Relationship of the Optical Coherence Tomography Signal to Underlying Retinal Histology in the Tree Shrew (Tupaia belangeri)

Abstract

purpose. To interpret the retinal origin of the optical coherence tomography (OCT) signal by objectively (i.e., minimal investigator bias) aligning in vivo OCT longitudinal reflectivity profiles (LRPs) with corresponding vertical histologic sections.

methods. The Zeiss StratusOCT system was used to obtain retinal B-scans in vivo in eyes from adult tree shrews. Subsequently, the retinas were fixed and embedded. Semithin vertical sections through the retina were obtained from the same locations as the LRPs. A statistical correlation procedure that accounted for axial tissue shrinkage determined the best relationship between features in the LRP and sublaminae boundaries in corresponding histology sections.

results. For the optimal relationship, the three regions of high reflectivity in the inner OCT signal corresponded to (1) the nerve fiber and ganglion cell layers, (2) the inner plexiform layer and amacrine cell somas, and (3) the outer plexiform layer. The two regions of low reflectivity in the inner OCT signal corresponded to (1) the somas of Müller, bipolar, and horizontal cells in the inner nuclear layer and (2) the outer nuclear layer. The outer OCT signal had a region of high reflectivity that corresponded to the photoreceptor inner and outer segments, the pigment epithelium, Bruch’s membrane, and at least part of the choriocapillaris.

conclusions. These results provide a clear interpretation for the OCT signal in terms of the underlying retinal anatomy. This interpretation can be used in vivo to identify sublaminae affected by retinal disease and has implications for the origin of the inner OCT signal in human retina.

Optical coherence tomography (OCT) is a noninvasive, high-resolution imaging tool used to study retinal structure. ^{1 2} OCT uses interferometry to measure the reflectivity of the retina, which varies as a function of depth due to laminar changes in refractive index and microstructure. Clinically, OCT is used in diagnosis of retinal disease and to monitor structural changes of the retina, both postsurgically and during disease progression. ^{3 4 5 6 7 8 9} Recently, OCT has also played an important role in studying the effectiveness of pharmacologic treatments in chorioretinal disease and in the identification of candidates for treatment. ^{10 11 12 13 14 15} In these studies, the interpretation of retinal structural change posttreatment critically depends on the relationship between OCT features and anatomic features. A number of studies have been conducted to investigate this relationship, but species differences (human, ^{1 2 16 17 18 19 20 21 22 23 24} monkey, ^{25 26 27} chick, ²⁸ swine, ^{29 30} bovine, ²² mouse, ^{31 32} and rat ³³ ) and whether OCT was conducted in vitro ^{1 22 23 27 30} or in vivo ^{2 16 17 18 19 20 21 24 25 28 29 31 32 33} have complicated the interpretation of the OCT signal. Retinal structure varies between species, and in vitro reflectivity profiles could differ from in vivo reflectivity profiles as the effects of fixation and perfusion of blood vessels may alter the optical properties of the retina. A further complication is that different studies use OCT instruments with different axial resolutions and more bands of reflectivity are apparent in the OCT signal from higher resolution instruments. In earlier studies laboratory-built and first-generation commercial instruments (axial resolution, 15–20 μm), ^{1 2 22 25 28 29 31 32 33} were used, whereas the StratusOCT (axial resolution 8–10 μm; Carl Zeiss Ophthalmic Systems, Dublin, CA), ^{18 20 21 23} and ultrahigh resolution OCT (axial resolution 2–3 μm) were used in later studies. ^{16 17 20 27 30} In addition, Fourier-domain OCT has been established recently, enabling high-resolution (2–3 μm) volumetric images of the retina without motion (eye movement) artifacts. ^{19 24 34 35 36 37 38} In clinical settings, the StratusOCT is currently the most common model of OCT, although commercial ultrahigh resolution instruments have recently become available. 

Another issue is that previous studies have used subjective methods relying on visual judgments and/or preexisting assumptions regarding alignment of retinal layers with features in the OCT signal. Most human studies have interpreted B-scans by visual inspection without correlation to the matching histology because of limited availability of donor tissue. ^{2 16 17 19 20 21 24} Also, in studies comparing the OCT signal with histology sections (mostly in animal models), visual alignment methods were used to match the section to either the B-scan ^{1 21 22 23 25 31 33} or longitudinal reflectivity profiles (LRPs or A-scans). ^{18 27 28 29 30 32} Interpretation of the OCT signal using LRPs is preferable to B-scans as subtle features are more readily identifiable. Furthermore, it is well accepted that tissue undergoes shrinkage during histologic processing. ^{39 40 41 42 43 44} Despite this, in most correlation studies shrinkage is not measured; instead, inferred shrinkage is calculated after resizing of the OCT B-scan to subjectively align it with the histology, ^{25 27 30} or it is not accounted for at all. ^{21 31 32 33} In some studies tissue was used that was not fixed or dehydrated, to avoid the shrinkage problem ^{28 29} ; however, the process of cutting unfixed tissue is still prone to mechanical artifacts, and so high quality tissue is not guaranteed with this approach. To date, in only two studies has a quantitative assessment been made of the OCT/histology relationship. ^{22 27}  

