cells/mm² in peripheral retina,⁵⁴ the second a peak density of 10 to 22 cells/mm² at about 7 degrees eccentricity dropping below 6 cells/mm² at 27 degrees,¹⁶ and the third a peak density in the central retina close to the fovea of 43 cells/mm² dropping in the peripheral retina to 6.9 cells/mm².⁵⁶ These densities are notably smaller than the bright, large soma densities we measured up to 8 degrees retinal eccentricity (average of 543 and 77 cells/mm² at 2 degrees and 8 degrees temporal in Fig. 5) and still smaller than the density we measured at our largest retinal eccentricity of 13 degrees (average of 38 cells/mm²). However, this difference should not be unexpected assuming the bright, large somas observed in our images include other dRGC types. ipdRGCs represent only a fraction of the total dRGC population, as for example, 20% reported in humans by Chandra et al.⁵³ Regardless of absolute count, the human histological measurements show a similar trend to our study, with higher density near the fovea. 

In addition, ipRGC somas are among the largest RGCs in primates,⁵⁴ and thus may be of interest due to the large size of the somas that we observed at the INL edge. In two recent human studies,^16,54 the size of os-ipRGC somas was reported to vary little with retinal eccentricity with examined ranges of 3.3 degrees to 63.3 degrees and 6.7 degrees to 63.3 degrees (converted from millimeters using 0.3 mm/degrees) (see Fig. 11). As shown in Figure 11, the average os-ipRGC soma sizes of 19 µm and 21 µm are similar to the average size (20 µm) of the large-sized cluster of dRGCs in macaque monkeys reported by Bunt and Minckler.²² They are also consistent with the higher-diameter mode somas in our data (see Fig. 6), such as the cluster of large somas at 13 degrees temporal that range in size from 18 to 21 µm. Our data show that the bright, large somas increase in size with retinal eccentricity, so our 13 degrees measurement (the largest retinal eccentricity we examined) is probably the most meaningful to compare to these other studies, although they measured as far out as 63.3 degrees. 

The ipdRGC soma sizes from a third recent human study are also plotted in Figure 11 at 3 retinal eccentricities: 5 degrees, 20 degrees, and 39 degrees. The average soma sizes of 17.6 µm and 19.6 µm at the 2 larger eccentricities are similar to the average size (20 µm) of the large-sized cluster of dRGCs in macaque monkeys reported by Bunt and Minckler,²² the average size (21 µm) of os-ipRGCs in humans in Nasir-Ahmad et al.,¹⁶ the average size of os-ipRGCs (19 µm) in humans in Liao et al.,⁵⁴ and the range in size from 18 to 21 µm of our cluster of large somas at 13 degrees temporal. In contrast, the average soma size of 24.8 µm at 5 degrees eccentricity in Chandra et al. is larger than even the largest sized soma (22 µm) we measured at a similar eccentricity (6 degrees, nasal and temporal). Although more work is needed to understand this discrepancy, it is worth noting that Chandra et al. included giant M1 ipdRGCs, which as reported by Hannabil et al.⁵⁶ are large in size (28.7 ± 1.0 µm), sparse, and displaced deep in the INL close to the outer plexiform layer. Our study may have missed these cells as we focused on the INL border with the inner plexiform layer. 

Taken together, the density, size, and brightness of our bright, large somas share trends with the overlying RGCs and are consistent with certain findings in the histological literature related to dRGCs and os-ipRGCs. Unfortunately, the absence of similar morphological measurements for dRGCs in the human literature hinders definitive confirmation. In contrast, os-ipRGCs in humans have undergone extensive study in the histological literature. When we compare our density and size measurements to the data reported in these studies, it suggests that os-ipRGCs may contribute to the bright, large somas we observe, albeit to a limited extent. Notably, their density is much lower than that of our bright, large somas, and their size corresponds only to our higher-diameter mode somas. 

A Subpopulation of Amacrine Cells May Contribute to the Bright, Large Somas

Unlike dRGCs, amacrine cells are more numerous and densely distributed along the IPL border of the INL. We can estimate their spatial density distributions in humans by converting marmoset histological measurements,⁵⁷ using retinal magnification factors of 0.15 mm/degrees for marmosets⁵⁸ and 0.3 mm/degrees for humans,⁵⁹ and accounting for marmosets’ 2 times lower visual acuity compared to humans.⁶⁰ We estimate that their density peaks at approximately 16,000 cells/mm² at 3 degrees to 4 degrees retinal eccentricity and gradually declines to 10,000 cells/mm² at 13 degrees retinal eccentricity, the maximum covered in our study. 

The high density of amacrine cells leads to the formation of multiple rows of stacked somas at the IPL border of the INL, which contrasts with the sparse distribution of isolated bright, large somas that we have observed. This discrepancy indicates that the entire amacrine cell population is not a good model for the bright, large soma distribution, but perhaps specific subpopulations could be. There are over 25 different types of amacrine cells in humans, each varying in morphology and density.⁶¹ For instance, the amacrine subtype identified with nitric oxide synthase labeling has an estimated peak density of 400 cells/mm²,⁵⁷ which is comparable to the density of our bright, large somas. Another amacrine cell subpopulation, dopaminergic amacrine cells, have a lower density (peaks at 50 cells/mm² near the fovea) but have large soma sizes (range = 11.2–16.6 µm) that are comparable to what we observe.⁶² Most amacrine cell subpopulations have smaller somas compared to RGCs,⁴² probably making them too small to account for the higher-diameter-mode somas that we measured (see Fig. 6). It would not be surprising that specific amacrine cell subpopulations contribute to the smaller-sized somas that we observe. 

