The Relationship between Retinal Ganglion Cell Axon Constituents and Retinal Nerve Fiber Layer Birefringence in the Primate

Abstract

Purpose.: To determine the degree of correlation between spatial characteristics of the retinal nerve fiber layer (RNFL) birefringence (Δn _RNFL) surrounding the optic nerve head (ONH) with the corresponding anatomy of retinal ganglion cell (RGC) axons and their respective organelles.

Methods.: RNFL phase retardation per unit depth (PR/UD, proportional to Δn _RNFL) was measured in two cynomolgus monkeys by enhanced polarization-sensitive optical coherence tomography (EPS-OCT). The monkeys were perfused with glutaraldehyde and the eyes were enucleated and prepared for transmission electron microscopy (TEM) histologic analysis. Morphologic measurements from TEM images were used to estimate neurotubule density (ρ_RNFL), axoplasmic area (A _x) mode, axon area (A _a) mode, slope (u) of the number of neurotubules versus axoplasmic area (neurotubule packing density), fractional area of axoplasm in the nerve fiber bundle (f), mitochondrial fractional area in the nerve fiber bundle (x _m), mitochondria-containing axon profile fraction (m _p), and length of axonal membrane profiles per unit of nerve fiber bundle area (L _am/A _b). Registered PR/UD and morphologic parameters from corresponding angular sections were then correlated by using Pearson's correlation and multilevel models.

Results.: In one eye there was a statistically significant correlation between PR/UD and ρ_RNFL (r = 0.67, P = 0.005) and between PR/UD and neurotubule packing density (r = 0.70, P = 0.002). Correlation coefficients of r = 0.81 (P = 0.01) and r = 0.50 (P = 0.05) were observed between the PR/UD and A _x modes for each respective subject.

Conclusions.: Neurotubules are the primary source of birefringence in the RNFL of the primate retina.

Glaucoma is a progressive optic nerve disease characterized by a loss of retinal ganglion cell (RGC) axons and thinning of the retinal nerve fiber layer (RNFL). Axon loss can be as much as 50% before the disease is clinically detectable. ¹ Imaging devices have been developed to quantitatively measure RNFL thinning and birefringence loss. For example, optical coherence tomography (OCT) produces cross-sectional images of the retina allowing measurement of RNFL thickness. ² Polarization-sensitive OCT (PS-OCT) ³ and scanning laser polarimetry (GDx; Carl Zeiss Meditec, Inc. Dublin, CA) ⁴ measure phase retardation in the RNFL that arises from anisotropic structures in the RNFL. ⁵ Changes in RNFL birefringence may precede RGC death. 

Birefringence (Δn) is a dimensionless measure of the anisotropy of the refractive index (n) of a material and is given by the difference between the extraordinary index (n _e) and ordinary index (n _o), where n _o and n _e are the refractive indices for polarization perpendicular and parallel to the axis of anisotropy respectively (Δn = n _e − n _o). Phase retardation is proportional to birefringence according to the expression Δn = λ₀PR/360°ΔZ where λ₀ is the free-space wavelength, PR is the phase retardation, and ΔZ is the thickness of the material. ⁶ In general, Δn of the RNFL is weakly wavelength dependent across visible and near infrared wavelengths. ⁷ The RNFL demonstrates form birefringence that originates from anisotropic cylindrically shaped cellular structures in the RGC axons. ^8,9 The major cylindrical structures of the RGC axonal cytoskeleton are neurotubules (NT), neurofilaments, and neurotubule-associated proteins. ¹⁰ Huang and Knighton ⁵ identify neurotubules as the source of birefringence from the RNFL. Theoretical analysis has attributed RNFL reflectance to the cylindrically shaped mitochondria and axonal membranes. ⁹  

It is postulated that either changes in neurotubule density of RNFL bundles or neurotubule packing densities within the axons themselves are responsible for the differences in birefringence surrounding the ONH. ¹¹ Neurotubule density (ρ_RNFL) is defined in this report as a scalar estimate of the number of neurotubules per unit area of RNFL tissue (NT/μm²). There is strong evidence that the regional characteristic ρ_RNFL surrounding the optic nerve head (ONH) is a source of birefringence, ^5,12,13 but it has not been clear whether ρ_RNFL is the only source of birefringence. ⁹ RNFL birefringence has been investigated as a possible diagnostic to detect early subcellular changes in glaucoma before there is any measurable change in RNFL thickness. ^5,12,13 Fortune et al. ¹³ injected colchicine into the vitreous of nonhuman primate eyes and observed neurotubule disruption with a reduction of RNFL birefringence without any accompanying change in RNFL thickness. These results show that decreases in the number of neurotubules occur before thinning of the RNFL, and in situ measurement of ρ_RNFL can be used as a diagnostic for cytoskeletal changes in RGC axons. ¹³ In this study, we compared the measured birefringence signal in the peripapillary retina with the corresponding measured anatomic features to explore the origin of the RNFL birefringence. 