Because of the subjective nature and the differences in design of the previous studies, there is still debate regarding the precise relationship between the OCT signal and the retinal layers. ^{23 37} If the origin of the OCT signal from retinal structure is understood in greater detail, the clinical usefulness of the OCT technique in detecting and monitoring microanatomical retinal disease will be improved. In the present study, we developed an objective method (i.e., with minimal investigator bias or assumptions) of correlating the OCT LRP features with established retinal sublaminae in corresponding histology sections, taking tissue shrinkage into account. The tree shrew is a mammal that has a well-characterized retinal structure ^{45 46 47 48 49 50} with similarities to that of the primate. ⁵¹ Thus, it is a useful model for studying the relationship between in vivo OCT scans and histology. 

Materials and Methods

Animals

We used two adult maternally reared tree shrews (Tupaia belangeri) from our breeding colony (TS1 and TS2). The animals were maintained on a 15:9-hour (light:dark) cycle and food and water were available ad libitum. The animals were fully anesthetized for procedures by an intramuscular injection of (90 mg/kg) ketamine hydrochloride with (10 mg/kg) xylazine, and body temperature was maintained with a heating pad. Pupils were dilated (1% tropicamide; Alcon, Frenchs Forest, Australia) to enable retinal imaging and ultrasonography. The cornea was anesthetized (0.5% proxymetacaine hydrochloride, Alcaine; Alcon Australia) for ultrasonography. Artificial tears (Artificial Tears; CIBA Vision, Baulkham Hills, NSW, Australia) were applied regularly to prevent corneal desiccation. A bite bar stabilized the animal’s head ⁵² and eyelids were retracted with a custom-made speculum. All animals were treated in accordance with the ARVO statement for the Use of Animals in Ophthalmic and Vision Research, and institutional ethics approval was obtained. 

Biometry

Biometric measures were performed in vivo to verify normal ocular development. Axial length measures were used to correct the lateral OCT B-scan length for the tree shrew eyes based on retinal magnification factor calculations. ⁵³ Refractive error was measured by streak retinoscopy (Keeler) to the nearest 0.5 D and recorded as the mean equivalent sphere at the cornea. Axial eye dimensions were determined by A-scan ultrasonography (10-MHz focused transducer; 9400 digital storage oscilloscope; LeCroy, Geneva, Switzerland). Further details of the biometric methodology have been described elsewhere. ^{52 54} The mean refraction (±SD) was +6.70 ± 0.24 D and the mean axial length (±SD) was 7.65 ± 0.03 mm (n = 4 eyes), which is similar to reported normal measures. ⁵²  

Optical Coherence Tomography

A third-generation OCT system (StratusOCT 3000; Carl Zeiss Ophthalmic Systems, Dublin, CA) was used to acquire B-scans (e.g., Fig. 1A ). There were 512 lines (A-scans or LRPs) per frame (B-scan) and the frame was 2 mm in length and height (depth). ⁵⁵ Therefore, the sampling density of the frame was 512 pixels per 2 mm in the longitudinal direction and 1024 pixels per 2 mm in the axial direction. Near-horizontal B-scans (n = 118–167) were taken over a total scanned area of approximately 4 mm² centered on the optic disc in each eye (e.g., Fig. 1B ). In the tree shrew the area centralis is located in the far temporal retina, close to the ora serrata. ⁵⁶ We were unable to image this region because of positional constraints imposed by the size of the StratusOCT scanning head and the long nose of the tree shrew. 

Groups of B-scans (n = 2–4 scans per group) were separated vertically by approximately 0.17 mm (e.g., Fig. 1B ). The total acquisition time per frame was approximately 1.28 seconds. ⁵⁵ StratusOCT has a z-offset range (axial distance between instrument and retina) designed for human eyes; however, tree shrew eyes are approximately one-third the size of human eyes. ^{52 57} Therefore, Carl Zeiss Ophthalmic Systems Inc. provided a StratusOCT calibration file that increased the available z-offset range so that images of tree shrew retina could be obtained. The focus knob was adjusted to obtain maximum signal strength across the depth of the retina (e.g., Fig. 1A). In the tree shrew eye, the axial resolution of the StratusOCT is 8 to 10 μm (assuming the refractive index of the retina in tree shrew is the same as in human). The transverse resolution of the StratusOCT in the tree shrew eye (approximately 60 μm based on the difference in retinal magnification between humans ⁵⁷ and tree shrews ⁵² ) is poorer than in the human eye (20 μm ⁵⁵ ). Nevertheless, the quality of the StratusOCT B-scan in the tree shrew (Fig. 1A)is comparable to that in the human. ⁵⁸  