Glaucoma Effect on the Bright, Large Somas

We found that the bright, large somas were dramatically reduced in the arcuate defects of the three glaucoma subjects, suggesting that these cells are highly vulnerable to the disease. Average soma density was 9.95 cells/mm², 22.9-fold lower than the 227.4 ± 88.2 cells/mm² that we measured in young, healthy individuals at an equivalent retinal location assuming radial symmetry (see Fig. 5) and corresponding to a 95.6% loss in RGCs. This loss is much larger than the expected 3.8% to 19.1% loss due to the age difference between the young, healthy subjects and glaucomatous subjects, virtually eliminating the possibility that this is primarily an age effect. Given that RGCs are the primary cell type that dies in glaucoma, this high loss rate is consistent with bright, large somas being RGCs. 

The structural effects of glaucoma on amacrine cells have been examined in numerous studies, although mostly in rodents,^63–68 exceedingly little in nonhuman primates,⁶⁹ and none that we know of in humans (Suresh Viswanathan, SUNY College of Optometry, personal communication, April 29, 2022). In retinal sections of the macaque monkey that underwent experimental glaucoma, Frishman et al. (1996) found no evidence of a loss of gamma-aminobutyric acid (GABA)-ergic amacrine cells, which comprise more than half of the amacrine cell population in the human retina, nor did they find loss of cells in the INL.⁶⁹ In a rat model of experimental glaucoma and optic nerve transection, Kielczewski et al. (2005) investigated loss of amacrine cells that were labeled primarily for GABA or glycine in the GCL and INL.⁶⁵ They found no definitive evidence that glaucomatous damage leads to loss of cells, other than RGCs, in the retina. In contrast, other mouse and rat studies found subpopulations of amacrine cells to be vulnerable.^63,64,66,67 More recently using a mouse model of experimental glaucoma, Akopian et al. (2019) found calretinin- and GABA-ergic amacrine cells were highly vulnerable to glaucomatous damage, whereas choline acetyltransferase (ChAT)-positive and glycinergic amacrine cells were unaffected, regardless of location (GCL or INL).⁶⁸ 

These studies are challenging to integrate and to extrapolate to humans due to their use of different animal models and varying reports of amacrine cell vulnerability. Collectively, however, these studies suggest that the overall population of amacrine cells is less vulnerable to glaucoma than are RGCs, and they do not support the notion of amacrine cells being lost at the high levels observed in our glaucoma subjects. 

It is interesting to speculate whether the degree to which the bright, large somas are lost in the arcuate defects of the glaucoma subjects follows that of the overlying GCL somas. Although we did not count somas in the GCL, we did measure the GCL thickness (see Fig. 9) and thus can use it as a surrogate of GCL soma density. Average GCL thickness in the bottom two volumes of the arcuate defects for all glaucoma subjects (6.3 µm) was, as expected, thinner than in age- and location-matched normals measured with Spectralis OCT (35–45 µm).⁴¹ Our pooled soma and thickness measurements across the three arcuate defects (see Fig. 9, left) and in the opposing hemifield of the same patients with glaucoma (see Fig. 9, right) confirmed a significant correlation between the density of bright, large somas and the thickness of the overlying GCL. However, considerable variation was also found. In particular, whereas a near-zero GCL thickness (<10 µm) was always associated with the presence of very few bright, large somas, the opposite was not always true. Numerous locations with a relatively thick GCL were found with soma densities no greater than that at locations with near-zero thickness. Although more work is needed to interpret this apparent oddity, the dramatic loss of these somas in our glaucoma subjects indicates a purely neural change that may complement OCT thickness measurements as a potential biomarker of disease progression. Unlike thickness measurements that require offsetting to correct for presumed accumulation of non-neural tissue in the RNFL and GCL (which OCT cannot distinguish from neural tissue),^70,71 soma densities can reach zero, which we indeed find in our measurements. 

Limitations of the Study

Key limitations of our study include the small number of subject participants and the age difference between our normal, healthy subjects and glaucoma subjects (26.6 ± 1.3 vs. 64.7 ± 4.7 years). Although the loss due to aging is expected to be much smaller than what we measured in the glaucoma subjects, this needs to be confirmed and is a subject for future work. 

Conclusions

This study demonstrates the utility of AO-OCT in visualizing and quantifying individual inner retinal neurons in both normal, healthy subjects and those with diseases. We measured the morphological properties of the bright, large somas at the inner edge of the INL in our AO-OCT images and compared them to the INL cell types reported in histological literature to determine the most likely attribution for these somas. We also observed significant loss of these bright, large somas in our glaucoma subjects. Taken together, our evidence points most strongly toward a dRGC contribution, but subpopulations of amacrine cells may also contribute. Our ability to identify these somas was ultimately limited by the lack of similar morphological measurements of dRGC (excluding the well-studied ipdRGCs) and amacrine cell somas in the human histological literature. 

Acknowledgments

This study was funded by NIH Grants EY018339 and EY029808. 

Disclosure: M.E. Marte, None; K. Kurokawa, (P); H. Jung, None; Y. Liu, None; M.T. Bernucci, None; B.J. King, None; D.T. Miller, (P)