Methods

Measurement of RNFL Birefringence

The retinas of two healthy 6-year-old female, 5.9 and 7.7 kg, cynomolgus monkeys (subjects 1853 and 57204) were imaged over a period of 62 days. Intraocular pressures and subjective visual function were normal. All experimental procedures were approved by The University of Texas at Austin Institutional Animal Care and Use Committee (IACUC, Protocol 05021401) and conformed to all United States Department of Agriculture (USDA), National Institutes of Health (NIH) guidelines, and the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. A combination of IM ketamine (10 mg/kg) and xylazine (0.25 mg/kg) was used to anesthetize the monkeys, and anesthesia levels were monitored by a certified veterinary technologist. A retrobulbar injection of 0.5% xylocaine and 0.375% marcaine was given to immobilize the eye that was imaged. Pupils were dilated by using 1 drop of 1% cyclopentolate and 1 drop of 1% tropicamide. One drop of 10% methylcellulose, to prevent dehydration, was placed in the eye to be imaged, and a contact lens was placed on that eye. A contact lens that rendered the monkey slightly myopic was used to ensure that light was focused on the internal limiting membrane (ILM). 

The peripapillary RNFL thickness (Z _RNFL) and single-pass phase retardation maps of the two cynomolgus monkeys were measured by using an enhanced polarization-sensitive optical coherence tomography (EPS-OCT) system described previously. ^11,14 EPS-OCT consists of a PS-OCT instrument combined with a nonlinear fitting algorithm to determine PR and Δn with high sensitivity in weakly birefringent tissues. The system utilizes a mode-locked Ti:Al₂O₃ laser source (λ₀ = 830 nm, Δλ = 55 nm) linearly polarized at 45°. The Z _RNFL and PR maps are used to construct phase retardation per unit depth (PR/UD) area maps given in units of degrees of retardation per 100 μm of RNFL thickness (degrees/per 100 μm) by dividing local PR by Z _RNFL. 

The PR map was measured in an area and ring scan configuration. Each area map consisted of 24 evenly spaced (15°) radial scans containing 2, 6, or 8 clusters distributed uniformly between 1.4 and 1.9 mm from the center of the optic nerve head. A cluster comprised 36 or 64 A-scans individually acquired over a respective 0.9- or 1.6-mm² area. Ring scans are derived from sector scans spread 5° apart. The right eye of each subject was imaged twice on separate days, to assess reproducibility of retinal birefringence measurements. 

A single peripapillary map was acquired in approximately 45 minutes. Laser power incident on the cornea was 2.8 mW during lateral scanning and 1.7 mW while stationary for both scan configurations. Approximate laser spot size at the retinal surface was 30 μm. Axial resolution was 5 μm (determined by the coherence length of the laser source in air). 

RGC Axon Organelle Sampling

The PR/UD sinusoid pattern ^11,15,16 surrounding the ONH was used to determine the angular interval around the ONH that RGC axon organelles must be sampled. Using the Nyquist criterion, RGC axon organelles were measured in eight angular sections (octants) of the peripapillary retina in the first eye sampled (1853 OS). Based on the correlation result between birefringence and ρ_RNFL of the first eye, it was determined that 16 angular interval sections should be sampled around the ONH of the second eye (decreasing bundle variance within a region) to increase the power to 80% and detect a significant correlation between ρ_RNFL and birefringence (Fig. 1). Each bundle within an angular section was chosen at random to eliminate bias. Nerve bundles were sampled from a 0.15-mm² rectangular area from each angular section, which contained approximately 15 to 20 nerve fiber bundles; therefore, 3 nerve fiber bundles approximate 15% to 20% and 7.5% to 10% of the sampled area of subjects 57204 and 1853, respectively. The exact number of bundles within each angular section was not known. 

Figure 1.

 

The peripapillary RNFL was divided into 8 and 16 angular sections. Each angular section is labeled by the symbols superimposed on the area surrounding the ONH (white circle) in subjects (A) 1853 OS and (B) 57204 OD, The uncertainty in RNFL bundle positions are ±22.5° and ±11.25° for subjects 1853 and 57204, respectively. Adapted with permission from Hogan MJ, Alvarado JA, Weddell JE. Histology of the Human Eye. Philadelphia: WB Saunders; 1971.