Image Processing

The StratusOCT software automatically corrects optical path lengths to physical distances based on the refractive index of human retina, with the assumption that the refractive index is the same for all retinal sublaminae. The OCT data were transferred to a computer for custom quantitative analysis (MatLab, ver. 7; The MathWorks, Natick, MA). The position of each B-scan was registered relative to the optic disc and blood vessels seen in its corresponding fundus photo and was coregistered relative to all other B-scans, to create an accurate in vivo map (Fig. 1B) . The scanned area was then overlaid with a square grid, with each square defining a bin of approximately 100 μm². Individual LRPs were assigned to bins based on their position, and then the LRPs in each bin were aligned and averaged to create a mean LRP for the area of retina encompassed by the bin. This averaging reduced speckle noise ⁵⁹ and improved the signal-to-noise ratio, making it easier to identify features of the LRP. Subsequent smoothing of the LRP with a median filter (kernel size of five pixels) enabled reliable autodetection of LRP features. 

Histologic Processing

The animals were administered a lethal dose of pentobarbital sodium (120 mg/kg) at the end of the OCT and biometry procedures. The eyes were enucleated, and the superior aspect marked with a permanent marker for orientation reference. Posterior eye cups were immersion fixed for 3 hours in 2.5% glutaraldehyde and 1% paraformaldehyde in 0.1 M phosphate buffer (pH 7.2), before rinsing in 5% sucrose solution in phosphate buffer. The retina was dissected from the sclera and choroid and then cut into quadrants along the horizontal and vertical midlines through the optic disc. For two retinas, the quadrants were cut into 1 × 2-mm pieces, dehydrated in an acetone series and embedded in epoxy resin. For the other two retinas, the quadrants were kept flat during dehydration and embedding and then cut into 1 × 2-mm pieces for vertical sectioning. Micrographs of all quadrants (e.g., Fig. 1C ), were obtained before and after dehydration. The lateral shrinkage (S _lateral) was determined by measuring the distance between landmarks (n = 20) such as blood vessels, optic disc, pigment epithelium defects, and distinctive edges of the quadrant, on the pre- and postdehydration whole mount micrographs and using the equation:  

\[S_{\mathrm{lateral}}(\%){=}(1{-}\frac{D_{\mathrm{post}}}{D_{\mathrm{pre}}}){\times}100,\]

where D _pre indicates the predehydration distance and D _post indicates the postdehydration distance. The mean (±SD) lateral shrinkage was 24.8% (±1.62%; n = 3 retinas), which is within the range of reported values (8%–29%). ^{39 40 41 42 43 44 60}  

Serial vertical sections (1 μm thick) were cut with a diamond knife on an ultramicrotome (Reichert, Vienna, Austria). The sections were stained with toluidine blue and micrographs were taken with differential interference contrast optics using ×10 and ×40 oil-immersion objectives on a Zeiss microscope. Sections were deemed vertical when single Müller cell processes were visible over most of the depth of the inner plexiform layer (Fig. 2) . 

Selection of Analysis Locations

To ensure that the LRP and histology pairs used in the correlation analysis were from the same retinal location, we superimposed the retinal blood vessel pattern from the predehydration histology wholemount and the OCT fundus photographs precisely by using linear scaling. This method created an accurate in vivo map conveying both LRP and histology location information (Fig. 1D) . Exact dimensions of the nominally 1 × 2-mm (after dehydration) retinal pieces, after accounting for the lateral shrinkage, were matched to the in vivo map. In addition, inner retinal blood vessels were identified in the micrographs of vertical sections, scaled for lateral shrinkage, and also matched to the in vivo map. This process determined the precise location of each vertical section relative to the OCT scanned area, and corresponding LRP bins and histology sections could be identified. Retinal locations analyzed avoided inner retinal blood vessels due to shadowing of the OCT signal (e.g., Fig. 1A , blood vessel labels). A sampling density of at least three OCT B-scans per bin was necessary, to achieve a signal-to-noise ratio that enabled the features within the mean LRP for each bin to be reliably identified. Given these restrictions, 35 locations were analyzed: seven locations from TS1 (right eye) and 28 locations from TS2 (13 from the right eye and 15 from the left eye). 

Correlation Procedure

Axial (radial) shrinkage in the retina due to tissue processing cannot easily be measured directly. Thus, we developed a statistical correlation procedure to determine the optimal relationship between LRP features and histology sublaminae, assuming a linear axial scale (described later). The assumption of a linear axial scale is based on the assumption that different retinal sublaminae shrink by similar amounts. The correlation method enabled the relationship between LRP features and histology boundaries to be determined with minimal investigator bias (i.e., without making assumptions regarding the alignment of histology and LRP features). 

The correlation procedure to account for axial shrinkage was conducted in the following way. Corresponding mean LRPs and histology micrographs were imported into a custom program (MatLab; The MathWorks). Ten sublaminae boundaries were identified in the histologic sections by using established anatomic criteria ^{45 50 61 62} (Figs. 2 3A) . A LRP shows changes in reflectivity across the retina, presumably due to change in microstructure and refractive indices of the different layers, ^{1 2} so it is likely that the maximum rate of change in reflectivity corresponds to a boundary between structurally different retinal sublaminae. ^{28 29} Based on this assumption, four maximum rates of change or transitions (T) in the LRP were identified (MatLab; The MathWorks). These four transitions were consistently found for all 35 locations analyzed (Fig. 3A)and were labeled T1, T2, T3a, and T3b according to previous customs. ²⁹ The distances between these transitions were found to be always greater than 40 μm, which is four times the resolution of the StratusOCT. 

Each of the four LRP boundaries is represented by a Gaussian distribution along the depth axis (Figs. 3A 3B)with a SD of 3 μm (based on the instrument resolution and the thickness of the thinnest layers to be imaged) and a maximum amplitude of 1.0. A normalized correlation value was determined by summing the height (amplitude) of the four distributions at depths corresponding to the 10 histology boundaries (indicated by the gray lines under the Gaussians in Figs. 3A 3B ) and dividing by the number of Gaussians (i.e., four). The normalized correlation value approaches 1.0 when the LRP and histology boundaries are closely aligned (i.e., a histology boundary is near the peak of each Gaussian distribution). Note that this definition of the normalized correlation value is not equivalent to a correlation coefficient. 

The normalized correlation value was calculated for a continuum of linear axial histology scale factors (e.g., Fig. 3C ). The optimal scale factor is the amount of linear stretch that needs to be applied to the histology to optimally match the histology boundaries with LRP transitions (Figs. 3A 3B) . Scale factors between 1.0 and 1.7 were chosen, as they represent 0% to 40% axial shrinkage of the histology (Fig. 3C) . The previously reported range of lateral shrinkage is 8% to 29% ^{39 40 41 42 43 44 60} ; therefore, our scaling criteria fully encompassed and extended beyond this range. The scale factor (F) can be converted to axial shrinkage (S _axial) using the equation:  

\[S_{\mathrm{axial}}(\%){=}100{-}\left(\frac{100}{F}\right).\]

 

During scaling, the LRP boundary T1 was maintained in alignment (±2 μm) with the inner limiting membrane (e.g., Figs. 3A 3B ). This produced a correlation function for each of the 35 retinal locations (e.g., Fig. 3C ). We assumed the maximum rate of change of the innermost edge of the LRP (T1) corresponded closely to the inner limiting membrane of the histology section, as they both represent the innermost boundary. Within a correlation function (e.g., Fig. 3C ), local minima (troughs) represent transitions of different relationships between the LRP and histology boundaries. The term “solution” is used to refer to a unique match between the LRP and histology boundaries; each different solution has at least one LRP boundary matching a different histology boundary. The transitions between different solutions are shown by gray vertical lines in Figure 3C . Local maxima (peaks) between these transitions represent the best correlation for each of the possible solutions. Local maxima from all 35 correlation functions and their corresponding scales were recorded (e.g., Fig. 3C ), along with the histology boundaries that most closely aligned with the four LRP boundaries. The scale factors that produced maxima corresponding to the same alignment solution over the 35 different retinal locations were averaged to find the mean scale factor and the mean axial shrinkage value for each possible solution. From equation 2and Figure 3Cit can be seen that scale factors and axial shrinkage values are interchangeable, but for clarity purposes we will now refer to axial shrinkage only. 

Estimating Differential Axial Shrinkage across Retinal Sublaminae

The correlation procedure described above assumes linear axial shrinkage of the retina due to fixation and dehydration. However, it is possible that retinal sublaminae shrink by different amounts because they vary in microstructure and probably also in water content. If the LRP and histology boundaries do not correlate perfectly, it is possible that errors in the alignment of boundaries between the LRP and histology section represent deviation from the linear axial shrinkage assumption. To assess whether nonlinear shrinkage may have occurred, we measured the thickness of the three axial zones (defined in Fig. 3A ) for both the LRP and the unscaled histology section after the alignment between the LRP and histology boundaries had been determined. The axial shrinkage of each zone (S _zone) was then calculated using:  

\[S_{\mathrm{zone}}(\%){=}100{-}\left(100\ \frac{T_{\mathrm{histo}}}{T_{\mathrm{LRP}}}\right),\]

where T _histo indicates the thickness of each zone derived from the unscaled histology section and T _LRP indicates the thickness of each zone derived from the LRP. The axial shrinkage results were averaged for each zone and compared with the mean overall axial shrinkage. Precise shrinkage of the individual retinal sublaminae identified in the histology section cannot be determined as only four LRP boundaries were consistently identified in the correlation procedure. 