Preparation of Ocular Tissue

Two days after EPS-OCT imaging was completed, both monkeys were anesthetized and transcardially perfused with a fixative solution (pH 7.4) consisting of 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M sodium phosphate buffer. Each animal was first brought down to a surgical level of anesthesia with a combination of IM ketamine (10 mg/kg) and xylazine (0.5 mg/kg). Pentobarbital sodium (10 mg/kg) was given first for surgical anesthesia and then an IV overdose was administered. 

After each perfusion was complete, all four eyes were enucleated and immersed in 180 mL of the primary fixative solution from 1 to 3 hours before dissection from the periorbital tissue and removal of the posterior hemisphere of the eye. A 360° incision was made at the ora serrata, and the anterior eye structures including the ciliary body, iris, lens, and cornea were removed from the posterior eye cup. A notch and suture placed in the nasal sclera before enucleation remained visible throughout the specimen preparation process to maintain proper orientation. All posterior eye cups were then simultaneously immersed in 180 mL of the primary fixative solution for approximately 1.5 hours. 

Eye cups were washed in three 30-minute changes of 0.1 M sodium phosphate buffer (180 mL each, pH 7.4) and postfixed in 90 mL of 2% osmium tetroxide (aqueous) solution for 1 hour. They were washed again two times for 30 minutes per wash in 0.1 M sodium phosphate buffer (180 mL each, pH 7.4) and then one more time in 180 mL of distilled water for 30 minutes. The eye cups were then dehydrated through a series of increasing ethanol concentration solutions (50% for 14 hours, 75% for 1 hour, 100% for 1 hour, and 100% for 1 hour, all in 180 mL) followed by two immersions in acetone (50 mL 100% acetone for 1 hour each). The posterior eye cups were then infused with 1:3 low-viscosity resin (EMBed-812; Electron Microscopy Sciences, Hatfield, PA) in acetone for 1 hour, after which they were covered and left in 2:3 resin for 19 hours overnight. The following day, the eye cups were placed into fresh solutions of 100% resin and left in uncovered containers for approximately 5 hours. They were then placed anterior side down into cubical embedding molds. The molds were put into a 60°C oven for 21 hours overnight, during which the resin solution polymerized, and the eye cups were removed the following day. 

Superficial lines intersecting at the center of the ONH profile were used to divide the cube mold of the embedded eye into wedge-shaped angular sections. Each sample wedge was carefully carved from the cube to prevent excess tissue removal surrounding the ONH. Glass knives were used to create a flat rectangular block face (500 × 300 μm). 

Transmission Electron Microscopy

Ultrathin and semithin sections were sampled 1.6 to 1.9 mm from the ONH in regions (Ultracut UCT Ultramicrotome; Leica Microsystems GmbH, Ernst-Leitz-Strasse, Germany). Semithin sections (0.5 μm) were cut and stained with toluidine blue before ultrathin sections (60 nm) to ensure that the block's orientation on the microtome would result in transverse cuts. Several ultrathin sections were placed on copper, α-numeric–indexed grids (Electron Microscopy Sciences) and stained with 2% uranium acetate and 0.3% lead citrate before TEM analysis. 

Gray-scale digital images were captured on a transmission electron microscope (EM208 TEM; Phillips, Eindhoven, The Netherlands) equipped with a digital camera system (Advantage HR 1 MB; AMT, Danvers, MA). Each bundle was photographed at low (2,200×–2,800×), medium (11,000×–14,000×), and high (28,000×–44,000×) magnification. 

Identification of RGC Axons and Organelles

All morphologic measurements were made using Image J (ver. 1.36; developed by Wayne Rasband, National Institutes of Health, Bethesda, MD; available at http://rsb.info.nih.gov/ij/index.html) on a laptop computer (Hewlett Packard Compaq, Palo Alto, CA). Low magnification TEM images were used to measure nerve fiber bundle area (A _b; Fig. 2). 

Figure 2.

 

Low-magnification (2200×) image of a RNFL bundle from subject 57204 in angular section N_SI. Scale bar, 2 μm.

A montage of each nerve fiber bundle was created from a series of medium-magnification images (Photoshop, ver. 8.0; Adobe Systems, San Jose, CA; Fig. 3). RGC axons were captured in high-magnification images to measure respective axonal cross-sectional area (A _a), neurotubule number (k), mitochondrial area (A _m), noncytoskeletal organelle areas (A _o), and unidentifiable features of the RNFL bundle. Neurotubules in each of the selected RGC axons were manually counted using the Cell Counter plug-in feature in Image J. A single person identified and counted the RGC organelles in subject 1853 and two different people identified and counted organelles for subject 57204. 

Figure 3.