Results

Correlation of OCT Features with Histology

Vertical sections of the tree shrew retina showed easily identifiable layers (Fig. 2) . The tree shrew retina resembles the peripheral retina of primate ⁵¹ in terms of the relative thickness of its inner layers. Within the inner nuclear layer, the amacrine, Müller, bipolar, and horizontal cells were identifiable, and the boundary from amacrine cells to Müller cells and bipolar cells was visualized. The outer retina in the tree shrew is thinner than the peripheral outer retina in primates, because the outer nuclear layer is a monolayer in tree shrew (Fig. 2)and multilayered in primates. ⁵¹ In addition, the photoreceptor outer segments are shorter in tree shrew than in primate. ^{49 50}  

Five local maxima were consistently observed in the correlation function from each of the 35 retinal locations analyzed with the correlation procedure (e.g., Fig. 3C ). The five maxima represent five possible solutions or relationships between an LRP and histology section that are structurally equivalent across all locations. Therefore, the five solutions can be depicted schematically by a representative LRP and histology section and are defined by their mean axial shrinkage values (Fig. 4) . Any one of these five solutions has the potential to be the true relationship between the LRP and histology section; however, they have been ordered from 1 to 5 in Figures 4B 4C 4D 4E 4Fbased on the mean normalized correlation results shown in Figure 5A . 

The OCT images from tree shrew retina revealed two highly reflective regions (peaks) in the outer LRP (Fig. 3B) . No part of the histology section was found to align with the outermost LRP peak for any of the solutions (Figs. 4B 4C 4D 4E 4F) , as low correlation values were found at the high scale values needed to align this peak with the retina. The histologic sections lacked choroid and sclera, and so a possible structural origin of the outermost peak could not be directly identified. However, as the depth between the two peaks in the outer LRP is variable (e.g., Fig. 1A ), the outermost peak is unlikely to originate from a structure distal to the retinal pigment epithelium. The outermost peak is more likely to arise from multiple internal reflections within the retina and will not be considered further. However, the other peak was found to align with the retina in solutions 1, 3, and 4 and will be referred to as the outer peak throughout the Results and Discussion sections. 

The mean normalized correlation results show that solution 1 gives the best match between the LRP and histology features, with a mean normalized correlation of 0.71 (Fig. 5A) . Similarly, solution 1 also shows the best match between the LRP and histology features when the mean normalized correlation value for each solution is obtained using only locations for which that solution was the best one (Fig. 5B) . In addition, the mean axial shrinkage determined for solution 1 lies within the lateral shrinkage 95% confidence interval (CI; Fig. 5C ). The fact that the indirectly measured axial shrinkage for solution 1 is not significantly different from the directly measured total lateral shrinkage shows that solution 1 is a sensible interpretation of the LRP. Taking all the information from Figure 5together, solution 1 represents the most likely correspondence between LRP and histology features. 

Inner OCT Signal

The best match between LRP and histology boundaries (solution 1) is shown in Figure 6 . The inner OCT signal incorporates zones 1, 2, and 3, which are defined by the four LRP boundaries. The broad peak in reflectivity making up zone 1 corresponds to the nerve fiber layer and ganglion cell layer; with T1 corresponding to the inner limiting membrane and T2 to the ganglion cell layer to the inner plexiform layer boundary. It is difficult to identify a consistent LRP feature that corresponds to the nerve fiber to ganglion cell layer boundary, perhaps because the precise boundary between the nerve fiber layer and the ganglion cell layer is an undulating line with ganglion cells protruding into the areas between nerve fiber bundles (Fig. 2) . Zone 2 primarily consists of a broad peak in reflectivity, although of lesser amplitude than in zone 1, and corresponds to the inner plexiform layer and the proximal region of the inner nuclear layer (amacrine cell somas). Boundary T3a corresponds to the transition from amacrine cell somas to the distal region of the inner nuclear layer. The LRP in zone 3 is composed of a trough and a small peak. The Müller, bipolar, and horizontal cell somas correspond to the trough, whereas the outer plexiform layer corresponds to the small peak. Boundary T3b aligns with the outer plexiform layer to outer nuclear layer boundary. The outer nuclear layer corresponds to the small trough immediately distal to T3b. 

Outer OCT Signal

In the outer OCT signal (defined in Fig. 6 ), there is one highly reflective peak that corresponds to the retina for solution 1. The inner slope of this outer peak is made up of two components: a small leading edge with a relatively shallow gradient and a main incline with a relatively steep gradient. The boundary between the outer nuclear layer and the inner segments of the photoreceptors (external limiting membrane) aligns closely to the maximum rate of change of the shallower incline and the boundary between the inner segments and the outer segments/retinal pigment epithelium complex occurs at the maximum rate of change on the steep incline. The retinal pigment epithelium to choriocapillaris boundary aligns with the top of the peak. Therefore, Bruch’s membrane and at least part of the choriocapillaris/choroid form the outer slope of the outer peak. 

Axial Shrinkage across Retinal Sublaminae

The mean axial shrinkages for zones 1, 2, and 3 (defined in Fig. 3A ) are shown in Figure 7and compared to the overall mean axial shrinkage found for solution 1. Zone 1 shows significantly less shrinkage than either zone 2 or 3 and also shrinks less than the mean axial shrinkage for solution 1. Zones 2 and 3 shrink more than the mean axial shrinkage for solution 1. This difference in axial shrinkage between zones suggests it is likely that shrinkage is nonlinear; however, further possibilities are discussed later. 