 

(A) A montage of a nerve fiber bundle from subject 57204 in angular section I_TL. Montages were transferred to a tablet PC, where cellular structures were traced, counted, and measured. The montage was used to keep track of photographed high-magnification images of RGC axons using the Image J's Region of Interest (ROI) Manager to ensure that there were not any repeat or missed images of any RGC axons within the nerve fiber bundle. (B) Medium magnification (11,000×–14,000×) TEM image used to create the montage in (A). Scale bar, (A) 2 μm; (B) 1 μm.

The RNFL anterior and posterior boundaries were defined as the ILM vitreous interface and the anterior-most RGC bodies, ¹⁷ respectively. If RGC bodies were not present directly posterior to the nerve fiber bundle, then the posterior boundary was defined at the points were RGC axons were absent. A line that bisected Müller cell processes and their footplates at the ILM defined boundaries between adjacent nerve fiber bundles. Axons that appeared pale and had substantially fewer neurotubules were identified as “sparse” axons and were recorded apart from the “nonsparse” axons (Fig. 4). Amacrine cells or glial cells were not included in RGC axons counts when criteria described previously were used. ¹⁸  

Figure 4.

 

High-magnification images (28,000×–44,000×) were used to measure RGC axon neurotubules and organelles. Sparse axons were rare in nerve fiber bundles except in the temporal macular region of the eye in both subjects. Axons that had a neurotubule density that was less than 25% percent of the average neurotubule density of the axon population were classified as sparse axons. (A) Sparse axons (S) appeared pale and had fewer neurotubules. (B) An RGC axon from nasal region of subject 57204. Scale bar, 0.5 μm.

Neurotubule Density Estimation

Nonsparse Axons.

RGC axoplasmic area (A_x) measurements measured from TEM images were used in a statistical algorithm to estimate ρ_RNFL. (A_x) was calculated as the difference between (A_a) and the total organelle area (A_m and A_o) for each RGC axon and was then plotted against the respective number of neurotubules (k) in a scatter plot (Excel; Microsoft, Redmond, WA) to determine k(A_x). Axoplasmic area probability distribution functions (p[A_x]) were then computed for each bundle using a kernel estimation method based on an Epanechnikov kernel function. A Freedman-Diaconis bin width rule was used to determine the bandwidth of the kernel smoothing window and was adjusted using Scott's skewness factor. The axoplasmic fractional area in the bundle (f_x) is determined by summing all A_x measurements in the bundle and then dividing the result by the area of the bundle (A_b):  

 

Calculations of k(A_x), p(A_x), and f_x are used to estimate RNFL neurotubule density in the nonsparse axons of the nerve fiber bundle according to the expression:  

 

where ρ_RNFL is given in units of number of neurotubules per unit RNFL area. 

The right side of equation 2 is essentially a twice-weighted sum of RGC axoplasmic neurotubule densities.  

 

which are estimated by the k(A _x) linear regression model. Individual estimates of neurotubule counts for each axon (ρ_x) are weighted by the probability of a given A _x in the nerve fiber bundle (p[A _x]dA _x) and by the fraction of axoplasmic areas in the bundle (f _x) which accounts for null ρ_RNFL values in non-RGC areas of the RNFL bundle and effectively normalizes ρ_x into ρ_RNFL. 

Sparse Axons.

A different statistical method is used to estimate ρ_RNFL in the temporal maculopapillary fibers (angular sections T_I, T_SI, and T_IS) for both subjects. The A _x range for the nonsparse axons in this region was a narrow range and resulted in a low R ² in linear fits of k versus A _x. 

One hundred nonsparse axons were selected from the angular regions. The morphologic characteristics of the selected axons resembled nonsparse axons in other regions of the eye. The sample set was used to generate an average nonsparse axoplasmic neurotubule density (ρ_nsx) from which a 25% neurotubule density threshold was set to classify sparse axons in the respective region. Once sparse axons were identified, a sparse axon neurotubule density (ρ_sx) was determined. The ρ_nsx and ρ_sx were individually weighted by their respective f _x and summed to get an estimate for ρ_RNFL in the respective section according to:  

 

 

Experimental Correlation between Birefringence and RGC Organelles

Radial area and ring scan PR/UD measurements falling within an angular section were averaged, resulting in an equal number of PR/UD values for every mean ρ_RNFL. The averaged PR/UD measurements for subjects 1853 and 57204 resulted in ±22.5° and ±11.25° of angular error, respectively. Average PR/UD measurement positions were optimally registered with angular RGC axon organelle mean measurements to compute the correlation between them (Fig. 5). The shift was used to avoid PR/UD and ρ_RNFL measurement positions taken at angular boundaries between sections in both subjects. 

Figure 5.