Discussion

Quantification of the Correlation

The mean normalized correlation for the best solution is 0.71, rather than the ideal 1.0 (Fig. 5A) . This correlation indicates good, but less-than-perfect, alignment between LRP and histologic boundaries. A key component of this imprecision is the apparent nonlinear axial shrinkage of the histology, which suggests that the nerve fiber and ganglion cell layers (zone 1) shrink less than the inner plexiform, inner nuclear, and outer plexiform layers (zones 2 and 3) during tissue processing (Fig. 7) . Additional noise components that would act to reduce the correlation include inherent irregularity of histologic boundaries (Fig. 2) , subjective judgment required in positioning histologic boundaries, variability in identifying the location of the LRP transitions, and variations in refractive index across the retina. 

Interpretation of the Inner OCT Signal Results

Our interpretation of the relationship between the inner OCT signal and the retinal histology (Fig. 6)largely concurs with the current literature that nerve fiber and plexiform layers are highly reflective, whereas inner and outer nuclear layers are less reflective. ^{2 16 17 18 19 20 25 27 28 29 30} However, we found that the innermost peak in reflectivity between T1 and T2 corresponds to the nerve fiber layer and the ganglion cell layer, whereas previous studies in humans ^{17 20 24} suggest that this innermost peak should correspond to the nerve fiber layer alone. This finding in the tree shrew also means that the ganglion cell layer appears with a similar, or sometimes even greater, reflectivity than the nerve fiber layer and always with higher reflectivity than the inner plexiform layer, which is again different from the interpretation proposed by studies in humans. As discussed earlier, the human studies have not correlated the OCT signal directly to matching histology and rely on subjective visual judgments. Therefore, as the tree shrew has an inner retinal structure similar to that of the peripheral retina in humans ^{50 51} and this study used objective interpretation methods that incorporate matching histology sections, it raises questions about whether the origin of the innermost peak in an LRP originates solely from the nerve fiber layer in humans. 

Studies in chicks ²⁸ and swine ²⁹ support the present results as direct correlation between the in vivo OCT signal and corresponding histology also find that both the nerve fiber layer and the ganglion cell layer correspond with the peak between T1 and T2. In addition, the study in swine ²⁹ also demonstrated an example of a double-peaked highly reflective region between T1 and T2 where the distal peak is more reflective than the proximal peak. Furthermore, in an in vitro study in nonhuman primates, ²⁷ the ganglion cell layer was also found to correspond to a relative peak in the LRP, although it had less amplitude than the nerve fiber layer peak. It is unclear why the ganglion cell layer should sometimes appear to be more reflective than the nerve fiber layer and always more reflective than the inner plexiform layer. However, as ganglion cells protrude between nerve fiber bundles, these two layers significantly overlap, and their reflectivity may be linked. 

Another discrepancy between this study and the current literature in humans ^{17 20 24} involves the interpretation of LRP boundary T3a. The tree shrew results showed that T3a corresponded to the border between amacrine and Müller cell somas within the inner nuclear layer (Fig. 6) . However, in humans, it has been suggested that T3a aligns with the border between the inner plexiform and inner nuclear layers. ^{17 20 24} The inner plexiform and inner nuclear layers in tree shrew retina (Fig. 2)resemble their counterparts in peripheral human retina, ⁵¹ making it likely that they should originate from similar features in the LRP. Therefore, this study also raises questions about whether the origin of the peak between T2 and T3a corresponds solely to the inner plexiform layer in humans. 

Studies in chick ²⁸ and swine ²⁹ retinas support the present results, as the inner plexiform-inner nuclear layer histologic boundary is usually proximal to the LRP boundary T3a. The alignment of T3a with the border between amacrine and Müller cell somas may be explained by the known properties of Müller cell somas, which are more polygonal than either amacrine or bipolar cell somas ^{61 62 63} and so may alter the backscattering properties of the retina. In addition, the Müller cell density in tree shrew retina has been found to be similar to primate in central retina (both greater than 20,000 cells per mm²), ^{63 64} so that Müller cells form a close-knit sublayer within the inner nuclear layer (Fig. 2) . Since the density of Müller cells is sufficient to form a line of polygonal nuclei through the inner nuclear layer, it is feasible that this anatomic transition may underlie the change in reflectivity of the OCT signal in this region. 