 

(A) Original positions of the mean ρ_RNFL and PR/UD radial area and ring scan measurements used for correlation with ρ_RNFL for subject 57204. Yellow and blue bars: the original position of angular section T_IS and the lowest PR/UD measurements within an angular section, respectively. (B) Radial area and ring scans of PR/UD falling within an angular section were averaged, resulting in an equal number of PR/UD measurements for every ρ_RNFL. Green bar: the optimally registered ρ_RNFL for angular section T_IS and PR/UD measurement. N, nasal; T, temporal; S, superior; I, inferior.

Two approaches were used to find a correlation between PR/UD and RGC organelles. The first approach was a Pearson product moment correlation. Correlation coefficients and significance values were computed (SAS ver. 9; SAS, Cary, NC). 

A multilevel modeling technique was used for the second approach, since it allows PR/UD and RGC organelles to be correlated on two levels so that finer angular resolution of PR/UD measurements could be preserved. Multilevel modeling was implemented through the procedure PROC MIXED within the statistical analysis software. The model accounts for the random effect of angular position of PR/UD measurements. 

Results

Primate RNFL Birefringence

Spatially resolved RNFL peripapillary PR/UD maps were obtained by EPS-OCT from four eyes of two subjects. Results from both subjects are presented throughout the results section to demonstrate the similarities and differences in birefringence and morphologic measurements. 

Radial area and ring scan measurements of PR/UD were relatively higher in the superior and inferior areas in both subjects (Fig. 6). Clusters of PR/UD radial area and ring scans within respective ρ_RNFL angular section boundaries were averaged, resulting in an equal number of PR/UD measurements for every ρ_RNFL value. 

Figure 6.

 

PR/UD (proportional to birefringence) measurements along a 360° circular sweep around the ONH. PR/UD measurements (ring or radial) in each eye are plotted with a unique symbol. Black squares: the average PR/UD of combined ring and radial scans for each angular section region that was used to determine neurotubule density. (A) Two different radial area and single ring scan for subject 57204. (B) Area scan and ring scan of 1853 OS. N, nasal; T, temporal; S, superior; I, inferior.

The PR/UD radial area and ring measurements are grouped into regional quadrants and averaged to give the mean PR/UD values listed in Table 1. One-way ANOVA was used to compare quadrant means. The temporal and inferior pair was the only quadrant pair that was not significantly different (P < 0.05) for subject 1853. All quadrant pairs excluding the inferior and nasal pair were significantly different (P < 0.05) for subject 57204. 

Table 1.

 

PR/UD (Proportional to Birefringence) Quadrant Means of Radial Area Scan and Ring Scan Measurements in One Eye of Each Subject

Scan Pattern	57204 OD						1853 OS
	Radial Area Scan 1		Radial Area Scan 2		Ring Scan		Radial Area Scan		Ring Scan
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Superior	19.1	2	18.3	2.6	17.1	4.2	14.1	4.3	17.6	3.2
Inferior	14.1	3.1	14.9	2.6	14.8	3.9	12.3	5.1	9.4	4.7
Temporal	9.4	3.3	9.0	3.3	9.6	4.3	8.5	3.6	9.3	3.3
Nasal	12.5	4.5	15.5	12.4	11.5	5.7	13.3	6.1	14.3	5.6

PR/UD measurement reproducibility was determined for both radial area scans and the ring scan of subject 57204 by calculating the standard error (SE) ¹⁵ for each of the PR/UD measurements that were averaged within ρ_RNFL angular section boundaries. To calculate the SE for each PR/UD measurement, the standard deviation (SD) of the cluster PR/UD measurements within angular section boundaries were divided by the square root of the number of clusters. The highest SE of all PR/UD measurements was ±7.1 deg/100 μm in the nasal region. The average SE of all three scans was ±1.9 deg/100 μm (Δn = 0.4 × 10⁻⁴) and was calculated by dividing the average SD of cluster PR/UD measurements from all three scans by the square root of the number of scans. 

Primate RNFL Neurotubule Density

Estimates of ρ_RNFL for all bundles in both eyes are given in Tables 2 and 3 and are shown graphically in Figure 7A. Neurotubule densities were highest in the superior and inferior quadrants and lowest in the temporal and nasal regions of both subjects. The Pearson product moment correlation was used to find the similarity between the octant mean ρ_RNFL between the two subjects (Fig. 7B). The result was a correlation coefficient of 0.97 (P < 0.001). 

Table 2.

 

Estimated Neurotubule Densities in Three Nerve Fiber Bundles of Each RNFL Octant

Section	Bundle 1	Bundle 2	Bundle 3	Mean	SD
NS	66.0	72.2	76.3	71.5	5.2
SN	110.1	69.8	64.6	81.5	24.9
ST	126.5	79.4	85.4	97.1	25.7
TS	69.9	64.7	69.6	68.1	2.9
TI	32.5	28.8	37.6	33.0	4.4
IT	80.5	96.3	69.9	82.2	13.3
IN	98.4	77.0	68.3	81.2	15.5
NI	49.6	74.6	90.1	71.4	20.4

Table 3.