Interpretation of the Outer OCT Signal Results

The thicknesses of the outer retinal layers in the tree shrew are close to the resolution limit of the StratusOCT (approximately 10 μm), and so understandably the outer peak represented a combination of signals from different retinal components (Fig. 6) . This study found the border between the outer nuclear layer and the inner segments of the photoreceptors (external limiting membrane) to correspond to the maximum rate of change in reflectivity of the shallower inner gradient of the outer peak. This result differs from previous subjective results in chicks ²⁸ and swine, ²⁹ which showed this border to correspond to the maximum rate of change in reflectivity of the steeper inner gradient. With respect to the border between the retinal pigment epithelium and choriocapillaris, the present results are in agreement with the results of the swine study ²⁹ and are consistent with the theory that a decrease in backscatter is expected from the choriocapillaris and choroid as hemoglobin absorbs light, thus attenuating the OCT signal. ^{58 65}  

The B-scan in the tree shrew has one highly reflective region representing the outer retina (Fig. 6) , while the B-scan in humans has three or four highly reflective regions representing the outer retina. ^{16 17 18 19 20 24 35 36} This can be explained by the fact that the photoreceptors are shorter in the tree shrew. ^{49 50} Thus, the outer OCT signal in tree shrews and humans cannot be directly compared. Several studies have provided subjective interpretations of the outer OCT signal in humans. ^{16 17 18 19 20 21 23 24 35 36} However, the precise structural origin of the outer OCT signal is still not well established. ^{23 37} An in vivo study in primates using methods similar to those developed in this study is needed to fully establish the origin of the outer OCT signal in humans. 

Conclusions

The relationship between the in vivo OCT signal and retinal structure in the tree shrew was established by using a novel method of correlating the OCT LRP features to established sublaminae in histology sections. This study differs from previous studies in that the correlation between the LRP and the corresponding histology has been performed objectively, with minimal investigator bias. This study can be considered objective, for three main reasons: First, the LRP boundaries (transitions) were automatically detected using computer software rather than by manual visual judgment. Second, there were no assumptions regarding which histology boundaries should align with which LRP transitions, aside from the assumption that the inner limiting membrane aligned with LRP boundary T1. Third, the correlation procedure used a mathematical equation to determine the best correlation rather than relying on a potentially biased visual inspection approach. In addition, the analysis methodology was thorough, with corresponding LRP bins and histology sections precisely matched and lateral and axial tissue shrinkage accounted for. 

The origin of the inner OCT signal can be compared between tree shrew and human as the inner retinal structure is similar. In this study, we identified two differences in the interpretation of the inner OCT signal compared with previous studies in humans. ^{17 20 24} The origin of the innermost peak of the LRP in the tree shrew was found to correspond to the nerve fiber and ganglion cell layers, rather than just the nerve fiber layer. In addition, the second innermost peak was found to correspond to the inner plexiform layer and the amacrine cell somas rather than solely the inner plexiform layer. Knowledge of the relationship between the OCT signal and retinal structure found in this study can be used to measure in vivo changes of sublaminae thicknesses in tree shrew models of retinal disease (e.g., pathologic myopia). In addition, although future studies using objective correlation methodology in primates are necessary to confirm the precise structural origin of the human LRP, these results will aid interpretation of the specific sublaminae affected in inner retinal disease clinically. 

 Supported by National Health and Medical Research Council of Australia Grant 454602 (NAM), a Retina Australia Grant (MJP), a Lions Vision Research Fellowship (UG), and an Ernst and Grace Matthaei Scholarship (CJA).

 Submitted for publication September 14, 2007; revised April 30 and July 7, 2008; accepted October 27, 2008.

 Disclosure: C.J. Abbott, None; N.A. McBrien, None; U. Grünert, None; M.J. Pianta, None

 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.

 

Figure 1.

 

Example of methods used to identify corresponding OCT LRP and histology section locations within 100 μm. (A) OCT B-scan (2 mm length) and mean LRP for the section of the B-scan shown by the yellow outline. Warm colors in the B-scan correspond to peaks of reflectivity in the LRP. (B) In vivo OCT map showing the location of all OCT B-scans and the location of bins analyzed for their mean LRPs. (C) Photomontage of retinal wholemount in 0.1 M phosphate buffer. The rectangle indicates the area scanned with OCT. (D) Combined in vivo map (OCT and histology) indicating the retinal pieces (gray outlines) and the matched pairs of OCT and histology locations that correspond within 100 μm. Orientations marked in (B) also apply to (C) and (D). Scale bars: (B, C, D) 500 μm (before dehydration). bv, blood vessel; S, superior; I, inferior; N, nasal; T, temporal.

Figure 2.

 

(A) Micrograph of the retina (TS2) in vertical section from a location ∼1.5 mm inferior-nasal to the optic disc. Lines: boundaries between layers. Short line: amacrine-Müller cell boundary. (B) High-power image from the region outlined in (A). Arrowheads: Müller cell somas, which form a close-knit layer within the inner nuclear layer. Arrows: Müller cell processes, the elongated presence of which demonstrates the section is close to vertical. The width of box in (A), 50 μm. NFL, nerve fiber layer; GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; AC, amacrine cells; MC, Müller cells; BC, bipolar cells; HC, horizontal cells; OPL, outer plexiform layer; ONL, outer nuclear layer; IS, photoreceptor inner segments; OS, photoreceptor outer segments; RPE, retinal pigment epithelium. Scale bar, 50 μm.

Figure 3.