 

Estimated Neurotubule Densities in Three Nerve Fiber Bundles of Each RNFL Section

Section	Bundle 1	Bundle 2	Bundle 3	Mean	SD
NSI	53.7	42.6	37.1	44.5	8.5
NSS	66.9	54.2	36.1	52.4	15.5
SNL	77.0	45.1	68.4	63.5	16.5
SNM	87.7	66.9	68.3	74.3	11.6
STM	45.1	59.8	45.8	50.2	8.3
STL	63.0	18.1	60.8	47.3	25.3
TSS	33.0	26.2	33.5	30.9	4.1
TSI	4.5	9.3	33.1	15.6	15.3
TIS	11.3	5.6	3.7	6.9	3.9
TII	46.5	35.0	42.3	41.3	5.8
ITL	72.6	24.5	83.5	60.2	31.4
ITM	50.6	65.6	47.2	54.5	9.8
INM	42.1	73.7	41.4	52.4	18.5
INL	73.7	22.3	25.6	40.5	28.8
NII	18.4	73.0	76.7	56.0	32.6
NIS	17.7	42.0	38.8	32.9	13.2

Figure 7.

 

(A) Mean ρ_RNFL estimates averaged over three nerve fiber bundles are presented for both subjects. (B) ρ_RNFL estimates for subject 57204 within angular section octants analogous to subject 1853 were further averaged to produce eight ρ_RNFL values for comparison with those in subject 1853. N, nasal; T, temporal; S, superior; I, inferior.

The linear relationship between k and A _x used in the ρ_RNFL estimate was similar across all angular sections for both subjects (Fig. 8A). As axon size increased, the neurotubule number also increased, but the number of neurotubules per unit axon area decreased. This finding is similar to other studies investigating axon size and neurotubule density in the optic nerve head. ¹⁹  

Figure 8.

 

(A) Axoplasmic area probability density function from angular section N_SS of 57204 OD used in the statistical algorithm to estimate ρ_RNFL. (B) Least-squares linear fits of neurotubule number (k) versus axoplasmic area (A _x).

Correlation between Birefringence and RGC Organelles

Subject 57204 had a correlation coefficient of 0.67 (P = 0.005) between PR/UD and ρ_RNFL (Table 4), and the multilevel model resulted in standardized regression coefficients of 0.50 (P = 0.04). The averaged pattern of PR/UD is shown superimposed on respective ρ_RNFL for both subjects in Figure 9. 

Table 4.

 

Correlation between PR/UD and ρ_RNFL in Subject 57204

Statistical Model	r	P
Pearson's correlation	0.67	0.005
Multilevel model	0.5	0.04

Figure 9.

 

Average birefringence measurements calculated from EPS-OCT PR/UD radial area and ring scans plotted with respective regional ρ_RNFL for subjects (A) 57204 (B) and 1853. Area and ring scans of PR/UD measured within angular regions that were sampled for neurotubule density surrounding the ONH were averaged within the respective angular section region, resulting in an equal number of PR/UD measurements for every ρ_RNFL value. N, nasal; T, temporal; S, superior; I, inferior.

Morphologic measurements from TEM images were used to estimate mean values of axoplasmic area (A _x) mode, axon area (A _a) mode, slope (u) of neurotubule number versus axoplasmic area (neurotubule packing density), fractional area of axoplasm in the nerve fiber bundle (f _x), mitochondrial fractional area in the nerve fiber bundle (x _m), mitochondria-containing axon profile fraction (m _p), and length of axonal membrane profiles per unit nerve fiber bundle (L _am/A _b). 

Pearson product moment correlation coefficients of the morphologic parameters with PR/UD are summarized in Table 5. Correlation coefficients of r = 0.81 (P = 0.01) and r = 0.5 (P = 0.05) were observed between PR/UD and mean A _x mode in subjects 1853 and 57204 (Fig. 10), respectively. Mean A _x mode values were similar in the superior and inferior portions of both eyes and were larger than temporal and nasal portions. Mean A _a and PR/UD yielded a significant correlation coefficient of r = 0.79 (P = 0.01) in subject 1853 (Fig. 10B). 

Table 5.

 

RGC Morphologic measurements Correlated with Phase Retardation Per Unit Depth (PR/UD, Proportional to Birefringence) for Subjects 57205 and 1853

RGC Axon Morphologic Measurement	Abbreviation	57204 OD		1853 OS
		r	P	r	P
Least-squares linear slope (number of NTs per unit Ax area), NT/μm2	Mean u	0.70*	0.0028	0.68	0.0941
Mean RGC axoplasmic area mode, μm2	Mean A x	0.50*	0.0500	0.81*	0.0145
Mean axon area mode, μm2	Mean A a	0.47	0.0600	0.79*	0.0184
Neurotubule density (number of NTs per unit RNFL area), NT/μm2	Mean ρRNFL	0.67*	0.0050	0.61	0.1077
Axoplasmic fractional area (per unit RNFL bundle area), %	Mean f	0.34	0.1924	0.51	0.1934
Mitochondrial fractional area (per unit RNFL bundle area), %	Mean x m	0.02	0.9500	—	—
Mitochondria-containing axon profile fraction (per unit RNFL bundle area), %	Mean m p	—	—	0.24	0.5708
Axonal membrane length (per unit RNFL bundle area), μm−1	Mean L am/A b	0.24	0.3670	0.23	0.5811

 RGC axoplasmic area (A _x) mode, neurotubule (NT) number (k) versus A _x least-squares linear slope (u), and fractional area of axoplasm per unit nerve fiber bundle (f _x) measurements are associated with nonsparse axons only. Axon area (A _a) mode, mitochondrial fractional area (x _m), mitochondria-containing axon profile fraction (m _p), and length of axonal membrane per unit nerve fiber bundle area (L _am/A _b) measurements are associated with both sparse and nonsparse axons.

*  Morphologic parameters that had a significant correlation with PR/UD.

Figure 10.

 

PR/UD (proportional to birefringence) measurements from radial area and ring scans in subject 57204 were averaged within its respective angular section around the (ONH) and superimposed on (A) mean (A _x) modes for subject 57204 and (B) mean (A _a) and (A _x) modes for subject 1853 measured in corresponding regions of the same eye. N, nasal; T, temporal; S, superior; I, inferior.

In subject 57204, a significant correlation coefficient of r = 0.7 (P = 0.002) was observed between section averaged PR/UD and the k versus A _x regression fit slope (u) that is used as a measure of neurotubule packing density (Fig. 11). Arcuate bundle regions had the highest packing density, followed by the nasal region, and last the papillomacular region. 

Figure 11.

 

PR/UD (proportional to birefringence) measurements from radial area and ring scans for subject 57204 were averaged within its respective angular section around the ONH and superimposed on mean (u; neurotubule packing density) data measured in corresponding regions of the same eye. N, nasal; T, temporal; S, superior; I, inferior.

Discussion

Primate RNFL Birefringence

PR/UD measurements were highest superior and inferior to the ONH for both subjects. In subject 57204, the average of the quadrant PR/UD values given in Table 1 for the superior and temporal region are 18.2 ± 1.0 deg/100 μm (Δn = 4.2 × 10⁻⁴) and 9.3 ± 0.3 deg/100 μm (Δn = 2.1 × 10⁻⁴). Cense et al. ¹⁵ measured the RNFL birefringence of two human subjects in vivo using PS-OCT at 840 nm. They also reported higher RNFL birefringence (4.1 × 10⁻⁴) in the superior and inferior areas with the lowest in the temporal region (1.2 × 10⁻⁴). Huang et al. ¹⁶ calculated similar results of averaged birefringence for the superior (4.2 × 10⁻⁴) and temporal region (2.5 × 10⁻⁴) of 12 human eyes. The SE was greatest in the nasal region for subject 57204 and may have resulted from a thinner RNFL area or eye movements. This finding was similar to others. ¹⁵  

Uncertainty in the PR/UD measurements comes from any error in the RNFL thickness (Z _RFNL) measured from EPS-OCT intensity images and phase retardation (PR) estimates from the EPS-OCT algorithm. Error in the Z _RFNL comes from the refractive index range for the RNFL (4%, n = 1.34–1.39) ²⁰ and uncertainty in the exact location of the RNFL boundaries due to the axial resolution limit of the EPS-OCT instrument (5 μm). Noise creates an uncertainty of approximately ±1° for PR estimates. Thus, uncertainty in PR is fractionally greater in regions with low PR such as the nasal and temporal regions. Radial and rotational error in position of the eye during imaging may have caused a registration error in PR/UD measurements. 

Neurotubule Density of the Primate RNFL

Sources of uncertainty for ρ_RNFL determination include: sampling RNFL bundles near the angular section boundaries, misclassified cells or cellular structures, different observers for the two eyes studied, and sampling location from the ONH for each eye. The higher ρ_RNFL values for subject 1853 can be explained by counter bias between the first and second eye. The mean difference (95% limits of agreement) between neurotubule counts of the same RGC axons of two counters for subject 57204 was −1.6(−8.8 to 5.6) neurotubules. 