 

Example of the method (correlation procedure) used to determine the best relationship between the LRP features and histology vertical section boundaries for the location shown by the yellow bin in Figures 1B1D . This method was applied to all 35 locations analyzed. (A) Mean LRP and unscaled histology section with their respective boundaries (circles: LRP boundaries; horizontal lines: on histology denote histology boundaries). Each LRP boundary determines the mean of a Gaussian distribution (amplitude = 1, SD = 3 μm). The correlation was determined by summing the amplitude of the four Gaussian distributions at depths corresponding to the 10 histology boundaries (gray lines under Gaussians). The only boundary constraint used was that LRP boundary T1 was within 2 μm of the inner limiting membrane histology boundary. Only one other histology boundary was aligned close to an LRP boundary (T3a) in this example, and so the normalized correlation when the histology was unscaled (i.e., scale factor = 1) was 0.5, as demonstrated in (C). The LRP was divided into inner and outer regions. The outer LRP consisted of two peaks. The inner LRP was divided into three zones, based on the four LRP boundaries. The unscaled histology section was divided into three zones ( ) after the histology boundaries that align with the LRP boundaries were identified (relevant in equation 3 , to calculate axial zone shrinkage). (B) Optimum alignment of the LRP and histology from (A) after axial scaling of the histology image to achieve the highest correlation value. Histology boundaries were closely aligned with all LRP boundaries, so the normalized correlation approached a value of 1 as shown in (C). Because the normalized correlation in (B) is greater than that in (A), it represents a better match between the LRP and histology boundaries. (C) Normalized correlation as a function of histology scale (bottom axis) and axial shrinkage (top axis) for the specific retinal location shown in (A) and (B). The five local maxima (labeled 1– 5) indicate the scale (and inferred shrinkage) values that produce the best unique alignments between the LRP and histology boundaries, and represent five different possible solutions. Gray lines: transitions between different solutions; a solution is deemed different when at least one LRP boundary aligns with a different histology boundary. Local maxima are ordered by peak correlation from 1 to 5, where 1 represents the highest correlation. The solution that gives the highest correlation (solution 1) is shown in (B). Dashed arrows: scale/shrinkage values for the LRP/histology relationships shown in (A) and (B). Scale and axial shrinkage are related according to equation 2 . LRP transitions (T1, T2, T3a, T3b), histology sublaminae as in Figure 2 . Scale bar, (A, B) 50 μm.

Figure 4.

 

Schematic diagrams of the relationship between a representative LRP and a representative histology section for the five possible solutions. The LRP/histology relationships for the five solutions were identified from local maxima (peaks) in the correlation functions (an example of which is shown in Fig. 3C ) from all the retinal locations analyzed. All five solutions were identified in all 35 locations analyzed. The order of the solutions is from highest to lowest mean normalized correlation (Fig. 5A) . (A) Micrograph of the histology section from Figure 3provides a reference for the schematic diagrams shown in (B–F). Dark gray, OS/RPE; light gray, nuclear layers; white, all other layers. (B–F) Representative LRP/histology relationships that give the best to fifth-best mean normalized correlations, respectively. Mean axial shrinkage is indicated for each relationship (n = 35 locations). LRP transitions (T1, T2, T3a, T3b), histology sublamina as in Figure 2 . Scale bars, 50 μm.

Figure 5.

 

Results of the correlation analysis. (A) Mean normalized correlation results for each solution (n = 35 locations for each solution). The solutions are labeled 1 through 5 in descending order of mean normalized correlation. (B) Mean normalized correlation results for each solution when it is the best solution only for each location (i.e., n < 35 locations for each solution as indicated; sum of n across the five solutions = 35 locations). (C) Mean axial shrinkage of the histology section for each solution (n = 35 locations for each solution). Horizontal line: measured average lateral shrinkage (n = 3 eyes). The gray zone represents the 95% CI of the lateral shrinkage estimate. The axial shrinkages of solutions 1 and 3 fall within the lateral shrinkage 95% CI. Error bars, SEM.

Figure 6.

 

Relationship between the histology (after accounting for shrinkage) and the in vivo LRP in tree shrew (i.e., solution 1). The inner LRP incorporates zones 1, 2, and 3, which are based on LRP boundaries. Only one peak in the outer LRP (outer peak) corresponds to the retina. LRP transitions as defined in Figure 3 , histology sublaminae as defined in Figure 2 ; BM, Bruch’s membrane; CC, choriocapillaris. Scale bars, 50 μm.

Figure 7.

 

Mean axial shrinkage of the retinal zones defined in Figure 3Aand calculated using equation 3 . n = 35 locations per zone. Horizontal line and shaded area: the overall mean axial shrinkage for solution 1 and its 95% CI. Zone 1 has significantly less shrinkage than zone 2 or 3 (t-test, ***P < 0.001). All zones have a mean shrinkage outside the 95% CI of the overall mean shrinkage, indicating that nonlinear shrinkage occurs in the axial direction.