Some angular sections had more variability in ρ_RNFL within nerve bundles than others. Most of the variability could be attributed to differences in RNFL nerve bundle area (A _b) measurements, since ρ_RNFL estimates are calculated using A _b. The standard deviation of RNFL thickness of human retina ranges from 15 μm in the temporal region to 26.5 μm in the inferior region. ²¹ Standard deviations of ρ_RNFL greater than ±20 NT/μm² occurred in the arcuate bundles and the bundles located in the inferior nasal portion of the RNFL near major blood vessels. Bundles lying close or adjacent to temporal blood vessels that approximately overlie the arcuate nerve bundles ²² were not excluded from the sampled population. The few histologic studies of primate RNFL thickness ^18,22–24 did not measure thickness near large blood vessels because RNFL borders could not be accurately determined. ²⁵ The RNFL bundle area variation near and away from blood vessels as well judgment as to the location of the RNFL borders would produce variability in ρ_RNFL estimates. 

Glial content (Müller cells and astrocytes) of nerve bundles would also contribute to bundle area variation seen within the same angular section since glial tissue is included in the A _b measurement. Ogden ²² reported that the proportion of RNFL bundle area occupied by glia is independent of RNFL thickness and does not vary regularly with distance from the ONH in cynomolgus and rhesus monkeys. Angular sections with the highest variability in ρ_RNFL estimates also had the largest variability in glial content percentage. 

Sections T_SI, T_IS, and T_I are sampled from the papillomacular bundle fibers and have the smallest RGC fibers. ²⁶ A low mean ρ_RNFL may be attributable to a combination of a thin RNFL ²³ and low number of RGC axons ²⁷ that have a significantly smaller mean neurotubule number. The sparse axons in the papillomacular bundles could be a fixation artifact if the smaller axons were not adequately preserved by the perfusion of fixative. A similar pattern of sparse axons was observed in the two eyes studied which should give some assurance that the pattern was not unique to a single eye; however, it is possible that the neurotubules in the small axons of the papillomacular bundle degenerated faster than the large axons found elsewhere. The neurotubules polymerize and depolymerize very quickly and the smaller axons may be more fragile than the larger ones. The sparse axons and fewer nonsparse axons within octant T_I and angular sections T_SI and T_IS could also be attributed to bilateral optic atrophy (BOA). ²⁸  

The packing density was greatest for sections S_NL, S_NM, S_TM, I_TM, and I_NM which are sampled from arcuate fibers. The arcuate portions of the RNFL are frequently involved in glaucoma and optic neuropathies. ²⁷ Higher birefringence in the superior and inferior region of the eye can be attributed to a combination of higher neurotubule packing density, larger axons, and less compartmentalization by glial tissue than other regions of the eye. ²⁷  

Correlation between Birefringence and RGC Axon Organelles

The high correlation between A _x and PR/UD in both eyes is interesting because axoplasmic regions of RGC axons are the onlyareas that neurotubules can inhabit. In addition, there was a significant correlation between A _a and PR/UD in subject 1853. These findings suggest that a structure within the axoplasm is the source of the birefringence signal. The mean (u) (neurotubule packing density) was more significantly correlated with PR/UD than the mean ρ_RNFL for subject 57204. This suggests that the birefringence surrounding the ONH results from a difference in NT packing density of the axons themselves and not from a difference in the packing density of axons within bundles. 

L _am/A _b failed to correlate with PR/UD in subject 57204, despite increased sampling and further supports the finding ²⁹ that cell membranes are not the source of birefringence. Mitochondrial fractional area (x _m) and the percentage of mitochondria-containing axon profiles (m _p) did not correlate with PR/UD for subject 57204 and 1853, respectively. 

Conclusion

Our findings further support the hypothesis that neurotubules are the primary source of birefringence in the RNFL. The strongest correlation between Δn and PR/UD (P = 0.0028) for all the morphologic parameters studied was the slope (u) of the regression fit of neurotubule number (k) versus axoplasmic area (A _x), which represents the neurotubule packing density. 

Footnotes

 Supported by Grant R01EY016462-01A1 from the National Eye Institute at the National Institutes of Health and Long Term Full Time Program Grant BV0700313000. Any opinions, interpretations, conclusions, and recommendations are not necessarily endorsed by the U.S. Air Force.

Footnotes

 Disclosure: G.M. Pocock, None; R.G. Aranibar, None; N.J. Kemp, None; C.S. Specht, None; M.K. Markey, None; H.G. Rylander III, None

Footnotes

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

The authors thank Matt Hersh and the Graduate Student's Fellows Program run by The University of Texas at Austin's Division of Statistics and Scientific Computation for their review of the manuscript and Maura Boyle for technical assistance.